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 void sd_eh_reset(struct scsi_cmnd *);
119 static int sd_eh_action(struct scsi_cmnd *, int);
120 static void sd_read_capacity(struct scsi_disk *sdkp, unsigned char *buffer);
121 static void scsi_disk_release(struct device *cdev);
122 static void sd_print_sense_hdr(struct scsi_disk *, struct scsi_sense_hdr *);
123 static void sd_print_result(const struct scsi_disk *, const char *, int);
125 static DEFINE_SPINLOCK(sd_index_lock);
126 static DEFINE_IDA(sd_index_ida);
128 /* This semaphore is used to mediate the 0->1 reference get in the
129 * face of object destruction (i.e. we can't allow a get on an
130 * object after last put) */
131 static DEFINE_MUTEX(sd_ref_mutex);
133 static struct kmem_cache *sd_cdb_cache;
134 static mempool_t *sd_cdb_pool;
136 static const char *sd_cache_types[] = {
137 "write through", "none", "write back",
138 "write back, no read (daft)"
141 static void sd_set_flush_flag(struct scsi_disk *sdkp)
143 bool wc = false, fua = false;
151 blk_queue_write_cache(sdkp->disk->queue, wc, fua);
155 cache_type_store(struct device *dev, struct device_attribute *attr,
156 const char *buf, size_t count)
158 int i, ct = -1, rcd, wce, sp;
159 struct scsi_disk *sdkp = to_scsi_disk(dev);
160 struct scsi_device *sdp = sdkp->device;
163 struct scsi_mode_data data;
164 struct scsi_sense_hdr sshdr;
165 static const char temp[] = "temporary ";
168 if (sdp->type != TYPE_DISK && sdp->type != TYPE_ZBC)
169 /* no cache control on RBC devices; theoretically they
170 * can do it, but there's probably so many exceptions
171 * it's not worth the risk */
174 if (strncmp(buf, temp, sizeof(temp) - 1) == 0) {
175 buf += sizeof(temp) - 1;
176 sdkp->cache_override = 1;
178 sdkp->cache_override = 0;
181 for (i = 0; i < ARRAY_SIZE(sd_cache_types); i++) {
182 len = strlen(sd_cache_types[i]);
183 if (strncmp(sd_cache_types[i], buf, len) == 0 &&
191 rcd = ct & 0x01 ? 1 : 0;
192 wce = (ct & 0x02) && !sdkp->write_prot ? 1 : 0;
194 if (sdkp->cache_override) {
197 sd_set_flush_flag(sdkp);
201 if (scsi_mode_sense(sdp, 0x08, 8, buffer, sizeof(buffer), SD_TIMEOUT,
202 SD_MAX_RETRIES, &data, NULL))
204 len = min_t(size_t, sizeof(buffer), data.length - data.header_length -
205 data.block_descriptor_length);
206 buffer_data = buffer + data.header_length +
207 data.block_descriptor_length;
208 buffer_data[2] &= ~0x05;
209 buffer_data[2] |= wce << 2 | rcd;
210 sp = buffer_data[0] & 0x80 ? 1 : 0;
211 buffer_data[0] &= ~0x80;
213 if (scsi_mode_select(sdp, 1, sp, 8, buffer_data, len, SD_TIMEOUT,
214 SD_MAX_RETRIES, &data, &sshdr)) {
215 if (scsi_sense_valid(&sshdr))
216 sd_print_sense_hdr(sdkp, &sshdr);
219 revalidate_disk(sdkp->disk);
224 manage_start_stop_show(struct device *dev, struct device_attribute *attr,
227 struct scsi_disk *sdkp = to_scsi_disk(dev);
228 struct scsi_device *sdp = sdkp->device;
230 return snprintf(buf, 20, "%u\n", sdp->manage_start_stop);
234 manage_start_stop_store(struct device *dev, struct device_attribute *attr,
235 const char *buf, size_t count)
237 struct scsi_disk *sdkp = to_scsi_disk(dev);
238 struct scsi_device *sdp = sdkp->device;
240 if (!capable(CAP_SYS_ADMIN))
243 sdp->manage_start_stop = simple_strtoul(buf, NULL, 10);
247 static DEVICE_ATTR_RW(manage_start_stop);
250 allow_restart_show(struct device *dev, struct device_attribute *attr, char *buf)
252 struct scsi_disk *sdkp = to_scsi_disk(dev);
254 return snprintf(buf, 40, "%d\n", sdkp->device->allow_restart);
258 allow_restart_store(struct device *dev, struct device_attribute *attr,
259 const char *buf, size_t count)
261 struct scsi_disk *sdkp = to_scsi_disk(dev);
262 struct scsi_device *sdp = sdkp->device;
264 if (!capable(CAP_SYS_ADMIN))
267 if (sdp->type != TYPE_DISK && sdp->type != TYPE_ZBC)
270 sdp->allow_restart = simple_strtoul(buf, NULL, 10);
274 static DEVICE_ATTR_RW(allow_restart);
277 cache_type_show(struct device *dev, struct device_attribute *attr, char *buf)
279 struct scsi_disk *sdkp = to_scsi_disk(dev);
280 int ct = sdkp->RCD + 2*sdkp->WCE;
282 return snprintf(buf, 40, "%s\n", sd_cache_types[ct]);
284 static DEVICE_ATTR_RW(cache_type);
287 FUA_show(struct device *dev, struct device_attribute *attr, char *buf)
289 struct scsi_disk *sdkp = to_scsi_disk(dev);
291 return snprintf(buf, 20, "%u\n", sdkp->DPOFUA);
293 static DEVICE_ATTR_RO(FUA);
296 protection_type_show(struct device *dev, struct device_attribute *attr,
299 struct scsi_disk *sdkp = to_scsi_disk(dev);
301 return snprintf(buf, 20, "%u\n", sdkp->protection_type);
305 protection_type_store(struct device *dev, struct device_attribute *attr,
306 const char *buf, size_t count)
308 struct scsi_disk *sdkp = to_scsi_disk(dev);
312 if (!capable(CAP_SYS_ADMIN))
315 err = kstrtouint(buf, 10, &val);
320 if (val >= 0 && val <= T10_PI_TYPE3_PROTECTION)
321 sdkp->protection_type = val;
325 static DEVICE_ATTR_RW(protection_type);
328 protection_mode_show(struct device *dev, struct device_attribute *attr,
331 struct scsi_disk *sdkp = to_scsi_disk(dev);
332 struct scsi_device *sdp = sdkp->device;
333 unsigned int dif, dix;
335 dif = scsi_host_dif_capable(sdp->host, sdkp->protection_type);
336 dix = scsi_host_dix_capable(sdp->host, sdkp->protection_type);
338 if (!dix && scsi_host_dix_capable(sdp->host, T10_PI_TYPE0_PROTECTION)) {
344 return snprintf(buf, 20, "none\n");
346 return snprintf(buf, 20, "%s%u\n", dix ? "dix" : "dif", dif);
348 static DEVICE_ATTR_RO(protection_mode);
351 app_tag_own_show(struct device *dev, struct device_attribute *attr, char *buf)
353 struct scsi_disk *sdkp = to_scsi_disk(dev);
355 return snprintf(buf, 20, "%u\n", sdkp->ATO);
357 static DEVICE_ATTR_RO(app_tag_own);
360 thin_provisioning_show(struct device *dev, struct device_attribute *attr,
363 struct scsi_disk *sdkp = to_scsi_disk(dev);
365 return snprintf(buf, 20, "%u\n", sdkp->lbpme);
367 static DEVICE_ATTR_RO(thin_provisioning);
369 static const char *lbp_mode[] = {
370 [SD_LBP_FULL] = "full",
371 [SD_LBP_UNMAP] = "unmap",
372 [SD_LBP_WS16] = "writesame_16",
373 [SD_LBP_WS10] = "writesame_10",
374 [SD_LBP_ZERO] = "writesame_zero",
375 [SD_LBP_DISABLE] = "disabled",
379 provisioning_mode_show(struct device *dev, struct device_attribute *attr,
382 struct scsi_disk *sdkp = to_scsi_disk(dev);
384 return snprintf(buf, 20, "%s\n", lbp_mode[sdkp->provisioning_mode]);
388 provisioning_mode_store(struct device *dev, struct device_attribute *attr,
389 const char *buf, size_t count)
391 struct scsi_disk *sdkp = to_scsi_disk(dev);
392 struct scsi_device *sdp = sdkp->device;
394 if (!capable(CAP_SYS_ADMIN))
397 if (sd_is_zoned(sdkp)) {
398 sd_config_discard(sdkp, SD_LBP_DISABLE);
402 if (sdp->type != TYPE_DISK)
405 if (!strncmp(buf, lbp_mode[SD_LBP_UNMAP], 20))
406 sd_config_discard(sdkp, SD_LBP_UNMAP);
407 else if (!strncmp(buf, lbp_mode[SD_LBP_WS16], 20))
408 sd_config_discard(sdkp, SD_LBP_WS16);
409 else if (!strncmp(buf, lbp_mode[SD_LBP_WS10], 20))
410 sd_config_discard(sdkp, SD_LBP_WS10);
411 else if (!strncmp(buf, lbp_mode[SD_LBP_ZERO], 20))
412 sd_config_discard(sdkp, SD_LBP_ZERO);
413 else if (!strncmp(buf, lbp_mode[SD_LBP_DISABLE], 20))
414 sd_config_discard(sdkp, SD_LBP_DISABLE);
420 static DEVICE_ATTR_RW(provisioning_mode);
422 static const char *zeroing_mode[] = {
423 [SD_ZERO_WRITE] = "write",
424 [SD_ZERO_WS] = "writesame",
425 [SD_ZERO_WS16_UNMAP] = "writesame_16_unmap",
426 [SD_ZERO_WS10_UNMAP] = "writesame_10_unmap",
430 zeroing_mode_show(struct device *dev, struct device_attribute *attr,
433 struct scsi_disk *sdkp = to_scsi_disk(dev);
435 return snprintf(buf, 20, "%s\n", zeroing_mode[sdkp->zeroing_mode]);
439 zeroing_mode_store(struct device *dev, struct device_attribute *attr,
440 const char *buf, size_t count)
442 struct scsi_disk *sdkp = to_scsi_disk(dev);
444 if (!capable(CAP_SYS_ADMIN))
447 if (!strncmp(buf, zeroing_mode[SD_ZERO_WRITE], 20))
448 sdkp->zeroing_mode = SD_ZERO_WRITE;
449 else if (!strncmp(buf, zeroing_mode[SD_ZERO_WS], 20))
450 sdkp->zeroing_mode = SD_ZERO_WS;
451 else if (!strncmp(buf, zeroing_mode[SD_ZERO_WS16_UNMAP], 20))
452 sdkp->zeroing_mode = SD_ZERO_WS16_UNMAP;
453 else if (!strncmp(buf, zeroing_mode[SD_ZERO_WS10_UNMAP], 20))
454 sdkp->zeroing_mode = SD_ZERO_WS10_UNMAP;
460 static DEVICE_ATTR_RW(zeroing_mode);
463 max_medium_access_timeouts_show(struct device *dev,
464 struct device_attribute *attr, char *buf)
466 struct scsi_disk *sdkp = to_scsi_disk(dev);
468 return snprintf(buf, 20, "%u\n", sdkp->max_medium_access_timeouts);
472 max_medium_access_timeouts_store(struct device *dev,
473 struct device_attribute *attr, const char *buf,
476 struct scsi_disk *sdkp = to_scsi_disk(dev);
479 if (!capable(CAP_SYS_ADMIN))
482 err = kstrtouint(buf, 10, &sdkp->max_medium_access_timeouts);
484 return err ? err : count;
486 static DEVICE_ATTR_RW(max_medium_access_timeouts);
489 max_write_same_blocks_show(struct device *dev, struct device_attribute *attr,
492 struct scsi_disk *sdkp = to_scsi_disk(dev);
494 return snprintf(buf, 20, "%u\n", sdkp->max_ws_blocks);
498 max_write_same_blocks_store(struct device *dev, struct device_attribute *attr,
499 const char *buf, size_t count)
501 struct scsi_disk *sdkp = to_scsi_disk(dev);
502 struct scsi_device *sdp = sdkp->device;
506 if (!capable(CAP_SYS_ADMIN))
509 if (sdp->type != TYPE_DISK && sdp->type != TYPE_ZBC)
512 err = kstrtoul(buf, 10, &max);
518 sdp->no_write_same = 1;
519 else if (max <= SD_MAX_WS16_BLOCKS) {
520 sdp->no_write_same = 0;
521 sdkp->max_ws_blocks = max;
524 sd_config_write_same(sdkp);
528 static DEVICE_ATTR_RW(max_write_same_blocks);
530 static struct attribute *sd_disk_attrs[] = {
531 &dev_attr_cache_type.attr,
533 &dev_attr_allow_restart.attr,
534 &dev_attr_manage_start_stop.attr,
535 &dev_attr_protection_type.attr,
536 &dev_attr_protection_mode.attr,
537 &dev_attr_app_tag_own.attr,
538 &dev_attr_thin_provisioning.attr,
539 &dev_attr_provisioning_mode.attr,
540 &dev_attr_zeroing_mode.attr,
541 &dev_attr_max_write_same_blocks.attr,
542 &dev_attr_max_medium_access_timeouts.attr,
545 ATTRIBUTE_GROUPS(sd_disk);
547 static struct class sd_disk_class = {
549 .owner = THIS_MODULE,
550 .dev_release = scsi_disk_release,
551 .dev_groups = sd_disk_groups,
554 static const struct dev_pm_ops sd_pm_ops = {
555 .suspend = sd_suspend_system,
557 .poweroff = sd_suspend_system,
558 .restore = sd_resume,
559 .runtime_suspend = sd_suspend_runtime,
560 .runtime_resume = sd_resume,
563 static struct scsi_driver sd_template = {
566 .owner = THIS_MODULE,
569 .shutdown = sd_shutdown,
573 .init_command = sd_init_command,
574 .uninit_command = sd_uninit_command,
576 .eh_action = sd_eh_action,
577 .eh_reset = sd_eh_reset,
581 * Dummy kobj_map->probe function.
582 * The default ->probe function will call modprobe, which is
583 * pointless as this module is already loaded.
585 static struct kobject *sd_default_probe(dev_t devt, int *partno, void *data)
591 * Device no to disk mapping:
593 * major disc2 disc p1
594 * |............|.............|....|....| <- dev_t
597 * Inside a major, we have 16k disks, however mapped non-
598 * contiguously. The first 16 disks are for major0, the next
599 * ones with major1, ... Disk 256 is for major0 again, disk 272
601 * As we stay compatible with our numbering scheme, we can reuse
602 * the well-know SCSI majors 8, 65--71, 136--143.
604 static int sd_major(int major_idx)
608 return SCSI_DISK0_MAJOR;
610 return SCSI_DISK1_MAJOR + major_idx - 1;
612 return SCSI_DISK8_MAJOR + major_idx - 8;
615 return 0; /* shut up gcc */
619 static struct scsi_disk *scsi_disk_get(struct gendisk *disk)
621 struct scsi_disk *sdkp = NULL;
623 mutex_lock(&sd_ref_mutex);
625 if (disk->private_data) {
626 sdkp = scsi_disk(disk);
627 if (scsi_device_get(sdkp->device) == 0)
628 get_device(&sdkp->dev);
632 mutex_unlock(&sd_ref_mutex);
636 static void scsi_disk_put(struct scsi_disk *sdkp)
638 struct scsi_device *sdev = sdkp->device;
640 mutex_lock(&sd_ref_mutex);
641 put_device(&sdkp->dev);
642 scsi_device_put(sdev);
643 mutex_unlock(&sd_ref_mutex);
646 static unsigned char sd_setup_protect_cmnd(struct scsi_cmnd *scmd,
647 unsigned int dix, unsigned int dif)
649 struct bio *bio = scmd->request->bio;
650 unsigned int prot_op = sd_prot_op(rq_data_dir(scmd->request), dix, dif);
651 unsigned int protect = 0;
653 if (dix) { /* DIX Type 0, 1, 2, 3 */
654 if (bio_integrity_flagged(bio, BIP_IP_CHECKSUM))
655 scmd->prot_flags |= SCSI_PROT_IP_CHECKSUM;
657 if (bio_integrity_flagged(bio, BIP_CTRL_NOCHECK) == false)
658 scmd->prot_flags |= SCSI_PROT_GUARD_CHECK;
661 if (dif != T10_PI_TYPE3_PROTECTION) { /* DIX/DIF Type 0, 1, 2 */
662 scmd->prot_flags |= SCSI_PROT_REF_INCREMENT;
664 if (bio_integrity_flagged(bio, BIP_CTRL_NOCHECK) == false)
665 scmd->prot_flags |= SCSI_PROT_REF_CHECK;
668 if (dif) { /* DIX/DIF Type 1, 2, 3 */
669 scmd->prot_flags |= SCSI_PROT_TRANSFER_PI;
671 if (bio_integrity_flagged(bio, BIP_DISK_NOCHECK))
672 protect = 3 << 5; /* Disable target PI checking */
674 protect = 1 << 5; /* Enable target PI checking */
677 scsi_set_prot_op(scmd, prot_op);
678 scsi_set_prot_type(scmd, dif);
679 scmd->prot_flags &= sd_prot_flag_mask(prot_op);
684 static void sd_config_discard(struct scsi_disk *sdkp, unsigned int mode)
686 struct request_queue *q = sdkp->disk->queue;
687 unsigned int logical_block_size = sdkp->device->sector_size;
688 unsigned int max_blocks = 0;
690 q->limits.discard_alignment =
691 sdkp->unmap_alignment * logical_block_size;
692 q->limits.discard_granularity =
693 max(sdkp->physical_block_size,
694 sdkp->unmap_granularity * logical_block_size);
695 sdkp->provisioning_mode = mode;
700 blk_queue_max_discard_sectors(q, 0);
701 queue_flag_clear_unlocked(QUEUE_FLAG_DISCARD, q);
705 max_blocks = min_not_zero(sdkp->max_unmap_blocks,
706 (u32)SD_MAX_WS16_BLOCKS);
710 max_blocks = min_not_zero(sdkp->max_ws_blocks,
711 (u32)SD_MAX_WS16_BLOCKS);
715 max_blocks = min_not_zero(sdkp->max_ws_blocks,
716 (u32)SD_MAX_WS10_BLOCKS);
720 max_blocks = min_not_zero(sdkp->max_ws_blocks,
721 (u32)SD_MAX_WS10_BLOCKS);
725 blk_queue_max_discard_sectors(q, max_blocks * (logical_block_size >> 9));
726 queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, q);
729 static int sd_setup_unmap_cmnd(struct scsi_cmnd *cmd)
731 struct scsi_device *sdp = cmd->device;
732 struct request *rq = cmd->request;
733 u64 sector = blk_rq_pos(rq) >> (ilog2(sdp->sector_size) - 9);
734 u32 nr_sectors = blk_rq_sectors(rq) >> (ilog2(sdp->sector_size) - 9);
735 unsigned int data_len = 24;
738 rq->special_vec.bv_page = alloc_page(GFP_ATOMIC | __GFP_ZERO);
739 if (!rq->special_vec.bv_page)
740 return BLKPREP_DEFER;
741 rq->special_vec.bv_offset = 0;
742 rq->special_vec.bv_len = data_len;
743 rq->rq_flags |= RQF_SPECIAL_PAYLOAD;
746 cmd->cmnd[0] = UNMAP;
749 buf = page_address(rq->special_vec.bv_page);
750 put_unaligned_be16(6 + 16, &buf[0]);
751 put_unaligned_be16(16, &buf[2]);
752 put_unaligned_be64(sector, &buf[8]);
753 put_unaligned_be32(nr_sectors, &buf[16]);
755 cmd->allowed = SD_MAX_RETRIES;
756 cmd->transfersize = data_len;
757 rq->timeout = SD_TIMEOUT;
758 scsi_req(rq)->resid_len = data_len;
760 return scsi_init_io(cmd);
763 static int sd_setup_write_same16_cmnd(struct scsi_cmnd *cmd, bool unmap)
765 struct scsi_device *sdp = cmd->device;
766 struct request *rq = cmd->request;
767 u64 sector = blk_rq_pos(rq) >> (ilog2(sdp->sector_size) - 9);
768 u32 nr_sectors = blk_rq_sectors(rq) >> (ilog2(sdp->sector_size) - 9);
769 u32 data_len = sdp->sector_size;
771 rq->special_vec.bv_page = alloc_page(GFP_ATOMIC | __GFP_ZERO);
772 if (!rq->special_vec.bv_page)
773 return BLKPREP_DEFER;
774 rq->special_vec.bv_offset = 0;
775 rq->special_vec.bv_len = data_len;
776 rq->rq_flags |= RQF_SPECIAL_PAYLOAD;
779 cmd->cmnd[0] = WRITE_SAME_16;
781 cmd->cmnd[1] = 0x8; /* UNMAP */
782 put_unaligned_be64(sector, &cmd->cmnd[2]);
783 put_unaligned_be32(nr_sectors, &cmd->cmnd[10]);
785 cmd->allowed = SD_MAX_RETRIES;
786 cmd->transfersize = data_len;
787 rq->timeout = unmap ? SD_TIMEOUT : SD_WRITE_SAME_TIMEOUT;
788 scsi_req(rq)->resid_len = data_len;
790 return scsi_init_io(cmd);
793 static int sd_setup_write_same10_cmnd(struct scsi_cmnd *cmd, bool unmap)
795 struct scsi_device *sdp = cmd->device;
796 struct request *rq = cmd->request;
797 u64 sector = blk_rq_pos(rq) >> (ilog2(sdp->sector_size) - 9);
798 u32 nr_sectors = blk_rq_sectors(rq) >> (ilog2(sdp->sector_size) - 9);
799 u32 data_len = sdp->sector_size;
801 rq->special_vec.bv_page = alloc_page(GFP_ATOMIC | __GFP_ZERO);
802 if (!rq->special_vec.bv_page)
803 return BLKPREP_DEFER;
804 rq->special_vec.bv_offset = 0;
805 rq->special_vec.bv_len = data_len;
806 rq->rq_flags |= RQF_SPECIAL_PAYLOAD;
809 cmd->cmnd[0] = WRITE_SAME;
811 cmd->cmnd[1] = 0x8; /* UNMAP */
812 put_unaligned_be32(sector, &cmd->cmnd[2]);
813 put_unaligned_be16(nr_sectors, &cmd->cmnd[7]);
815 cmd->allowed = SD_MAX_RETRIES;
816 cmd->transfersize = data_len;
817 rq->timeout = unmap ? SD_TIMEOUT : SD_WRITE_SAME_TIMEOUT;
818 scsi_req(rq)->resid_len = data_len;
820 return scsi_init_io(cmd);
823 static int sd_setup_write_zeroes_cmnd(struct scsi_cmnd *cmd)
825 struct request *rq = cmd->request;
826 struct scsi_device *sdp = cmd->device;
827 struct scsi_disk *sdkp = scsi_disk(rq->rq_disk);
828 u64 sector = blk_rq_pos(rq) >> (ilog2(sdp->sector_size) - 9);
829 u32 nr_sectors = blk_rq_sectors(rq) >> (ilog2(sdp->sector_size) - 9);
832 if (!(rq->cmd_flags & REQ_NOUNMAP)) {
833 switch (sdkp->zeroing_mode) {
834 case SD_ZERO_WS16_UNMAP:
835 ret = sd_setup_write_same16_cmnd(cmd, true);
837 case SD_ZERO_WS10_UNMAP:
838 ret = sd_setup_write_same10_cmnd(cmd, true);
843 if (sdp->no_write_same)
844 return BLKPREP_INVALID;
846 if (sdkp->ws16 || sector > 0xffffffff || nr_sectors > 0xffff)
847 ret = sd_setup_write_same16_cmnd(cmd, false);
849 ret = sd_setup_write_same10_cmnd(cmd, false);
852 if (sd_is_zoned(sdkp) && ret == BLKPREP_OK)
853 return sd_zbc_write_lock_zone(cmd);
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 set up either WRITE SAME(10) or WRITE SAME(16) depending on
905 * the preference indicated by the 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_write_lock_zone(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;
963 if (sd_is_zoned(sdkp) && ret != BLKPREP_OK)
964 sd_zbc_write_unlock_zone(cmd);
969 static int sd_setup_flush_cmnd(struct scsi_cmnd *cmd)
971 struct request *rq = cmd->request;
973 /* flush requests don't perform I/O, zero the S/G table */
974 memset(&cmd->sdb, 0, sizeof(cmd->sdb));
976 cmd->cmnd[0] = SYNCHRONIZE_CACHE;
978 cmd->transfersize = 0;
979 cmd->allowed = SD_MAX_RETRIES;
981 rq->timeout = rq->q->rq_timeout * SD_FLUSH_TIMEOUT_MULTIPLIER;
985 static int sd_setup_read_write_cmnd(struct scsi_cmnd *SCpnt)
987 struct request *rq = SCpnt->request;
988 struct scsi_device *sdp = SCpnt->device;
989 struct gendisk *disk = rq->rq_disk;
990 struct scsi_disk *sdkp = scsi_disk(disk);
991 sector_t block = blk_rq_pos(rq);
993 unsigned int this_count = blk_rq_sectors(rq);
994 unsigned int dif, dix;
995 bool zoned_write = sd_is_zoned(sdkp) && rq_data_dir(rq) == WRITE;
997 unsigned char protect;
1000 ret = sd_zbc_write_lock_zone(SCpnt);
1001 if (ret != BLKPREP_OK)
1005 ret = scsi_init_io(SCpnt);
1006 if (ret != BLKPREP_OK)
1008 SCpnt = rq->special;
1010 /* from here on until we're complete, any goto out
1011 * is used for a killable error condition */
1015 scmd_printk(KERN_INFO, SCpnt,
1016 "%s: block=%llu, count=%d\n",
1017 __func__, (unsigned long long)block, this_count));
1019 if (!sdp || !scsi_device_online(sdp) ||
1020 block + blk_rq_sectors(rq) > get_capacity(disk)) {
1021 SCSI_LOG_HLQUEUE(2, scmd_printk(KERN_INFO, SCpnt,
1022 "Finishing %u sectors\n",
1023 blk_rq_sectors(rq)));
1024 SCSI_LOG_HLQUEUE(2, scmd_printk(KERN_INFO, SCpnt,
1025 "Retry with 0x%p\n", SCpnt));
1031 * quietly refuse to do anything to a changed disc until
1032 * the changed bit has been reset
1034 /* printk("SCSI disk has been changed or is not present. Prohibiting further I/O.\n"); */
1039 * Some SD card readers can't handle multi-sector accesses which touch
1040 * the last one or two hardware sectors. Split accesses as needed.
1042 threshold = get_capacity(disk) - SD_LAST_BUGGY_SECTORS *
1043 (sdp->sector_size / 512);
1045 if (unlikely(sdp->last_sector_bug && block + this_count > threshold)) {
1046 if (block < threshold) {
1047 /* Access up to the threshold but not beyond */
1048 this_count = threshold - block;
1050 /* Access only a single hardware sector */
1051 this_count = sdp->sector_size / 512;
1055 SCSI_LOG_HLQUEUE(2, scmd_printk(KERN_INFO, SCpnt, "block=%llu\n",
1056 (unsigned long long)block));
1059 * If we have a 1K hardware sectorsize, prevent access to single
1060 * 512 byte sectors. In theory we could handle this - in fact
1061 * the scsi cdrom driver must be able to handle this because
1062 * we typically use 1K blocksizes, and cdroms typically have
1063 * 2K hardware sectorsizes. Of course, things are simpler
1064 * with the cdrom, since it is read-only. For performance
1065 * reasons, the filesystems should be able to handle this
1066 * and not force the scsi disk driver to use bounce buffers
1069 if (sdp->sector_size == 1024) {
1070 if ((block & 1) || (blk_rq_sectors(rq) & 1)) {
1071 scmd_printk(KERN_ERR, SCpnt,
1072 "Bad block number requested\n");
1076 this_count = this_count >> 1;
1079 if (sdp->sector_size == 2048) {
1080 if ((block & 3) || (blk_rq_sectors(rq) & 3)) {
1081 scmd_printk(KERN_ERR, SCpnt,
1082 "Bad block number requested\n");
1086 this_count = this_count >> 2;
1089 if (sdp->sector_size == 4096) {
1090 if ((block & 7) || (blk_rq_sectors(rq) & 7)) {
1091 scmd_printk(KERN_ERR, SCpnt,
1092 "Bad block number requested\n");
1096 this_count = this_count >> 3;
1099 if (rq_data_dir(rq) == WRITE) {
1100 SCpnt->cmnd[0] = WRITE_6;
1102 if (blk_integrity_rq(rq))
1103 sd_dif_prepare(SCpnt);
1105 } else if (rq_data_dir(rq) == READ) {
1106 SCpnt->cmnd[0] = READ_6;
1108 scmd_printk(KERN_ERR, SCpnt, "Unknown command %d\n", req_op(rq));
1112 SCSI_LOG_HLQUEUE(2, scmd_printk(KERN_INFO, SCpnt,
1113 "%s %d/%u 512 byte blocks.\n",
1114 (rq_data_dir(rq) == WRITE) ?
1115 "writing" : "reading", this_count,
1116 blk_rq_sectors(rq)));
1118 dix = scsi_prot_sg_count(SCpnt);
1119 dif = scsi_host_dif_capable(SCpnt->device->host, sdkp->protection_type);
1122 protect = sd_setup_protect_cmnd(SCpnt, dix, dif);
1126 if (protect && sdkp->protection_type == T10_PI_TYPE2_PROTECTION) {
1127 SCpnt->cmnd = mempool_alloc(sd_cdb_pool, GFP_ATOMIC);
1129 if (unlikely(SCpnt->cmnd == NULL)) {
1130 ret = BLKPREP_DEFER;
1134 SCpnt->cmd_len = SD_EXT_CDB_SIZE;
1135 memset(SCpnt->cmnd, 0, SCpnt->cmd_len);
1136 SCpnt->cmnd[0] = VARIABLE_LENGTH_CMD;
1137 SCpnt->cmnd[7] = 0x18;
1138 SCpnt->cmnd[9] = (rq_data_dir(rq) == READ) ? READ_32 : WRITE_32;
1139 SCpnt->cmnd[10] = protect | ((rq->cmd_flags & REQ_FUA) ? 0x8 : 0);
1142 SCpnt->cmnd[12] = sizeof(block) > 4 ? (unsigned char) (block >> 56) & 0xff : 0;
1143 SCpnt->cmnd[13] = sizeof(block) > 4 ? (unsigned char) (block >> 48) & 0xff : 0;
1144 SCpnt->cmnd[14] = sizeof(block) > 4 ? (unsigned char) (block >> 40) & 0xff : 0;
1145 SCpnt->cmnd[15] = sizeof(block) > 4 ? (unsigned char) (block >> 32) & 0xff : 0;
1146 SCpnt->cmnd[16] = (unsigned char) (block >> 24) & 0xff;
1147 SCpnt->cmnd[17] = (unsigned char) (block >> 16) & 0xff;
1148 SCpnt->cmnd[18] = (unsigned char) (block >> 8) & 0xff;
1149 SCpnt->cmnd[19] = (unsigned char) block & 0xff;
1151 /* Expected Indirect LBA */
1152 SCpnt->cmnd[20] = (unsigned char) (block >> 24) & 0xff;
1153 SCpnt->cmnd[21] = (unsigned char) (block >> 16) & 0xff;
1154 SCpnt->cmnd[22] = (unsigned char) (block >> 8) & 0xff;
1155 SCpnt->cmnd[23] = (unsigned char) block & 0xff;
1157 /* Transfer length */
1158 SCpnt->cmnd[28] = (unsigned char) (this_count >> 24) & 0xff;
1159 SCpnt->cmnd[29] = (unsigned char) (this_count >> 16) & 0xff;
1160 SCpnt->cmnd[30] = (unsigned char) (this_count >> 8) & 0xff;
1161 SCpnt->cmnd[31] = (unsigned char) this_count & 0xff;
1162 } else if (sdp->use_16_for_rw || (this_count > 0xffff)) {
1163 SCpnt->cmnd[0] += READ_16 - READ_6;
1164 SCpnt->cmnd[1] = protect | ((rq->cmd_flags & REQ_FUA) ? 0x8 : 0);
1165 SCpnt->cmnd[2] = sizeof(block) > 4 ? (unsigned char) (block >> 56) & 0xff : 0;
1166 SCpnt->cmnd[3] = sizeof(block) > 4 ? (unsigned char) (block >> 48) & 0xff : 0;
1167 SCpnt->cmnd[4] = sizeof(block) > 4 ? (unsigned char) (block >> 40) & 0xff : 0;
1168 SCpnt->cmnd[5] = sizeof(block) > 4 ? (unsigned char) (block >> 32) & 0xff : 0;
1169 SCpnt->cmnd[6] = (unsigned char) (block >> 24) & 0xff;
1170 SCpnt->cmnd[7] = (unsigned char) (block >> 16) & 0xff;
1171 SCpnt->cmnd[8] = (unsigned char) (block >> 8) & 0xff;
1172 SCpnt->cmnd[9] = (unsigned char) block & 0xff;
1173 SCpnt->cmnd[10] = (unsigned char) (this_count >> 24) & 0xff;
1174 SCpnt->cmnd[11] = (unsigned char) (this_count >> 16) & 0xff;
1175 SCpnt->cmnd[12] = (unsigned char) (this_count >> 8) & 0xff;
1176 SCpnt->cmnd[13] = (unsigned char) this_count & 0xff;
1177 SCpnt->cmnd[14] = SCpnt->cmnd[15] = 0;
1178 } else if ((this_count > 0xff) || (block > 0x1fffff) ||
1179 scsi_device_protection(SCpnt->device) ||
1180 SCpnt->device->use_10_for_rw) {
1181 SCpnt->cmnd[0] += READ_10 - READ_6;
1182 SCpnt->cmnd[1] = protect | ((rq->cmd_flags & REQ_FUA) ? 0x8 : 0);
1183 SCpnt->cmnd[2] = (unsigned char) (block >> 24) & 0xff;
1184 SCpnt->cmnd[3] = (unsigned char) (block >> 16) & 0xff;
1185 SCpnt->cmnd[4] = (unsigned char) (block >> 8) & 0xff;
1186 SCpnt->cmnd[5] = (unsigned char) block & 0xff;
1187 SCpnt->cmnd[6] = SCpnt->cmnd[9] = 0;
1188 SCpnt->cmnd[7] = (unsigned char) (this_count >> 8) & 0xff;
1189 SCpnt->cmnd[8] = (unsigned char) this_count & 0xff;
1191 if (unlikely(rq->cmd_flags & REQ_FUA)) {
1193 * This happens only if this drive failed
1194 * 10byte rw command with ILLEGAL_REQUEST
1195 * during operation and thus turned off
1198 scmd_printk(KERN_ERR, SCpnt,
1199 "FUA write on READ/WRITE(6) drive\n");
1203 SCpnt->cmnd[1] |= (unsigned char) ((block >> 16) & 0x1f);
1204 SCpnt->cmnd[2] = (unsigned char) ((block >> 8) & 0xff);
1205 SCpnt->cmnd[3] = (unsigned char) block & 0xff;
1206 SCpnt->cmnd[4] = (unsigned char) this_count;
1209 SCpnt->sdb.length = this_count * sdp->sector_size;
1212 * We shouldn't disconnect in the middle of a sector, so with a dumb
1213 * host adapter, it's safe to assume that we can at least transfer
1214 * this many bytes between each connect / disconnect.
1216 SCpnt->transfersize = sdp->sector_size;
1217 SCpnt->underflow = this_count << 9;
1218 SCpnt->allowed = SD_MAX_RETRIES;
1221 * This indicates that the command is ready from our end to be
1226 if (zoned_write && ret != BLKPREP_OK)
1227 sd_zbc_write_unlock_zone(SCpnt);
1232 static int sd_init_command(struct scsi_cmnd *cmd)
1234 struct request *rq = cmd->request;
1236 switch (req_op(rq)) {
1237 case REQ_OP_DISCARD:
1238 switch (scsi_disk(rq->rq_disk)->provisioning_mode) {
1240 return sd_setup_unmap_cmnd(cmd);
1242 return sd_setup_write_same16_cmnd(cmd, true);
1244 return sd_setup_write_same10_cmnd(cmd, true);
1246 return sd_setup_write_same10_cmnd(cmd, false);
1248 return BLKPREP_INVALID;
1250 case REQ_OP_WRITE_ZEROES:
1251 return sd_setup_write_zeroes_cmnd(cmd);
1252 case REQ_OP_WRITE_SAME:
1253 return sd_setup_write_same_cmnd(cmd);
1255 return sd_setup_flush_cmnd(cmd);
1258 return sd_setup_read_write_cmnd(cmd);
1259 case REQ_OP_ZONE_REPORT:
1260 return sd_zbc_setup_report_cmnd(cmd);
1261 case REQ_OP_ZONE_RESET:
1262 return sd_zbc_setup_reset_cmnd(cmd);
1268 static void sd_uninit_command(struct scsi_cmnd *SCpnt)
1270 struct request *rq = SCpnt->request;
1272 if (rq->rq_flags & RQF_SPECIAL_PAYLOAD)
1273 __free_page(rq->special_vec.bv_page);
1275 if (SCpnt->cmnd != scsi_req(rq)->cmd) {
1276 mempool_free(SCpnt->cmnd, sd_cdb_pool);
1283 * sd_open - open a scsi disk device
1284 * @bdev: Block device of the scsi disk to open
1285 * @mode: FMODE_* mask
1287 * Returns 0 if successful. Returns a negated errno value in case
1290 * Note: This can be called from a user context (e.g. fsck(1) )
1291 * or from within the kernel (e.g. as a result of a mount(1) ).
1292 * In the latter case @inode and @filp carry an abridged amount
1293 * of information as noted above.
1295 * Locking: called with bdev->bd_mutex held.
1297 static int sd_open(struct block_device *bdev, fmode_t mode)
1299 struct scsi_disk *sdkp = scsi_disk_get(bdev->bd_disk);
1300 struct scsi_device *sdev;
1306 SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp, "sd_open\n"));
1308 sdev = sdkp->device;
1311 * If the device is in error recovery, wait until it is done.
1312 * If the device is offline, then disallow any access to it.
1315 if (!scsi_block_when_processing_errors(sdev))
1318 if (sdev->removable || sdkp->write_prot)
1319 check_disk_change(bdev);
1322 * If the drive is empty, just let the open fail.
1324 retval = -ENOMEDIUM;
1325 if (sdev->removable && !sdkp->media_present && !(mode & FMODE_NDELAY))
1329 * If the device has the write protect tab set, have the open fail
1330 * if the user expects to be able to write to the thing.
1333 if (sdkp->write_prot && (mode & FMODE_WRITE))
1337 * It is possible that the disk changing stuff resulted in
1338 * the device being taken offline. If this is the case,
1339 * report this to the user, and don't pretend that the
1340 * open actually succeeded.
1343 if (!scsi_device_online(sdev))
1346 if ((atomic_inc_return(&sdkp->openers) == 1) && sdev->removable) {
1347 if (scsi_block_when_processing_errors(sdev))
1348 scsi_set_medium_removal(sdev, SCSI_REMOVAL_PREVENT);
1354 scsi_disk_put(sdkp);
1359 * sd_release - invoked when the (last) close(2) is called on this
1361 * @disk: disk to release
1362 * @mode: FMODE_* mask
1366 * Note: may block (uninterruptible) if error recovery is underway
1369 * Locking: called with bdev->bd_mutex held.
1371 static void sd_release(struct gendisk *disk, fmode_t mode)
1373 struct scsi_disk *sdkp = scsi_disk(disk);
1374 struct scsi_device *sdev = sdkp->device;
1376 SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp, "sd_release\n"));
1378 if (atomic_dec_return(&sdkp->openers) == 0 && sdev->removable) {
1379 if (scsi_block_when_processing_errors(sdev))
1380 scsi_set_medium_removal(sdev, SCSI_REMOVAL_ALLOW);
1384 * XXX and what if there are packets in flight and this close()
1385 * XXX is followed by a "rmmod sd_mod"?
1388 scsi_disk_put(sdkp);
1391 static int sd_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1393 struct scsi_disk *sdkp = scsi_disk(bdev->bd_disk);
1394 struct scsi_device *sdp = sdkp->device;
1395 struct Scsi_Host *host = sdp->host;
1396 sector_t capacity = logical_to_sectors(sdp, sdkp->capacity);
1399 /* default to most commonly used values */
1400 diskinfo[0] = 0x40; /* 1 << 6 */
1401 diskinfo[1] = 0x20; /* 1 << 5 */
1402 diskinfo[2] = capacity >> 11;
1404 /* override with calculated, extended default, or driver values */
1405 if (host->hostt->bios_param)
1406 host->hostt->bios_param(sdp, bdev, capacity, diskinfo);
1408 scsicam_bios_param(bdev, capacity, diskinfo);
1410 geo->heads = diskinfo[0];
1411 geo->sectors = diskinfo[1];
1412 geo->cylinders = diskinfo[2];
1417 * sd_ioctl - process an ioctl
1418 * @bdev: target block device
1419 * @mode: FMODE_* mask
1420 * @cmd: ioctl command number
1421 * @arg: this is third argument given to ioctl(2) system call.
1422 * Often contains a pointer.
1424 * Returns 0 if successful (some ioctls return positive numbers on
1425 * success as well). Returns a negated errno value in case of error.
1427 * Note: most ioctls are forward onto the block subsystem or further
1428 * down in the scsi subsystem.
1430 static int sd_ioctl(struct block_device *bdev, fmode_t mode,
1431 unsigned int cmd, unsigned long arg)
1433 struct gendisk *disk = bdev->bd_disk;
1434 struct scsi_disk *sdkp = scsi_disk(disk);
1435 struct scsi_device *sdp = sdkp->device;
1436 void __user *p = (void __user *)arg;
1439 SCSI_LOG_IOCTL(1, sd_printk(KERN_INFO, sdkp, "sd_ioctl: disk=%s, "
1440 "cmd=0x%x\n", disk->disk_name, cmd));
1442 error = scsi_verify_blk_ioctl(bdev, cmd);
1447 * If we are in the middle of error recovery, don't let anyone
1448 * else try and use this device. Also, if error recovery fails, it
1449 * may try and take the device offline, in which case all further
1450 * access to the device is prohibited.
1452 error = scsi_ioctl_block_when_processing_errors(sdp, cmd,
1453 (mode & FMODE_NDELAY) != 0);
1458 * Send SCSI addressing ioctls directly to mid level, send other
1459 * ioctls to block level and then onto mid level if they can't be
1463 case SCSI_IOCTL_GET_IDLUN:
1464 case SCSI_IOCTL_GET_BUS_NUMBER:
1465 error = scsi_ioctl(sdp, cmd, p);
1468 error = scsi_cmd_blk_ioctl(bdev, mode, cmd, p);
1469 if (error != -ENOTTY)
1471 error = scsi_ioctl(sdp, cmd, p);
1478 static void set_media_not_present(struct scsi_disk *sdkp)
1480 if (sdkp->media_present)
1481 sdkp->device->changed = 1;
1483 if (sdkp->device->removable) {
1484 sdkp->media_present = 0;
1489 static int media_not_present(struct scsi_disk *sdkp,
1490 struct scsi_sense_hdr *sshdr)
1492 if (!scsi_sense_valid(sshdr))
1495 /* not invoked for commands that could return deferred errors */
1496 switch (sshdr->sense_key) {
1497 case UNIT_ATTENTION:
1499 /* medium not present */
1500 if (sshdr->asc == 0x3A) {
1501 set_media_not_present(sdkp);
1509 * sd_check_events - check media events
1510 * @disk: kernel device descriptor
1511 * @clearing: disk events currently being cleared
1513 * Returns mask of DISK_EVENT_*.
1515 * Note: this function is invoked from the block subsystem.
1517 static unsigned int sd_check_events(struct gendisk *disk, unsigned int clearing)
1519 struct scsi_disk *sdkp = scsi_disk_get(disk);
1520 struct scsi_device *sdp;
1527 SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp, "sd_check_events\n"));
1530 * If the device is offline, don't send any commands - just pretend as
1531 * if the command failed. If the device ever comes back online, we
1532 * can deal with it then. It is only because of unrecoverable errors
1533 * that we would ever take a device offline in the first place.
1535 if (!scsi_device_online(sdp)) {
1536 set_media_not_present(sdkp);
1541 * Using TEST_UNIT_READY enables differentiation between drive with
1542 * no cartridge loaded - NOT READY, drive with changed cartridge -
1543 * UNIT ATTENTION, or with same cartridge - GOOD STATUS.
1545 * Drives that auto spin down. eg iomega jaz 1G, will be started
1546 * by sd_spinup_disk() from sd_revalidate_disk(), which happens whenever
1547 * sd_revalidate() is called.
1549 if (scsi_block_when_processing_errors(sdp)) {
1550 struct scsi_sense_hdr sshdr = { 0, };
1552 retval = scsi_test_unit_ready(sdp, SD_TIMEOUT, SD_MAX_RETRIES,
1555 /* failed to execute TUR, assume media not present */
1556 if (host_byte(retval)) {
1557 set_media_not_present(sdkp);
1561 if (media_not_present(sdkp, &sshdr))
1566 * For removable scsi disk we have to recognise the presence
1567 * of a disk in the drive.
1569 if (!sdkp->media_present)
1571 sdkp->media_present = 1;
1574 * sdp->changed is set under the following conditions:
1576 * Medium present state has changed in either direction.
1577 * Device has indicated UNIT_ATTENTION.
1579 retval = sdp->changed ? DISK_EVENT_MEDIA_CHANGE : 0;
1581 scsi_disk_put(sdkp);
1585 static int sd_sync_cache(struct scsi_disk *sdkp, struct scsi_sense_hdr *sshdr)
1588 struct scsi_device *sdp = sdkp->device;
1589 const int timeout = sdp->request_queue->rq_timeout
1590 * SD_FLUSH_TIMEOUT_MULTIPLIER;
1591 struct scsi_sense_hdr my_sshdr;
1593 if (!scsi_device_online(sdp))
1596 /* caller might not be interested in sense, but we need it */
1600 for (retries = 3; retries > 0; --retries) {
1601 unsigned char cmd[10] = { 0 };
1603 cmd[0] = SYNCHRONIZE_CACHE;
1605 * Leave the rest of the command zero to indicate
1608 res = scsi_execute(sdp, cmd, DMA_NONE, NULL, 0, NULL, sshdr,
1609 timeout, SD_MAX_RETRIES, 0, RQF_PM, NULL);
1615 sd_print_result(sdkp, "Synchronize Cache(10) failed", res);
1617 if (driver_byte(res) & DRIVER_SENSE)
1618 sd_print_sense_hdr(sdkp, sshdr);
1620 /* we need to evaluate the error return */
1621 if (scsi_sense_valid(sshdr) &&
1622 (sshdr->asc == 0x3a || /* medium not present */
1623 sshdr->asc == 0x20)) /* invalid command */
1624 /* this is no error here */
1627 switch (host_byte(res)) {
1628 /* ignore errors due to racing a disconnection */
1629 case DID_BAD_TARGET:
1630 case DID_NO_CONNECT:
1632 /* signal the upper layer it might try again */
1636 case DID_SOFT_ERROR:
1645 static void sd_rescan(struct device *dev)
1647 struct scsi_disk *sdkp = dev_get_drvdata(dev);
1649 revalidate_disk(sdkp->disk);
1653 #ifdef CONFIG_COMPAT
1655 * This gets directly called from VFS. When the ioctl
1656 * is not recognized we go back to the other translation paths.
1658 static int sd_compat_ioctl(struct block_device *bdev, fmode_t mode,
1659 unsigned int cmd, unsigned long arg)
1661 struct scsi_device *sdev = scsi_disk(bdev->bd_disk)->device;
1664 error = scsi_ioctl_block_when_processing_errors(sdev, cmd,
1665 (mode & FMODE_NDELAY) != 0);
1670 * Let the static ioctl translation table take care of it.
1672 if (!sdev->host->hostt->compat_ioctl)
1673 return -ENOIOCTLCMD;
1674 return sdev->host->hostt->compat_ioctl(sdev, cmd, (void __user *)arg);
1678 static char sd_pr_type(enum pr_type type)
1681 case PR_WRITE_EXCLUSIVE:
1683 case PR_EXCLUSIVE_ACCESS:
1685 case PR_WRITE_EXCLUSIVE_REG_ONLY:
1687 case PR_EXCLUSIVE_ACCESS_REG_ONLY:
1689 case PR_WRITE_EXCLUSIVE_ALL_REGS:
1691 case PR_EXCLUSIVE_ACCESS_ALL_REGS:
1698 static int sd_pr_command(struct block_device *bdev, u8 sa,
1699 u64 key, u64 sa_key, u8 type, u8 flags)
1701 struct scsi_device *sdev = scsi_disk(bdev->bd_disk)->device;
1702 struct scsi_sense_hdr sshdr;
1704 u8 cmd[16] = { 0, };
1705 u8 data[24] = { 0, };
1707 cmd[0] = PERSISTENT_RESERVE_OUT;
1710 put_unaligned_be32(sizeof(data), &cmd[5]);
1712 put_unaligned_be64(key, &data[0]);
1713 put_unaligned_be64(sa_key, &data[8]);
1716 result = scsi_execute_req(sdev, cmd, DMA_TO_DEVICE, &data, sizeof(data),
1717 &sshdr, SD_TIMEOUT, SD_MAX_RETRIES, NULL);
1719 if ((driver_byte(result) & DRIVER_SENSE) &&
1720 (scsi_sense_valid(&sshdr))) {
1721 sdev_printk(KERN_INFO, sdev, "PR command failed: %d\n", result);
1722 scsi_print_sense_hdr(sdev, NULL, &sshdr);
1728 static int sd_pr_register(struct block_device *bdev, u64 old_key, u64 new_key,
1731 if (flags & ~PR_FL_IGNORE_KEY)
1733 return sd_pr_command(bdev, (flags & PR_FL_IGNORE_KEY) ? 0x06 : 0x00,
1734 old_key, new_key, 0,
1735 (1 << 0) /* APTPL */);
1738 static int sd_pr_reserve(struct block_device *bdev, u64 key, enum pr_type type,
1743 return sd_pr_command(bdev, 0x01, key, 0, sd_pr_type(type), 0);
1746 static int sd_pr_release(struct block_device *bdev, u64 key, enum pr_type type)
1748 return sd_pr_command(bdev, 0x02, key, 0, sd_pr_type(type), 0);
1751 static int sd_pr_preempt(struct block_device *bdev, u64 old_key, u64 new_key,
1752 enum pr_type type, bool abort)
1754 return sd_pr_command(bdev, abort ? 0x05 : 0x04, old_key, new_key,
1755 sd_pr_type(type), 0);
1758 static int sd_pr_clear(struct block_device *bdev, u64 key)
1760 return sd_pr_command(bdev, 0x03, key, 0, 0, 0);
1763 static const struct pr_ops sd_pr_ops = {
1764 .pr_register = sd_pr_register,
1765 .pr_reserve = sd_pr_reserve,
1766 .pr_release = sd_pr_release,
1767 .pr_preempt = sd_pr_preempt,
1768 .pr_clear = sd_pr_clear,
1771 static const struct block_device_operations sd_fops = {
1772 .owner = THIS_MODULE,
1774 .release = sd_release,
1776 .getgeo = sd_getgeo,
1777 #ifdef CONFIG_COMPAT
1778 .compat_ioctl = sd_compat_ioctl,
1780 .check_events = sd_check_events,
1781 .revalidate_disk = sd_revalidate_disk,
1782 .unlock_native_capacity = sd_unlock_native_capacity,
1783 .pr_ops = &sd_pr_ops,
1787 * sd_eh_reset - reset error handling callback
1788 * @scmd: sd-issued command that has failed
1790 * This function is called by the SCSI midlayer before starting
1791 * SCSI EH. When counting medium access failures we have to be
1792 * careful to register it only only once per device and SCSI EH run;
1793 * there might be several timed out commands which will cause the
1794 * 'max_medium_access_timeouts' counter to trigger after the first
1795 * SCSI EH run already and set the device to offline.
1796 * So this function resets the internal counter before starting SCSI EH.
1798 static void sd_eh_reset(struct scsi_cmnd *scmd)
1800 struct scsi_disk *sdkp = scsi_disk(scmd->request->rq_disk);
1802 /* New SCSI EH run, reset gate variable */
1803 sdkp->ignore_medium_access_errors = false;
1807 * sd_eh_action - error handling callback
1808 * @scmd: sd-issued command that has failed
1809 * @eh_disp: The recovery disposition suggested by the midlayer
1811 * This function is called by the SCSI midlayer upon completion of an
1812 * error test command (currently TEST UNIT READY). The result of sending
1813 * the eh command is passed in eh_disp. We're looking for devices that
1814 * fail medium access commands but are OK with non access commands like
1815 * test unit ready (so wrongly see the device as having a successful
1818 static int sd_eh_action(struct scsi_cmnd *scmd, int eh_disp)
1820 struct scsi_disk *sdkp = scsi_disk(scmd->request->rq_disk);
1821 struct scsi_device *sdev = scmd->device;
1823 if (!scsi_device_online(sdev) ||
1824 !scsi_medium_access_command(scmd) ||
1825 host_byte(scmd->result) != DID_TIME_OUT ||
1830 * The device has timed out executing a medium access command.
1831 * However, the TEST UNIT READY command sent during error
1832 * handling completed successfully. Either the device is in the
1833 * process of recovering or has it suffered an internal failure
1834 * that prevents access to the storage medium.
1836 if (!sdkp->ignore_medium_access_errors) {
1837 sdkp->medium_access_timed_out++;
1838 sdkp->ignore_medium_access_errors = true;
1842 * If the device keeps failing read/write commands but TEST UNIT
1843 * READY always completes successfully we assume that medium
1844 * access is no longer possible and take the device offline.
1846 if (sdkp->medium_access_timed_out >= sdkp->max_medium_access_timeouts) {
1847 scmd_printk(KERN_ERR, scmd,
1848 "Medium access timeout failure. Offlining disk!\n");
1849 mutex_lock(&sdev->state_mutex);
1850 scsi_device_set_state(sdev, SDEV_OFFLINE);
1851 mutex_unlock(&sdev->state_mutex);
1859 static unsigned int sd_completed_bytes(struct scsi_cmnd *scmd)
1861 struct request *req = scmd->request;
1862 struct scsi_device *sdev = scmd->device;
1863 unsigned int transferred, good_bytes;
1864 u64 start_lba, end_lba, bad_lba;
1867 * Some commands have a payload smaller than the device logical
1868 * block size (e.g. INQUIRY on a 4K disk).
1870 if (scsi_bufflen(scmd) <= sdev->sector_size)
1873 /* Check if we have a 'bad_lba' information */
1874 if (!scsi_get_sense_info_fld(scmd->sense_buffer,
1875 SCSI_SENSE_BUFFERSIZE,
1880 * If the bad lba was reported incorrectly, we have no idea where
1883 start_lba = sectors_to_logical(sdev, blk_rq_pos(req));
1884 end_lba = start_lba + bytes_to_logical(sdev, scsi_bufflen(scmd));
1885 if (bad_lba < start_lba || bad_lba >= end_lba)
1889 * resid is optional but mostly filled in. When it's unused,
1890 * its value is zero, so we assume the whole buffer transferred
1892 transferred = scsi_bufflen(scmd) - scsi_get_resid(scmd);
1894 /* This computation should always be done in terms of the
1895 * resolution of the device's medium.
1897 good_bytes = logical_to_bytes(sdev, bad_lba - start_lba);
1899 return min(good_bytes, transferred);
1903 * sd_done - bottom half handler: called when the lower level
1904 * driver has completed (successfully or otherwise) a scsi command.
1905 * @SCpnt: mid-level's per command structure.
1907 * Note: potentially run from within an ISR. Must not block.
1909 static int sd_done(struct scsi_cmnd *SCpnt)
1911 int result = SCpnt->result;
1912 unsigned int good_bytes = result ? 0 : scsi_bufflen(SCpnt);
1913 unsigned int sector_size = SCpnt->device->sector_size;
1915 struct scsi_sense_hdr sshdr;
1916 struct scsi_disk *sdkp = scsi_disk(SCpnt->request->rq_disk);
1917 struct request *req = SCpnt->request;
1918 int sense_valid = 0;
1919 int sense_deferred = 0;
1921 switch (req_op(req)) {
1922 case REQ_OP_DISCARD:
1923 case REQ_OP_WRITE_ZEROES:
1924 case REQ_OP_WRITE_SAME:
1925 case REQ_OP_ZONE_RESET:
1927 good_bytes = blk_rq_bytes(req);
1928 scsi_set_resid(SCpnt, 0);
1931 scsi_set_resid(SCpnt, blk_rq_bytes(req));
1934 case REQ_OP_ZONE_REPORT:
1936 good_bytes = scsi_bufflen(SCpnt)
1937 - scsi_get_resid(SCpnt);
1938 scsi_set_resid(SCpnt, 0);
1941 scsi_set_resid(SCpnt, blk_rq_bytes(req));
1946 * In case of bogus fw or device, we could end up having
1947 * an unaligned partial completion. Check this here and force
1950 resid = scsi_get_resid(SCpnt);
1951 if (resid & (sector_size - 1)) {
1952 sd_printk(KERN_INFO, sdkp,
1953 "Unaligned partial completion (resid=%u, sector_sz=%u)\n",
1954 resid, sector_size);
1955 resid = min(scsi_bufflen(SCpnt),
1956 round_up(resid, sector_size));
1957 scsi_set_resid(SCpnt, resid);
1962 sense_valid = scsi_command_normalize_sense(SCpnt, &sshdr);
1964 sense_deferred = scsi_sense_is_deferred(&sshdr);
1966 sdkp->medium_access_timed_out = 0;
1968 if (driver_byte(result) != DRIVER_SENSE &&
1969 (!sense_valid || sense_deferred))
1972 switch (sshdr.sense_key) {
1973 case HARDWARE_ERROR:
1975 good_bytes = sd_completed_bytes(SCpnt);
1977 case RECOVERED_ERROR:
1978 good_bytes = scsi_bufflen(SCpnt);
1981 /* This indicates a false check condition, so ignore it. An
1982 * unknown amount of data was transferred so treat it as an
1986 memset(SCpnt->sense_buffer, 0, SCSI_SENSE_BUFFERSIZE);
1988 case ABORTED_COMMAND:
1989 if (sshdr.asc == 0x10) /* DIF: Target detected corruption */
1990 good_bytes = sd_completed_bytes(SCpnt);
1992 case ILLEGAL_REQUEST:
1993 switch (sshdr.asc) {
1994 case 0x10: /* DIX: Host detected corruption */
1995 good_bytes = sd_completed_bytes(SCpnt);
1997 case 0x20: /* INVALID COMMAND OPCODE */
1998 case 0x24: /* INVALID FIELD IN CDB */
1999 switch (SCpnt->cmnd[0]) {
2001 sd_config_discard(sdkp, SD_LBP_DISABLE);
2005 if (SCpnt->cmnd[1] & 8) { /* UNMAP */
2006 sd_config_discard(sdkp, SD_LBP_DISABLE);
2008 sdkp->device->no_write_same = 1;
2009 sd_config_write_same(sdkp);
2010 req->__data_len = blk_rq_bytes(req);
2011 req->rq_flags |= RQF_QUIET;
2022 if (sd_is_zoned(sdkp))
2023 sd_zbc_complete(SCpnt, good_bytes, &sshdr);
2025 SCSI_LOG_HLCOMPLETE(1, scmd_printk(KERN_INFO, SCpnt,
2026 "sd_done: completed %d of %d bytes\n",
2027 good_bytes, scsi_bufflen(SCpnt)));
2029 if (rq_data_dir(SCpnt->request) == READ && scsi_prot_sg_count(SCpnt))
2030 sd_dif_complete(SCpnt, good_bytes);
2036 * spinup disk - called only in sd_revalidate_disk()
2039 sd_spinup_disk(struct scsi_disk *sdkp)
2041 unsigned char cmd[10];
2042 unsigned long spintime_expire = 0;
2043 int retries, spintime;
2044 unsigned int the_result;
2045 struct scsi_sense_hdr sshdr;
2046 int sense_valid = 0;
2050 /* Spin up drives, as required. Only do this at boot time */
2051 /* Spinup needs to be done for module loads too. */
2056 cmd[0] = TEST_UNIT_READY;
2057 memset((void *) &cmd[1], 0, 9);
2059 the_result = scsi_execute_req(sdkp->device, cmd,
2062 SD_MAX_RETRIES, NULL);
2065 * If the drive has indicated to us that it
2066 * doesn't have any media in it, don't bother
2067 * with any more polling.
2069 if (media_not_present(sdkp, &sshdr))
2073 sense_valid = scsi_sense_valid(&sshdr);
2075 } while (retries < 3 &&
2076 (!scsi_status_is_good(the_result) ||
2077 ((driver_byte(the_result) & DRIVER_SENSE) &&
2078 sense_valid && sshdr.sense_key == UNIT_ATTENTION)));
2080 if ((driver_byte(the_result) & DRIVER_SENSE) == 0) {
2081 /* no sense, TUR either succeeded or failed
2082 * with a status error */
2083 if(!spintime && !scsi_status_is_good(the_result)) {
2084 sd_print_result(sdkp, "Test Unit Ready failed",
2091 * The device does not want the automatic start to be issued.
2093 if (sdkp->device->no_start_on_add)
2096 if (sense_valid && sshdr.sense_key == NOT_READY) {
2097 if (sshdr.asc == 4 && sshdr.ascq == 3)
2098 break; /* manual intervention required */
2099 if (sshdr.asc == 4 && sshdr.ascq == 0xb)
2100 break; /* standby */
2101 if (sshdr.asc == 4 && sshdr.ascq == 0xc)
2102 break; /* unavailable */
2104 * Issue command to spin up drive when not ready
2107 sd_printk(KERN_NOTICE, sdkp, "Spinning up disk...");
2108 cmd[0] = START_STOP;
2109 cmd[1] = 1; /* Return immediately */
2110 memset((void *) &cmd[2], 0, 8);
2111 cmd[4] = 1; /* Start spin cycle */
2112 if (sdkp->device->start_stop_pwr_cond)
2114 scsi_execute_req(sdkp->device, cmd, DMA_NONE,
2116 SD_TIMEOUT, SD_MAX_RETRIES,
2118 spintime_expire = jiffies + 100 * HZ;
2121 /* Wait 1 second for next try */
2126 * Wait for USB flash devices with slow firmware.
2127 * Yes, this sense key/ASC combination shouldn't
2128 * occur here. It's characteristic of these devices.
2130 } else if (sense_valid &&
2131 sshdr.sense_key == UNIT_ATTENTION &&
2132 sshdr.asc == 0x28) {
2134 spintime_expire = jiffies + 5 * HZ;
2137 /* Wait 1 second for next try */
2140 /* we don't understand the sense code, so it's
2141 * probably pointless to loop */
2143 sd_printk(KERN_NOTICE, sdkp, "Unit Not Ready\n");
2144 sd_print_sense_hdr(sdkp, &sshdr);
2149 } while (spintime && time_before_eq(jiffies, spintime_expire));
2152 if (scsi_status_is_good(the_result))
2155 printk("not responding...\n");
2160 * Determine whether disk supports Data Integrity Field.
2162 static int sd_read_protection_type(struct scsi_disk *sdkp, unsigned char *buffer)
2164 struct scsi_device *sdp = sdkp->device;
2168 if (scsi_device_protection(sdp) == 0 || (buffer[12] & 1) == 0)
2171 type = ((buffer[12] >> 1) & 7) + 1; /* P_TYPE 0 = Type 1 */
2173 if (type > T10_PI_TYPE3_PROTECTION)
2175 else if (scsi_host_dif_capable(sdp->host, type))
2178 if (sdkp->first_scan || type != sdkp->protection_type)
2181 sd_printk(KERN_ERR, sdkp, "formatted with unsupported" \
2182 " protection type %u. Disabling disk!\n",
2186 sd_printk(KERN_NOTICE, sdkp,
2187 "Enabling DIF Type %u protection\n", type);
2190 sd_printk(KERN_NOTICE, sdkp,
2191 "Disabling DIF Type %u protection\n", type);
2195 sdkp->protection_type = type;
2200 static void read_capacity_error(struct scsi_disk *sdkp, struct scsi_device *sdp,
2201 struct scsi_sense_hdr *sshdr, int sense_valid,
2204 if (driver_byte(the_result) & DRIVER_SENSE)
2205 sd_print_sense_hdr(sdkp, sshdr);
2207 sd_printk(KERN_NOTICE, sdkp, "Sense not available.\n");
2210 * Set dirty bit for removable devices if not ready -
2211 * sometimes drives will not report this properly.
2213 if (sdp->removable &&
2214 sense_valid && sshdr->sense_key == NOT_READY)
2215 set_media_not_present(sdkp);
2218 * We used to set media_present to 0 here to indicate no media
2219 * in the drive, but some drives fail read capacity even with
2220 * media present, so we can't do that.
2222 sdkp->capacity = 0; /* unknown mapped to zero - as usual */
2226 #if RC16_LEN > SD_BUF_SIZE
2227 #error RC16_LEN must not be more than SD_BUF_SIZE
2230 #define READ_CAPACITY_RETRIES_ON_RESET 10
2233 * Ensure that we don't overflow sector_t when CONFIG_LBDAF is not set
2234 * and the reported logical block size is bigger than 512 bytes. Note
2235 * that last_sector is a u64 and therefore logical_to_sectors() is not
2238 static bool sd_addressable_capacity(u64 lba, unsigned int sector_size)
2240 u64 last_sector = (lba + 1ULL) << (ilog2(sector_size) - 9);
2242 if (sizeof(sector_t) == 4 && last_sector > U32_MAX)
2248 static int read_capacity_16(struct scsi_disk *sdkp, struct scsi_device *sdp,
2249 unsigned char *buffer)
2251 unsigned char cmd[16];
2252 struct scsi_sense_hdr sshdr;
2253 int sense_valid = 0;
2255 int retries = 3, reset_retries = READ_CAPACITY_RETRIES_ON_RESET;
2256 unsigned int alignment;
2257 unsigned long long lba;
2258 unsigned sector_size;
2260 if (sdp->no_read_capacity_16)
2265 cmd[0] = SERVICE_ACTION_IN_16;
2266 cmd[1] = SAI_READ_CAPACITY_16;
2268 memset(buffer, 0, RC16_LEN);
2270 the_result = scsi_execute_req(sdp, cmd, DMA_FROM_DEVICE,
2271 buffer, RC16_LEN, &sshdr,
2272 SD_TIMEOUT, SD_MAX_RETRIES, NULL);
2274 if (media_not_present(sdkp, &sshdr))
2278 sense_valid = scsi_sense_valid(&sshdr);
2280 sshdr.sense_key == ILLEGAL_REQUEST &&
2281 (sshdr.asc == 0x20 || sshdr.asc == 0x24) &&
2283 /* Invalid Command Operation Code or
2284 * Invalid Field in CDB, just retry
2285 * silently with RC10 */
2288 sshdr.sense_key == UNIT_ATTENTION &&
2289 sshdr.asc == 0x29 && sshdr.ascq == 0x00)
2290 /* Device reset might occur several times,
2291 * give it one more chance */
2292 if (--reset_retries > 0)
2297 } while (the_result && retries);
2300 sd_print_result(sdkp, "Read Capacity(16) failed", the_result);
2301 read_capacity_error(sdkp, sdp, &sshdr, sense_valid, the_result);
2305 sector_size = get_unaligned_be32(&buffer[8]);
2306 lba = get_unaligned_be64(&buffer[0]);
2308 if (sd_read_protection_type(sdkp, buffer) < 0) {
2313 if (!sd_addressable_capacity(lba, sector_size)) {
2314 sd_printk(KERN_ERR, sdkp, "Too big for this kernel. Use a "
2315 "kernel compiled with support for large block "
2321 /* Logical blocks per physical block exponent */
2322 sdkp->physical_block_size = (1 << (buffer[13] & 0xf)) * sector_size;
2325 sdkp->rc_basis = (buffer[12] >> 4) & 0x3;
2327 /* Lowest aligned logical block */
2328 alignment = ((buffer[14] & 0x3f) << 8 | buffer[15]) * sector_size;
2329 blk_queue_alignment_offset(sdp->request_queue, alignment);
2330 if (alignment && sdkp->first_scan)
2331 sd_printk(KERN_NOTICE, sdkp,
2332 "physical block alignment offset: %u\n", alignment);
2334 if (buffer[14] & 0x80) { /* LBPME */
2337 if (buffer[14] & 0x40) /* LBPRZ */
2340 sd_config_discard(sdkp, SD_LBP_WS16);
2343 sdkp->capacity = lba + 1;
2347 static int read_capacity_10(struct scsi_disk *sdkp, struct scsi_device *sdp,
2348 unsigned char *buffer)
2350 unsigned char cmd[16];
2351 struct scsi_sense_hdr sshdr;
2352 int sense_valid = 0;
2354 int retries = 3, reset_retries = READ_CAPACITY_RETRIES_ON_RESET;
2356 unsigned sector_size;
2359 cmd[0] = READ_CAPACITY;
2360 memset(&cmd[1], 0, 9);
2361 memset(buffer, 0, 8);
2363 the_result = scsi_execute_req(sdp, cmd, DMA_FROM_DEVICE,
2365 SD_TIMEOUT, SD_MAX_RETRIES, NULL);
2367 if (media_not_present(sdkp, &sshdr))
2371 sense_valid = scsi_sense_valid(&sshdr);
2373 sshdr.sense_key == UNIT_ATTENTION &&
2374 sshdr.asc == 0x29 && sshdr.ascq == 0x00)
2375 /* Device reset might occur several times,
2376 * give it one more chance */
2377 if (--reset_retries > 0)
2382 } while (the_result && retries);
2385 sd_print_result(sdkp, "Read Capacity(10) failed", the_result);
2386 read_capacity_error(sdkp, sdp, &sshdr, sense_valid, the_result);
2390 sector_size = get_unaligned_be32(&buffer[4]);
2391 lba = get_unaligned_be32(&buffer[0]);
2393 if (sdp->no_read_capacity_16 && (lba == 0xffffffff)) {
2394 /* Some buggy (usb cardreader) devices return an lba of
2395 0xffffffff when the want to report a size of 0 (with
2396 which they really mean no media is present) */
2398 sdkp->physical_block_size = sector_size;
2402 if (!sd_addressable_capacity(lba, sector_size)) {
2403 sd_printk(KERN_ERR, sdkp, "Too big for this kernel. Use a "
2404 "kernel compiled with support for large block "
2410 sdkp->capacity = lba + 1;
2411 sdkp->physical_block_size = sector_size;
2415 static int sd_try_rc16_first(struct scsi_device *sdp)
2417 if (sdp->host->max_cmd_len < 16)
2419 if (sdp->try_rc_10_first)
2421 if (sdp->scsi_level > SCSI_SPC_2)
2423 if (scsi_device_protection(sdp))
2429 * read disk capacity
2432 sd_read_capacity(struct scsi_disk *sdkp, unsigned char *buffer)
2435 struct scsi_device *sdp = sdkp->device;
2437 if (sd_try_rc16_first(sdp)) {
2438 sector_size = read_capacity_16(sdkp, sdp, buffer);
2439 if (sector_size == -EOVERFLOW)
2441 if (sector_size == -ENODEV)
2443 if (sector_size < 0)
2444 sector_size = read_capacity_10(sdkp, sdp, buffer);
2445 if (sector_size < 0)
2448 sector_size = read_capacity_10(sdkp, sdp, buffer);
2449 if (sector_size == -EOVERFLOW)
2451 if (sector_size < 0)
2453 if ((sizeof(sdkp->capacity) > 4) &&
2454 (sdkp->capacity > 0xffffffffULL)) {
2455 int old_sector_size = sector_size;
2456 sd_printk(KERN_NOTICE, sdkp, "Very big device. "
2457 "Trying to use READ CAPACITY(16).\n");
2458 sector_size = read_capacity_16(sdkp, sdp, buffer);
2459 if (sector_size < 0) {
2460 sd_printk(KERN_NOTICE, sdkp,
2461 "Using 0xffffffff as device size\n");
2462 sdkp->capacity = 1 + (sector_t) 0xffffffff;
2463 sector_size = old_sector_size;
2469 /* Some devices are known to return the total number of blocks,
2470 * not the highest block number. Some devices have versions
2471 * which do this and others which do not. Some devices we might
2472 * suspect of doing this but we don't know for certain.
2474 * If we know the reported capacity is wrong, decrement it. If
2475 * we can only guess, then assume the number of blocks is even
2476 * (usually true but not always) and err on the side of lowering
2479 if (sdp->fix_capacity ||
2480 (sdp->guess_capacity && (sdkp->capacity & 0x01))) {
2481 sd_printk(KERN_INFO, sdkp, "Adjusting the sector count "
2482 "from its reported value: %llu\n",
2483 (unsigned long long) sdkp->capacity);
2488 if (sector_size == 0) {
2490 sd_printk(KERN_NOTICE, sdkp, "Sector size 0 reported, "
2494 if (sector_size != 512 &&
2495 sector_size != 1024 &&
2496 sector_size != 2048 &&
2497 sector_size != 4096) {
2498 sd_printk(KERN_NOTICE, sdkp, "Unsupported sector size %d.\n",
2501 * The user might want to re-format the drive with
2502 * a supported sectorsize. Once this happens, it
2503 * would be relatively trivial to set the thing up.
2504 * For this reason, we leave the thing in the table.
2508 * set a bogus sector size so the normal read/write
2509 * logic in the block layer will eventually refuse any
2510 * request on this device without tripping over power
2511 * of two sector size assumptions
2515 blk_queue_logical_block_size(sdp->request_queue, sector_size);
2516 blk_queue_physical_block_size(sdp->request_queue,
2517 sdkp->physical_block_size);
2518 sdkp->device->sector_size = sector_size;
2520 if (sdkp->capacity > 0xffffffff)
2521 sdp->use_16_for_rw = 1;
2526 * Print disk capacity
2529 sd_print_capacity(struct scsi_disk *sdkp,
2530 sector_t old_capacity)
2532 int sector_size = sdkp->device->sector_size;
2533 char cap_str_2[10], cap_str_10[10];
2535 string_get_size(sdkp->capacity, sector_size,
2536 STRING_UNITS_2, cap_str_2, sizeof(cap_str_2));
2537 string_get_size(sdkp->capacity, sector_size,
2538 STRING_UNITS_10, cap_str_10,
2539 sizeof(cap_str_10));
2541 if (sdkp->first_scan || old_capacity != sdkp->capacity) {
2542 sd_printk(KERN_NOTICE, sdkp,
2543 "%llu %d-byte logical blocks: (%s/%s)\n",
2544 (unsigned long long)sdkp->capacity,
2545 sector_size, cap_str_10, cap_str_2);
2547 if (sdkp->physical_block_size != sector_size)
2548 sd_printk(KERN_NOTICE, sdkp,
2549 "%u-byte physical blocks\n",
2550 sdkp->physical_block_size);
2552 sd_zbc_print_zones(sdkp);
2556 /* called with buffer of length 512 */
2558 sd_do_mode_sense(struct scsi_device *sdp, int dbd, int modepage,
2559 unsigned char *buffer, int len, struct scsi_mode_data *data,
2560 struct scsi_sense_hdr *sshdr)
2562 return scsi_mode_sense(sdp, dbd, modepage, buffer, len,
2563 SD_TIMEOUT, SD_MAX_RETRIES, data,
2568 * read write protect setting, if possible - called only in sd_revalidate_disk()
2569 * called with buffer of length SD_BUF_SIZE
2572 sd_read_write_protect_flag(struct scsi_disk *sdkp, unsigned char *buffer)
2575 struct scsi_device *sdp = sdkp->device;
2576 struct scsi_mode_data data;
2577 int old_wp = sdkp->write_prot;
2579 set_disk_ro(sdkp->disk, 0);
2580 if (sdp->skip_ms_page_3f) {
2581 sd_first_printk(KERN_NOTICE, sdkp, "Assuming Write Enabled\n");
2585 if (sdp->use_192_bytes_for_3f) {
2586 res = sd_do_mode_sense(sdp, 0, 0x3F, buffer, 192, &data, NULL);
2589 * First attempt: ask for all pages (0x3F), but only 4 bytes.
2590 * We have to start carefully: some devices hang if we ask
2591 * for more than is available.
2593 res = sd_do_mode_sense(sdp, 0, 0x3F, buffer, 4, &data, NULL);
2596 * Second attempt: ask for page 0 When only page 0 is
2597 * implemented, a request for page 3F may return Sense Key
2598 * 5: Illegal Request, Sense Code 24: Invalid field in
2601 if (!scsi_status_is_good(res))
2602 res = sd_do_mode_sense(sdp, 0, 0, buffer, 4, &data, NULL);
2605 * Third attempt: ask 255 bytes, as we did earlier.
2607 if (!scsi_status_is_good(res))
2608 res = sd_do_mode_sense(sdp, 0, 0x3F, buffer, 255,
2612 if (!scsi_status_is_good(res)) {
2613 sd_first_printk(KERN_WARNING, sdkp,
2614 "Test WP failed, assume Write Enabled\n");
2616 sdkp->write_prot = ((data.device_specific & 0x80) != 0);
2617 set_disk_ro(sdkp->disk, sdkp->write_prot);
2618 if (sdkp->first_scan || old_wp != sdkp->write_prot) {
2619 sd_printk(KERN_NOTICE, sdkp, "Write Protect is %s\n",
2620 sdkp->write_prot ? "on" : "off");
2621 sd_printk(KERN_DEBUG, sdkp, "Mode Sense: %4ph\n", buffer);
2627 * sd_read_cache_type - called only from sd_revalidate_disk()
2628 * called with buffer of length SD_BUF_SIZE
2631 sd_read_cache_type(struct scsi_disk *sdkp, unsigned char *buffer)
2634 struct scsi_device *sdp = sdkp->device;
2639 struct scsi_mode_data data;
2640 struct scsi_sense_hdr sshdr;
2641 int old_wce = sdkp->WCE;
2642 int old_rcd = sdkp->RCD;
2643 int old_dpofua = sdkp->DPOFUA;
2646 if (sdkp->cache_override)
2650 if (sdp->skip_ms_page_8) {
2651 if (sdp->type == TYPE_RBC)
2654 if (sdp->skip_ms_page_3f)
2657 if (sdp->use_192_bytes_for_3f)
2661 } else if (sdp->type == TYPE_RBC) {
2669 /* cautiously ask */
2670 res = sd_do_mode_sense(sdp, dbd, modepage, buffer, first_len,
2673 if (!scsi_status_is_good(res))
2676 if (!data.header_length) {
2679 sd_first_printk(KERN_ERR, sdkp,
2680 "Missing header in MODE_SENSE response\n");
2683 /* that went OK, now ask for the proper length */
2687 * We're only interested in the first three bytes, actually.
2688 * But the data cache page is defined for the first 20.
2692 else if (len > SD_BUF_SIZE) {
2693 sd_first_printk(KERN_NOTICE, sdkp, "Truncating mode parameter "
2694 "data from %d to %d bytes\n", len, SD_BUF_SIZE);
2697 if (modepage == 0x3F && sdp->use_192_bytes_for_3f)
2701 if (len > first_len)
2702 res = sd_do_mode_sense(sdp, dbd, modepage, buffer, len,
2705 if (scsi_status_is_good(res)) {
2706 int offset = data.header_length + data.block_descriptor_length;
2708 while (offset < len) {
2709 u8 page_code = buffer[offset] & 0x3F;
2710 u8 spf = buffer[offset] & 0x40;
2712 if (page_code == 8 || page_code == 6) {
2713 /* We're interested only in the first 3 bytes.
2715 if (len - offset <= 2) {
2716 sd_first_printk(KERN_ERR, sdkp,
2717 "Incomplete mode parameter "
2721 modepage = page_code;
2725 /* Go to the next page */
2726 if (spf && len - offset > 3)
2727 offset += 4 + (buffer[offset+2] << 8) +
2729 else if (!spf && len - offset > 1)
2730 offset += 2 + buffer[offset+1];
2732 sd_first_printk(KERN_ERR, sdkp,
2734 "parameter data\n");
2740 sd_first_printk(KERN_ERR, sdkp, "No Caching mode page found\n");
2744 if (modepage == 8) {
2745 sdkp->WCE = ((buffer[offset + 2] & 0x04) != 0);
2746 sdkp->RCD = ((buffer[offset + 2] & 0x01) != 0);
2748 sdkp->WCE = ((buffer[offset + 2] & 0x01) == 0);
2752 sdkp->DPOFUA = (data.device_specific & 0x10) != 0;
2753 if (sdp->broken_fua) {
2754 sd_first_printk(KERN_NOTICE, sdkp, "Disabling FUA\n");
2756 } else if (sdkp->DPOFUA && !sdkp->device->use_10_for_rw &&
2757 !sdkp->device->use_16_for_rw) {
2758 sd_first_printk(KERN_NOTICE, sdkp,
2759 "Uses READ/WRITE(6), disabling FUA\n");
2763 /* No cache flush allowed for write protected devices */
2764 if (sdkp->WCE && sdkp->write_prot)
2767 if (sdkp->first_scan || old_wce != sdkp->WCE ||
2768 old_rcd != sdkp->RCD || old_dpofua != sdkp->DPOFUA)
2769 sd_printk(KERN_NOTICE, sdkp,
2770 "Write cache: %s, read cache: %s, %s\n",
2771 sdkp->WCE ? "enabled" : "disabled",
2772 sdkp->RCD ? "disabled" : "enabled",
2773 sdkp->DPOFUA ? "supports DPO and FUA"
2774 : "doesn't support DPO or FUA");
2780 if (scsi_sense_valid(&sshdr) &&
2781 sshdr.sense_key == ILLEGAL_REQUEST &&
2782 sshdr.asc == 0x24 && sshdr.ascq == 0x0)
2783 /* Invalid field in CDB */
2784 sd_first_printk(KERN_NOTICE, sdkp, "Cache data unavailable\n");
2786 sd_first_printk(KERN_ERR, sdkp,
2787 "Asking for cache data failed\n");
2790 if (sdp->wce_default_on) {
2791 sd_first_printk(KERN_NOTICE, sdkp,
2792 "Assuming drive cache: write back\n");
2795 sd_first_printk(KERN_ERR, sdkp,
2796 "Assuming drive cache: write through\n");
2804 * The ATO bit indicates whether the DIF application tag is available
2805 * for use by the operating system.
2807 static void sd_read_app_tag_own(struct scsi_disk *sdkp, unsigned char *buffer)
2810 struct scsi_device *sdp = sdkp->device;
2811 struct scsi_mode_data data;
2812 struct scsi_sense_hdr sshdr;
2814 if (sdp->type != TYPE_DISK && sdp->type != TYPE_ZBC)
2817 if (sdkp->protection_type == 0)
2820 res = scsi_mode_sense(sdp, 1, 0x0a, buffer, 36, SD_TIMEOUT,
2821 SD_MAX_RETRIES, &data, &sshdr);
2823 if (!scsi_status_is_good(res) || !data.header_length ||
2825 sd_first_printk(KERN_WARNING, sdkp,
2826 "getting Control mode page failed, assume no ATO\n");
2828 if (scsi_sense_valid(&sshdr))
2829 sd_print_sense_hdr(sdkp, &sshdr);
2834 offset = data.header_length + data.block_descriptor_length;
2836 if ((buffer[offset] & 0x3f) != 0x0a) {
2837 sd_first_printk(KERN_ERR, sdkp, "ATO Got wrong page\n");
2841 if ((buffer[offset + 5] & 0x80) == 0)
2850 * sd_read_block_limits - Query disk device for preferred I/O sizes.
2851 * @sdkp: disk to query
2853 static void sd_read_block_limits(struct scsi_disk *sdkp)
2855 unsigned int sector_sz = sdkp->device->sector_size;
2856 const int vpd_len = 64;
2857 unsigned char *buffer = kmalloc(vpd_len, GFP_KERNEL);
2860 /* Block Limits VPD */
2861 scsi_get_vpd_page(sdkp->device, 0xb0, buffer, vpd_len))
2864 blk_queue_io_min(sdkp->disk->queue,
2865 get_unaligned_be16(&buffer[6]) * sector_sz);
2867 sdkp->max_xfer_blocks = get_unaligned_be32(&buffer[8]);
2868 sdkp->opt_xfer_blocks = get_unaligned_be32(&buffer[12]);
2870 if (buffer[3] == 0x3c) {
2871 unsigned int lba_count, desc_count;
2873 sdkp->max_ws_blocks = (u32)get_unaligned_be64(&buffer[36]);
2878 lba_count = get_unaligned_be32(&buffer[20]);
2879 desc_count = get_unaligned_be32(&buffer[24]);
2881 if (lba_count && desc_count)
2882 sdkp->max_unmap_blocks = lba_count;
2884 sdkp->unmap_granularity = get_unaligned_be32(&buffer[28]);
2886 if (buffer[32] & 0x80)
2887 sdkp->unmap_alignment =
2888 get_unaligned_be32(&buffer[32]) & ~(1 << 31);
2890 if (!sdkp->lbpvpd) { /* LBP VPD page not provided */
2892 if (sdkp->max_unmap_blocks)
2893 sd_config_discard(sdkp, SD_LBP_UNMAP);
2895 sd_config_discard(sdkp, SD_LBP_WS16);
2897 } else { /* LBP VPD page tells us what to use */
2898 if (sdkp->lbpu && sdkp->max_unmap_blocks)
2899 sd_config_discard(sdkp, SD_LBP_UNMAP);
2900 else if (sdkp->lbpws)
2901 sd_config_discard(sdkp, SD_LBP_WS16);
2902 else if (sdkp->lbpws10)
2903 sd_config_discard(sdkp, SD_LBP_WS10);
2904 else if (sdkp->lbpu && sdkp->max_unmap_blocks)
2905 sd_config_discard(sdkp, SD_LBP_UNMAP);
2907 sd_config_discard(sdkp, SD_LBP_DISABLE);
2916 * sd_read_block_characteristics - Query block dev. characteristics
2917 * @sdkp: disk to query
2919 static void sd_read_block_characteristics(struct scsi_disk *sdkp)
2921 struct request_queue *q = sdkp->disk->queue;
2922 unsigned char *buffer;
2924 const int vpd_len = 64;
2926 buffer = kmalloc(vpd_len, GFP_KERNEL);
2929 /* Block Device Characteristics VPD */
2930 scsi_get_vpd_page(sdkp->device, 0xb1, buffer, vpd_len))
2933 rot = get_unaligned_be16(&buffer[4]);
2936 queue_flag_set_unlocked(QUEUE_FLAG_NONROT, q);
2937 queue_flag_clear_unlocked(QUEUE_FLAG_ADD_RANDOM, q);
2940 if (sdkp->device->type == TYPE_ZBC) {
2942 q->limits.zoned = BLK_ZONED_HM;
2944 sdkp->zoned = (buffer[8] >> 4) & 3;
2945 if (sdkp->zoned == 1)
2947 q->limits.zoned = BLK_ZONED_HA;
2950 * Treat drive-managed devices as
2951 * regular block devices.
2953 q->limits.zoned = BLK_ZONED_NONE;
2955 if (blk_queue_is_zoned(q) && sdkp->first_scan)
2956 sd_printk(KERN_NOTICE, sdkp, "Host-%s zoned block device\n",
2957 q->limits.zoned == BLK_ZONED_HM ? "managed" : "aware");
2964 * sd_read_block_provisioning - Query provisioning VPD page
2965 * @sdkp: disk to query
2967 static void sd_read_block_provisioning(struct scsi_disk *sdkp)
2969 unsigned char *buffer;
2970 const int vpd_len = 8;
2972 if (sdkp->lbpme == 0)
2975 buffer = kmalloc(vpd_len, GFP_KERNEL);
2977 if (!buffer || scsi_get_vpd_page(sdkp->device, 0xb2, buffer, vpd_len))
2981 sdkp->lbpu = (buffer[5] >> 7) & 1; /* UNMAP */
2982 sdkp->lbpws = (buffer[5] >> 6) & 1; /* WRITE SAME(16) with UNMAP */
2983 sdkp->lbpws10 = (buffer[5] >> 5) & 1; /* WRITE SAME(10) with UNMAP */
2989 static void sd_read_write_same(struct scsi_disk *sdkp, unsigned char *buffer)
2991 struct scsi_device *sdev = sdkp->device;
2993 if (sdev->host->no_write_same) {
2994 sdev->no_write_same = 1;
2999 if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE, INQUIRY) < 0) {
3000 /* too large values might cause issues with arcmsr */
3001 int vpd_buf_len = 64;
3003 sdev->no_report_opcodes = 1;
3005 /* Disable WRITE SAME if REPORT SUPPORTED OPERATION
3006 * CODES is unsupported and the device has an ATA
3007 * Information VPD page (SAT).
3009 if (!scsi_get_vpd_page(sdev, 0x89, buffer, vpd_buf_len))
3010 sdev->no_write_same = 1;
3013 if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE, WRITE_SAME_16) == 1)
3016 if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE, WRITE_SAME) == 1)
3021 * sd_revalidate_disk - called the first time a new disk is seen,
3022 * performs disk spin up, read_capacity, etc.
3023 * @disk: struct gendisk we care about
3025 static int sd_revalidate_disk(struct gendisk *disk)
3027 struct scsi_disk *sdkp = scsi_disk(disk);
3028 struct scsi_device *sdp = sdkp->device;
3029 struct request_queue *q = sdkp->disk->queue;
3030 sector_t old_capacity = sdkp->capacity;
3031 unsigned char *buffer;
3032 unsigned int dev_max, rw_max;
3034 SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp,
3035 "sd_revalidate_disk\n"));
3038 * If the device is offline, don't try and read capacity or any
3039 * of the other niceties.
3041 if (!scsi_device_online(sdp))
3044 buffer = kmalloc(SD_BUF_SIZE, GFP_KERNEL);
3046 sd_printk(KERN_WARNING, sdkp, "sd_revalidate_disk: Memory "
3047 "allocation failure.\n");
3051 sd_spinup_disk(sdkp);
3054 * Without media there is no reason to ask; moreover, some devices
3055 * react badly if we do.
3057 if (sdkp->media_present) {
3058 sd_read_capacity(sdkp, buffer);
3060 if (scsi_device_supports_vpd(sdp)) {
3061 sd_read_block_provisioning(sdkp);
3062 sd_read_block_limits(sdkp);
3063 sd_read_block_characteristics(sdkp);
3064 sd_zbc_read_zones(sdkp, buffer);
3067 sd_print_capacity(sdkp, old_capacity);
3069 sd_read_write_protect_flag(sdkp, buffer);
3070 sd_read_cache_type(sdkp, buffer);
3071 sd_read_app_tag_own(sdkp, buffer);
3072 sd_read_write_same(sdkp, buffer);
3075 sdkp->first_scan = 0;
3078 * We now have all cache related info, determine how we deal
3079 * with flush requests.
3081 sd_set_flush_flag(sdkp);
3083 /* Initial block count limit based on CDB TRANSFER LENGTH field size. */
3084 dev_max = sdp->use_16_for_rw ? SD_MAX_XFER_BLOCKS : SD_DEF_XFER_BLOCKS;
3086 /* Some devices report a maximum block count for READ/WRITE requests. */
3087 dev_max = min_not_zero(dev_max, sdkp->max_xfer_blocks);
3088 q->limits.max_dev_sectors = logical_to_sectors(sdp, dev_max);
3091 * Use the device's preferred I/O size for reads and writes
3092 * unless the reported value is unreasonably small, large, or
3095 if (sdkp->opt_xfer_blocks &&
3096 sdkp->opt_xfer_blocks <= dev_max &&
3097 sdkp->opt_xfer_blocks <= SD_DEF_XFER_BLOCKS &&
3098 logical_to_bytes(sdp, sdkp->opt_xfer_blocks) >= PAGE_SIZE) {
3099 q->limits.io_opt = logical_to_bytes(sdp, sdkp->opt_xfer_blocks);
3100 rw_max = logical_to_sectors(sdp, sdkp->opt_xfer_blocks);
3102 rw_max = min_not_zero(logical_to_sectors(sdp, dev_max),
3103 (sector_t)BLK_DEF_MAX_SECTORS);
3105 /* Combine with controller limits */
3106 q->limits.max_sectors = min(rw_max, queue_max_hw_sectors(q));
3108 set_capacity(disk, logical_to_sectors(sdp, sdkp->capacity));
3109 sd_config_write_same(sdkp);
3117 * sd_unlock_native_capacity - unlock native capacity
3118 * @disk: struct gendisk to set capacity for
3120 * Block layer calls this function if it detects that partitions
3121 * on @disk reach beyond the end of the device. If the SCSI host
3122 * implements ->unlock_native_capacity() method, it's invoked to
3123 * give it a chance to adjust the device capacity.
3126 * Defined by block layer. Might sleep.
3128 static void sd_unlock_native_capacity(struct gendisk *disk)
3130 struct scsi_device *sdev = scsi_disk(disk)->device;
3132 if (sdev->host->hostt->unlock_native_capacity)
3133 sdev->host->hostt->unlock_native_capacity(sdev);
3137 * sd_format_disk_name - format disk name
3138 * @prefix: name prefix - ie. "sd" for SCSI disks
3139 * @index: index of the disk to format name for
3140 * @buf: output buffer
3141 * @buflen: length of the output buffer
3143 * SCSI disk names starts at sda. The 26th device is sdz and the
3144 * 27th is sdaa. The last one for two lettered suffix is sdzz
3145 * which is followed by sdaaa.
3147 * This is basically 26 base counting with one extra 'nil' entry
3148 * at the beginning from the second digit on and can be
3149 * determined using similar method as 26 base conversion with the
3150 * index shifted -1 after each digit is computed.
3156 * 0 on success, -errno on failure.
3158 static int sd_format_disk_name(char *prefix, int index, char *buf, int buflen)
3160 const int base = 'z' - 'a' + 1;
3161 char *begin = buf + strlen(prefix);
3162 char *end = buf + buflen;
3172 *--p = 'a' + (index % unit);
3173 index = (index / unit) - 1;
3174 } while (index >= 0);
3176 memmove(begin, p, end - p);
3177 memcpy(buf, prefix, strlen(prefix));
3183 * The asynchronous part of sd_probe
3185 static void sd_probe_async(void *data, async_cookie_t cookie)
3187 struct scsi_disk *sdkp = data;
3188 struct scsi_device *sdp;
3195 index = sdkp->index;
3196 dev = &sdp->sdev_gendev;
3198 gd->major = sd_major((index & 0xf0) >> 4);
3199 gd->first_minor = ((index & 0xf) << 4) | (index & 0xfff00);
3200 gd->minors = SD_MINORS;
3202 gd->fops = &sd_fops;
3203 gd->private_data = &sdkp->driver;
3204 gd->queue = sdkp->device->request_queue;
3206 /* defaults, until the device tells us otherwise */
3207 sdp->sector_size = 512;
3209 sdkp->media_present = 1;
3210 sdkp->write_prot = 0;
3211 sdkp->cache_override = 0;
3215 sdkp->first_scan = 1;
3216 sdkp->max_medium_access_timeouts = SD_MAX_MEDIUM_TIMEOUTS;
3218 sd_revalidate_disk(gd);
3220 gd->flags = GENHD_FL_EXT_DEVT;
3221 if (sdp->removable) {
3222 gd->flags |= GENHD_FL_REMOVABLE;
3223 gd->events |= DISK_EVENT_MEDIA_CHANGE;
3226 blk_pm_runtime_init(sdp->request_queue, dev);
3227 device_add_disk(dev, gd);
3229 sd_dif_config_host(sdkp);
3231 sd_revalidate_disk(gd);
3233 sd_printk(KERN_NOTICE, sdkp, "Attached SCSI %sdisk\n",
3234 sdp->removable ? "removable " : "");
3235 scsi_autopm_put_device(sdp);
3236 put_device(&sdkp->dev);
3240 * sd_probe - called during driver initialization and whenever a
3241 * new scsi device is attached to the system. It is called once
3242 * for each scsi device (not just disks) present.
3243 * @dev: pointer to device object
3245 * Returns 0 if successful (or not interested in this scsi device
3246 * (e.g. scanner)); 1 when there is an error.
3248 * Note: this function is invoked from the scsi mid-level.
3249 * This function sets up the mapping between a given
3250 * <host,channel,id,lun> (found in sdp) and new device name
3251 * (e.g. /dev/sda). More precisely it is the block device major
3252 * and minor number that is chosen here.
3254 * Assume sd_probe is not re-entrant (for time being)
3255 * Also think about sd_probe() and sd_remove() running coincidentally.
3257 static int sd_probe(struct device *dev)
3259 struct scsi_device *sdp = to_scsi_device(dev);
3260 struct scsi_disk *sdkp;
3265 scsi_autopm_get_device(sdp);
3267 if (sdp->type != TYPE_DISK &&
3268 sdp->type != TYPE_ZBC &&
3269 sdp->type != TYPE_MOD &&
3270 sdp->type != TYPE_RBC)
3273 #ifndef CONFIG_BLK_DEV_ZONED
3274 if (sdp->type == TYPE_ZBC)
3277 SCSI_LOG_HLQUEUE(3, sdev_printk(KERN_INFO, sdp,
3281 sdkp = kzalloc(sizeof(*sdkp), GFP_KERNEL);
3285 gd = alloc_disk(SD_MINORS);
3290 if (!ida_pre_get(&sd_index_ida, GFP_KERNEL))
3293 spin_lock(&sd_index_lock);
3294 error = ida_get_new(&sd_index_ida, &index);
3295 spin_unlock(&sd_index_lock);
3296 } while (error == -EAGAIN);
3299 sdev_printk(KERN_WARNING, sdp, "sd_probe: memory exhausted.\n");
3303 error = sd_format_disk_name("sd", index, gd->disk_name, DISK_NAME_LEN);
3305 sdev_printk(KERN_WARNING, sdp, "SCSI disk (sd) name length exceeded.\n");
3306 goto out_free_index;
3310 sdkp->driver = &sd_template;
3312 sdkp->index = index;
3313 atomic_set(&sdkp->openers, 0);
3314 atomic_set(&sdkp->device->ioerr_cnt, 0);
3316 if (!sdp->request_queue->rq_timeout) {
3317 if (sdp->type != TYPE_MOD)
3318 blk_queue_rq_timeout(sdp->request_queue, SD_TIMEOUT);
3320 blk_queue_rq_timeout(sdp->request_queue,
3324 device_initialize(&sdkp->dev);
3325 sdkp->dev.parent = dev;
3326 sdkp->dev.class = &sd_disk_class;
3327 dev_set_name(&sdkp->dev, "%s", dev_name(dev));
3329 error = device_add(&sdkp->dev);
3331 goto out_free_index;
3334 dev_set_drvdata(dev, sdkp);
3336 get_device(&sdkp->dev); /* prevent release before async_schedule */
3337 async_schedule_domain(sd_probe_async, sdkp, &scsi_sd_probe_domain);
3342 spin_lock(&sd_index_lock);
3343 ida_remove(&sd_index_ida, index);
3344 spin_unlock(&sd_index_lock);
3350 scsi_autopm_put_device(sdp);
3355 * sd_remove - called whenever a scsi disk (previously recognized by
3356 * sd_probe) is detached from the system. It is called (potentially
3357 * multiple times) during sd module unload.
3358 * @dev: pointer to device object
3360 * Note: this function is invoked from the scsi mid-level.
3361 * This function potentially frees up a device name (e.g. /dev/sdc)
3362 * that could be re-used by a subsequent sd_probe().
3363 * This function is not called when the built-in sd driver is "exit-ed".
3365 static int sd_remove(struct device *dev)
3367 struct scsi_disk *sdkp;
3370 sdkp = dev_get_drvdata(dev);
3371 devt = disk_devt(sdkp->disk);
3372 scsi_autopm_get_device(sdkp->device);
3374 async_synchronize_full_domain(&scsi_sd_pm_domain);
3375 async_synchronize_full_domain(&scsi_sd_probe_domain);
3376 device_del(&sdkp->dev);
3377 del_gendisk(sdkp->disk);
3380 sd_zbc_remove(sdkp);
3382 blk_register_region(devt, SD_MINORS, NULL,
3383 sd_default_probe, NULL, NULL);
3385 mutex_lock(&sd_ref_mutex);
3386 dev_set_drvdata(dev, NULL);
3387 put_device(&sdkp->dev);
3388 mutex_unlock(&sd_ref_mutex);
3394 * scsi_disk_release - Called to free the scsi_disk structure
3395 * @dev: pointer to embedded class device
3397 * sd_ref_mutex must be held entering this routine. Because it is
3398 * called on last put, you should always use the scsi_disk_get()
3399 * scsi_disk_put() helpers which manipulate the semaphore directly
3400 * and never do a direct put_device.
3402 static void scsi_disk_release(struct device *dev)
3404 struct scsi_disk *sdkp = to_scsi_disk(dev);
3405 struct gendisk *disk = sdkp->disk;
3407 spin_lock(&sd_index_lock);
3408 ida_remove(&sd_index_ida, sdkp->index);
3409 spin_unlock(&sd_index_lock);
3411 disk->private_data = NULL;
3413 put_device(&sdkp->device->sdev_gendev);
3418 static int sd_start_stop_device(struct scsi_disk *sdkp, int start)
3420 unsigned char cmd[6] = { START_STOP }; /* START_VALID */
3421 struct scsi_sense_hdr sshdr;
3422 struct scsi_device *sdp = sdkp->device;
3426 cmd[4] |= 1; /* START */
3428 if (sdp->start_stop_pwr_cond)
3429 cmd[4] |= start ? 1 << 4 : 3 << 4; /* Active or Standby */
3431 if (!scsi_device_online(sdp))
3434 res = scsi_execute(sdp, cmd, DMA_NONE, NULL, 0, NULL, &sshdr,
3435 SD_TIMEOUT, SD_MAX_RETRIES, 0, RQF_PM, NULL);
3437 sd_print_result(sdkp, "Start/Stop Unit failed", res);
3438 if (driver_byte(res) & DRIVER_SENSE)
3439 sd_print_sense_hdr(sdkp, &sshdr);
3440 if (scsi_sense_valid(&sshdr) &&
3441 /* 0x3a is medium not present */
3446 /* SCSI error codes must not go to the generic layer */
3454 * Send a SYNCHRONIZE CACHE instruction down to the device through
3455 * the normal SCSI command structure. Wait for the command to
3458 static void sd_shutdown(struct device *dev)
3460 struct scsi_disk *sdkp = dev_get_drvdata(dev);
3463 return; /* this can happen */
3465 if (pm_runtime_suspended(dev))
3468 if (sdkp->WCE && sdkp->media_present) {
3469 sd_printk(KERN_NOTICE, sdkp, "Synchronizing SCSI cache\n");
3470 sd_sync_cache(sdkp, NULL);
3473 if (system_state != SYSTEM_RESTART && sdkp->device->manage_start_stop) {
3474 sd_printk(KERN_NOTICE, sdkp, "Stopping disk\n");
3475 sd_start_stop_device(sdkp, 0);
3479 static int sd_suspend_common(struct device *dev, bool ignore_stop_errors)
3481 struct scsi_disk *sdkp = dev_get_drvdata(dev);
3482 struct scsi_sense_hdr sshdr;
3485 if (!sdkp) /* E.g.: runtime suspend following sd_remove() */
3488 if (sdkp->WCE && sdkp->media_present) {
3489 sd_printk(KERN_NOTICE, sdkp, "Synchronizing SCSI cache\n");
3490 ret = sd_sync_cache(sdkp, &sshdr);
3493 /* ignore OFFLINE device */
3497 if (!scsi_sense_valid(&sshdr) ||
3498 sshdr.sense_key != ILLEGAL_REQUEST)
3502 * sshdr.sense_key == ILLEGAL_REQUEST means this drive
3503 * doesn't support sync. There's not much to do and
3504 * suspend shouldn't fail.
3510 if (sdkp->device->manage_start_stop) {
3511 sd_printk(KERN_NOTICE, sdkp, "Stopping disk\n");
3512 /* an error is not worth aborting a system sleep */
3513 ret = sd_start_stop_device(sdkp, 0);
3514 if (ignore_stop_errors)
3521 static int sd_suspend_system(struct device *dev)
3523 return sd_suspend_common(dev, true);
3526 static int sd_suspend_runtime(struct device *dev)
3528 return sd_suspend_common(dev, false);
3531 static int sd_resume(struct device *dev)
3533 struct scsi_disk *sdkp = dev_get_drvdata(dev);
3535 if (!sdkp) /* E.g.: runtime resume at the start of sd_probe() */
3538 if (!sdkp->device->manage_start_stop)
3541 sd_printk(KERN_NOTICE, sdkp, "Starting disk\n");
3542 return sd_start_stop_device(sdkp, 1);
3546 * init_sd - entry point for this driver (both when built in or when
3549 * Note: this function registers this driver with the scsi mid-level.
3551 static int __init init_sd(void)
3553 int majors = 0, i, err;
3555 SCSI_LOG_HLQUEUE(3, printk("init_sd: sd driver entry point\n"));
3557 for (i = 0; i < SD_MAJORS; i++) {
3558 if (register_blkdev(sd_major(i), "sd") != 0)
3561 blk_register_region(sd_major(i), SD_MINORS, NULL,
3562 sd_default_probe, NULL, NULL);
3568 err = class_register(&sd_disk_class);
3572 sd_cdb_cache = kmem_cache_create("sd_ext_cdb", SD_EXT_CDB_SIZE,
3574 if (!sd_cdb_cache) {
3575 printk(KERN_ERR "sd: can't init extended cdb cache\n");
3580 sd_cdb_pool = mempool_create_slab_pool(SD_MEMPOOL_SIZE, sd_cdb_cache);
3582 printk(KERN_ERR "sd: can't init extended cdb pool\n");
3587 err = scsi_register_driver(&sd_template.gendrv);
3589 goto err_out_driver;
3594 mempool_destroy(sd_cdb_pool);
3597 kmem_cache_destroy(sd_cdb_cache);
3600 class_unregister(&sd_disk_class);
3602 for (i = 0; i < SD_MAJORS; i++)
3603 unregister_blkdev(sd_major(i), "sd");
3608 * exit_sd - exit point for this driver (when it is a module).
3610 * Note: this function unregisters this driver from the scsi mid-level.
3612 static void __exit exit_sd(void)
3616 SCSI_LOG_HLQUEUE(3, printk("exit_sd: exiting sd driver\n"));
3618 scsi_unregister_driver(&sd_template.gendrv);
3619 mempool_destroy(sd_cdb_pool);
3620 kmem_cache_destroy(sd_cdb_cache);
3622 class_unregister(&sd_disk_class);
3624 for (i = 0; i < SD_MAJORS; i++) {
3625 blk_unregister_region(sd_major(i), SD_MINORS);
3626 unregister_blkdev(sd_major(i), "sd");
3630 module_init(init_sd);
3631 module_exit(exit_sd);
3633 static void sd_print_sense_hdr(struct scsi_disk *sdkp,
3634 struct scsi_sense_hdr *sshdr)
3636 scsi_print_sense_hdr(sdkp->device,
3637 sdkp->disk ? sdkp->disk->disk_name : NULL, sshdr);
3640 static void sd_print_result(const struct scsi_disk *sdkp, const char *msg,
3643 const char *hb_string = scsi_hostbyte_string(result);
3644 const char *db_string = scsi_driverbyte_string(result);
3646 if (hb_string || db_string)
3647 sd_printk(KERN_INFO, sdkp,
3648 "%s: Result: hostbyte=%s driverbyte=%s\n", msg,
3649 hb_string ? hb_string : "invalid",
3650 db_string ? db_string : "invalid");
3652 sd_printk(KERN_INFO, sdkp,
3653 "%s: Result: hostbyte=0x%02x driverbyte=0x%02x\n",
3654 msg, host_byte(result), driver_byte(result));