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);
421 static const char *zeroing_mode[] = {
422 [SD_ZERO_WRITE] = "write",
423 [SD_ZERO_WS] = "writesame",
424 [SD_ZERO_WS16_UNMAP] = "writesame_16_unmap",
425 [SD_ZERO_WS10_UNMAP] = "writesame_10_unmap",
429 zeroing_mode_show(struct device *dev, struct device_attribute *attr,
432 struct scsi_disk *sdkp = to_scsi_disk(dev);
434 return snprintf(buf, 20, "%s\n", zeroing_mode[sdkp->zeroing_mode]);
438 zeroing_mode_store(struct device *dev, struct device_attribute *attr,
439 const char *buf, size_t count)
441 struct scsi_disk *sdkp = to_scsi_disk(dev);
443 if (!capable(CAP_SYS_ADMIN))
446 if (!strncmp(buf, zeroing_mode[SD_ZERO_WRITE], 20))
447 sdkp->zeroing_mode = SD_ZERO_WRITE;
448 else if (!strncmp(buf, zeroing_mode[SD_ZERO_WS], 20))
449 sdkp->zeroing_mode = SD_ZERO_WS;
450 else if (!strncmp(buf, zeroing_mode[SD_ZERO_WS16_UNMAP], 20))
451 sdkp->zeroing_mode = SD_ZERO_WS16_UNMAP;
452 else if (!strncmp(buf, zeroing_mode[SD_ZERO_WS10_UNMAP], 20))
453 sdkp->zeroing_mode = SD_ZERO_WS10_UNMAP;
459 static DEVICE_ATTR_RW(zeroing_mode);
462 max_medium_access_timeouts_show(struct device *dev,
463 struct device_attribute *attr, char *buf)
465 struct scsi_disk *sdkp = to_scsi_disk(dev);
467 return snprintf(buf, 20, "%u\n", sdkp->max_medium_access_timeouts);
471 max_medium_access_timeouts_store(struct device *dev,
472 struct device_attribute *attr, const char *buf,
475 struct scsi_disk *sdkp = to_scsi_disk(dev);
478 if (!capable(CAP_SYS_ADMIN))
481 err = kstrtouint(buf, 10, &sdkp->max_medium_access_timeouts);
483 return err ? err : count;
485 static DEVICE_ATTR_RW(max_medium_access_timeouts);
488 max_write_same_blocks_show(struct device *dev, struct device_attribute *attr,
491 struct scsi_disk *sdkp = to_scsi_disk(dev);
493 return snprintf(buf, 20, "%u\n", sdkp->max_ws_blocks);
497 max_write_same_blocks_store(struct device *dev, struct device_attribute *attr,
498 const char *buf, size_t count)
500 struct scsi_disk *sdkp = to_scsi_disk(dev);
501 struct scsi_device *sdp = sdkp->device;
505 if (!capable(CAP_SYS_ADMIN))
508 if (sdp->type != TYPE_DISK && sdp->type != TYPE_ZBC)
511 err = kstrtoul(buf, 10, &max);
517 sdp->no_write_same = 1;
518 else if (max <= SD_MAX_WS16_BLOCKS) {
519 sdp->no_write_same = 0;
520 sdkp->max_ws_blocks = max;
523 sd_config_write_same(sdkp);
527 static DEVICE_ATTR_RW(max_write_same_blocks);
529 static struct attribute *sd_disk_attrs[] = {
530 &dev_attr_cache_type.attr,
532 &dev_attr_allow_restart.attr,
533 &dev_attr_manage_start_stop.attr,
534 &dev_attr_protection_type.attr,
535 &dev_attr_protection_mode.attr,
536 &dev_attr_app_tag_own.attr,
537 &dev_attr_thin_provisioning.attr,
538 &dev_attr_provisioning_mode.attr,
539 &dev_attr_zeroing_mode.attr,
540 &dev_attr_max_write_same_blocks.attr,
541 &dev_attr_max_medium_access_timeouts.attr,
544 ATTRIBUTE_GROUPS(sd_disk);
546 static struct class sd_disk_class = {
548 .owner = THIS_MODULE,
549 .dev_release = scsi_disk_release,
550 .dev_groups = sd_disk_groups,
553 static const struct dev_pm_ops sd_pm_ops = {
554 .suspend = sd_suspend_system,
556 .poweroff = sd_suspend_system,
557 .restore = sd_resume,
558 .runtime_suspend = sd_suspend_runtime,
559 .runtime_resume = sd_resume,
562 static struct scsi_driver sd_template = {
565 .owner = THIS_MODULE,
568 .shutdown = sd_shutdown,
572 .init_command = sd_init_command,
573 .uninit_command = sd_uninit_command,
575 .eh_action = sd_eh_action,
579 * Dummy kobj_map->probe function.
580 * The default ->probe function will call modprobe, which is
581 * pointless as this module is already loaded.
583 static struct kobject *sd_default_probe(dev_t devt, int *partno, void *data)
589 * Device no to disk mapping:
591 * major disc2 disc p1
592 * |............|.............|....|....| <- dev_t
595 * Inside a major, we have 16k disks, however mapped non-
596 * contiguously. The first 16 disks are for major0, the next
597 * ones with major1, ... Disk 256 is for major0 again, disk 272
599 * As we stay compatible with our numbering scheme, we can reuse
600 * the well-know SCSI majors 8, 65--71, 136--143.
602 static int sd_major(int major_idx)
606 return SCSI_DISK0_MAJOR;
608 return SCSI_DISK1_MAJOR + major_idx - 1;
610 return SCSI_DISK8_MAJOR + major_idx - 8;
613 return 0; /* shut up gcc */
617 static struct scsi_disk *scsi_disk_get(struct gendisk *disk)
619 struct scsi_disk *sdkp = NULL;
621 mutex_lock(&sd_ref_mutex);
623 if (disk->private_data) {
624 sdkp = scsi_disk(disk);
625 if (scsi_device_get(sdkp->device) == 0)
626 get_device(&sdkp->dev);
630 mutex_unlock(&sd_ref_mutex);
634 static void scsi_disk_put(struct scsi_disk *sdkp)
636 struct scsi_device *sdev = sdkp->device;
638 mutex_lock(&sd_ref_mutex);
639 put_device(&sdkp->dev);
640 scsi_device_put(sdev);
641 mutex_unlock(&sd_ref_mutex);
644 static unsigned char sd_setup_protect_cmnd(struct scsi_cmnd *scmd,
645 unsigned int dix, unsigned int dif)
647 struct bio *bio = scmd->request->bio;
648 unsigned int prot_op = sd_prot_op(rq_data_dir(scmd->request), dix, dif);
649 unsigned int protect = 0;
651 if (dix) { /* DIX Type 0, 1, 2, 3 */
652 if (bio_integrity_flagged(bio, BIP_IP_CHECKSUM))
653 scmd->prot_flags |= SCSI_PROT_IP_CHECKSUM;
655 if (bio_integrity_flagged(bio, BIP_CTRL_NOCHECK) == false)
656 scmd->prot_flags |= SCSI_PROT_GUARD_CHECK;
659 if (dif != T10_PI_TYPE3_PROTECTION) { /* DIX/DIF Type 0, 1, 2 */
660 scmd->prot_flags |= SCSI_PROT_REF_INCREMENT;
662 if (bio_integrity_flagged(bio, BIP_CTRL_NOCHECK) == false)
663 scmd->prot_flags |= SCSI_PROT_REF_CHECK;
666 if (dif) { /* DIX/DIF Type 1, 2, 3 */
667 scmd->prot_flags |= SCSI_PROT_TRANSFER_PI;
669 if (bio_integrity_flagged(bio, BIP_DISK_NOCHECK))
670 protect = 3 << 5; /* Disable target PI checking */
672 protect = 1 << 5; /* Enable target PI checking */
675 scsi_set_prot_op(scmd, prot_op);
676 scsi_set_prot_type(scmd, dif);
677 scmd->prot_flags &= sd_prot_flag_mask(prot_op);
682 static void sd_config_discard(struct scsi_disk *sdkp, unsigned int mode)
684 struct request_queue *q = sdkp->disk->queue;
685 unsigned int logical_block_size = sdkp->device->sector_size;
686 unsigned int max_blocks = 0;
689 * When LBPRZ is reported, discard alignment and granularity
690 * must be fixed to the logical block size. Otherwise the block
691 * layer will drop misaligned portions of the request which can
692 * lead to data corruption. If LBPRZ is not set, we honor the
696 q->limits.discard_alignment = 0;
697 q->limits.discard_granularity = logical_block_size;
699 q->limits.discard_alignment = sdkp->unmap_alignment *
701 q->limits.discard_granularity =
702 max(sdkp->physical_block_size,
703 sdkp->unmap_granularity * logical_block_size);
706 sdkp->provisioning_mode = mode;
711 blk_queue_max_discard_sectors(q, 0);
712 queue_flag_clear_unlocked(QUEUE_FLAG_DISCARD, q);
716 max_blocks = min_not_zero(sdkp->max_unmap_blocks,
717 (u32)SD_MAX_WS16_BLOCKS);
721 max_blocks = min_not_zero(sdkp->max_ws_blocks,
722 (u32)SD_MAX_WS16_BLOCKS);
726 max_blocks = min_not_zero(sdkp->max_ws_blocks,
727 (u32)SD_MAX_WS10_BLOCKS);
731 max_blocks = min_not_zero(sdkp->max_ws_blocks,
732 (u32)SD_MAX_WS10_BLOCKS);
736 blk_queue_max_discard_sectors(q, max_blocks * (logical_block_size >> 9));
737 queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, q);
740 static int sd_setup_unmap_cmnd(struct scsi_cmnd *cmd)
742 struct scsi_device *sdp = cmd->device;
743 struct request *rq = cmd->request;
744 u64 sector = blk_rq_pos(rq) >> (ilog2(sdp->sector_size) - 9);
745 u32 nr_sectors = blk_rq_sectors(rq) >> (ilog2(sdp->sector_size) - 9);
746 unsigned int data_len = 24;
749 rq->special_vec.bv_page = alloc_page(GFP_ATOMIC | __GFP_ZERO);
750 if (!rq->special_vec.bv_page)
751 return BLKPREP_DEFER;
752 rq->special_vec.bv_offset = 0;
753 rq->special_vec.bv_len = data_len;
754 rq->rq_flags |= RQF_SPECIAL_PAYLOAD;
757 cmd->cmnd[0] = UNMAP;
760 buf = page_address(rq->special_vec.bv_page);
761 put_unaligned_be16(6 + 16, &buf[0]);
762 put_unaligned_be16(16, &buf[2]);
763 put_unaligned_be64(sector, &buf[8]);
764 put_unaligned_be32(nr_sectors, &buf[16]);
766 cmd->allowed = SD_MAX_RETRIES;
767 cmd->transfersize = data_len;
768 rq->timeout = SD_TIMEOUT;
769 scsi_req(rq)->resid_len = data_len;
771 return scsi_init_io(cmd);
774 static int sd_setup_write_same16_cmnd(struct scsi_cmnd *cmd, bool unmap)
776 struct scsi_device *sdp = cmd->device;
777 struct request *rq = cmd->request;
778 u64 sector = blk_rq_pos(rq) >> (ilog2(sdp->sector_size) - 9);
779 u32 nr_sectors = blk_rq_sectors(rq) >> (ilog2(sdp->sector_size) - 9);
780 u32 data_len = sdp->sector_size;
782 rq->special_vec.bv_page = alloc_page(GFP_ATOMIC | __GFP_ZERO);
783 if (!rq->special_vec.bv_page)
784 return BLKPREP_DEFER;
785 rq->special_vec.bv_offset = 0;
786 rq->special_vec.bv_len = data_len;
787 rq->rq_flags |= RQF_SPECIAL_PAYLOAD;
790 cmd->cmnd[0] = WRITE_SAME_16;
792 cmd->cmnd[1] = 0x8; /* UNMAP */
793 put_unaligned_be64(sector, &cmd->cmnd[2]);
794 put_unaligned_be32(nr_sectors, &cmd->cmnd[10]);
796 cmd->allowed = SD_MAX_RETRIES;
797 cmd->transfersize = data_len;
798 rq->timeout = unmap ? SD_TIMEOUT : SD_WRITE_SAME_TIMEOUT;
799 scsi_req(rq)->resid_len = data_len;
801 return scsi_init_io(cmd);
804 static int sd_setup_write_same10_cmnd(struct scsi_cmnd *cmd, bool unmap)
806 struct scsi_device *sdp = cmd->device;
807 struct request *rq = cmd->request;
808 u64 sector = blk_rq_pos(rq) >> (ilog2(sdp->sector_size) - 9);
809 u32 nr_sectors = blk_rq_sectors(rq) >> (ilog2(sdp->sector_size) - 9);
810 u32 data_len = sdp->sector_size;
812 rq->special_vec.bv_page = alloc_page(GFP_ATOMIC | __GFP_ZERO);
813 if (!rq->special_vec.bv_page)
814 return BLKPREP_DEFER;
815 rq->special_vec.bv_offset = 0;
816 rq->special_vec.bv_len = data_len;
817 rq->rq_flags |= RQF_SPECIAL_PAYLOAD;
820 cmd->cmnd[0] = WRITE_SAME;
822 cmd->cmnd[1] = 0x8; /* UNMAP */
823 put_unaligned_be32(sector, &cmd->cmnd[2]);
824 put_unaligned_be16(nr_sectors, &cmd->cmnd[7]);
826 cmd->allowed = SD_MAX_RETRIES;
827 cmd->transfersize = data_len;
828 rq->timeout = unmap ? SD_TIMEOUT : SD_WRITE_SAME_TIMEOUT;
829 scsi_req(rq)->resid_len = data_len;
831 return scsi_init_io(cmd);
834 static int sd_setup_write_zeroes_cmnd(struct scsi_cmnd *cmd)
836 struct request *rq = cmd->request;
837 struct scsi_device *sdp = cmd->device;
838 struct scsi_disk *sdkp = scsi_disk(rq->rq_disk);
839 u64 sector = blk_rq_pos(rq) >> (ilog2(sdp->sector_size) - 9);
840 u32 nr_sectors = blk_rq_sectors(rq) >> (ilog2(sdp->sector_size) - 9);
842 if (!(rq->cmd_flags & REQ_NOUNMAP)) {
843 switch (sdkp->zeroing_mode) {
844 case SD_ZERO_WS16_UNMAP:
845 return sd_setup_write_same16_cmnd(cmd, true);
846 case SD_ZERO_WS10_UNMAP:
847 return sd_setup_write_same10_cmnd(cmd, true);
851 if (sdp->no_write_same)
852 return BLKPREP_INVALID;
853 if (sdkp->ws16 || sector > 0xffffffff || nr_sectors > 0xffff)
854 return sd_setup_write_same16_cmnd(cmd, false);
855 return sd_setup_write_same10_cmnd(cmd, false);
858 static void sd_config_write_same(struct scsi_disk *sdkp)
860 struct request_queue *q = sdkp->disk->queue;
861 unsigned int logical_block_size = sdkp->device->sector_size;
863 if (sdkp->device->no_write_same) {
864 sdkp->max_ws_blocks = 0;
868 /* Some devices can not handle block counts above 0xffff despite
869 * supporting WRITE SAME(16). Consequently we default to 64k
870 * blocks per I/O unless the device explicitly advertises a
873 if (sdkp->max_ws_blocks > SD_MAX_WS10_BLOCKS)
874 sdkp->max_ws_blocks = min_not_zero(sdkp->max_ws_blocks,
875 (u32)SD_MAX_WS16_BLOCKS);
876 else if (sdkp->ws16 || sdkp->ws10 || sdkp->device->no_report_opcodes)
877 sdkp->max_ws_blocks = min_not_zero(sdkp->max_ws_blocks,
878 (u32)SD_MAX_WS10_BLOCKS);
880 sdkp->device->no_write_same = 1;
881 sdkp->max_ws_blocks = 0;
884 if (sdkp->lbprz && sdkp->lbpws)
885 sdkp->zeroing_mode = SD_ZERO_WS16_UNMAP;
886 else if (sdkp->lbprz && sdkp->lbpws10)
887 sdkp->zeroing_mode = SD_ZERO_WS10_UNMAP;
888 else if (sdkp->max_ws_blocks)
889 sdkp->zeroing_mode = SD_ZERO_WS;
891 sdkp->zeroing_mode = SD_ZERO_WRITE;
894 blk_queue_max_write_same_sectors(q, sdkp->max_ws_blocks *
895 (logical_block_size >> 9));
896 blk_queue_max_write_zeroes_sectors(q, sdkp->max_ws_blocks *
897 (logical_block_size >> 9));
901 * sd_setup_write_same_cmnd - write the same data to multiple blocks
902 * @cmd: command to prepare
904 * Will issue either WRITE SAME(10) or WRITE SAME(16) depending on
905 * preference indicated by target device.
907 static int sd_setup_write_same_cmnd(struct scsi_cmnd *cmd)
909 struct request *rq = cmd->request;
910 struct scsi_device *sdp = cmd->device;
911 struct scsi_disk *sdkp = scsi_disk(rq->rq_disk);
912 struct bio *bio = rq->bio;
913 sector_t sector = blk_rq_pos(rq);
914 unsigned int nr_sectors = blk_rq_sectors(rq);
915 unsigned int nr_bytes = blk_rq_bytes(rq);
918 if (sdkp->device->no_write_same)
919 return BLKPREP_INVALID;
921 BUG_ON(bio_offset(bio) || bio_iovec(bio).bv_len != sdp->sector_size);
923 if (sd_is_zoned(sdkp)) {
924 ret = sd_zbc_setup_write_cmnd(cmd);
925 if (ret != BLKPREP_OK)
929 sector >>= ilog2(sdp->sector_size) - 9;
930 nr_sectors >>= ilog2(sdp->sector_size) - 9;
932 rq->timeout = SD_WRITE_SAME_TIMEOUT;
934 if (sdkp->ws16 || sector > 0xffffffff || nr_sectors > 0xffff) {
936 cmd->cmnd[0] = WRITE_SAME_16;
937 put_unaligned_be64(sector, &cmd->cmnd[2]);
938 put_unaligned_be32(nr_sectors, &cmd->cmnd[10]);
941 cmd->cmnd[0] = WRITE_SAME;
942 put_unaligned_be32(sector, &cmd->cmnd[2]);
943 put_unaligned_be16(nr_sectors, &cmd->cmnd[7]);
946 cmd->transfersize = sdp->sector_size;
947 cmd->allowed = SD_MAX_RETRIES;
950 * For WRITE SAME the data transferred via the DATA OUT buffer is
951 * different from the amount of data actually written to the target.
953 * We set up __data_len to the amount of data transferred via the
954 * DATA OUT buffer so that blk_rq_map_sg sets up the proper S/G list
955 * to transfer a single sector of data first, but then reset it to
956 * the amount of data to be written right after so that the I/O path
957 * knows how much to actually write.
959 rq->__data_len = sdp->sector_size;
960 ret = scsi_init_io(cmd);
961 rq->__data_len = nr_bytes;
965 static int sd_setup_flush_cmnd(struct scsi_cmnd *cmd)
967 struct request *rq = cmd->request;
969 /* flush requests don't perform I/O, zero the S/G table */
970 memset(&cmd->sdb, 0, sizeof(cmd->sdb));
972 cmd->cmnd[0] = SYNCHRONIZE_CACHE;
974 cmd->transfersize = 0;
975 cmd->allowed = SD_MAX_RETRIES;
977 rq->timeout = rq->q->rq_timeout * SD_FLUSH_TIMEOUT_MULTIPLIER;
981 static int sd_setup_read_write_cmnd(struct scsi_cmnd *SCpnt)
983 struct request *rq = SCpnt->request;
984 struct scsi_device *sdp = SCpnt->device;
985 struct gendisk *disk = rq->rq_disk;
986 struct scsi_disk *sdkp = scsi_disk(disk);
987 sector_t block = blk_rq_pos(rq);
989 unsigned int this_count = blk_rq_sectors(rq);
990 unsigned int dif, dix;
991 bool zoned_write = sd_is_zoned(sdkp) && rq_data_dir(rq) == WRITE;
993 unsigned char protect;
996 ret = sd_zbc_setup_write_cmnd(SCpnt);
997 if (ret != BLKPREP_OK)
1001 ret = scsi_init_io(SCpnt);
1002 if (ret != BLKPREP_OK)
1004 SCpnt = rq->special;
1006 /* from here on until we're complete, any goto out
1007 * is used for a killable error condition */
1011 scmd_printk(KERN_INFO, SCpnt,
1012 "%s: block=%llu, count=%d\n",
1013 __func__, (unsigned long long)block, this_count));
1015 if (!sdp || !scsi_device_online(sdp) ||
1016 block + blk_rq_sectors(rq) > get_capacity(disk)) {
1017 SCSI_LOG_HLQUEUE(2, scmd_printk(KERN_INFO, SCpnt,
1018 "Finishing %u sectors\n",
1019 blk_rq_sectors(rq)));
1020 SCSI_LOG_HLQUEUE(2, scmd_printk(KERN_INFO, SCpnt,
1021 "Retry with 0x%p\n", SCpnt));
1027 * quietly refuse to do anything to a changed disc until
1028 * the changed bit has been reset
1030 /* printk("SCSI disk has been changed or is not present. Prohibiting further I/O.\n"); */
1035 * Some SD card readers can't handle multi-sector accesses which touch
1036 * the last one or two hardware sectors. Split accesses as needed.
1038 threshold = get_capacity(disk) - SD_LAST_BUGGY_SECTORS *
1039 (sdp->sector_size / 512);
1041 if (unlikely(sdp->last_sector_bug && block + this_count > threshold)) {
1042 if (block < threshold) {
1043 /* Access up to the threshold but not beyond */
1044 this_count = threshold - block;
1046 /* Access only a single hardware sector */
1047 this_count = sdp->sector_size / 512;
1051 SCSI_LOG_HLQUEUE(2, scmd_printk(KERN_INFO, SCpnt, "block=%llu\n",
1052 (unsigned long long)block));
1055 * If we have a 1K hardware sectorsize, prevent access to single
1056 * 512 byte sectors. In theory we could handle this - in fact
1057 * the scsi cdrom driver must be able to handle this because
1058 * we typically use 1K blocksizes, and cdroms typically have
1059 * 2K hardware sectorsizes. Of course, things are simpler
1060 * with the cdrom, since it is read-only. For performance
1061 * reasons, the filesystems should be able to handle this
1062 * and not force the scsi disk driver to use bounce buffers
1065 if (sdp->sector_size == 1024) {
1066 if ((block & 1) || (blk_rq_sectors(rq) & 1)) {
1067 scmd_printk(KERN_ERR, SCpnt,
1068 "Bad block number requested\n");
1072 this_count = this_count >> 1;
1075 if (sdp->sector_size == 2048) {
1076 if ((block & 3) || (blk_rq_sectors(rq) & 3)) {
1077 scmd_printk(KERN_ERR, SCpnt,
1078 "Bad block number requested\n");
1082 this_count = this_count >> 2;
1085 if (sdp->sector_size == 4096) {
1086 if ((block & 7) || (blk_rq_sectors(rq) & 7)) {
1087 scmd_printk(KERN_ERR, SCpnt,
1088 "Bad block number requested\n");
1092 this_count = this_count >> 3;
1095 if (rq_data_dir(rq) == WRITE) {
1096 SCpnt->cmnd[0] = WRITE_6;
1098 if (blk_integrity_rq(rq))
1099 sd_dif_prepare(SCpnt);
1101 } else if (rq_data_dir(rq) == READ) {
1102 SCpnt->cmnd[0] = READ_6;
1104 scmd_printk(KERN_ERR, SCpnt, "Unknown command %d\n", req_op(rq));
1108 SCSI_LOG_HLQUEUE(2, scmd_printk(KERN_INFO, SCpnt,
1109 "%s %d/%u 512 byte blocks.\n",
1110 (rq_data_dir(rq) == WRITE) ?
1111 "writing" : "reading", this_count,
1112 blk_rq_sectors(rq)));
1114 dix = scsi_prot_sg_count(SCpnt);
1115 dif = scsi_host_dif_capable(SCpnt->device->host, sdkp->protection_type);
1118 protect = sd_setup_protect_cmnd(SCpnt, dix, dif);
1122 if (protect && sdkp->protection_type == T10_PI_TYPE2_PROTECTION) {
1123 SCpnt->cmnd = mempool_alloc(sd_cdb_pool, GFP_ATOMIC);
1125 if (unlikely(SCpnt->cmnd == NULL)) {
1126 ret = BLKPREP_DEFER;
1130 SCpnt->cmd_len = SD_EXT_CDB_SIZE;
1131 memset(SCpnt->cmnd, 0, SCpnt->cmd_len);
1132 SCpnt->cmnd[0] = VARIABLE_LENGTH_CMD;
1133 SCpnt->cmnd[7] = 0x18;
1134 SCpnt->cmnd[9] = (rq_data_dir(rq) == READ) ? READ_32 : WRITE_32;
1135 SCpnt->cmnd[10] = protect | ((rq->cmd_flags & REQ_FUA) ? 0x8 : 0);
1138 SCpnt->cmnd[12] = sizeof(block) > 4 ? (unsigned char) (block >> 56) & 0xff : 0;
1139 SCpnt->cmnd[13] = sizeof(block) > 4 ? (unsigned char) (block >> 48) & 0xff : 0;
1140 SCpnt->cmnd[14] = sizeof(block) > 4 ? (unsigned char) (block >> 40) & 0xff : 0;
1141 SCpnt->cmnd[15] = sizeof(block) > 4 ? (unsigned char) (block >> 32) & 0xff : 0;
1142 SCpnt->cmnd[16] = (unsigned char) (block >> 24) & 0xff;
1143 SCpnt->cmnd[17] = (unsigned char) (block >> 16) & 0xff;
1144 SCpnt->cmnd[18] = (unsigned char) (block >> 8) & 0xff;
1145 SCpnt->cmnd[19] = (unsigned char) block & 0xff;
1147 /* Expected Indirect LBA */
1148 SCpnt->cmnd[20] = (unsigned char) (block >> 24) & 0xff;
1149 SCpnt->cmnd[21] = (unsigned char) (block >> 16) & 0xff;
1150 SCpnt->cmnd[22] = (unsigned char) (block >> 8) & 0xff;
1151 SCpnt->cmnd[23] = (unsigned char) block & 0xff;
1153 /* Transfer length */
1154 SCpnt->cmnd[28] = (unsigned char) (this_count >> 24) & 0xff;
1155 SCpnt->cmnd[29] = (unsigned char) (this_count >> 16) & 0xff;
1156 SCpnt->cmnd[30] = (unsigned char) (this_count >> 8) & 0xff;
1157 SCpnt->cmnd[31] = (unsigned char) this_count & 0xff;
1158 } else if (sdp->use_16_for_rw || (this_count > 0xffff)) {
1159 SCpnt->cmnd[0] += READ_16 - READ_6;
1160 SCpnt->cmnd[1] = protect | ((rq->cmd_flags & REQ_FUA) ? 0x8 : 0);
1161 SCpnt->cmnd[2] = sizeof(block) > 4 ? (unsigned char) (block >> 56) & 0xff : 0;
1162 SCpnt->cmnd[3] = sizeof(block) > 4 ? (unsigned char) (block >> 48) & 0xff : 0;
1163 SCpnt->cmnd[4] = sizeof(block) > 4 ? (unsigned char) (block >> 40) & 0xff : 0;
1164 SCpnt->cmnd[5] = sizeof(block) > 4 ? (unsigned char) (block >> 32) & 0xff : 0;
1165 SCpnt->cmnd[6] = (unsigned char) (block >> 24) & 0xff;
1166 SCpnt->cmnd[7] = (unsigned char) (block >> 16) & 0xff;
1167 SCpnt->cmnd[8] = (unsigned char) (block >> 8) & 0xff;
1168 SCpnt->cmnd[9] = (unsigned char) block & 0xff;
1169 SCpnt->cmnd[10] = (unsigned char) (this_count >> 24) & 0xff;
1170 SCpnt->cmnd[11] = (unsigned char) (this_count >> 16) & 0xff;
1171 SCpnt->cmnd[12] = (unsigned char) (this_count >> 8) & 0xff;
1172 SCpnt->cmnd[13] = (unsigned char) this_count & 0xff;
1173 SCpnt->cmnd[14] = SCpnt->cmnd[15] = 0;
1174 } else if ((this_count > 0xff) || (block > 0x1fffff) ||
1175 scsi_device_protection(SCpnt->device) ||
1176 SCpnt->device->use_10_for_rw) {
1177 SCpnt->cmnd[0] += READ_10 - READ_6;
1178 SCpnt->cmnd[1] = protect | ((rq->cmd_flags & REQ_FUA) ? 0x8 : 0);
1179 SCpnt->cmnd[2] = (unsigned char) (block >> 24) & 0xff;
1180 SCpnt->cmnd[3] = (unsigned char) (block >> 16) & 0xff;
1181 SCpnt->cmnd[4] = (unsigned char) (block >> 8) & 0xff;
1182 SCpnt->cmnd[5] = (unsigned char) block & 0xff;
1183 SCpnt->cmnd[6] = SCpnt->cmnd[9] = 0;
1184 SCpnt->cmnd[7] = (unsigned char) (this_count >> 8) & 0xff;
1185 SCpnt->cmnd[8] = (unsigned char) this_count & 0xff;
1187 if (unlikely(rq->cmd_flags & REQ_FUA)) {
1189 * This happens only if this drive failed
1190 * 10byte rw command with ILLEGAL_REQUEST
1191 * during operation and thus turned off
1194 scmd_printk(KERN_ERR, SCpnt,
1195 "FUA write on READ/WRITE(6) drive\n");
1199 SCpnt->cmnd[1] |= (unsigned char) ((block >> 16) & 0x1f);
1200 SCpnt->cmnd[2] = (unsigned char) ((block >> 8) & 0xff);
1201 SCpnt->cmnd[3] = (unsigned char) block & 0xff;
1202 SCpnt->cmnd[4] = (unsigned char) this_count;
1205 SCpnt->sdb.length = this_count * sdp->sector_size;
1208 * We shouldn't disconnect in the middle of a sector, so with a dumb
1209 * host adapter, it's safe to assume that we can at least transfer
1210 * this many bytes between each connect / disconnect.
1212 SCpnt->transfersize = sdp->sector_size;
1213 SCpnt->underflow = this_count << 9;
1214 SCpnt->allowed = SD_MAX_RETRIES;
1217 * This indicates that the command is ready from our end to be
1222 if (zoned_write && ret != BLKPREP_OK)
1223 sd_zbc_cancel_write_cmnd(SCpnt);
1228 static int sd_init_command(struct scsi_cmnd *cmd)
1230 struct request *rq = cmd->request;
1232 switch (req_op(rq)) {
1233 case REQ_OP_DISCARD:
1234 switch (scsi_disk(rq->rq_disk)->provisioning_mode) {
1236 return sd_setup_unmap_cmnd(cmd);
1238 return sd_setup_write_same16_cmnd(cmd, true);
1240 return sd_setup_write_same10_cmnd(cmd, true);
1242 return sd_setup_write_same10_cmnd(cmd, false);
1244 return BLKPREP_INVALID;
1246 case REQ_OP_WRITE_ZEROES:
1247 return sd_setup_write_zeroes_cmnd(cmd);
1248 case REQ_OP_WRITE_SAME:
1249 return sd_setup_write_same_cmnd(cmd);
1251 return sd_setup_flush_cmnd(cmd);
1254 return sd_setup_read_write_cmnd(cmd);
1255 case REQ_OP_ZONE_REPORT:
1256 return sd_zbc_setup_report_cmnd(cmd);
1257 case REQ_OP_ZONE_RESET:
1258 return sd_zbc_setup_reset_cmnd(cmd);
1264 static void sd_uninit_command(struct scsi_cmnd *SCpnt)
1266 struct request *rq = SCpnt->request;
1268 if (rq->rq_flags & RQF_SPECIAL_PAYLOAD)
1269 __free_page(rq->special_vec.bv_page);
1271 if (SCpnt->cmnd != scsi_req(rq)->cmd) {
1272 mempool_free(SCpnt->cmnd, sd_cdb_pool);
1279 * sd_open - open a scsi disk device
1280 * @inode: only i_rdev member may be used
1281 * @filp: only f_mode and f_flags may be used
1283 * Returns 0 if successful. Returns a negated errno value in case
1286 * Note: This can be called from a user context (e.g. fsck(1) )
1287 * or from within the kernel (e.g. as a result of a mount(1) ).
1288 * In the latter case @inode and @filp carry an abridged amount
1289 * of information as noted above.
1291 * Locking: called with bdev->bd_mutex held.
1293 static int sd_open(struct block_device *bdev, fmode_t mode)
1295 struct scsi_disk *sdkp = scsi_disk_get(bdev->bd_disk);
1296 struct scsi_device *sdev;
1302 SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp, "sd_open\n"));
1304 sdev = sdkp->device;
1307 * If the device is in error recovery, wait until it is done.
1308 * If the device is offline, then disallow any access to it.
1311 if (!scsi_block_when_processing_errors(sdev))
1314 if (sdev->removable || sdkp->write_prot)
1315 check_disk_change(bdev);
1318 * If the drive is empty, just let the open fail.
1320 retval = -ENOMEDIUM;
1321 if (sdev->removable && !sdkp->media_present && !(mode & FMODE_NDELAY))
1325 * If the device has the write protect tab set, have the open fail
1326 * if the user expects to be able to write to the thing.
1329 if (sdkp->write_prot && (mode & FMODE_WRITE))
1333 * It is possible that the disk changing stuff resulted in
1334 * the device being taken offline. If this is the case,
1335 * report this to the user, and don't pretend that the
1336 * open actually succeeded.
1339 if (!scsi_device_online(sdev))
1342 if ((atomic_inc_return(&sdkp->openers) == 1) && sdev->removable) {
1343 if (scsi_block_when_processing_errors(sdev))
1344 scsi_set_medium_removal(sdev, SCSI_REMOVAL_PREVENT);
1350 scsi_disk_put(sdkp);
1355 * sd_release - invoked when the (last) close(2) is called on this
1357 * @inode: only i_rdev member may be used
1358 * @filp: only f_mode and f_flags may be used
1362 * Note: may block (uninterruptible) if error recovery is underway
1365 * Locking: called with bdev->bd_mutex held.
1367 static void sd_release(struct gendisk *disk, fmode_t mode)
1369 struct scsi_disk *sdkp = scsi_disk(disk);
1370 struct scsi_device *sdev = sdkp->device;
1372 SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp, "sd_release\n"));
1374 if (atomic_dec_return(&sdkp->openers) == 0 && sdev->removable) {
1375 if (scsi_block_when_processing_errors(sdev))
1376 scsi_set_medium_removal(sdev, SCSI_REMOVAL_ALLOW);
1380 * XXX and what if there are packets in flight and this close()
1381 * XXX is followed by a "rmmod sd_mod"?
1384 scsi_disk_put(sdkp);
1387 static int sd_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1389 struct scsi_disk *sdkp = scsi_disk(bdev->bd_disk);
1390 struct scsi_device *sdp = sdkp->device;
1391 struct Scsi_Host *host = sdp->host;
1392 sector_t capacity = logical_to_sectors(sdp, sdkp->capacity);
1395 /* default to most commonly used values */
1396 diskinfo[0] = 0x40; /* 1 << 6 */
1397 diskinfo[1] = 0x20; /* 1 << 5 */
1398 diskinfo[2] = capacity >> 11;
1400 /* override with calculated, extended default, or driver values */
1401 if (host->hostt->bios_param)
1402 host->hostt->bios_param(sdp, bdev, capacity, diskinfo);
1404 scsicam_bios_param(bdev, capacity, diskinfo);
1406 geo->heads = diskinfo[0];
1407 geo->sectors = diskinfo[1];
1408 geo->cylinders = diskinfo[2];
1413 * sd_ioctl - process an ioctl
1414 * @inode: only i_rdev/i_bdev members may be used
1415 * @filp: only f_mode and f_flags may be used
1416 * @cmd: ioctl command number
1417 * @arg: this is third argument given to ioctl(2) system call.
1418 * Often contains a pointer.
1420 * Returns 0 if successful (some ioctls return positive numbers on
1421 * success as well). Returns a negated errno value in case of error.
1423 * Note: most ioctls are forward onto the block subsystem or further
1424 * down in the scsi subsystem.
1426 static int sd_ioctl(struct block_device *bdev, fmode_t mode,
1427 unsigned int cmd, unsigned long arg)
1429 struct gendisk *disk = bdev->bd_disk;
1430 struct scsi_disk *sdkp = scsi_disk(disk);
1431 struct scsi_device *sdp = sdkp->device;
1432 void __user *p = (void __user *)arg;
1435 SCSI_LOG_IOCTL(1, sd_printk(KERN_INFO, sdkp, "sd_ioctl: disk=%s, "
1436 "cmd=0x%x\n", disk->disk_name, cmd));
1438 error = scsi_verify_blk_ioctl(bdev, cmd);
1443 * If we are in the middle of error recovery, don't let anyone
1444 * else try and use this device. Also, if error recovery fails, it
1445 * may try and take the device offline, in which case all further
1446 * access to the device is prohibited.
1448 error = scsi_ioctl_block_when_processing_errors(sdp, cmd,
1449 (mode & FMODE_NDELAY) != 0);
1454 * Send SCSI addressing ioctls directly to mid level, send other
1455 * ioctls to block level and then onto mid level if they can't be
1459 case SCSI_IOCTL_GET_IDLUN:
1460 case SCSI_IOCTL_GET_BUS_NUMBER:
1461 error = scsi_ioctl(sdp, cmd, p);
1464 error = scsi_cmd_blk_ioctl(bdev, mode, cmd, p);
1465 if (error != -ENOTTY)
1467 error = scsi_ioctl(sdp, cmd, p);
1474 static void set_media_not_present(struct scsi_disk *sdkp)
1476 if (sdkp->media_present)
1477 sdkp->device->changed = 1;
1479 if (sdkp->device->removable) {
1480 sdkp->media_present = 0;
1485 static int media_not_present(struct scsi_disk *sdkp,
1486 struct scsi_sense_hdr *sshdr)
1488 if (!scsi_sense_valid(sshdr))
1491 /* not invoked for commands that could return deferred errors */
1492 switch (sshdr->sense_key) {
1493 case UNIT_ATTENTION:
1495 /* medium not present */
1496 if (sshdr->asc == 0x3A) {
1497 set_media_not_present(sdkp);
1505 * sd_check_events - check media events
1506 * @disk: kernel device descriptor
1507 * @clearing: disk events currently being cleared
1509 * Returns mask of DISK_EVENT_*.
1511 * Note: this function is invoked from the block subsystem.
1513 static unsigned int sd_check_events(struct gendisk *disk, unsigned int clearing)
1515 struct scsi_disk *sdkp = scsi_disk_get(disk);
1516 struct scsi_device *sdp;
1523 SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp, "sd_check_events\n"));
1526 * If the device is offline, don't send any commands - just pretend as
1527 * if the command failed. If the device ever comes back online, we
1528 * can deal with it then. It is only because of unrecoverable errors
1529 * that we would ever take a device offline in the first place.
1531 if (!scsi_device_online(sdp)) {
1532 set_media_not_present(sdkp);
1537 * Using TEST_UNIT_READY enables differentiation between drive with
1538 * no cartridge loaded - NOT READY, drive with changed cartridge -
1539 * UNIT ATTENTION, or with same cartridge - GOOD STATUS.
1541 * Drives that auto spin down. eg iomega jaz 1G, will be started
1542 * by sd_spinup_disk() from sd_revalidate_disk(), which happens whenever
1543 * sd_revalidate() is called.
1545 if (scsi_block_when_processing_errors(sdp)) {
1546 struct scsi_sense_hdr sshdr = { 0, };
1548 retval = scsi_test_unit_ready(sdp, SD_TIMEOUT, SD_MAX_RETRIES,
1551 /* failed to execute TUR, assume media not present */
1552 if (host_byte(retval)) {
1553 set_media_not_present(sdkp);
1557 if (media_not_present(sdkp, &sshdr))
1562 * For removable scsi disk we have to recognise the presence
1563 * of a disk in the drive.
1565 if (!sdkp->media_present)
1567 sdkp->media_present = 1;
1570 * sdp->changed is set under the following conditions:
1572 * Medium present state has changed in either direction.
1573 * Device has indicated UNIT_ATTENTION.
1575 retval = sdp->changed ? DISK_EVENT_MEDIA_CHANGE : 0;
1577 scsi_disk_put(sdkp);
1581 static int sd_sync_cache(struct scsi_disk *sdkp)
1584 struct scsi_device *sdp = sdkp->device;
1585 const int timeout = sdp->request_queue->rq_timeout
1586 * SD_FLUSH_TIMEOUT_MULTIPLIER;
1587 struct scsi_sense_hdr sshdr;
1589 if (!scsi_device_online(sdp))
1592 for (retries = 3; retries > 0; --retries) {
1593 unsigned char cmd[10] = { 0 };
1595 cmd[0] = SYNCHRONIZE_CACHE;
1597 * Leave the rest of the command zero to indicate
1600 res = scsi_execute(sdp, cmd, DMA_NONE, NULL, 0, NULL, &sshdr,
1601 timeout, SD_MAX_RETRIES, 0, RQF_PM, NULL);
1607 sd_print_result(sdkp, "Synchronize Cache(10) failed", res);
1609 if (driver_byte(res) & DRIVER_SENSE)
1610 sd_print_sense_hdr(sdkp, &sshdr);
1611 /* we need to evaluate the error return */
1612 if (scsi_sense_valid(&sshdr) &&
1613 (sshdr.asc == 0x3a || /* medium not present */
1614 sshdr.asc == 0x20)) /* invalid command */
1615 /* this is no error here */
1618 switch (host_byte(res)) {
1619 /* ignore errors due to racing a disconnection */
1620 case DID_BAD_TARGET:
1621 case DID_NO_CONNECT:
1623 /* signal the upper layer it might try again */
1627 case DID_SOFT_ERROR:
1636 static void sd_rescan(struct device *dev)
1638 struct scsi_disk *sdkp = dev_get_drvdata(dev);
1640 revalidate_disk(sdkp->disk);
1644 #ifdef CONFIG_COMPAT
1646 * This gets directly called from VFS. When the ioctl
1647 * is not recognized we go back to the other translation paths.
1649 static int sd_compat_ioctl(struct block_device *bdev, fmode_t mode,
1650 unsigned int cmd, unsigned long arg)
1652 struct scsi_device *sdev = scsi_disk(bdev->bd_disk)->device;
1655 error = scsi_ioctl_block_when_processing_errors(sdev, cmd,
1656 (mode & FMODE_NDELAY) != 0);
1661 * Let the static ioctl translation table take care of it.
1663 if (!sdev->host->hostt->compat_ioctl)
1664 return -ENOIOCTLCMD;
1665 return sdev->host->hostt->compat_ioctl(sdev, cmd, (void __user *)arg);
1669 static char sd_pr_type(enum pr_type type)
1672 case PR_WRITE_EXCLUSIVE:
1674 case PR_EXCLUSIVE_ACCESS:
1676 case PR_WRITE_EXCLUSIVE_REG_ONLY:
1678 case PR_EXCLUSIVE_ACCESS_REG_ONLY:
1680 case PR_WRITE_EXCLUSIVE_ALL_REGS:
1682 case PR_EXCLUSIVE_ACCESS_ALL_REGS:
1689 static int sd_pr_command(struct block_device *bdev, u8 sa,
1690 u64 key, u64 sa_key, u8 type, u8 flags)
1692 struct scsi_device *sdev = scsi_disk(bdev->bd_disk)->device;
1693 struct scsi_sense_hdr sshdr;
1695 u8 cmd[16] = { 0, };
1696 u8 data[24] = { 0, };
1698 cmd[0] = PERSISTENT_RESERVE_OUT;
1701 put_unaligned_be32(sizeof(data), &cmd[5]);
1703 put_unaligned_be64(key, &data[0]);
1704 put_unaligned_be64(sa_key, &data[8]);
1707 result = scsi_execute_req(sdev, cmd, DMA_TO_DEVICE, &data, sizeof(data),
1708 &sshdr, SD_TIMEOUT, SD_MAX_RETRIES, NULL);
1710 if ((driver_byte(result) & DRIVER_SENSE) &&
1711 (scsi_sense_valid(&sshdr))) {
1712 sdev_printk(KERN_INFO, sdev, "PR command failed: %d\n", result);
1713 scsi_print_sense_hdr(sdev, NULL, &sshdr);
1719 static int sd_pr_register(struct block_device *bdev, u64 old_key, u64 new_key,
1722 if (flags & ~PR_FL_IGNORE_KEY)
1724 return sd_pr_command(bdev, (flags & PR_FL_IGNORE_KEY) ? 0x06 : 0x00,
1725 old_key, new_key, 0,
1726 (1 << 0) /* APTPL */);
1729 static int sd_pr_reserve(struct block_device *bdev, u64 key, enum pr_type type,
1734 return sd_pr_command(bdev, 0x01, key, 0, sd_pr_type(type), 0);
1737 static int sd_pr_release(struct block_device *bdev, u64 key, enum pr_type type)
1739 return sd_pr_command(bdev, 0x02, key, 0, sd_pr_type(type), 0);
1742 static int sd_pr_preempt(struct block_device *bdev, u64 old_key, u64 new_key,
1743 enum pr_type type, bool abort)
1745 return sd_pr_command(bdev, abort ? 0x05 : 0x04, old_key, new_key,
1746 sd_pr_type(type), 0);
1749 static int sd_pr_clear(struct block_device *bdev, u64 key)
1751 return sd_pr_command(bdev, 0x03, key, 0, 0, 0);
1754 static const struct pr_ops sd_pr_ops = {
1755 .pr_register = sd_pr_register,
1756 .pr_reserve = sd_pr_reserve,
1757 .pr_release = sd_pr_release,
1758 .pr_preempt = sd_pr_preempt,
1759 .pr_clear = sd_pr_clear,
1762 static const struct block_device_operations sd_fops = {
1763 .owner = THIS_MODULE,
1765 .release = sd_release,
1767 .getgeo = sd_getgeo,
1768 #ifdef CONFIG_COMPAT
1769 .compat_ioctl = sd_compat_ioctl,
1771 .check_events = sd_check_events,
1772 .revalidate_disk = sd_revalidate_disk,
1773 .unlock_native_capacity = sd_unlock_native_capacity,
1774 .pr_ops = &sd_pr_ops,
1778 * sd_eh_action - error handling callback
1779 * @scmd: sd-issued command that has failed
1780 * @eh_disp: The recovery disposition suggested by the midlayer
1782 * This function is called by the SCSI midlayer upon completion of an
1783 * error test command (currently TEST UNIT READY). The result of sending
1784 * the eh command is passed in eh_disp. We're looking for devices that
1785 * fail medium access commands but are OK with non access commands like
1786 * test unit ready (so wrongly see the device as having a successful
1789 static int sd_eh_action(struct scsi_cmnd *scmd, int eh_disp)
1791 struct scsi_disk *sdkp = scsi_disk(scmd->request->rq_disk);
1793 if (!scsi_device_online(scmd->device) ||
1794 !scsi_medium_access_command(scmd) ||
1795 host_byte(scmd->result) != DID_TIME_OUT ||
1800 * The device has timed out executing a medium access command.
1801 * However, the TEST UNIT READY command sent during error
1802 * handling completed successfully. Either the device is in the
1803 * process of recovering or has it suffered an internal failure
1804 * that prevents access to the storage medium.
1806 sdkp->medium_access_timed_out++;
1809 * If the device keeps failing read/write commands but TEST UNIT
1810 * READY always completes successfully we assume that medium
1811 * access is no longer possible and take the device offline.
1813 if (sdkp->medium_access_timed_out >= sdkp->max_medium_access_timeouts) {
1814 scmd_printk(KERN_ERR, scmd,
1815 "Medium access timeout failure. Offlining disk!\n");
1816 scsi_device_set_state(scmd->device, SDEV_OFFLINE);
1824 static unsigned int sd_completed_bytes(struct scsi_cmnd *scmd)
1826 u64 start_lba = blk_rq_pos(scmd->request);
1827 u64 end_lba = blk_rq_pos(scmd->request) + (scsi_bufflen(scmd) / 512);
1828 u64 factor = scmd->device->sector_size / 512;
1832 * resid is optional but mostly filled in. When it's unused,
1833 * its value is zero, so we assume the whole buffer transferred
1835 unsigned int transferred = scsi_bufflen(scmd) - scsi_get_resid(scmd);
1836 unsigned int good_bytes;
1838 info_valid = scsi_get_sense_info_fld(scmd->sense_buffer,
1839 SCSI_SENSE_BUFFERSIZE,
1844 if (scsi_bufflen(scmd) <= scmd->device->sector_size)
1847 /* be careful ... don't want any overflows */
1848 do_div(start_lba, factor);
1849 do_div(end_lba, factor);
1851 /* The bad lba was reported incorrectly, we have no idea where
1854 if (bad_lba < start_lba || bad_lba >= end_lba)
1857 /* This computation should always be done in terms of
1858 * the resolution of the device's medium.
1860 good_bytes = (bad_lba - start_lba) * scmd->device->sector_size;
1861 return min(good_bytes, transferred);
1865 * sd_done - bottom half handler: called when the lower level
1866 * driver has completed (successfully or otherwise) a scsi command.
1867 * @SCpnt: mid-level's per command structure.
1869 * Note: potentially run from within an ISR. Must not block.
1871 static int sd_done(struct scsi_cmnd *SCpnt)
1873 int result = SCpnt->result;
1874 unsigned int good_bytes = result ? 0 : scsi_bufflen(SCpnt);
1875 unsigned int sector_size = SCpnt->device->sector_size;
1877 struct scsi_sense_hdr sshdr;
1878 struct scsi_disk *sdkp = scsi_disk(SCpnt->request->rq_disk);
1879 struct request *req = SCpnt->request;
1880 int sense_valid = 0;
1881 int sense_deferred = 0;
1882 unsigned char op = SCpnt->cmnd[0];
1883 unsigned char unmap = SCpnt->cmnd[1] & 8;
1885 switch (req_op(req)) {
1886 case REQ_OP_DISCARD:
1887 case REQ_OP_WRITE_ZEROES:
1888 case REQ_OP_WRITE_SAME:
1889 case REQ_OP_ZONE_RESET:
1891 good_bytes = blk_rq_bytes(req);
1892 scsi_set_resid(SCpnt, 0);
1895 scsi_set_resid(SCpnt, blk_rq_bytes(req));
1898 case REQ_OP_ZONE_REPORT:
1900 good_bytes = scsi_bufflen(SCpnt)
1901 - scsi_get_resid(SCpnt);
1902 scsi_set_resid(SCpnt, 0);
1905 scsi_set_resid(SCpnt, blk_rq_bytes(req));
1910 * In case of bogus fw or device, we could end up having
1911 * an unaligned partial completion. Check this here and force
1914 resid = scsi_get_resid(SCpnt);
1915 if (resid & (sector_size - 1)) {
1916 sd_printk(KERN_INFO, sdkp,
1917 "Unaligned partial completion (resid=%u, sector_sz=%u)\n",
1918 resid, sector_size);
1919 resid = min(scsi_bufflen(SCpnt),
1920 round_up(resid, sector_size));
1921 scsi_set_resid(SCpnt, resid);
1926 sense_valid = scsi_command_normalize_sense(SCpnt, &sshdr);
1928 sense_deferred = scsi_sense_is_deferred(&sshdr);
1930 sdkp->medium_access_timed_out = 0;
1932 if (driver_byte(result) != DRIVER_SENSE &&
1933 (!sense_valid || sense_deferred))
1936 switch (sshdr.sense_key) {
1937 case HARDWARE_ERROR:
1939 good_bytes = sd_completed_bytes(SCpnt);
1941 case RECOVERED_ERROR:
1942 good_bytes = scsi_bufflen(SCpnt);
1945 /* This indicates a false check condition, so ignore it. An
1946 * unknown amount of data was transferred so treat it as an
1950 memset(SCpnt->sense_buffer, 0, SCSI_SENSE_BUFFERSIZE);
1952 case ABORTED_COMMAND:
1953 if (sshdr.asc == 0x10) /* DIF: Target detected corruption */
1954 good_bytes = sd_completed_bytes(SCpnt);
1956 case ILLEGAL_REQUEST:
1957 if (sshdr.asc == 0x10) /* DIX: Host detected corruption */
1958 good_bytes = sd_completed_bytes(SCpnt);
1959 /* INVALID COMMAND OPCODE or INVALID FIELD IN CDB */
1960 if (sshdr.asc == 0x20 || sshdr.asc == 0x24) {
1963 sd_config_discard(sdkp, SD_LBP_DISABLE);
1968 sd_config_discard(sdkp, SD_LBP_DISABLE);
1970 sdkp->device->no_write_same = 1;
1971 sd_config_write_same(sdkp);
1974 req->__data_len = blk_rq_bytes(req);
1975 req->rq_flags |= RQF_QUIET;
1985 if (sd_is_zoned(sdkp))
1986 sd_zbc_complete(SCpnt, good_bytes, &sshdr);
1988 SCSI_LOG_HLCOMPLETE(1, scmd_printk(KERN_INFO, SCpnt,
1989 "sd_done: completed %d of %d bytes\n",
1990 good_bytes, scsi_bufflen(SCpnt)));
1992 if (rq_data_dir(SCpnt->request) == READ && scsi_prot_sg_count(SCpnt))
1993 sd_dif_complete(SCpnt, good_bytes);
1999 * spinup disk - called only in sd_revalidate_disk()
2002 sd_spinup_disk(struct scsi_disk *sdkp)
2004 unsigned char cmd[10];
2005 unsigned long spintime_expire = 0;
2006 int retries, spintime;
2007 unsigned int the_result;
2008 struct scsi_sense_hdr sshdr;
2009 int sense_valid = 0;
2013 /* Spin up drives, as required. Only do this at boot time */
2014 /* Spinup needs to be done for module loads too. */
2019 cmd[0] = TEST_UNIT_READY;
2020 memset((void *) &cmd[1], 0, 9);
2022 the_result = scsi_execute_req(sdkp->device, cmd,
2025 SD_MAX_RETRIES, NULL);
2028 * If the drive has indicated to us that it
2029 * doesn't have any media in it, don't bother
2030 * with any more polling.
2032 if (media_not_present(sdkp, &sshdr))
2036 sense_valid = scsi_sense_valid(&sshdr);
2038 } while (retries < 3 &&
2039 (!scsi_status_is_good(the_result) ||
2040 ((driver_byte(the_result) & DRIVER_SENSE) &&
2041 sense_valid && sshdr.sense_key == UNIT_ATTENTION)));
2043 if ((driver_byte(the_result) & DRIVER_SENSE) == 0) {
2044 /* no sense, TUR either succeeded or failed
2045 * with a status error */
2046 if(!spintime && !scsi_status_is_good(the_result)) {
2047 sd_print_result(sdkp, "Test Unit Ready failed",
2054 * The device does not want the automatic start to be issued.
2056 if (sdkp->device->no_start_on_add)
2059 if (sense_valid && sshdr.sense_key == NOT_READY) {
2060 if (sshdr.asc == 4 && sshdr.ascq == 3)
2061 break; /* manual intervention required */
2062 if (sshdr.asc == 4 && sshdr.ascq == 0xb)
2063 break; /* standby */
2064 if (sshdr.asc == 4 && sshdr.ascq == 0xc)
2065 break; /* unavailable */
2067 * Issue command to spin up drive when not ready
2070 sd_printk(KERN_NOTICE, sdkp, "Spinning up disk...");
2071 cmd[0] = START_STOP;
2072 cmd[1] = 1; /* Return immediately */
2073 memset((void *) &cmd[2], 0, 8);
2074 cmd[4] = 1; /* Start spin cycle */
2075 if (sdkp->device->start_stop_pwr_cond)
2077 scsi_execute_req(sdkp->device, cmd, DMA_NONE,
2079 SD_TIMEOUT, SD_MAX_RETRIES,
2081 spintime_expire = jiffies + 100 * HZ;
2084 /* Wait 1 second for next try */
2089 * Wait for USB flash devices with slow firmware.
2090 * Yes, this sense key/ASC combination shouldn't
2091 * occur here. It's characteristic of these devices.
2093 } else if (sense_valid &&
2094 sshdr.sense_key == UNIT_ATTENTION &&
2095 sshdr.asc == 0x28) {
2097 spintime_expire = jiffies + 5 * HZ;
2100 /* Wait 1 second for next try */
2103 /* we don't understand the sense code, so it's
2104 * probably pointless to loop */
2106 sd_printk(KERN_NOTICE, sdkp, "Unit Not Ready\n");
2107 sd_print_sense_hdr(sdkp, &sshdr);
2112 } while (spintime && time_before_eq(jiffies, spintime_expire));
2115 if (scsi_status_is_good(the_result))
2118 printk("not responding...\n");
2123 * Determine whether disk supports Data Integrity Field.
2125 static int sd_read_protection_type(struct scsi_disk *sdkp, unsigned char *buffer)
2127 struct scsi_device *sdp = sdkp->device;
2131 if (scsi_device_protection(sdp) == 0 || (buffer[12] & 1) == 0)
2134 type = ((buffer[12] >> 1) & 7) + 1; /* P_TYPE 0 = Type 1 */
2136 if (type > T10_PI_TYPE3_PROTECTION)
2138 else if (scsi_host_dif_capable(sdp->host, type))
2141 if (sdkp->first_scan || type != sdkp->protection_type)
2144 sd_printk(KERN_ERR, sdkp, "formatted with unsupported" \
2145 " protection type %u. Disabling disk!\n",
2149 sd_printk(KERN_NOTICE, sdkp,
2150 "Enabling DIF Type %u protection\n", type);
2153 sd_printk(KERN_NOTICE, sdkp,
2154 "Disabling DIF Type %u protection\n", type);
2158 sdkp->protection_type = type;
2163 static void read_capacity_error(struct scsi_disk *sdkp, struct scsi_device *sdp,
2164 struct scsi_sense_hdr *sshdr, int sense_valid,
2167 if (driver_byte(the_result) & DRIVER_SENSE)
2168 sd_print_sense_hdr(sdkp, sshdr);
2170 sd_printk(KERN_NOTICE, sdkp, "Sense not available.\n");
2173 * Set dirty bit for removable devices if not ready -
2174 * sometimes drives will not report this properly.
2176 if (sdp->removable &&
2177 sense_valid && sshdr->sense_key == NOT_READY)
2178 set_media_not_present(sdkp);
2181 * We used to set media_present to 0 here to indicate no media
2182 * in the drive, but some drives fail read capacity even with
2183 * media present, so we can't do that.
2185 sdkp->capacity = 0; /* unknown mapped to zero - as usual */
2189 #if RC16_LEN > SD_BUF_SIZE
2190 #error RC16_LEN must not be more than SD_BUF_SIZE
2193 #define READ_CAPACITY_RETRIES_ON_RESET 10
2195 static int read_capacity_16(struct scsi_disk *sdkp, struct scsi_device *sdp,
2196 unsigned char *buffer)
2198 unsigned char cmd[16];
2199 struct scsi_sense_hdr sshdr;
2200 int sense_valid = 0;
2202 int retries = 3, reset_retries = READ_CAPACITY_RETRIES_ON_RESET;
2203 unsigned int alignment;
2204 unsigned long long lba;
2205 unsigned sector_size;
2207 if (sdp->no_read_capacity_16)
2212 cmd[0] = SERVICE_ACTION_IN_16;
2213 cmd[1] = SAI_READ_CAPACITY_16;
2215 memset(buffer, 0, RC16_LEN);
2217 the_result = scsi_execute_req(sdp, cmd, DMA_FROM_DEVICE,
2218 buffer, RC16_LEN, &sshdr,
2219 SD_TIMEOUT, SD_MAX_RETRIES, NULL);
2221 if (media_not_present(sdkp, &sshdr))
2225 sense_valid = scsi_sense_valid(&sshdr);
2227 sshdr.sense_key == ILLEGAL_REQUEST &&
2228 (sshdr.asc == 0x20 || sshdr.asc == 0x24) &&
2230 /* Invalid Command Operation Code or
2231 * Invalid Field in CDB, just retry
2232 * silently with RC10 */
2235 sshdr.sense_key == UNIT_ATTENTION &&
2236 sshdr.asc == 0x29 && sshdr.ascq == 0x00)
2237 /* Device reset might occur several times,
2238 * give it one more chance */
2239 if (--reset_retries > 0)
2244 } while (the_result && retries);
2247 sd_print_result(sdkp, "Read Capacity(16) failed", the_result);
2248 read_capacity_error(sdkp, sdp, &sshdr, sense_valid, the_result);
2252 sector_size = get_unaligned_be32(&buffer[8]);
2253 lba = get_unaligned_be64(&buffer[0]);
2255 if (sd_read_protection_type(sdkp, buffer) < 0) {
2260 if ((sizeof(sdkp->capacity) == 4) && (lba >= 0xffffffffULL)) {
2261 sd_printk(KERN_ERR, sdkp, "Too big for this kernel. Use a "
2262 "kernel compiled with support for large block "
2268 /* Logical blocks per physical block exponent */
2269 sdkp->physical_block_size = (1 << (buffer[13] & 0xf)) * sector_size;
2272 sdkp->rc_basis = (buffer[12] >> 4) & 0x3;
2274 /* Lowest aligned logical block */
2275 alignment = ((buffer[14] & 0x3f) << 8 | buffer[15]) * sector_size;
2276 blk_queue_alignment_offset(sdp->request_queue, alignment);
2277 if (alignment && sdkp->first_scan)
2278 sd_printk(KERN_NOTICE, sdkp,
2279 "physical block alignment offset: %u\n", alignment);
2281 if (buffer[14] & 0x80) { /* LBPME */
2284 if (buffer[14] & 0x40) /* LBPRZ */
2287 sd_config_discard(sdkp, SD_LBP_WS16);
2290 sdkp->capacity = lba + 1;
2294 static int read_capacity_10(struct scsi_disk *sdkp, struct scsi_device *sdp,
2295 unsigned char *buffer)
2297 unsigned char cmd[16];
2298 struct scsi_sense_hdr sshdr;
2299 int sense_valid = 0;
2301 int retries = 3, reset_retries = READ_CAPACITY_RETRIES_ON_RESET;
2303 unsigned sector_size;
2306 cmd[0] = READ_CAPACITY;
2307 memset(&cmd[1], 0, 9);
2308 memset(buffer, 0, 8);
2310 the_result = scsi_execute_req(sdp, cmd, DMA_FROM_DEVICE,
2312 SD_TIMEOUT, SD_MAX_RETRIES, NULL);
2314 if (media_not_present(sdkp, &sshdr))
2318 sense_valid = scsi_sense_valid(&sshdr);
2320 sshdr.sense_key == UNIT_ATTENTION &&
2321 sshdr.asc == 0x29 && sshdr.ascq == 0x00)
2322 /* Device reset might occur several times,
2323 * give it one more chance */
2324 if (--reset_retries > 0)
2329 } while (the_result && retries);
2332 sd_print_result(sdkp, "Read Capacity(10) failed", the_result);
2333 read_capacity_error(sdkp, sdp, &sshdr, sense_valid, the_result);
2337 sector_size = get_unaligned_be32(&buffer[4]);
2338 lba = get_unaligned_be32(&buffer[0]);
2340 if (sdp->no_read_capacity_16 && (lba == 0xffffffff)) {
2341 /* Some buggy (usb cardreader) devices return an lba of
2342 0xffffffff when the want to report a size of 0 (with
2343 which they really mean no media is present) */
2345 sdkp->physical_block_size = sector_size;
2349 if ((sizeof(sdkp->capacity) == 4) && (lba == 0xffffffff)) {
2350 sd_printk(KERN_ERR, sdkp, "Too big for this kernel. Use a "
2351 "kernel compiled with support for large block "
2357 sdkp->capacity = lba + 1;
2358 sdkp->physical_block_size = sector_size;
2362 static int sd_try_rc16_first(struct scsi_device *sdp)
2364 if (sdp->host->max_cmd_len < 16)
2366 if (sdp->try_rc_10_first)
2368 if (sdp->scsi_level > SCSI_SPC_2)
2370 if (scsi_device_protection(sdp))
2376 * read disk capacity
2379 sd_read_capacity(struct scsi_disk *sdkp, unsigned char *buffer)
2382 struct scsi_device *sdp = sdkp->device;
2384 if (sd_try_rc16_first(sdp)) {
2385 sector_size = read_capacity_16(sdkp, sdp, buffer);
2386 if (sector_size == -EOVERFLOW)
2388 if (sector_size == -ENODEV)
2390 if (sector_size < 0)
2391 sector_size = read_capacity_10(sdkp, sdp, buffer);
2392 if (sector_size < 0)
2395 sector_size = read_capacity_10(sdkp, sdp, buffer);
2396 if (sector_size == -EOVERFLOW)
2398 if (sector_size < 0)
2400 if ((sizeof(sdkp->capacity) > 4) &&
2401 (sdkp->capacity > 0xffffffffULL)) {
2402 int old_sector_size = sector_size;
2403 sd_printk(KERN_NOTICE, sdkp, "Very big device. "
2404 "Trying to use READ CAPACITY(16).\n");
2405 sector_size = read_capacity_16(sdkp, sdp, buffer);
2406 if (sector_size < 0) {
2407 sd_printk(KERN_NOTICE, sdkp,
2408 "Using 0xffffffff as device size\n");
2409 sdkp->capacity = 1 + (sector_t) 0xffffffff;
2410 sector_size = old_sector_size;
2416 /* Some devices are known to return the total number of blocks,
2417 * not the highest block number. Some devices have versions
2418 * which do this and others which do not. Some devices we might
2419 * suspect of doing this but we don't know for certain.
2421 * If we know the reported capacity is wrong, decrement it. If
2422 * we can only guess, then assume the number of blocks is even
2423 * (usually true but not always) and err on the side of lowering
2426 if (sdp->fix_capacity ||
2427 (sdp->guess_capacity && (sdkp->capacity & 0x01))) {
2428 sd_printk(KERN_INFO, sdkp, "Adjusting the sector count "
2429 "from its reported value: %llu\n",
2430 (unsigned long long) sdkp->capacity);
2435 if (sector_size == 0) {
2437 sd_printk(KERN_NOTICE, sdkp, "Sector size 0 reported, "
2441 if (sector_size != 512 &&
2442 sector_size != 1024 &&
2443 sector_size != 2048 &&
2444 sector_size != 4096) {
2445 sd_printk(KERN_NOTICE, sdkp, "Unsupported sector size %d.\n",
2448 * The user might want to re-format the drive with
2449 * a supported sectorsize. Once this happens, it
2450 * would be relatively trivial to set the thing up.
2451 * For this reason, we leave the thing in the table.
2455 * set a bogus sector size so the normal read/write
2456 * logic in the block layer will eventually refuse any
2457 * request on this device without tripping over power
2458 * of two sector size assumptions
2462 blk_queue_logical_block_size(sdp->request_queue, sector_size);
2463 blk_queue_physical_block_size(sdp->request_queue,
2464 sdkp->physical_block_size);
2465 sdkp->device->sector_size = sector_size;
2467 if (sdkp->capacity > 0xffffffff)
2468 sdp->use_16_for_rw = 1;
2473 * Print disk capacity
2476 sd_print_capacity(struct scsi_disk *sdkp,
2477 sector_t old_capacity)
2479 int sector_size = sdkp->device->sector_size;
2480 char cap_str_2[10], cap_str_10[10];
2482 string_get_size(sdkp->capacity, sector_size,
2483 STRING_UNITS_2, cap_str_2, sizeof(cap_str_2));
2484 string_get_size(sdkp->capacity, sector_size,
2485 STRING_UNITS_10, cap_str_10,
2486 sizeof(cap_str_10));
2488 if (sdkp->first_scan || old_capacity != sdkp->capacity) {
2489 sd_printk(KERN_NOTICE, sdkp,
2490 "%llu %d-byte logical blocks: (%s/%s)\n",
2491 (unsigned long long)sdkp->capacity,
2492 sector_size, cap_str_10, cap_str_2);
2494 if (sdkp->physical_block_size != sector_size)
2495 sd_printk(KERN_NOTICE, sdkp,
2496 "%u-byte physical blocks\n",
2497 sdkp->physical_block_size);
2499 sd_zbc_print_zones(sdkp);
2503 /* called with buffer of length 512 */
2505 sd_do_mode_sense(struct scsi_device *sdp, int dbd, int modepage,
2506 unsigned char *buffer, int len, struct scsi_mode_data *data,
2507 struct scsi_sense_hdr *sshdr)
2509 return scsi_mode_sense(sdp, dbd, modepage, buffer, len,
2510 SD_TIMEOUT, SD_MAX_RETRIES, data,
2515 * read write protect setting, if possible - called only in sd_revalidate_disk()
2516 * called with buffer of length SD_BUF_SIZE
2519 sd_read_write_protect_flag(struct scsi_disk *sdkp, unsigned char *buffer)
2522 struct scsi_device *sdp = sdkp->device;
2523 struct scsi_mode_data data;
2524 int old_wp = sdkp->write_prot;
2526 set_disk_ro(sdkp->disk, 0);
2527 if (sdp->skip_ms_page_3f) {
2528 sd_first_printk(KERN_NOTICE, sdkp, "Assuming Write Enabled\n");
2532 if (sdp->use_192_bytes_for_3f) {
2533 res = sd_do_mode_sense(sdp, 0, 0x3F, buffer, 192, &data, NULL);
2536 * First attempt: ask for all pages (0x3F), but only 4 bytes.
2537 * We have to start carefully: some devices hang if we ask
2538 * for more than is available.
2540 res = sd_do_mode_sense(sdp, 0, 0x3F, buffer, 4, &data, NULL);
2543 * Second attempt: ask for page 0 When only page 0 is
2544 * implemented, a request for page 3F may return Sense Key
2545 * 5: Illegal Request, Sense Code 24: Invalid field in
2548 if (!scsi_status_is_good(res))
2549 res = sd_do_mode_sense(sdp, 0, 0, buffer, 4, &data, NULL);
2552 * Third attempt: ask 255 bytes, as we did earlier.
2554 if (!scsi_status_is_good(res))
2555 res = sd_do_mode_sense(sdp, 0, 0x3F, buffer, 255,
2559 if (!scsi_status_is_good(res)) {
2560 sd_first_printk(KERN_WARNING, sdkp,
2561 "Test WP failed, assume Write Enabled\n");
2563 sdkp->write_prot = ((data.device_specific & 0x80) != 0);
2564 set_disk_ro(sdkp->disk, sdkp->write_prot);
2565 if (sdkp->first_scan || old_wp != sdkp->write_prot) {
2566 sd_printk(KERN_NOTICE, sdkp, "Write Protect is %s\n",
2567 sdkp->write_prot ? "on" : "off");
2568 sd_printk(KERN_DEBUG, sdkp, "Mode Sense: %4ph\n", buffer);
2574 * sd_read_cache_type - called only from sd_revalidate_disk()
2575 * called with buffer of length SD_BUF_SIZE
2578 sd_read_cache_type(struct scsi_disk *sdkp, unsigned char *buffer)
2581 struct scsi_device *sdp = sdkp->device;
2586 struct scsi_mode_data data;
2587 struct scsi_sense_hdr sshdr;
2588 int old_wce = sdkp->WCE;
2589 int old_rcd = sdkp->RCD;
2590 int old_dpofua = sdkp->DPOFUA;
2593 if (sdkp->cache_override)
2597 if (sdp->skip_ms_page_8) {
2598 if (sdp->type == TYPE_RBC)
2601 if (sdp->skip_ms_page_3f)
2604 if (sdp->use_192_bytes_for_3f)
2608 } else if (sdp->type == TYPE_RBC) {
2616 /* cautiously ask */
2617 res = sd_do_mode_sense(sdp, dbd, modepage, buffer, first_len,
2620 if (!scsi_status_is_good(res))
2623 if (!data.header_length) {
2626 sd_first_printk(KERN_ERR, sdkp,
2627 "Missing header in MODE_SENSE response\n");
2630 /* that went OK, now ask for the proper length */
2634 * We're only interested in the first three bytes, actually.
2635 * But the data cache page is defined for the first 20.
2639 else if (len > SD_BUF_SIZE) {
2640 sd_first_printk(KERN_NOTICE, sdkp, "Truncating mode parameter "
2641 "data from %d to %d bytes\n", len, SD_BUF_SIZE);
2644 if (modepage == 0x3F && sdp->use_192_bytes_for_3f)
2648 if (len > first_len)
2649 res = sd_do_mode_sense(sdp, dbd, modepage, buffer, len,
2652 if (scsi_status_is_good(res)) {
2653 int offset = data.header_length + data.block_descriptor_length;
2655 while (offset < len) {
2656 u8 page_code = buffer[offset] & 0x3F;
2657 u8 spf = buffer[offset] & 0x40;
2659 if (page_code == 8 || page_code == 6) {
2660 /* We're interested only in the first 3 bytes.
2662 if (len - offset <= 2) {
2663 sd_first_printk(KERN_ERR, sdkp,
2664 "Incomplete mode parameter "
2668 modepage = page_code;
2672 /* Go to the next page */
2673 if (spf && len - offset > 3)
2674 offset += 4 + (buffer[offset+2] << 8) +
2676 else if (!spf && len - offset > 1)
2677 offset += 2 + buffer[offset+1];
2679 sd_first_printk(KERN_ERR, sdkp,
2681 "parameter data\n");
2687 sd_first_printk(KERN_ERR, sdkp, "No Caching mode page found\n");
2691 if (modepage == 8) {
2692 sdkp->WCE = ((buffer[offset + 2] & 0x04) != 0);
2693 sdkp->RCD = ((buffer[offset + 2] & 0x01) != 0);
2695 sdkp->WCE = ((buffer[offset + 2] & 0x01) == 0);
2699 sdkp->DPOFUA = (data.device_specific & 0x10) != 0;
2700 if (sdp->broken_fua) {
2701 sd_first_printk(KERN_NOTICE, sdkp, "Disabling FUA\n");
2703 } else if (sdkp->DPOFUA && !sdkp->device->use_10_for_rw &&
2704 !sdkp->device->use_16_for_rw) {
2705 sd_first_printk(KERN_NOTICE, sdkp,
2706 "Uses READ/WRITE(6), disabling FUA\n");
2710 /* No cache flush allowed for write protected devices */
2711 if (sdkp->WCE && sdkp->write_prot)
2714 if (sdkp->first_scan || old_wce != sdkp->WCE ||
2715 old_rcd != sdkp->RCD || old_dpofua != sdkp->DPOFUA)
2716 sd_printk(KERN_NOTICE, sdkp,
2717 "Write cache: %s, read cache: %s, %s\n",
2718 sdkp->WCE ? "enabled" : "disabled",
2719 sdkp->RCD ? "disabled" : "enabled",
2720 sdkp->DPOFUA ? "supports DPO and FUA"
2721 : "doesn't support DPO or FUA");
2727 if (scsi_sense_valid(&sshdr) &&
2728 sshdr.sense_key == ILLEGAL_REQUEST &&
2729 sshdr.asc == 0x24 && sshdr.ascq == 0x0)
2730 /* Invalid field in CDB */
2731 sd_first_printk(KERN_NOTICE, sdkp, "Cache data unavailable\n");
2733 sd_first_printk(KERN_ERR, sdkp,
2734 "Asking for cache data failed\n");
2737 if (sdp->wce_default_on) {
2738 sd_first_printk(KERN_NOTICE, sdkp,
2739 "Assuming drive cache: write back\n");
2742 sd_first_printk(KERN_ERR, sdkp,
2743 "Assuming drive cache: write through\n");
2751 * The ATO bit indicates whether the DIF application tag is available
2752 * for use by the operating system.
2754 static void sd_read_app_tag_own(struct scsi_disk *sdkp, unsigned char *buffer)
2757 struct scsi_device *sdp = sdkp->device;
2758 struct scsi_mode_data data;
2759 struct scsi_sense_hdr sshdr;
2761 if (sdp->type != TYPE_DISK && sdp->type != TYPE_ZBC)
2764 if (sdkp->protection_type == 0)
2767 res = scsi_mode_sense(sdp, 1, 0x0a, buffer, 36, SD_TIMEOUT,
2768 SD_MAX_RETRIES, &data, &sshdr);
2770 if (!scsi_status_is_good(res) || !data.header_length ||
2772 sd_first_printk(KERN_WARNING, sdkp,
2773 "getting Control mode page failed, assume no ATO\n");
2775 if (scsi_sense_valid(&sshdr))
2776 sd_print_sense_hdr(sdkp, &sshdr);
2781 offset = data.header_length + data.block_descriptor_length;
2783 if ((buffer[offset] & 0x3f) != 0x0a) {
2784 sd_first_printk(KERN_ERR, sdkp, "ATO Got wrong page\n");
2788 if ((buffer[offset + 5] & 0x80) == 0)
2797 * sd_read_block_limits - Query disk device for preferred I/O sizes.
2798 * @disk: disk to query
2800 static void sd_read_block_limits(struct scsi_disk *sdkp)
2802 unsigned int sector_sz = sdkp->device->sector_size;
2803 const int vpd_len = 64;
2804 unsigned char *buffer = kmalloc(vpd_len, GFP_KERNEL);
2807 /* Block Limits VPD */
2808 scsi_get_vpd_page(sdkp->device, 0xb0, buffer, vpd_len))
2811 blk_queue_io_min(sdkp->disk->queue,
2812 get_unaligned_be16(&buffer[6]) * sector_sz);
2814 sdkp->max_xfer_blocks = get_unaligned_be32(&buffer[8]);
2815 sdkp->opt_xfer_blocks = get_unaligned_be32(&buffer[12]);
2817 if (buffer[3] == 0x3c) {
2818 unsigned int lba_count, desc_count;
2820 sdkp->max_ws_blocks = (u32)get_unaligned_be64(&buffer[36]);
2825 lba_count = get_unaligned_be32(&buffer[20]);
2826 desc_count = get_unaligned_be32(&buffer[24]);
2828 if (lba_count && desc_count)
2829 sdkp->max_unmap_blocks = lba_count;
2831 sdkp->unmap_granularity = get_unaligned_be32(&buffer[28]);
2833 if (buffer[32] & 0x80)
2834 sdkp->unmap_alignment =
2835 get_unaligned_be32(&buffer[32]) & ~(1 << 31);
2837 if (!sdkp->lbpvpd) { /* LBP VPD page not provided */
2839 if (sdkp->max_unmap_blocks)
2840 sd_config_discard(sdkp, SD_LBP_UNMAP);
2842 sd_config_discard(sdkp, SD_LBP_WS16);
2844 } else { /* LBP VPD page tells us what to use */
2845 if (sdkp->lbpu && sdkp->max_unmap_blocks && !sdkp->lbprz)
2846 sd_config_discard(sdkp, SD_LBP_UNMAP);
2847 else if (sdkp->lbpws)
2848 sd_config_discard(sdkp, SD_LBP_WS16);
2849 else if (sdkp->lbpws10)
2850 sd_config_discard(sdkp, SD_LBP_WS10);
2851 else if (sdkp->lbpu && sdkp->max_unmap_blocks)
2852 sd_config_discard(sdkp, SD_LBP_UNMAP);
2854 sd_config_discard(sdkp, SD_LBP_DISABLE);
2863 * sd_read_block_characteristics - Query block dev. characteristics
2864 * @disk: disk to query
2866 static void sd_read_block_characteristics(struct scsi_disk *sdkp)
2868 struct request_queue *q = sdkp->disk->queue;
2869 unsigned char *buffer;
2871 const int vpd_len = 64;
2873 buffer = kmalloc(vpd_len, GFP_KERNEL);
2876 /* Block Device Characteristics VPD */
2877 scsi_get_vpd_page(sdkp->device, 0xb1, buffer, vpd_len))
2880 rot = get_unaligned_be16(&buffer[4]);
2883 queue_flag_set_unlocked(QUEUE_FLAG_NONROT, q);
2884 queue_flag_clear_unlocked(QUEUE_FLAG_ADD_RANDOM, q);
2887 if (sdkp->device->type == TYPE_ZBC) {
2889 q->limits.zoned = BLK_ZONED_HM;
2891 sdkp->zoned = (buffer[8] >> 4) & 3;
2892 if (sdkp->zoned == 1)
2894 q->limits.zoned = BLK_ZONED_HA;
2897 * Treat drive-managed devices as
2898 * regular block devices.
2900 q->limits.zoned = BLK_ZONED_NONE;
2902 if (blk_queue_is_zoned(q) && sdkp->first_scan)
2903 sd_printk(KERN_NOTICE, sdkp, "Host-%s zoned block device\n",
2904 q->limits.zoned == BLK_ZONED_HM ? "managed" : "aware");
2911 * sd_read_block_provisioning - Query provisioning VPD page
2912 * @disk: disk to query
2914 static void sd_read_block_provisioning(struct scsi_disk *sdkp)
2916 unsigned char *buffer;
2917 const int vpd_len = 8;
2919 if (sdkp->lbpme == 0)
2922 buffer = kmalloc(vpd_len, GFP_KERNEL);
2924 if (!buffer || scsi_get_vpd_page(sdkp->device, 0xb2, buffer, vpd_len))
2928 sdkp->lbpu = (buffer[5] >> 7) & 1; /* UNMAP */
2929 sdkp->lbpws = (buffer[5] >> 6) & 1; /* WRITE SAME(16) with UNMAP */
2930 sdkp->lbpws10 = (buffer[5] >> 5) & 1; /* WRITE SAME(10) with UNMAP */
2936 static void sd_read_write_same(struct scsi_disk *sdkp, unsigned char *buffer)
2938 struct scsi_device *sdev = sdkp->device;
2940 if (sdev->host->no_write_same) {
2941 sdev->no_write_same = 1;
2946 if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE, INQUIRY) < 0) {
2947 /* too large values might cause issues with arcmsr */
2948 int vpd_buf_len = 64;
2950 sdev->no_report_opcodes = 1;
2952 /* Disable WRITE SAME if REPORT SUPPORTED OPERATION
2953 * CODES is unsupported and the device has an ATA
2954 * Information VPD page (SAT).
2956 if (!scsi_get_vpd_page(sdev, 0x89, buffer, vpd_buf_len))
2957 sdev->no_write_same = 1;
2960 if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE, WRITE_SAME_16) == 1)
2963 if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE, WRITE_SAME) == 1)
2968 * sd_revalidate_disk - called the first time a new disk is seen,
2969 * performs disk spin up, read_capacity, etc.
2970 * @disk: struct gendisk we care about
2972 static int sd_revalidate_disk(struct gendisk *disk)
2974 struct scsi_disk *sdkp = scsi_disk(disk);
2975 struct scsi_device *sdp = sdkp->device;
2976 struct request_queue *q = sdkp->disk->queue;
2977 sector_t old_capacity = sdkp->capacity;
2978 unsigned char *buffer;
2979 unsigned int dev_max, rw_max;
2981 SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp,
2982 "sd_revalidate_disk\n"));
2985 * If the device is offline, don't try and read capacity or any
2986 * of the other niceties.
2988 if (!scsi_device_online(sdp))
2991 buffer = kmalloc(SD_BUF_SIZE, GFP_KERNEL);
2993 sd_printk(KERN_WARNING, sdkp, "sd_revalidate_disk: Memory "
2994 "allocation failure.\n");
2998 sd_spinup_disk(sdkp);
3001 * Without media there is no reason to ask; moreover, some devices
3002 * react badly if we do.
3004 if (sdkp->media_present) {
3005 sd_read_capacity(sdkp, buffer);
3007 if (scsi_device_supports_vpd(sdp)) {
3008 sd_read_block_provisioning(sdkp);
3009 sd_read_block_limits(sdkp);
3010 sd_read_block_characteristics(sdkp);
3011 sd_zbc_read_zones(sdkp, buffer);
3014 sd_print_capacity(sdkp, old_capacity);
3016 sd_read_write_protect_flag(sdkp, buffer);
3017 sd_read_cache_type(sdkp, buffer);
3018 sd_read_app_tag_own(sdkp, buffer);
3019 sd_read_write_same(sdkp, buffer);
3022 sdkp->first_scan = 0;
3025 * We now have all cache related info, determine how we deal
3026 * with flush requests.
3028 sd_set_flush_flag(sdkp);
3030 /* Initial block count limit based on CDB TRANSFER LENGTH field size. */
3031 dev_max = sdp->use_16_for_rw ? SD_MAX_XFER_BLOCKS : SD_DEF_XFER_BLOCKS;
3033 /* Some devices report a maximum block count for READ/WRITE requests. */
3034 dev_max = min_not_zero(dev_max, sdkp->max_xfer_blocks);
3035 q->limits.max_dev_sectors = logical_to_sectors(sdp, dev_max);
3038 * Use the device's preferred I/O size for reads and writes
3039 * unless the reported value is unreasonably small, large, or
3042 if (sdkp->opt_xfer_blocks &&
3043 sdkp->opt_xfer_blocks <= dev_max &&
3044 sdkp->opt_xfer_blocks <= SD_DEF_XFER_BLOCKS &&
3045 logical_to_bytes(sdp, sdkp->opt_xfer_blocks) >= PAGE_SIZE) {
3046 q->limits.io_opt = logical_to_bytes(sdp, sdkp->opt_xfer_blocks);
3047 rw_max = logical_to_sectors(sdp, sdkp->opt_xfer_blocks);
3049 rw_max = BLK_DEF_MAX_SECTORS;
3051 /* Combine with controller limits */
3052 q->limits.max_sectors = min(rw_max, queue_max_hw_sectors(q));
3054 set_capacity(disk, logical_to_sectors(sdp, sdkp->capacity));
3055 sd_config_write_same(sdkp);
3063 * sd_unlock_native_capacity - unlock native capacity
3064 * @disk: struct gendisk to set capacity for
3066 * Block layer calls this function if it detects that partitions
3067 * on @disk reach beyond the end of the device. If the SCSI host
3068 * implements ->unlock_native_capacity() method, it's invoked to
3069 * give it a chance to adjust the device capacity.
3072 * Defined by block layer. Might sleep.
3074 static void sd_unlock_native_capacity(struct gendisk *disk)
3076 struct scsi_device *sdev = scsi_disk(disk)->device;
3078 if (sdev->host->hostt->unlock_native_capacity)
3079 sdev->host->hostt->unlock_native_capacity(sdev);
3083 * sd_format_disk_name - format disk name
3084 * @prefix: name prefix - ie. "sd" for SCSI disks
3085 * @index: index of the disk to format name for
3086 * @buf: output buffer
3087 * @buflen: length of the output buffer
3089 * SCSI disk names starts at sda. The 26th device is sdz and the
3090 * 27th is sdaa. The last one for two lettered suffix is sdzz
3091 * which is followed by sdaaa.
3093 * This is basically 26 base counting with one extra 'nil' entry
3094 * at the beginning from the second digit on and can be
3095 * determined using similar method as 26 base conversion with the
3096 * index shifted -1 after each digit is computed.
3102 * 0 on success, -errno on failure.
3104 static int sd_format_disk_name(char *prefix, int index, char *buf, int buflen)
3106 const int base = 'z' - 'a' + 1;
3107 char *begin = buf + strlen(prefix);
3108 char *end = buf + buflen;
3118 *--p = 'a' + (index % unit);
3119 index = (index / unit) - 1;
3120 } while (index >= 0);
3122 memmove(begin, p, end - p);
3123 memcpy(buf, prefix, strlen(prefix));
3129 * The asynchronous part of sd_probe
3131 static void sd_probe_async(void *data, async_cookie_t cookie)
3133 struct scsi_disk *sdkp = data;
3134 struct scsi_device *sdp;
3141 index = sdkp->index;
3142 dev = &sdp->sdev_gendev;
3144 gd->major = sd_major((index & 0xf0) >> 4);
3145 gd->first_minor = ((index & 0xf) << 4) | (index & 0xfff00);
3146 gd->minors = SD_MINORS;
3148 gd->fops = &sd_fops;
3149 gd->private_data = &sdkp->driver;
3150 gd->queue = sdkp->device->request_queue;
3152 /* defaults, until the device tells us otherwise */
3153 sdp->sector_size = 512;
3155 sdkp->media_present = 1;
3156 sdkp->write_prot = 0;
3157 sdkp->cache_override = 0;
3161 sdkp->first_scan = 1;
3162 sdkp->max_medium_access_timeouts = SD_MAX_MEDIUM_TIMEOUTS;
3164 sd_revalidate_disk(gd);
3166 gd->flags = GENHD_FL_EXT_DEVT;
3167 if (sdp->removable) {
3168 gd->flags |= GENHD_FL_REMOVABLE;
3169 gd->events |= DISK_EVENT_MEDIA_CHANGE;
3172 blk_pm_runtime_init(sdp->request_queue, dev);
3173 device_add_disk(dev, gd);
3175 sd_dif_config_host(sdkp);
3177 sd_revalidate_disk(gd);
3179 sd_printk(KERN_NOTICE, sdkp, "Attached SCSI %sdisk\n",
3180 sdp->removable ? "removable " : "");
3181 scsi_autopm_put_device(sdp);
3182 put_device(&sdkp->dev);
3186 * sd_probe - called during driver initialization and whenever a
3187 * new scsi device is attached to the system. It is called once
3188 * for each scsi device (not just disks) present.
3189 * @dev: pointer to device object
3191 * Returns 0 if successful (or not interested in this scsi device
3192 * (e.g. scanner)); 1 when there is an error.
3194 * Note: this function is invoked from the scsi mid-level.
3195 * This function sets up the mapping between a given
3196 * <host,channel,id,lun> (found in sdp) and new device name
3197 * (e.g. /dev/sda). More precisely it is the block device major
3198 * and minor number that is chosen here.
3200 * Assume sd_probe is not re-entrant (for time being)
3201 * Also think about sd_probe() and sd_remove() running coincidentally.
3203 static int sd_probe(struct device *dev)
3205 struct scsi_device *sdp = to_scsi_device(dev);
3206 struct scsi_disk *sdkp;
3211 scsi_autopm_get_device(sdp);
3213 if (sdp->type != TYPE_DISK &&
3214 sdp->type != TYPE_ZBC &&
3215 sdp->type != TYPE_MOD &&
3216 sdp->type != TYPE_RBC)
3219 #ifndef CONFIG_BLK_DEV_ZONED
3220 if (sdp->type == TYPE_ZBC)
3223 SCSI_LOG_HLQUEUE(3, sdev_printk(KERN_INFO, sdp,
3227 sdkp = kzalloc(sizeof(*sdkp), GFP_KERNEL);
3231 gd = alloc_disk(SD_MINORS);
3236 if (!ida_pre_get(&sd_index_ida, GFP_KERNEL))
3239 spin_lock(&sd_index_lock);
3240 error = ida_get_new(&sd_index_ida, &index);
3241 spin_unlock(&sd_index_lock);
3242 } while (error == -EAGAIN);
3245 sdev_printk(KERN_WARNING, sdp, "sd_probe: memory exhausted.\n");
3249 error = sd_format_disk_name("sd", index, gd->disk_name, DISK_NAME_LEN);
3251 sdev_printk(KERN_WARNING, sdp, "SCSI disk (sd) name length exceeded.\n");
3252 goto out_free_index;
3256 sdkp->driver = &sd_template;
3258 sdkp->index = index;
3259 atomic_set(&sdkp->openers, 0);
3260 atomic_set(&sdkp->device->ioerr_cnt, 0);
3262 if (!sdp->request_queue->rq_timeout) {
3263 if (sdp->type != TYPE_MOD)
3264 blk_queue_rq_timeout(sdp->request_queue, SD_TIMEOUT);
3266 blk_queue_rq_timeout(sdp->request_queue,
3270 device_initialize(&sdkp->dev);
3271 sdkp->dev.parent = dev;
3272 sdkp->dev.class = &sd_disk_class;
3273 dev_set_name(&sdkp->dev, "%s", dev_name(dev));
3275 error = device_add(&sdkp->dev);
3277 goto out_free_index;
3280 dev_set_drvdata(dev, sdkp);
3282 get_device(&sdkp->dev); /* prevent release before async_schedule */
3283 async_schedule_domain(sd_probe_async, sdkp, &scsi_sd_probe_domain);
3288 spin_lock(&sd_index_lock);
3289 ida_remove(&sd_index_ida, index);
3290 spin_unlock(&sd_index_lock);
3296 scsi_autopm_put_device(sdp);
3301 * sd_remove - called whenever a scsi disk (previously recognized by
3302 * sd_probe) is detached from the system. It is called (potentially
3303 * multiple times) during sd module unload.
3304 * @dev: pointer to device object
3306 * Note: this function is invoked from the scsi mid-level.
3307 * This function potentially frees up a device name (e.g. /dev/sdc)
3308 * that could be re-used by a subsequent sd_probe().
3309 * This function is not called when the built-in sd driver is "exit-ed".
3311 static int sd_remove(struct device *dev)
3313 struct scsi_disk *sdkp;
3316 sdkp = dev_get_drvdata(dev);
3317 devt = disk_devt(sdkp->disk);
3318 scsi_autopm_get_device(sdkp->device);
3320 async_synchronize_full_domain(&scsi_sd_pm_domain);
3321 async_synchronize_full_domain(&scsi_sd_probe_domain);
3322 device_del(&sdkp->dev);
3323 del_gendisk(sdkp->disk);
3326 sd_zbc_remove(sdkp);
3328 blk_register_region(devt, SD_MINORS, NULL,
3329 sd_default_probe, NULL, NULL);
3331 mutex_lock(&sd_ref_mutex);
3332 dev_set_drvdata(dev, NULL);
3333 put_device(&sdkp->dev);
3334 mutex_unlock(&sd_ref_mutex);
3340 * scsi_disk_release - Called to free the scsi_disk structure
3341 * @dev: pointer to embedded class device
3343 * sd_ref_mutex must be held entering this routine. Because it is
3344 * called on last put, you should always use the scsi_disk_get()
3345 * scsi_disk_put() helpers which manipulate the semaphore directly
3346 * and never do a direct put_device.
3348 static void scsi_disk_release(struct device *dev)
3350 struct scsi_disk *sdkp = to_scsi_disk(dev);
3351 struct gendisk *disk = sdkp->disk;
3353 spin_lock(&sd_index_lock);
3354 ida_remove(&sd_index_ida, sdkp->index);
3355 spin_unlock(&sd_index_lock);
3357 disk->private_data = NULL;
3359 put_device(&sdkp->device->sdev_gendev);
3364 static int sd_start_stop_device(struct scsi_disk *sdkp, int start)
3366 unsigned char cmd[6] = { START_STOP }; /* START_VALID */
3367 struct scsi_sense_hdr sshdr;
3368 struct scsi_device *sdp = sdkp->device;
3372 cmd[4] |= 1; /* START */
3374 if (sdp->start_stop_pwr_cond)
3375 cmd[4] |= start ? 1 << 4 : 3 << 4; /* Active or Standby */
3377 if (!scsi_device_online(sdp))
3380 res = scsi_execute(sdp, cmd, DMA_NONE, NULL, 0, NULL, &sshdr,
3381 SD_TIMEOUT, SD_MAX_RETRIES, 0, RQF_PM, NULL);
3383 sd_print_result(sdkp, "Start/Stop Unit failed", res);
3384 if (driver_byte(res) & DRIVER_SENSE)
3385 sd_print_sense_hdr(sdkp, &sshdr);
3386 if (scsi_sense_valid(&sshdr) &&
3387 /* 0x3a is medium not present */
3392 /* SCSI error codes must not go to the generic layer */
3400 * Send a SYNCHRONIZE CACHE instruction down to the device through
3401 * the normal SCSI command structure. Wait for the command to
3404 static void sd_shutdown(struct device *dev)
3406 struct scsi_disk *sdkp = dev_get_drvdata(dev);
3409 return; /* this can happen */
3411 if (pm_runtime_suspended(dev))
3414 if (sdkp->WCE && sdkp->media_present) {
3415 sd_printk(KERN_NOTICE, sdkp, "Synchronizing SCSI cache\n");
3416 sd_sync_cache(sdkp);
3419 if (system_state != SYSTEM_RESTART && sdkp->device->manage_start_stop) {
3420 sd_printk(KERN_NOTICE, sdkp, "Stopping disk\n");
3421 sd_start_stop_device(sdkp, 0);
3425 static int sd_suspend_common(struct device *dev, bool ignore_stop_errors)
3427 struct scsi_disk *sdkp = dev_get_drvdata(dev);
3430 if (!sdkp) /* E.g.: runtime suspend following sd_remove() */
3433 if (sdkp->WCE && sdkp->media_present) {
3434 sd_printk(KERN_NOTICE, sdkp, "Synchronizing SCSI cache\n");
3435 ret = sd_sync_cache(sdkp);
3437 /* ignore OFFLINE device */
3444 if (sdkp->device->manage_start_stop) {
3445 sd_printk(KERN_NOTICE, sdkp, "Stopping disk\n");
3446 /* an error is not worth aborting a system sleep */
3447 ret = sd_start_stop_device(sdkp, 0);
3448 if (ignore_stop_errors)
3456 static int sd_suspend_system(struct device *dev)
3458 return sd_suspend_common(dev, true);
3461 static int sd_suspend_runtime(struct device *dev)
3463 return sd_suspend_common(dev, false);
3466 static int sd_resume(struct device *dev)
3468 struct scsi_disk *sdkp = dev_get_drvdata(dev);
3470 if (!sdkp) /* E.g.: runtime resume at the start of sd_probe() */
3473 if (!sdkp->device->manage_start_stop)
3476 sd_printk(KERN_NOTICE, sdkp, "Starting disk\n");
3477 return sd_start_stop_device(sdkp, 1);
3481 * init_sd - entry point for this driver (both when built in or when
3484 * Note: this function registers this driver with the scsi mid-level.
3486 static int __init init_sd(void)
3488 int majors = 0, i, err;
3490 SCSI_LOG_HLQUEUE(3, printk("init_sd: sd driver entry point\n"));
3492 for (i = 0; i < SD_MAJORS; i++) {
3493 if (register_blkdev(sd_major(i), "sd") != 0)
3496 blk_register_region(sd_major(i), SD_MINORS, NULL,
3497 sd_default_probe, NULL, NULL);
3503 err = class_register(&sd_disk_class);
3507 sd_cdb_cache = kmem_cache_create("sd_ext_cdb", SD_EXT_CDB_SIZE,
3509 if (!sd_cdb_cache) {
3510 printk(KERN_ERR "sd: can't init extended cdb cache\n");
3515 sd_cdb_pool = mempool_create_slab_pool(SD_MEMPOOL_SIZE, sd_cdb_cache);
3517 printk(KERN_ERR "sd: can't init extended cdb pool\n");
3522 err = scsi_register_driver(&sd_template.gendrv);
3524 goto err_out_driver;
3529 mempool_destroy(sd_cdb_pool);
3532 kmem_cache_destroy(sd_cdb_cache);
3535 class_unregister(&sd_disk_class);
3537 for (i = 0; i < SD_MAJORS; i++)
3538 unregister_blkdev(sd_major(i), "sd");
3543 * exit_sd - exit point for this driver (when it is a module).
3545 * Note: this function unregisters this driver from the scsi mid-level.
3547 static void __exit exit_sd(void)
3551 SCSI_LOG_HLQUEUE(3, printk("exit_sd: exiting sd driver\n"));
3553 scsi_unregister_driver(&sd_template.gendrv);
3554 mempool_destroy(sd_cdb_pool);
3555 kmem_cache_destroy(sd_cdb_cache);
3557 class_unregister(&sd_disk_class);
3559 for (i = 0; i < SD_MAJORS; i++) {
3560 blk_unregister_region(sd_major(i), SD_MINORS);
3561 unregister_blkdev(sd_major(i), "sd");
3565 module_init(init_sd);
3566 module_exit(exit_sd);
3568 static void sd_print_sense_hdr(struct scsi_disk *sdkp,
3569 struct scsi_sense_hdr *sshdr)
3571 scsi_print_sense_hdr(sdkp->device,
3572 sdkp->disk ? sdkp->disk->disk_name : NULL, sshdr);
3575 static void sd_print_result(const struct scsi_disk *sdkp, const char *msg,
3578 const char *hb_string = scsi_hostbyte_string(result);
3579 const char *db_string = scsi_driverbyte_string(result);
3581 if (hb_string || db_string)
3582 sd_printk(KERN_INFO, sdkp,
3583 "%s: Result: hostbyte=%s driverbyte=%s\n", msg,
3584 hb_string ? hb_string : "invalid",
3585 db_string ? db_string : "invalid");
3587 sd_printk(KERN_INFO, sdkp,
3588 "%s: Result: hostbyte=0x%02x driverbyte=0x%02x\n",
3589 msg, host_byte(result), driver_byte(result));