2 * sd.c Copyright (C) 1992 Drew Eckhardt
3 * Copyright (C) 1993, 1994, 1995, 1999 Eric Youngdale
5 * Linux scsi disk driver
6 * Initial versions: Drew Eckhardt
7 * Subsequent revisions: Eric Youngdale
8 * Modification history:
9 * - Drew Eckhardt <drew@colorado.edu> original
10 * - Eric Youngdale <eric@andante.org> add scatter-gather, multiple
11 * outstanding request, and other enhancements.
12 * Support loadable low-level scsi drivers.
13 * - Jirka Hanika <geo@ff.cuni.cz> support more scsi disks using
14 * eight major numbers.
15 * - Richard Gooch <rgooch@atnf.csiro.au> support devfs.
16 * - Torben Mathiasen <tmm@image.dk> Resource allocation fixes in
17 * sd_init and cleanups.
18 * - Alex Davis <letmein@erols.com> Fix problem where partition info
19 * not being read in sd_open. Fix problem where removable media
20 * could be ejected after sd_open.
21 * - Douglas Gilbert <dgilbert@interlog.com> cleanup for lk 2.5.x
22 * - Badari Pulavarty <pbadari@us.ibm.com>, Matthew Wilcox
23 * <willy@debian.org>, Kurt Garloff <garloff@suse.de>:
24 * Support 32k/1M disks.
26 * Logging policy (needs CONFIG_SCSI_LOGGING defined):
27 * - setting up transfer: SCSI_LOG_HLQUEUE levels 1 and 2
28 * - end of transfer (bh + scsi_lib): SCSI_LOG_HLCOMPLETE level 1
29 * - entering sd_ioctl: SCSI_LOG_IOCTL level 1
30 * - entering other commands: SCSI_LOG_HLQUEUE level 3
31 * Note: when the logging level is set by the user, it must be greater
32 * than the level indicated above to trigger output.
35 #include <linux/module.h>
37 #include <linux/kernel.h>
39 #include <linux/bio.h>
40 #include <linux/genhd.h>
41 #include <linux/hdreg.h>
42 #include <linux/errno.h>
43 #include <linux/idr.h>
44 #include <linux/interrupt.h>
45 #include <linux/init.h>
46 #include <linux/blkdev.h>
47 #include <linux/blkpg.h>
48 #include <linux/delay.h>
49 #include <linux/mutex.h>
50 #include <linux/string_helpers.h>
51 #include <linux/async.h>
52 #include <linux/slab.h>
53 #include <linux/sed-opal.h>
54 #include <linux/pm_runtime.h>
56 #include <linux/t10-pi.h>
57 #include <linux/uaccess.h>
58 #include <asm/unaligned.h>
60 #include <scsi/scsi.h>
61 #include <scsi/scsi_cmnd.h>
62 #include <scsi/scsi_dbg.h>
63 #include <scsi/scsi_device.h>
64 #include <scsi/scsi_driver.h>
65 #include <scsi/scsi_eh.h>
66 #include <scsi/scsi_host.h>
67 #include <scsi/scsi_ioctl.h>
68 #include <scsi/scsicam.h>
71 #include "scsi_priv.h"
72 #include "scsi_logging.h"
74 MODULE_AUTHOR("Eric Youngdale");
75 MODULE_DESCRIPTION("SCSI disk (sd) driver");
76 MODULE_LICENSE("GPL");
78 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK0_MAJOR);
79 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK1_MAJOR);
80 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK2_MAJOR);
81 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK3_MAJOR);
82 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK4_MAJOR);
83 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK5_MAJOR);
84 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK6_MAJOR);
85 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK7_MAJOR);
86 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK8_MAJOR);
87 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK9_MAJOR);
88 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK10_MAJOR);
89 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK11_MAJOR);
90 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK12_MAJOR);
91 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK13_MAJOR);
92 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK14_MAJOR);
93 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK15_MAJOR);
94 MODULE_ALIAS_SCSI_DEVICE(TYPE_DISK);
95 MODULE_ALIAS_SCSI_DEVICE(TYPE_MOD);
96 MODULE_ALIAS_SCSI_DEVICE(TYPE_RBC);
97 MODULE_ALIAS_SCSI_DEVICE(TYPE_ZBC);
99 #if !defined(CONFIG_DEBUG_BLOCK_EXT_DEVT)
105 static void sd_config_discard(struct scsi_disk *, unsigned int);
106 static void sd_config_write_same(struct scsi_disk *);
107 static int sd_revalidate_disk(struct gendisk *);
108 static void sd_unlock_native_capacity(struct gendisk *disk);
109 static int sd_probe(struct device *);
110 static int sd_remove(struct device *);
111 static void sd_shutdown(struct device *);
112 static int sd_suspend_system(struct device *);
113 static int sd_suspend_runtime(struct device *);
114 static int sd_resume(struct device *);
115 static void sd_rescan(struct device *);
116 static int sd_init_command(struct scsi_cmnd *SCpnt);
117 static void sd_uninit_command(struct scsi_cmnd *SCpnt);
118 static int sd_done(struct scsi_cmnd *);
119 static void sd_eh_reset(struct scsi_cmnd *);
120 static int sd_eh_action(struct scsi_cmnd *, int);
121 static void sd_read_capacity(struct scsi_disk *sdkp, unsigned char *buffer);
122 static void scsi_disk_release(struct device *cdev);
123 static void sd_print_sense_hdr(struct scsi_disk *, struct scsi_sense_hdr *);
124 static void sd_print_result(const struct scsi_disk *, const char *, int);
126 static DEFINE_SPINLOCK(sd_index_lock);
127 static DEFINE_IDA(sd_index_ida);
129 /* This semaphore is used to mediate the 0->1 reference get in the
130 * face of object destruction (i.e. we can't allow a get on an
131 * object after last put) */
132 static DEFINE_MUTEX(sd_ref_mutex);
134 static struct kmem_cache *sd_cdb_cache;
135 static mempool_t *sd_cdb_pool;
137 static const char *sd_cache_types[] = {
138 "write through", "none", "write back",
139 "write back, no read (daft)"
142 static void sd_set_flush_flag(struct scsi_disk *sdkp)
144 bool wc = false, fua = false;
152 blk_queue_write_cache(sdkp->disk->queue, wc, fua);
156 cache_type_store(struct device *dev, struct device_attribute *attr,
157 const char *buf, size_t count)
159 int i, ct = -1, rcd, wce, sp;
160 struct scsi_disk *sdkp = to_scsi_disk(dev);
161 struct scsi_device *sdp = sdkp->device;
164 struct scsi_mode_data data;
165 struct scsi_sense_hdr sshdr;
166 static const char temp[] = "temporary ";
169 if (sdp->type != TYPE_DISK && sdp->type != TYPE_ZBC)
170 /* no cache control on RBC devices; theoretically they
171 * can do it, but there's probably so many exceptions
172 * it's not worth the risk */
175 if (strncmp(buf, temp, sizeof(temp) - 1) == 0) {
176 buf += sizeof(temp) - 1;
177 sdkp->cache_override = 1;
179 sdkp->cache_override = 0;
182 for (i = 0; i < ARRAY_SIZE(sd_cache_types); i++) {
183 len = strlen(sd_cache_types[i]);
184 if (strncmp(sd_cache_types[i], buf, len) == 0 &&
192 rcd = ct & 0x01 ? 1 : 0;
193 wce = (ct & 0x02) && !sdkp->write_prot ? 1 : 0;
195 if (sdkp->cache_override) {
198 sd_set_flush_flag(sdkp);
202 if (scsi_mode_sense(sdp, 0x08, 8, buffer, sizeof(buffer), SD_TIMEOUT,
203 SD_MAX_RETRIES, &data, NULL))
205 len = min_t(size_t, sizeof(buffer), data.length - data.header_length -
206 data.block_descriptor_length);
207 buffer_data = buffer + data.header_length +
208 data.block_descriptor_length;
209 buffer_data[2] &= ~0x05;
210 buffer_data[2] |= wce << 2 | rcd;
211 sp = buffer_data[0] & 0x80 ? 1 : 0;
212 buffer_data[0] &= ~0x80;
214 if (scsi_mode_select(sdp, 1, sp, 8, buffer_data, len, SD_TIMEOUT,
215 SD_MAX_RETRIES, &data, &sshdr)) {
216 if (scsi_sense_valid(&sshdr))
217 sd_print_sense_hdr(sdkp, &sshdr);
220 revalidate_disk(sdkp->disk);
225 manage_start_stop_show(struct device *dev, struct device_attribute *attr,
228 struct scsi_disk *sdkp = to_scsi_disk(dev);
229 struct scsi_device *sdp = sdkp->device;
231 return snprintf(buf, 20, "%u\n", sdp->manage_start_stop);
235 manage_start_stop_store(struct device *dev, struct device_attribute *attr,
236 const char *buf, size_t count)
238 struct scsi_disk *sdkp = to_scsi_disk(dev);
239 struct scsi_device *sdp = sdkp->device;
241 if (!capable(CAP_SYS_ADMIN))
244 sdp->manage_start_stop = simple_strtoul(buf, NULL, 10);
248 static DEVICE_ATTR_RW(manage_start_stop);
251 allow_restart_show(struct device *dev, struct device_attribute *attr, char *buf)
253 struct scsi_disk *sdkp = to_scsi_disk(dev);
255 return snprintf(buf, 40, "%d\n", sdkp->device->allow_restart);
259 allow_restart_store(struct device *dev, struct device_attribute *attr,
260 const char *buf, size_t count)
262 struct scsi_disk *sdkp = to_scsi_disk(dev);
263 struct scsi_device *sdp = sdkp->device;
265 if (!capable(CAP_SYS_ADMIN))
268 if (sdp->type != TYPE_DISK && sdp->type != TYPE_ZBC)
271 sdp->allow_restart = simple_strtoul(buf, NULL, 10);
275 static DEVICE_ATTR_RW(allow_restart);
278 cache_type_show(struct device *dev, struct device_attribute *attr, char *buf)
280 struct scsi_disk *sdkp = to_scsi_disk(dev);
281 int ct = sdkp->RCD + 2*sdkp->WCE;
283 return snprintf(buf, 40, "%s\n", sd_cache_types[ct]);
285 static DEVICE_ATTR_RW(cache_type);
288 FUA_show(struct device *dev, struct device_attribute *attr, char *buf)
290 struct scsi_disk *sdkp = to_scsi_disk(dev);
292 return snprintf(buf, 20, "%u\n", sdkp->DPOFUA);
294 static DEVICE_ATTR_RO(FUA);
297 protection_type_show(struct device *dev, struct device_attribute *attr,
300 struct scsi_disk *sdkp = to_scsi_disk(dev);
302 return snprintf(buf, 20, "%u\n", sdkp->protection_type);
306 protection_type_store(struct device *dev, struct device_attribute *attr,
307 const char *buf, size_t count)
309 struct scsi_disk *sdkp = to_scsi_disk(dev);
313 if (!capable(CAP_SYS_ADMIN))
316 err = kstrtouint(buf, 10, &val);
321 if (val >= 0 && val <= T10_PI_TYPE3_PROTECTION)
322 sdkp->protection_type = val;
326 static DEVICE_ATTR_RW(protection_type);
329 protection_mode_show(struct device *dev, struct device_attribute *attr,
332 struct scsi_disk *sdkp = to_scsi_disk(dev);
333 struct scsi_device *sdp = sdkp->device;
334 unsigned int dif, dix;
336 dif = scsi_host_dif_capable(sdp->host, sdkp->protection_type);
337 dix = scsi_host_dix_capable(sdp->host, sdkp->protection_type);
339 if (!dix && scsi_host_dix_capable(sdp->host, T10_PI_TYPE0_PROTECTION)) {
345 return snprintf(buf, 20, "none\n");
347 return snprintf(buf, 20, "%s%u\n", dix ? "dix" : "dif", dif);
349 static DEVICE_ATTR_RO(protection_mode);
352 app_tag_own_show(struct device *dev, struct device_attribute *attr, char *buf)
354 struct scsi_disk *sdkp = to_scsi_disk(dev);
356 return snprintf(buf, 20, "%u\n", sdkp->ATO);
358 static DEVICE_ATTR_RO(app_tag_own);
361 thin_provisioning_show(struct device *dev, struct device_attribute *attr,
364 struct scsi_disk *sdkp = to_scsi_disk(dev);
366 return snprintf(buf, 20, "%u\n", sdkp->lbpme);
368 static DEVICE_ATTR_RO(thin_provisioning);
370 static const char *lbp_mode[] = {
371 [SD_LBP_FULL] = "full",
372 [SD_LBP_UNMAP] = "unmap",
373 [SD_LBP_WS16] = "writesame_16",
374 [SD_LBP_WS10] = "writesame_10",
375 [SD_LBP_ZERO] = "writesame_zero",
376 [SD_LBP_DISABLE] = "disabled",
380 provisioning_mode_show(struct device *dev, struct device_attribute *attr,
383 struct scsi_disk *sdkp = to_scsi_disk(dev);
385 return snprintf(buf, 20, "%s\n", lbp_mode[sdkp->provisioning_mode]);
389 provisioning_mode_store(struct device *dev, struct device_attribute *attr,
390 const char *buf, size_t count)
392 struct scsi_disk *sdkp = to_scsi_disk(dev);
393 struct scsi_device *sdp = sdkp->device;
395 if (!capable(CAP_SYS_ADMIN))
398 if (sd_is_zoned(sdkp)) {
399 sd_config_discard(sdkp, SD_LBP_DISABLE);
403 if (sdp->type != TYPE_DISK)
406 if (!strncmp(buf, lbp_mode[SD_LBP_UNMAP], 20))
407 sd_config_discard(sdkp, SD_LBP_UNMAP);
408 else if (!strncmp(buf, lbp_mode[SD_LBP_WS16], 20))
409 sd_config_discard(sdkp, SD_LBP_WS16);
410 else if (!strncmp(buf, lbp_mode[SD_LBP_WS10], 20))
411 sd_config_discard(sdkp, SD_LBP_WS10);
412 else if (!strncmp(buf, lbp_mode[SD_LBP_ZERO], 20))
413 sd_config_discard(sdkp, SD_LBP_ZERO);
414 else if (!strncmp(buf, lbp_mode[SD_LBP_DISABLE], 20))
415 sd_config_discard(sdkp, SD_LBP_DISABLE);
421 static DEVICE_ATTR_RW(provisioning_mode);
423 static const char *zeroing_mode[] = {
424 [SD_ZERO_WRITE] = "write",
425 [SD_ZERO_WS] = "writesame",
426 [SD_ZERO_WS16_UNMAP] = "writesame_16_unmap",
427 [SD_ZERO_WS10_UNMAP] = "writesame_10_unmap",
431 zeroing_mode_show(struct device *dev, struct device_attribute *attr,
434 struct scsi_disk *sdkp = to_scsi_disk(dev);
436 return snprintf(buf, 20, "%s\n", zeroing_mode[sdkp->zeroing_mode]);
440 zeroing_mode_store(struct device *dev, struct device_attribute *attr,
441 const char *buf, size_t count)
443 struct scsi_disk *sdkp = to_scsi_disk(dev);
445 if (!capable(CAP_SYS_ADMIN))
448 if (!strncmp(buf, zeroing_mode[SD_ZERO_WRITE], 20))
449 sdkp->zeroing_mode = SD_ZERO_WRITE;
450 else if (!strncmp(buf, zeroing_mode[SD_ZERO_WS], 20))
451 sdkp->zeroing_mode = SD_ZERO_WS;
452 else if (!strncmp(buf, zeroing_mode[SD_ZERO_WS16_UNMAP], 20))
453 sdkp->zeroing_mode = SD_ZERO_WS16_UNMAP;
454 else if (!strncmp(buf, zeroing_mode[SD_ZERO_WS10_UNMAP], 20))
455 sdkp->zeroing_mode = SD_ZERO_WS10_UNMAP;
461 static DEVICE_ATTR_RW(zeroing_mode);
464 max_medium_access_timeouts_show(struct device *dev,
465 struct device_attribute *attr, char *buf)
467 struct scsi_disk *sdkp = to_scsi_disk(dev);
469 return snprintf(buf, 20, "%u\n", sdkp->max_medium_access_timeouts);
473 max_medium_access_timeouts_store(struct device *dev,
474 struct device_attribute *attr, const char *buf,
477 struct scsi_disk *sdkp = to_scsi_disk(dev);
480 if (!capable(CAP_SYS_ADMIN))
483 err = kstrtouint(buf, 10, &sdkp->max_medium_access_timeouts);
485 return err ? err : count;
487 static DEVICE_ATTR_RW(max_medium_access_timeouts);
490 max_write_same_blocks_show(struct device *dev, struct device_attribute *attr,
493 struct scsi_disk *sdkp = to_scsi_disk(dev);
495 return snprintf(buf, 20, "%u\n", sdkp->max_ws_blocks);
499 max_write_same_blocks_store(struct device *dev, struct device_attribute *attr,
500 const char *buf, size_t count)
502 struct scsi_disk *sdkp = to_scsi_disk(dev);
503 struct scsi_device *sdp = sdkp->device;
507 if (!capable(CAP_SYS_ADMIN))
510 if (sdp->type != TYPE_DISK && sdp->type != TYPE_ZBC)
513 err = kstrtoul(buf, 10, &max);
519 sdp->no_write_same = 1;
520 else if (max <= SD_MAX_WS16_BLOCKS) {
521 sdp->no_write_same = 0;
522 sdkp->max_ws_blocks = max;
525 sd_config_write_same(sdkp);
529 static DEVICE_ATTR_RW(max_write_same_blocks);
531 static struct attribute *sd_disk_attrs[] = {
532 &dev_attr_cache_type.attr,
534 &dev_attr_allow_restart.attr,
535 &dev_attr_manage_start_stop.attr,
536 &dev_attr_protection_type.attr,
537 &dev_attr_protection_mode.attr,
538 &dev_attr_app_tag_own.attr,
539 &dev_attr_thin_provisioning.attr,
540 &dev_attr_provisioning_mode.attr,
541 &dev_attr_zeroing_mode.attr,
542 &dev_attr_max_write_same_blocks.attr,
543 &dev_attr_max_medium_access_timeouts.attr,
546 ATTRIBUTE_GROUPS(sd_disk);
548 static struct class sd_disk_class = {
550 .owner = THIS_MODULE,
551 .dev_release = scsi_disk_release,
552 .dev_groups = sd_disk_groups,
555 static const struct dev_pm_ops sd_pm_ops = {
556 .suspend = sd_suspend_system,
558 .poweroff = sd_suspend_system,
559 .restore = sd_resume,
560 .runtime_suspend = sd_suspend_runtime,
561 .runtime_resume = sd_resume,
564 static struct scsi_driver sd_template = {
567 .owner = THIS_MODULE,
570 .shutdown = sd_shutdown,
574 .init_command = sd_init_command,
575 .uninit_command = sd_uninit_command,
577 .eh_action = sd_eh_action,
578 .eh_reset = sd_eh_reset,
582 * Dummy kobj_map->probe function.
583 * The default ->probe function will call modprobe, which is
584 * pointless as this module is already loaded.
586 static struct kobject *sd_default_probe(dev_t devt, int *partno, void *data)
592 * Device no to disk mapping:
594 * major disc2 disc p1
595 * |............|.............|....|....| <- dev_t
598 * Inside a major, we have 16k disks, however mapped non-
599 * contiguously. The first 16 disks are for major0, the next
600 * ones with major1, ... Disk 256 is for major0 again, disk 272
602 * As we stay compatible with our numbering scheme, we can reuse
603 * the well-know SCSI majors 8, 65--71, 136--143.
605 static int sd_major(int major_idx)
609 return SCSI_DISK0_MAJOR;
611 return SCSI_DISK1_MAJOR + major_idx - 1;
613 return SCSI_DISK8_MAJOR + major_idx - 8;
616 return 0; /* shut up gcc */
620 static struct scsi_disk *scsi_disk_get(struct gendisk *disk)
622 struct scsi_disk *sdkp = NULL;
624 mutex_lock(&sd_ref_mutex);
626 if (disk->private_data) {
627 sdkp = scsi_disk(disk);
628 if (scsi_device_get(sdkp->device) == 0)
629 get_device(&sdkp->dev);
633 mutex_unlock(&sd_ref_mutex);
637 static void scsi_disk_put(struct scsi_disk *sdkp)
639 struct scsi_device *sdev = sdkp->device;
641 mutex_lock(&sd_ref_mutex);
642 put_device(&sdkp->dev);
643 scsi_device_put(sdev);
644 mutex_unlock(&sd_ref_mutex);
647 #ifdef CONFIG_BLK_SED_OPAL
648 static int sd_sec_submit(void *data, u16 spsp, u8 secp, void *buffer,
649 size_t len, bool send)
651 struct scsi_device *sdev = data;
655 cdb[0] = send ? SECURITY_PROTOCOL_OUT : SECURITY_PROTOCOL_IN;
657 put_unaligned_be16(spsp, &cdb[2]);
658 put_unaligned_be32(len, &cdb[6]);
660 ret = scsi_execute_req(sdev, cdb,
661 send ? DMA_TO_DEVICE : DMA_FROM_DEVICE,
662 buffer, len, NULL, SD_TIMEOUT, SD_MAX_RETRIES, NULL);
663 return ret <= 0 ? ret : -EIO;
665 #endif /* CONFIG_BLK_SED_OPAL */
667 static unsigned char sd_setup_protect_cmnd(struct scsi_cmnd *scmd,
668 unsigned int dix, unsigned int dif)
670 struct bio *bio = scmd->request->bio;
671 unsigned int prot_op = sd_prot_op(rq_data_dir(scmd->request), dix, dif);
672 unsigned int protect = 0;
674 if (dix) { /* DIX Type 0, 1, 2, 3 */
675 if (bio_integrity_flagged(bio, BIP_IP_CHECKSUM))
676 scmd->prot_flags |= SCSI_PROT_IP_CHECKSUM;
678 if (bio_integrity_flagged(bio, BIP_CTRL_NOCHECK) == false)
679 scmd->prot_flags |= SCSI_PROT_GUARD_CHECK;
682 if (dif != T10_PI_TYPE3_PROTECTION) { /* DIX/DIF Type 0, 1, 2 */
683 scmd->prot_flags |= SCSI_PROT_REF_INCREMENT;
685 if (bio_integrity_flagged(bio, BIP_CTRL_NOCHECK) == false)
686 scmd->prot_flags |= SCSI_PROT_REF_CHECK;
689 if (dif) { /* DIX/DIF Type 1, 2, 3 */
690 scmd->prot_flags |= SCSI_PROT_TRANSFER_PI;
692 if (bio_integrity_flagged(bio, BIP_DISK_NOCHECK))
693 protect = 3 << 5; /* Disable target PI checking */
695 protect = 1 << 5; /* Enable target PI checking */
698 scsi_set_prot_op(scmd, prot_op);
699 scsi_set_prot_type(scmd, dif);
700 scmd->prot_flags &= sd_prot_flag_mask(prot_op);
705 static void sd_config_discard(struct scsi_disk *sdkp, unsigned int mode)
707 struct request_queue *q = sdkp->disk->queue;
708 unsigned int logical_block_size = sdkp->device->sector_size;
709 unsigned int max_blocks = 0;
711 q->limits.discard_alignment =
712 sdkp->unmap_alignment * logical_block_size;
713 q->limits.discard_granularity =
714 max(sdkp->physical_block_size,
715 sdkp->unmap_granularity * logical_block_size);
716 sdkp->provisioning_mode = mode;
721 blk_queue_max_discard_sectors(q, 0);
722 queue_flag_clear_unlocked(QUEUE_FLAG_DISCARD, q);
726 max_blocks = min_not_zero(sdkp->max_unmap_blocks,
727 (u32)SD_MAX_WS16_BLOCKS);
731 max_blocks = min_not_zero(sdkp->max_ws_blocks,
732 (u32)SD_MAX_WS16_BLOCKS);
736 max_blocks = min_not_zero(sdkp->max_ws_blocks,
737 (u32)SD_MAX_WS10_BLOCKS);
741 max_blocks = min_not_zero(sdkp->max_ws_blocks,
742 (u32)SD_MAX_WS10_BLOCKS);
746 blk_queue_max_discard_sectors(q, max_blocks * (logical_block_size >> 9));
747 queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, q);
750 static int sd_setup_unmap_cmnd(struct scsi_cmnd *cmd)
752 struct scsi_device *sdp = cmd->device;
753 struct request *rq = cmd->request;
754 u64 sector = blk_rq_pos(rq) >> (ilog2(sdp->sector_size) - 9);
755 u32 nr_sectors = blk_rq_sectors(rq) >> (ilog2(sdp->sector_size) - 9);
756 unsigned int data_len = 24;
759 rq->special_vec.bv_page = alloc_page(GFP_ATOMIC | __GFP_ZERO);
760 if (!rq->special_vec.bv_page)
761 return BLKPREP_DEFER;
762 rq->special_vec.bv_offset = 0;
763 rq->special_vec.bv_len = data_len;
764 rq->rq_flags |= RQF_SPECIAL_PAYLOAD;
767 cmd->cmnd[0] = UNMAP;
770 buf = page_address(rq->special_vec.bv_page);
771 put_unaligned_be16(6 + 16, &buf[0]);
772 put_unaligned_be16(16, &buf[2]);
773 put_unaligned_be64(sector, &buf[8]);
774 put_unaligned_be32(nr_sectors, &buf[16]);
776 cmd->allowed = SD_MAX_RETRIES;
777 cmd->transfersize = data_len;
778 rq->timeout = SD_TIMEOUT;
779 scsi_req(rq)->resid_len = data_len;
781 return scsi_init_io(cmd);
784 static int sd_setup_write_same16_cmnd(struct scsi_cmnd *cmd, bool unmap)
786 struct scsi_device *sdp = cmd->device;
787 struct request *rq = cmd->request;
788 u64 sector = blk_rq_pos(rq) >> (ilog2(sdp->sector_size) - 9);
789 u32 nr_sectors = blk_rq_sectors(rq) >> (ilog2(sdp->sector_size) - 9);
790 u32 data_len = sdp->sector_size;
792 rq->special_vec.bv_page = alloc_page(GFP_ATOMIC | __GFP_ZERO);
793 if (!rq->special_vec.bv_page)
794 return BLKPREP_DEFER;
795 rq->special_vec.bv_offset = 0;
796 rq->special_vec.bv_len = data_len;
797 rq->rq_flags |= RQF_SPECIAL_PAYLOAD;
800 cmd->cmnd[0] = WRITE_SAME_16;
802 cmd->cmnd[1] = 0x8; /* UNMAP */
803 put_unaligned_be64(sector, &cmd->cmnd[2]);
804 put_unaligned_be32(nr_sectors, &cmd->cmnd[10]);
806 cmd->allowed = SD_MAX_RETRIES;
807 cmd->transfersize = data_len;
808 rq->timeout = unmap ? SD_TIMEOUT : SD_WRITE_SAME_TIMEOUT;
809 scsi_req(rq)->resid_len = data_len;
811 return scsi_init_io(cmd);
814 static int sd_setup_write_same10_cmnd(struct scsi_cmnd *cmd, bool unmap)
816 struct scsi_device *sdp = cmd->device;
817 struct request *rq = cmd->request;
818 u64 sector = blk_rq_pos(rq) >> (ilog2(sdp->sector_size) - 9);
819 u32 nr_sectors = blk_rq_sectors(rq) >> (ilog2(sdp->sector_size) - 9);
820 u32 data_len = sdp->sector_size;
822 rq->special_vec.bv_page = alloc_page(GFP_ATOMIC | __GFP_ZERO);
823 if (!rq->special_vec.bv_page)
824 return BLKPREP_DEFER;
825 rq->special_vec.bv_offset = 0;
826 rq->special_vec.bv_len = data_len;
827 rq->rq_flags |= RQF_SPECIAL_PAYLOAD;
830 cmd->cmnd[0] = WRITE_SAME;
832 cmd->cmnd[1] = 0x8; /* UNMAP */
833 put_unaligned_be32(sector, &cmd->cmnd[2]);
834 put_unaligned_be16(nr_sectors, &cmd->cmnd[7]);
836 cmd->allowed = SD_MAX_RETRIES;
837 cmd->transfersize = data_len;
838 rq->timeout = unmap ? SD_TIMEOUT : SD_WRITE_SAME_TIMEOUT;
839 scsi_req(rq)->resid_len = data_len;
841 return scsi_init_io(cmd);
844 static int sd_setup_write_zeroes_cmnd(struct scsi_cmnd *cmd)
846 struct request *rq = cmd->request;
847 struct scsi_device *sdp = cmd->device;
848 struct scsi_disk *sdkp = scsi_disk(rq->rq_disk);
849 u64 sector = blk_rq_pos(rq) >> (ilog2(sdp->sector_size) - 9);
850 u32 nr_sectors = blk_rq_sectors(rq) >> (ilog2(sdp->sector_size) - 9);
852 if (!(rq->cmd_flags & REQ_NOUNMAP)) {
853 switch (sdkp->zeroing_mode) {
854 case SD_ZERO_WS16_UNMAP:
855 return sd_setup_write_same16_cmnd(cmd, true);
856 case SD_ZERO_WS10_UNMAP:
857 return sd_setup_write_same10_cmnd(cmd, true);
861 if (sdp->no_write_same)
862 return BLKPREP_INVALID;
863 if (sdkp->ws16 || sector > 0xffffffff || nr_sectors > 0xffff)
864 return sd_setup_write_same16_cmnd(cmd, false);
865 return sd_setup_write_same10_cmnd(cmd, false);
868 static void sd_config_write_same(struct scsi_disk *sdkp)
870 struct request_queue *q = sdkp->disk->queue;
871 unsigned int logical_block_size = sdkp->device->sector_size;
873 if (sdkp->device->no_write_same) {
874 sdkp->max_ws_blocks = 0;
878 /* Some devices can not handle block counts above 0xffff despite
879 * supporting WRITE SAME(16). Consequently we default to 64k
880 * blocks per I/O unless the device explicitly advertises a
883 if (sdkp->max_ws_blocks > SD_MAX_WS10_BLOCKS)
884 sdkp->max_ws_blocks = min_not_zero(sdkp->max_ws_blocks,
885 (u32)SD_MAX_WS16_BLOCKS);
886 else if (sdkp->ws16 || sdkp->ws10 || sdkp->device->no_report_opcodes)
887 sdkp->max_ws_blocks = min_not_zero(sdkp->max_ws_blocks,
888 (u32)SD_MAX_WS10_BLOCKS);
890 sdkp->device->no_write_same = 1;
891 sdkp->max_ws_blocks = 0;
894 if (sdkp->lbprz && sdkp->lbpws)
895 sdkp->zeroing_mode = SD_ZERO_WS16_UNMAP;
896 else if (sdkp->lbprz && sdkp->lbpws10)
897 sdkp->zeroing_mode = SD_ZERO_WS10_UNMAP;
898 else if (sdkp->max_ws_blocks)
899 sdkp->zeroing_mode = SD_ZERO_WS;
901 sdkp->zeroing_mode = SD_ZERO_WRITE;
904 blk_queue_max_write_same_sectors(q, sdkp->max_ws_blocks *
905 (logical_block_size >> 9));
906 blk_queue_max_write_zeroes_sectors(q, sdkp->max_ws_blocks *
907 (logical_block_size >> 9));
911 * sd_setup_write_same_cmnd - write the same data to multiple blocks
912 * @cmd: command to prepare
914 * Will set up either WRITE SAME(10) or WRITE SAME(16) depending on
915 * the preference indicated by the target device.
917 static int sd_setup_write_same_cmnd(struct scsi_cmnd *cmd)
919 struct request *rq = cmd->request;
920 struct scsi_device *sdp = cmd->device;
921 struct scsi_disk *sdkp = scsi_disk(rq->rq_disk);
922 struct bio *bio = rq->bio;
923 sector_t sector = blk_rq_pos(rq);
924 unsigned int nr_sectors = blk_rq_sectors(rq);
925 unsigned int nr_bytes = blk_rq_bytes(rq);
928 if (sdkp->device->no_write_same)
929 return BLKPREP_INVALID;
931 BUG_ON(bio_offset(bio) || bio_iovec(bio).bv_len != sdp->sector_size);
933 if (sd_is_zoned(sdkp)) {
934 ret = sd_zbc_write_lock_zone(cmd);
935 if (ret != BLKPREP_OK)
939 sector >>= ilog2(sdp->sector_size) - 9;
940 nr_sectors >>= ilog2(sdp->sector_size) - 9;
942 rq->timeout = SD_WRITE_SAME_TIMEOUT;
944 if (sdkp->ws16 || sector > 0xffffffff || nr_sectors > 0xffff) {
946 cmd->cmnd[0] = WRITE_SAME_16;
947 put_unaligned_be64(sector, &cmd->cmnd[2]);
948 put_unaligned_be32(nr_sectors, &cmd->cmnd[10]);
951 cmd->cmnd[0] = WRITE_SAME;
952 put_unaligned_be32(sector, &cmd->cmnd[2]);
953 put_unaligned_be16(nr_sectors, &cmd->cmnd[7]);
956 cmd->transfersize = sdp->sector_size;
957 cmd->allowed = SD_MAX_RETRIES;
960 * For WRITE SAME the data transferred via the DATA OUT buffer is
961 * different from the amount of data actually written to the target.
963 * We set up __data_len to the amount of data transferred via the
964 * DATA OUT buffer so that blk_rq_map_sg sets up the proper S/G list
965 * to transfer a single sector of data first, but then reset it to
966 * the amount of data to be written right after so that the I/O path
967 * knows how much to actually write.
969 rq->__data_len = sdp->sector_size;
970 ret = scsi_init_io(cmd);
971 rq->__data_len = nr_bytes;
975 static int sd_setup_flush_cmnd(struct scsi_cmnd *cmd)
977 struct request *rq = cmd->request;
979 /* flush requests don't perform I/O, zero the S/G table */
980 memset(&cmd->sdb, 0, sizeof(cmd->sdb));
982 cmd->cmnd[0] = SYNCHRONIZE_CACHE;
984 cmd->transfersize = 0;
985 cmd->allowed = SD_MAX_RETRIES;
987 rq->timeout = rq->q->rq_timeout * SD_FLUSH_TIMEOUT_MULTIPLIER;
991 static int sd_setup_read_write_cmnd(struct scsi_cmnd *SCpnt)
993 struct request *rq = SCpnt->request;
994 struct scsi_device *sdp = SCpnt->device;
995 struct gendisk *disk = rq->rq_disk;
996 struct scsi_disk *sdkp = scsi_disk(disk);
997 sector_t block = blk_rq_pos(rq);
999 unsigned int this_count = blk_rq_sectors(rq);
1000 unsigned int dif, dix;
1001 bool zoned_write = sd_is_zoned(sdkp) && rq_data_dir(rq) == WRITE;
1003 unsigned char protect;
1006 ret = sd_zbc_write_lock_zone(SCpnt);
1007 if (ret != BLKPREP_OK)
1011 ret = scsi_init_io(SCpnt);
1012 if (ret != BLKPREP_OK)
1014 SCpnt = rq->special;
1016 /* from here on until we're complete, any goto out
1017 * is used for a killable error condition */
1021 scmd_printk(KERN_INFO, SCpnt,
1022 "%s: block=%llu, count=%d\n",
1023 __func__, (unsigned long long)block, this_count));
1025 if (!sdp || !scsi_device_online(sdp) ||
1026 block + blk_rq_sectors(rq) > get_capacity(disk)) {
1027 SCSI_LOG_HLQUEUE(2, scmd_printk(KERN_INFO, SCpnt,
1028 "Finishing %u sectors\n",
1029 blk_rq_sectors(rq)));
1030 SCSI_LOG_HLQUEUE(2, scmd_printk(KERN_INFO, SCpnt,
1031 "Retry with 0x%p\n", SCpnt));
1037 * quietly refuse to do anything to a changed disc until
1038 * the changed bit has been reset
1040 /* printk("SCSI disk has been changed or is not present. Prohibiting further I/O.\n"); */
1045 * Some SD card readers can't handle multi-sector accesses which touch
1046 * the last one or two hardware sectors. Split accesses as needed.
1048 threshold = get_capacity(disk) - SD_LAST_BUGGY_SECTORS *
1049 (sdp->sector_size / 512);
1051 if (unlikely(sdp->last_sector_bug && block + this_count > threshold)) {
1052 if (block < threshold) {
1053 /* Access up to the threshold but not beyond */
1054 this_count = threshold - block;
1056 /* Access only a single hardware sector */
1057 this_count = sdp->sector_size / 512;
1061 SCSI_LOG_HLQUEUE(2, scmd_printk(KERN_INFO, SCpnt, "block=%llu\n",
1062 (unsigned long long)block));
1065 * If we have a 1K hardware sectorsize, prevent access to single
1066 * 512 byte sectors. In theory we could handle this - in fact
1067 * the scsi cdrom driver must be able to handle this because
1068 * we typically use 1K blocksizes, and cdroms typically have
1069 * 2K hardware sectorsizes. Of course, things are simpler
1070 * with the cdrom, since it is read-only. For performance
1071 * reasons, the filesystems should be able to handle this
1072 * and not force the scsi disk driver to use bounce buffers
1075 if (sdp->sector_size == 1024) {
1076 if ((block & 1) || (blk_rq_sectors(rq) & 1)) {
1077 scmd_printk(KERN_ERR, SCpnt,
1078 "Bad block number requested\n");
1082 this_count = this_count >> 1;
1085 if (sdp->sector_size == 2048) {
1086 if ((block & 3) || (blk_rq_sectors(rq) & 3)) {
1087 scmd_printk(KERN_ERR, SCpnt,
1088 "Bad block number requested\n");
1092 this_count = this_count >> 2;
1095 if (sdp->sector_size == 4096) {
1096 if ((block & 7) || (blk_rq_sectors(rq) & 7)) {
1097 scmd_printk(KERN_ERR, SCpnt,
1098 "Bad block number requested\n");
1102 this_count = this_count >> 3;
1105 if (rq_data_dir(rq) == WRITE) {
1106 SCpnt->cmnd[0] = WRITE_6;
1108 if (blk_integrity_rq(rq))
1109 sd_dif_prepare(SCpnt);
1111 } else if (rq_data_dir(rq) == READ) {
1112 SCpnt->cmnd[0] = READ_6;
1114 scmd_printk(KERN_ERR, SCpnt, "Unknown command %d\n", req_op(rq));
1118 SCSI_LOG_HLQUEUE(2, scmd_printk(KERN_INFO, SCpnt,
1119 "%s %d/%u 512 byte blocks.\n",
1120 (rq_data_dir(rq) == WRITE) ?
1121 "writing" : "reading", this_count,
1122 blk_rq_sectors(rq)));
1124 dix = scsi_prot_sg_count(SCpnt);
1125 dif = scsi_host_dif_capable(SCpnt->device->host, sdkp->protection_type);
1128 protect = sd_setup_protect_cmnd(SCpnt, dix, dif);
1132 if (protect && sdkp->protection_type == T10_PI_TYPE2_PROTECTION) {
1133 SCpnt->cmnd = mempool_alloc(sd_cdb_pool, GFP_ATOMIC);
1135 if (unlikely(SCpnt->cmnd == NULL)) {
1136 ret = BLKPREP_DEFER;
1140 SCpnt->cmd_len = SD_EXT_CDB_SIZE;
1141 memset(SCpnt->cmnd, 0, SCpnt->cmd_len);
1142 SCpnt->cmnd[0] = VARIABLE_LENGTH_CMD;
1143 SCpnt->cmnd[7] = 0x18;
1144 SCpnt->cmnd[9] = (rq_data_dir(rq) == READ) ? READ_32 : WRITE_32;
1145 SCpnt->cmnd[10] = protect | ((rq->cmd_flags & REQ_FUA) ? 0x8 : 0);
1148 SCpnt->cmnd[12] = sizeof(block) > 4 ? (unsigned char) (block >> 56) & 0xff : 0;
1149 SCpnt->cmnd[13] = sizeof(block) > 4 ? (unsigned char) (block >> 48) & 0xff : 0;
1150 SCpnt->cmnd[14] = sizeof(block) > 4 ? (unsigned char) (block >> 40) & 0xff : 0;
1151 SCpnt->cmnd[15] = sizeof(block) > 4 ? (unsigned char) (block >> 32) & 0xff : 0;
1152 SCpnt->cmnd[16] = (unsigned char) (block >> 24) & 0xff;
1153 SCpnt->cmnd[17] = (unsigned char) (block >> 16) & 0xff;
1154 SCpnt->cmnd[18] = (unsigned char) (block >> 8) & 0xff;
1155 SCpnt->cmnd[19] = (unsigned char) block & 0xff;
1157 /* Expected Indirect LBA */
1158 SCpnt->cmnd[20] = (unsigned char) (block >> 24) & 0xff;
1159 SCpnt->cmnd[21] = (unsigned char) (block >> 16) & 0xff;
1160 SCpnt->cmnd[22] = (unsigned char) (block >> 8) & 0xff;
1161 SCpnt->cmnd[23] = (unsigned char) block & 0xff;
1163 /* Transfer length */
1164 SCpnt->cmnd[28] = (unsigned char) (this_count >> 24) & 0xff;
1165 SCpnt->cmnd[29] = (unsigned char) (this_count >> 16) & 0xff;
1166 SCpnt->cmnd[30] = (unsigned char) (this_count >> 8) & 0xff;
1167 SCpnt->cmnd[31] = (unsigned char) this_count & 0xff;
1168 } else if (sdp->use_16_for_rw || (this_count > 0xffff)) {
1169 SCpnt->cmnd[0] += READ_16 - READ_6;
1170 SCpnt->cmnd[1] = protect | ((rq->cmd_flags & REQ_FUA) ? 0x8 : 0);
1171 SCpnt->cmnd[2] = sizeof(block) > 4 ? (unsigned char) (block >> 56) & 0xff : 0;
1172 SCpnt->cmnd[3] = sizeof(block) > 4 ? (unsigned char) (block >> 48) & 0xff : 0;
1173 SCpnt->cmnd[4] = sizeof(block) > 4 ? (unsigned char) (block >> 40) & 0xff : 0;
1174 SCpnt->cmnd[5] = sizeof(block) > 4 ? (unsigned char) (block >> 32) & 0xff : 0;
1175 SCpnt->cmnd[6] = (unsigned char) (block >> 24) & 0xff;
1176 SCpnt->cmnd[7] = (unsigned char) (block >> 16) & 0xff;
1177 SCpnt->cmnd[8] = (unsigned char) (block >> 8) & 0xff;
1178 SCpnt->cmnd[9] = (unsigned char) block & 0xff;
1179 SCpnt->cmnd[10] = (unsigned char) (this_count >> 24) & 0xff;
1180 SCpnt->cmnd[11] = (unsigned char) (this_count >> 16) & 0xff;
1181 SCpnt->cmnd[12] = (unsigned char) (this_count >> 8) & 0xff;
1182 SCpnt->cmnd[13] = (unsigned char) this_count & 0xff;
1183 SCpnt->cmnd[14] = SCpnt->cmnd[15] = 0;
1184 } else if ((this_count > 0xff) || (block > 0x1fffff) ||
1185 scsi_device_protection(SCpnt->device) ||
1186 SCpnt->device->use_10_for_rw) {
1187 SCpnt->cmnd[0] += READ_10 - READ_6;
1188 SCpnt->cmnd[1] = protect | ((rq->cmd_flags & REQ_FUA) ? 0x8 : 0);
1189 SCpnt->cmnd[2] = (unsigned char) (block >> 24) & 0xff;
1190 SCpnt->cmnd[3] = (unsigned char) (block >> 16) & 0xff;
1191 SCpnt->cmnd[4] = (unsigned char) (block >> 8) & 0xff;
1192 SCpnt->cmnd[5] = (unsigned char) block & 0xff;
1193 SCpnt->cmnd[6] = SCpnt->cmnd[9] = 0;
1194 SCpnt->cmnd[7] = (unsigned char) (this_count >> 8) & 0xff;
1195 SCpnt->cmnd[8] = (unsigned char) this_count & 0xff;
1197 if (unlikely(rq->cmd_flags & REQ_FUA)) {
1199 * This happens only if this drive failed
1200 * 10byte rw command with ILLEGAL_REQUEST
1201 * during operation and thus turned off
1204 scmd_printk(KERN_ERR, SCpnt,
1205 "FUA write on READ/WRITE(6) drive\n");
1209 SCpnt->cmnd[1] |= (unsigned char) ((block >> 16) & 0x1f);
1210 SCpnt->cmnd[2] = (unsigned char) ((block >> 8) & 0xff);
1211 SCpnt->cmnd[3] = (unsigned char) block & 0xff;
1212 SCpnt->cmnd[4] = (unsigned char) this_count;
1215 SCpnt->sdb.length = this_count * sdp->sector_size;
1218 * We shouldn't disconnect in the middle of a sector, so with a dumb
1219 * host adapter, it's safe to assume that we can at least transfer
1220 * this many bytes between each connect / disconnect.
1222 SCpnt->transfersize = sdp->sector_size;
1223 SCpnt->underflow = this_count << 9;
1224 SCpnt->allowed = SD_MAX_RETRIES;
1227 * This indicates that the command is ready from our end to be
1232 if (zoned_write && ret != BLKPREP_OK)
1233 sd_zbc_write_unlock_zone(SCpnt);
1238 static int sd_init_command(struct scsi_cmnd *cmd)
1240 struct request *rq = cmd->request;
1242 switch (req_op(rq)) {
1243 case REQ_OP_DISCARD:
1244 switch (scsi_disk(rq->rq_disk)->provisioning_mode) {
1246 return sd_setup_unmap_cmnd(cmd);
1248 return sd_setup_write_same16_cmnd(cmd, true);
1250 return sd_setup_write_same10_cmnd(cmd, true);
1252 return sd_setup_write_same10_cmnd(cmd, false);
1254 return BLKPREP_INVALID;
1256 case REQ_OP_WRITE_ZEROES:
1257 return sd_setup_write_zeroes_cmnd(cmd);
1258 case REQ_OP_WRITE_SAME:
1259 return sd_setup_write_same_cmnd(cmd);
1261 return sd_setup_flush_cmnd(cmd);
1264 return sd_setup_read_write_cmnd(cmd);
1265 case REQ_OP_ZONE_REPORT:
1266 return sd_zbc_setup_report_cmnd(cmd);
1267 case REQ_OP_ZONE_RESET:
1268 return sd_zbc_setup_reset_cmnd(cmd);
1274 static void sd_uninit_command(struct scsi_cmnd *SCpnt)
1276 struct request *rq = SCpnt->request;
1278 if (rq->rq_flags & RQF_SPECIAL_PAYLOAD)
1279 __free_page(rq->special_vec.bv_page);
1281 if (SCpnt->cmnd != scsi_req(rq)->cmd) {
1282 mempool_free(SCpnt->cmnd, sd_cdb_pool);
1289 * sd_open - open a scsi disk device
1290 * @bdev: Block device of the scsi disk to open
1291 * @mode: FMODE_* mask
1293 * Returns 0 if successful. Returns a negated errno value in case
1296 * Note: This can be called from a user context (e.g. fsck(1) )
1297 * or from within the kernel (e.g. as a result of a mount(1) ).
1298 * In the latter case @inode and @filp carry an abridged amount
1299 * of information as noted above.
1301 * Locking: called with bdev->bd_mutex held.
1303 static int sd_open(struct block_device *bdev, fmode_t mode)
1305 struct scsi_disk *sdkp = scsi_disk_get(bdev->bd_disk);
1306 struct scsi_device *sdev;
1312 SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp, "sd_open\n"));
1314 sdev = sdkp->device;
1317 * If the device is in error recovery, wait until it is done.
1318 * If the device is offline, then disallow any access to it.
1321 if (!scsi_block_when_processing_errors(sdev))
1324 if (sdev->removable || sdkp->write_prot)
1325 check_disk_change(bdev);
1328 * If the drive is empty, just let the open fail.
1330 retval = -ENOMEDIUM;
1331 if (sdev->removable && !sdkp->media_present && !(mode & FMODE_NDELAY))
1335 * If the device has the write protect tab set, have the open fail
1336 * if the user expects to be able to write to the thing.
1339 if (sdkp->write_prot && (mode & FMODE_WRITE))
1343 * It is possible that the disk changing stuff resulted in
1344 * the device being taken offline. If this is the case,
1345 * report this to the user, and don't pretend that the
1346 * open actually succeeded.
1349 if (!scsi_device_online(sdev))
1352 if ((atomic_inc_return(&sdkp->openers) == 1) && sdev->removable) {
1353 if (scsi_block_when_processing_errors(sdev))
1354 scsi_set_medium_removal(sdev, SCSI_REMOVAL_PREVENT);
1360 scsi_disk_put(sdkp);
1365 * sd_release - invoked when the (last) close(2) is called on this
1367 * @disk: disk to release
1368 * @mode: FMODE_* mask
1372 * Note: may block (uninterruptible) if error recovery is underway
1375 * Locking: called with bdev->bd_mutex held.
1377 static void sd_release(struct gendisk *disk, fmode_t mode)
1379 struct scsi_disk *sdkp = scsi_disk(disk);
1380 struct scsi_device *sdev = sdkp->device;
1382 SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp, "sd_release\n"));
1384 if (atomic_dec_return(&sdkp->openers) == 0 && sdev->removable) {
1385 if (scsi_block_when_processing_errors(sdev))
1386 scsi_set_medium_removal(sdev, SCSI_REMOVAL_ALLOW);
1390 * XXX and what if there are packets in flight and this close()
1391 * XXX is followed by a "rmmod sd_mod"?
1394 scsi_disk_put(sdkp);
1397 static int sd_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1399 struct scsi_disk *sdkp = scsi_disk(bdev->bd_disk);
1400 struct scsi_device *sdp = sdkp->device;
1401 struct Scsi_Host *host = sdp->host;
1402 sector_t capacity = logical_to_sectors(sdp, sdkp->capacity);
1405 /* default to most commonly used values */
1406 diskinfo[0] = 0x40; /* 1 << 6 */
1407 diskinfo[1] = 0x20; /* 1 << 5 */
1408 diskinfo[2] = capacity >> 11;
1410 /* override with calculated, extended default, or driver values */
1411 if (host->hostt->bios_param)
1412 host->hostt->bios_param(sdp, bdev, capacity, diskinfo);
1414 scsicam_bios_param(bdev, capacity, diskinfo);
1416 geo->heads = diskinfo[0];
1417 geo->sectors = diskinfo[1];
1418 geo->cylinders = diskinfo[2];
1423 * sd_ioctl - process an ioctl
1424 * @bdev: target block device
1425 * @mode: FMODE_* mask
1426 * @cmd: ioctl command number
1427 * @arg: this is third argument given to ioctl(2) system call.
1428 * Often contains a pointer.
1430 * Returns 0 if successful (some ioctls return positive numbers on
1431 * success as well). Returns a negated errno value in case of error.
1433 * Note: most ioctls are forward onto the block subsystem or further
1434 * down in the scsi subsystem.
1436 static int sd_ioctl(struct block_device *bdev, fmode_t mode,
1437 unsigned int cmd, unsigned long arg)
1439 struct gendisk *disk = bdev->bd_disk;
1440 struct scsi_disk *sdkp = scsi_disk(disk);
1441 struct scsi_device *sdp = sdkp->device;
1442 void __user *p = (void __user *)arg;
1445 SCSI_LOG_IOCTL(1, sd_printk(KERN_INFO, sdkp, "sd_ioctl: disk=%s, "
1446 "cmd=0x%x\n", disk->disk_name, cmd));
1448 error = scsi_verify_blk_ioctl(bdev, cmd);
1453 * If we are in the middle of error recovery, don't let anyone
1454 * else try and use this device. Also, if error recovery fails, it
1455 * may try and take the device offline, in which case all further
1456 * access to the device is prohibited.
1458 error = scsi_ioctl_block_when_processing_errors(sdp, cmd,
1459 (mode & FMODE_NDELAY) != 0);
1463 if (is_sed_ioctl(cmd))
1464 return sed_ioctl(sdkp->opal_dev, cmd, p);
1467 * Send SCSI addressing ioctls directly to mid level, send other
1468 * ioctls to block level and then onto mid level if they can't be
1472 case SCSI_IOCTL_GET_IDLUN:
1473 case SCSI_IOCTL_GET_BUS_NUMBER:
1474 error = scsi_ioctl(sdp, cmd, p);
1477 error = scsi_cmd_blk_ioctl(bdev, mode, cmd, p);
1478 if (error != -ENOTTY)
1480 error = scsi_ioctl(sdp, cmd, p);
1487 static void set_media_not_present(struct scsi_disk *sdkp)
1489 if (sdkp->media_present)
1490 sdkp->device->changed = 1;
1492 if (sdkp->device->removable) {
1493 sdkp->media_present = 0;
1498 static int media_not_present(struct scsi_disk *sdkp,
1499 struct scsi_sense_hdr *sshdr)
1501 if (!scsi_sense_valid(sshdr))
1504 /* not invoked for commands that could return deferred errors */
1505 switch (sshdr->sense_key) {
1506 case UNIT_ATTENTION:
1508 /* medium not present */
1509 if (sshdr->asc == 0x3A) {
1510 set_media_not_present(sdkp);
1518 * sd_check_events - check media events
1519 * @disk: kernel device descriptor
1520 * @clearing: disk events currently being cleared
1522 * Returns mask of DISK_EVENT_*.
1524 * Note: this function is invoked from the block subsystem.
1526 static unsigned int sd_check_events(struct gendisk *disk, unsigned int clearing)
1528 struct scsi_disk *sdkp = scsi_disk_get(disk);
1529 struct scsi_device *sdp;
1536 SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp, "sd_check_events\n"));
1539 * If the device is offline, don't send any commands - just pretend as
1540 * if the command failed. If the device ever comes back online, we
1541 * can deal with it then. It is only because of unrecoverable errors
1542 * that we would ever take a device offline in the first place.
1544 if (!scsi_device_online(sdp)) {
1545 set_media_not_present(sdkp);
1550 * Using TEST_UNIT_READY enables differentiation between drive with
1551 * no cartridge loaded - NOT READY, drive with changed cartridge -
1552 * UNIT ATTENTION, or with same cartridge - GOOD STATUS.
1554 * Drives that auto spin down. eg iomega jaz 1G, will be started
1555 * by sd_spinup_disk() from sd_revalidate_disk(), which happens whenever
1556 * sd_revalidate() is called.
1558 if (scsi_block_when_processing_errors(sdp)) {
1559 struct scsi_sense_hdr sshdr = { 0, };
1561 retval = scsi_test_unit_ready(sdp, SD_TIMEOUT, SD_MAX_RETRIES,
1564 /* failed to execute TUR, assume media not present */
1565 if (host_byte(retval)) {
1566 set_media_not_present(sdkp);
1570 if (media_not_present(sdkp, &sshdr))
1575 * For removable scsi disk we have to recognise the presence
1576 * of a disk in the drive.
1578 if (!sdkp->media_present)
1580 sdkp->media_present = 1;
1583 * sdp->changed is set under the following conditions:
1585 * Medium present state has changed in either direction.
1586 * Device has indicated UNIT_ATTENTION.
1588 retval = sdp->changed ? DISK_EVENT_MEDIA_CHANGE : 0;
1590 scsi_disk_put(sdkp);
1594 static int sd_sync_cache(struct scsi_disk *sdkp)
1597 struct scsi_device *sdp = sdkp->device;
1598 const int timeout = sdp->request_queue->rq_timeout
1599 * SD_FLUSH_TIMEOUT_MULTIPLIER;
1600 struct scsi_sense_hdr sshdr;
1602 if (!scsi_device_online(sdp))
1605 for (retries = 3; retries > 0; --retries) {
1606 unsigned char cmd[10] = { 0 };
1608 cmd[0] = SYNCHRONIZE_CACHE;
1610 * Leave the rest of the command zero to indicate
1613 res = scsi_execute(sdp, cmd, DMA_NONE, NULL, 0, NULL, &sshdr,
1614 timeout, SD_MAX_RETRIES, 0, RQF_PM, NULL);
1620 sd_print_result(sdkp, "Synchronize Cache(10) failed", res);
1622 if (driver_byte(res) & DRIVER_SENSE)
1623 sd_print_sense_hdr(sdkp, &sshdr);
1624 /* we need to evaluate the error return */
1625 if (scsi_sense_valid(&sshdr) &&
1626 (sshdr.asc == 0x3a || /* medium not present */
1627 sshdr.asc == 0x20)) /* invalid command */
1628 /* this is no error here */
1631 switch (host_byte(res)) {
1632 /* ignore errors due to racing a disconnection */
1633 case DID_BAD_TARGET:
1634 case DID_NO_CONNECT:
1636 /* signal the upper layer it might try again */
1640 case DID_SOFT_ERROR:
1649 static void sd_rescan(struct device *dev)
1651 struct scsi_disk *sdkp = dev_get_drvdata(dev);
1653 revalidate_disk(sdkp->disk);
1657 #ifdef CONFIG_COMPAT
1659 * This gets directly called from VFS. When the ioctl
1660 * is not recognized we go back to the other translation paths.
1662 static int sd_compat_ioctl(struct block_device *bdev, fmode_t mode,
1663 unsigned int cmd, unsigned long arg)
1665 struct scsi_device *sdev = scsi_disk(bdev->bd_disk)->device;
1668 error = scsi_ioctl_block_when_processing_errors(sdev, cmd,
1669 (mode & FMODE_NDELAY) != 0);
1674 * Let the static ioctl translation table take care of it.
1676 if (!sdev->host->hostt->compat_ioctl)
1677 return -ENOIOCTLCMD;
1678 return sdev->host->hostt->compat_ioctl(sdev, cmd, (void __user *)arg);
1682 static char sd_pr_type(enum pr_type type)
1685 case PR_WRITE_EXCLUSIVE:
1687 case PR_EXCLUSIVE_ACCESS:
1689 case PR_WRITE_EXCLUSIVE_REG_ONLY:
1691 case PR_EXCLUSIVE_ACCESS_REG_ONLY:
1693 case PR_WRITE_EXCLUSIVE_ALL_REGS:
1695 case PR_EXCLUSIVE_ACCESS_ALL_REGS:
1702 static int sd_pr_command(struct block_device *bdev, u8 sa,
1703 u64 key, u64 sa_key, u8 type, u8 flags)
1705 struct scsi_device *sdev = scsi_disk(bdev->bd_disk)->device;
1706 struct scsi_sense_hdr sshdr;
1708 u8 cmd[16] = { 0, };
1709 u8 data[24] = { 0, };
1711 cmd[0] = PERSISTENT_RESERVE_OUT;
1714 put_unaligned_be32(sizeof(data), &cmd[5]);
1716 put_unaligned_be64(key, &data[0]);
1717 put_unaligned_be64(sa_key, &data[8]);
1720 result = scsi_execute_req(sdev, cmd, DMA_TO_DEVICE, &data, sizeof(data),
1721 &sshdr, SD_TIMEOUT, SD_MAX_RETRIES, NULL);
1723 if ((driver_byte(result) & DRIVER_SENSE) &&
1724 (scsi_sense_valid(&sshdr))) {
1725 sdev_printk(KERN_INFO, sdev, "PR command failed: %d\n", result);
1726 scsi_print_sense_hdr(sdev, NULL, &sshdr);
1732 static int sd_pr_register(struct block_device *bdev, u64 old_key, u64 new_key,
1735 if (flags & ~PR_FL_IGNORE_KEY)
1737 return sd_pr_command(bdev, (flags & PR_FL_IGNORE_KEY) ? 0x06 : 0x00,
1738 old_key, new_key, 0,
1739 (1 << 0) /* APTPL */);
1742 static int sd_pr_reserve(struct block_device *bdev, u64 key, enum pr_type type,
1747 return sd_pr_command(bdev, 0x01, key, 0, sd_pr_type(type), 0);
1750 static int sd_pr_release(struct block_device *bdev, u64 key, enum pr_type type)
1752 return sd_pr_command(bdev, 0x02, key, 0, sd_pr_type(type), 0);
1755 static int sd_pr_preempt(struct block_device *bdev, u64 old_key, u64 new_key,
1756 enum pr_type type, bool abort)
1758 return sd_pr_command(bdev, abort ? 0x05 : 0x04, old_key, new_key,
1759 sd_pr_type(type), 0);
1762 static int sd_pr_clear(struct block_device *bdev, u64 key)
1764 return sd_pr_command(bdev, 0x03, key, 0, 0, 0);
1767 static const struct pr_ops sd_pr_ops = {
1768 .pr_register = sd_pr_register,
1769 .pr_reserve = sd_pr_reserve,
1770 .pr_release = sd_pr_release,
1771 .pr_preempt = sd_pr_preempt,
1772 .pr_clear = sd_pr_clear,
1775 static const struct block_device_operations sd_fops = {
1776 .owner = THIS_MODULE,
1778 .release = sd_release,
1780 .getgeo = sd_getgeo,
1781 #ifdef CONFIG_COMPAT
1782 .compat_ioctl = sd_compat_ioctl,
1784 .check_events = sd_check_events,
1785 .revalidate_disk = sd_revalidate_disk,
1786 .unlock_native_capacity = sd_unlock_native_capacity,
1787 .pr_ops = &sd_pr_ops,
1791 * sd_eh_reset - reset error handling callback
1792 * @scmd: sd-issued command that has failed
1794 * This function is called by the SCSI midlayer before starting
1795 * SCSI EH. When counting medium access failures we have to be
1796 * careful to register it only only once per device and SCSI EH run;
1797 * there might be several timed out commands which will cause the
1798 * 'max_medium_access_timeouts' counter to trigger after the first
1799 * SCSI EH run already and set the device to offline.
1800 * So this function resets the internal counter before starting SCSI EH.
1802 static void sd_eh_reset(struct scsi_cmnd *scmd)
1804 struct scsi_disk *sdkp = scsi_disk(scmd->request->rq_disk);
1806 /* New SCSI EH run, reset gate variable */
1807 sdkp->ignore_medium_access_errors = false;
1811 * sd_eh_action - error handling callback
1812 * @scmd: sd-issued command that has failed
1813 * @eh_disp: The recovery disposition suggested by the midlayer
1815 * This function is called by the SCSI midlayer upon completion of an
1816 * error test command (currently TEST UNIT READY). The result of sending
1817 * the eh command is passed in eh_disp. We're looking for devices that
1818 * fail medium access commands but are OK with non access commands like
1819 * test unit ready (so wrongly see the device as having a successful
1822 static int sd_eh_action(struct scsi_cmnd *scmd, int eh_disp)
1824 struct scsi_disk *sdkp = scsi_disk(scmd->request->rq_disk);
1826 if (!scsi_device_online(scmd->device) ||
1827 !scsi_medium_access_command(scmd) ||
1828 host_byte(scmd->result) != DID_TIME_OUT ||
1833 * The device has timed out executing a medium access command.
1834 * However, the TEST UNIT READY command sent during error
1835 * handling completed successfully. Either the device is in the
1836 * process of recovering or has it suffered an internal failure
1837 * that prevents access to the storage medium.
1839 if (!sdkp->ignore_medium_access_errors) {
1840 sdkp->medium_access_timed_out++;
1841 sdkp->ignore_medium_access_errors = true;
1845 * If the device keeps failing read/write commands but TEST UNIT
1846 * READY always completes successfully we assume that medium
1847 * access is no longer possible and take the device offline.
1849 if (sdkp->medium_access_timed_out >= sdkp->max_medium_access_timeouts) {
1850 scmd_printk(KERN_ERR, scmd,
1851 "Medium access timeout failure. Offlining disk!\n");
1852 scsi_device_set_state(scmd->device, SDEV_OFFLINE);
1860 static unsigned int sd_completed_bytes(struct scsi_cmnd *scmd)
1862 struct request *req = scmd->request;
1863 struct scsi_device *sdev = scmd->device;
1864 unsigned int transferred, good_bytes;
1865 u64 start_lba, end_lba, bad_lba;
1868 * Some commands have a payload smaller than the device logical
1869 * block size (e.g. INQUIRY on a 4K disk).
1871 if (scsi_bufflen(scmd) <= sdev->sector_size)
1874 /* Check if we have a 'bad_lba' information */
1875 if (!scsi_get_sense_info_fld(scmd->sense_buffer,
1876 SCSI_SENSE_BUFFERSIZE,
1881 * If the bad lba was reported incorrectly, we have no idea where
1884 start_lba = sectors_to_logical(sdev, blk_rq_pos(req));
1885 end_lba = start_lba + bytes_to_logical(sdev, scsi_bufflen(scmd));
1886 if (bad_lba < start_lba || bad_lba >= end_lba)
1890 * resid is optional but mostly filled in. When it's unused,
1891 * its value is zero, so we assume the whole buffer transferred
1893 transferred = scsi_bufflen(scmd) - scsi_get_resid(scmd);
1895 /* This computation should always be done in terms of the
1896 * resolution of the device's medium.
1898 good_bytes = logical_to_bytes(sdev, bad_lba - start_lba);
1900 return min(good_bytes, transferred);
1904 * sd_done - bottom half handler: called when the lower level
1905 * driver has completed (successfully or otherwise) a scsi command.
1906 * @SCpnt: mid-level's per command structure.
1908 * Note: potentially run from within an ISR. Must not block.
1910 static int sd_done(struct scsi_cmnd *SCpnt)
1912 int result = SCpnt->result;
1913 unsigned int good_bytes = result ? 0 : scsi_bufflen(SCpnt);
1914 unsigned int sector_size = SCpnt->device->sector_size;
1916 struct scsi_sense_hdr sshdr;
1917 struct scsi_disk *sdkp = scsi_disk(SCpnt->request->rq_disk);
1918 struct request *req = SCpnt->request;
1919 int sense_valid = 0;
1920 int sense_deferred = 0;
1922 switch (req_op(req)) {
1923 case REQ_OP_DISCARD:
1924 case REQ_OP_WRITE_ZEROES:
1925 case REQ_OP_WRITE_SAME:
1926 case REQ_OP_ZONE_RESET:
1928 good_bytes = blk_rq_bytes(req);
1929 scsi_set_resid(SCpnt, 0);
1932 scsi_set_resid(SCpnt, blk_rq_bytes(req));
1935 case REQ_OP_ZONE_REPORT:
1937 good_bytes = scsi_bufflen(SCpnt)
1938 - scsi_get_resid(SCpnt);
1939 scsi_set_resid(SCpnt, 0);
1942 scsi_set_resid(SCpnt, blk_rq_bytes(req));
1947 * In case of bogus fw or device, we could end up having
1948 * an unaligned partial completion. Check this here and force
1951 resid = scsi_get_resid(SCpnt);
1952 if (resid & (sector_size - 1)) {
1953 sd_printk(KERN_INFO, sdkp,
1954 "Unaligned partial completion (resid=%u, sector_sz=%u)\n",
1955 resid, sector_size);
1956 resid = min(scsi_bufflen(SCpnt),
1957 round_up(resid, sector_size));
1958 scsi_set_resid(SCpnt, resid);
1963 sense_valid = scsi_command_normalize_sense(SCpnt, &sshdr);
1965 sense_deferred = scsi_sense_is_deferred(&sshdr);
1967 sdkp->medium_access_timed_out = 0;
1969 if (driver_byte(result) != DRIVER_SENSE &&
1970 (!sense_valid || sense_deferred))
1973 switch (sshdr.sense_key) {
1974 case HARDWARE_ERROR:
1976 good_bytes = sd_completed_bytes(SCpnt);
1978 case RECOVERED_ERROR:
1979 good_bytes = scsi_bufflen(SCpnt);
1982 /* This indicates a false check condition, so ignore it. An
1983 * unknown amount of data was transferred so treat it as an
1987 memset(SCpnt->sense_buffer, 0, SCSI_SENSE_BUFFERSIZE);
1989 case ABORTED_COMMAND:
1990 if (sshdr.asc == 0x10) /* DIF: Target detected corruption */
1991 good_bytes = sd_completed_bytes(SCpnt);
1993 case ILLEGAL_REQUEST:
1994 switch (sshdr.asc) {
1995 case 0x10: /* DIX: Host detected corruption */
1996 good_bytes = sd_completed_bytes(SCpnt);
1998 case 0x20: /* INVALID COMMAND OPCODE */
1999 case 0x24: /* INVALID FIELD IN CDB */
2000 switch (SCpnt->cmnd[0]) {
2002 sd_config_discard(sdkp, SD_LBP_DISABLE);
2006 if (SCpnt->cmnd[1] & 8) { /* UNMAP */
2007 sd_config_discard(sdkp, SD_LBP_DISABLE);
2009 sdkp->device->no_write_same = 1;
2010 sd_config_write_same(sdkp);
2011 req->__data_len = blk_rq_bytes(req);
2012 req->rq_flags |= RQF_QUIET;
2023 if (sd_is_zoned(sdkp))
2024 sd_zbc_complete(SCpnt, good_bytes, &sshdr);
2026 SCSI_LOG_HLCOMPLETE(1, scmd_printk(KERN_INFO, SCpnt,
2027 "sd_done: completed %d of %d bytes\n",
2028 good_bytes, scsi_bufflen(SCpnt)));
2030 if (rq_data_dir(SCpnt->request) == READ && scsi_prot_sg_count(SCpnt))
2031 sd_dif_complete(SCpnt, good_bytes);
2037 * spinup disk - called only in sd_revalidate_disk()
2040 sd_spinup_disk(struct scsi_disk *sdkp)
2042 unsigned char cmd[10];
2043 unsigned long spintime_expire = 0;
2044 int retries, spintime;
2045 unsigned int the_result;
2046 struct scsi_sense_hdr sshdr;
2047 int sense_valid = 0;
2051 /* Spin up drives, as required. Only do this at boot time */
2052 /* Spinup needs to be done for module loads too. */
2057 cmd[0] = TEST_UNIT_READY;
2058 memset((void *) &cmd[1], 0, 9);
2060 the_result = scsi_execute_req(sdkp->device, cmd,
2063 SD_MAX_RETRIES, NULL);
2066 * If the drive has indicated to us that it
2067 * doesn't have any media in it, don't bother
2068 * with any more polling.
2070 if (media_not_present(sdkp, &sshdr))
2074 sense_valid = scsi_sense_valid(&sshdr);
2076 } while (retries < 3 &&
2077 (!scsi_status_is_good(the_result) ||
2078 ((driver_byte(the_result) & DRIVER_SENSE) &&
2079 sense_valid && sshdr.sense_key == UNIT_ATTENTION)));
2081 if ((driver_byte(the_result) & DRIVER_SENSE) == 0) {
2082 /* no sense, TUR either succeeded or failed
2083 * with a status error */
2084 if(!spintime && !scsi_status_is_good(the_result)) {
2085 sd_print_result(sdkp, "Test Unit Ready failed",
2092 * The device does not want the automatic start to be issued.
2094 if (sdkp->device->no_start_on_add)
2097 if (sense_valid && sshdr.sense_key == NOT_READY) {
2098 if (sshdr.asc == 4 && sshdr.ascq == 3)
2099 break; /* manual intervention required */
2100 if (sshdr.asc == 4 && sshdr.ascq == 0xb)
2101 break; /* standby */
2102 if (sshdr.asc == 4 && sshdr.ascq == 0xc)
2103 break; /* unavailable */
2105 * Issue command to spin up drive when not ready
2108 sd_printk(KERN_NOTICE, sdkp, "Spinning up disk...");
2109 cmd[0] = START_STOP;
2110 cmd[1] = 1; /* Return immediately */
2111 memset((void *) &cmd[2], 0, 8);
2112 cmd[4] = 1; /* Start spin cycle */
2113 if (sdkp->device->start_stop_pwr_cond)
2115 scsi_execute_req(sdkp->device, cmd, DMA_NONE,
2117 SD_TIMEOUT, SD_MAX_RETRIES,
2119 spintime_expire = jiffies + 100 * HZ;
2122 /* Wait 1 second for next try */
2127 * Wait for USB flash devices with slow firmware.
2128 * Yes, this sense key/ASC combination shouldn't
2129 * occur here. It's characteristic of these devices.
2131 } else if (sense_valid &&
2132 sshdr.sense_key == UNIT_ATTENTION &&
2133 sshdr.asc == 0x28) {
2135 spintime_expire = jiffies + 5 * HZ;
2138 /* Wait 1 second for next try */
2141 /* we don't understand the sense code, so it's
2142 * probably pointless to loop */
2144 sd_printk(KERN_NOTICE, sdkp, "Unit Not Ready\n");
2145 sd_print_sense_hdr(sdkp, &sshdr);
2150 } while (spintime && time_before_eq(jiffies, spintime_expire));
2153 if (scsi_status_is_good(the_result))
2156 printk("not responding...\n");
2161 * Determine whether disk supports Data Integrity Field.
2163 static int sd_read_protection_type(struct scsi_disk *sdkp, unsigned char *buffer)
2165 struct scsi_device *sdp = sdkp->device;
2169 if (scsi_device_protection(sdp) == 0 || (buffer[12] & 1) == 0)
2172 type = ((buffer[12] >> 1) & 7) + 1; /* P_TYPE 0 = Type 1 */
2174 if (type > T10_PI_TYPE3_PROTECTION)
2176 else if (scsi_host_dif_capable(sdp->host, type))
2179 if (sdkp->first_scan || type != sdkp->protection_type)
2182 sd_printk(KERN_ERR, sdkp, "formatted with unsupported" \
2183 " protection type %u. Disabling disk!\n",
2187 sd_printk(KERN_NOTICE, sdkp,
2188 "Enabling DIF Type %u protection\n", type);
2191 sd_printk(KERN_NOTICE, sdkp,
2192 "Disabling DIF Type %u protection\n", type);
2196 sdkp->protection_type = type;
2201 static void read_capacity_error(struct scsi_disk *sdkp, struct scsi_device *sdp,
2202 struct scsi_sense_hdr *sshdr, int sense_valid,
2205 if (driver_byte(the_result) & DRIVER_SENSE)
2206 sd_print_sense_hdr(sdkp, sshdr);
2208 sd_printk(KERN_NOTICE, sdkp, "Sense not available.\n");
2211 * Set dirty bit for removable devices if not ready -
2212 * sometimes drives will not report this properly.
2214 if (sdp->removable &&
2215 sense_valid && sshdr->sense_key == NOT_READY)
2216 set_media_not_present(sdkp);
2219 * We used to set media_present to 0 here to indicate no media
2220 * in the drive, but some drives fail read capacity even with
2221 * media present, so we can't do that.
2223 sdkp->capacity = 0; /* unknown mapped to zero - as usual */
2227 #if RC16_LEN > SD_BUF_SIZE
2228 #error RC16_LEN must not be more than SD_BUF_SIZE
2231 #define READ_CAPACITY_RETRIES_ON_RESET 10
2234 * Ensure that we don't overflow sector_t when CONFIG_LBDAF is not set
2235 * and the reported logical block size is bigger than 512 bytes. Note
2236 * that last_sector is a u64 and therefore logical_to_sectors() is not
2239 static bool sd_addressable_capacity(u64 lba, unsigned int sector_size)
2241 u64 last_sector = (lba + 1ULL) << (ilog2(sector_size) - 9);
2243 if (sizeof(sector_t) == 4 && last_sector > U32_MAX)
2249 static int read_capacity_16(struct scsi_disk *sdkp, struct scsi_device *sdp,
2250 unsigned char *buffer)
2252 unsigned char cmd[16];
2253 struct scsi_sense_hdr sshdr;
2254 int sense_valid = 0;
2256 int retries = 3, reset_retries = READ_CAPACITY_RETRIES_ON_RESET;
2257 unsigned int alignment;
2258 unsigned long long lba;
2259 unsigned sector_size;
2261 if (sdp->no_read_capacity_16)
2266 cmd[0] = SERVICE_ACTION_IN_16;
2267 cmd[1] = SAI_READ_CAPACITY_16;
2269 memset(buffer, 0, RC16_LEN);
2271 the_result = scsi_execute_req(sdp, cmd, DMA_FROM_DEVICE,
2272 buffer, RC16_LEN, &sshdr,
2273 SD_TIMEOUT, SD_MAX_RETRIES, NULL);
2275 if (media_not_present(sdkp, &sshdr))
2279 sense_valid = scsi_sense_valid(&sshdr);
2281 sshdr.sense_key == ILLEGAL_REQUEST &&
2282 (sshdr.asc == 0x20 || sshdr.asc == 0x24) &&
2284 /* Invalid Command Operation Code or
2285 * Invalid Field in CDB, just retry
2286 * silently with RC10 */
2289 sshdr.sense_key == UNIT_ATTENTION &&
2290 sshdr.asc == 0x29 && sshdr.ascq == 0x00)
2291 /* Device reset might occur several times,
2292 * give it one more chance */
2293 if (--reset_retries > 0)
2298 } while (the_result && retries);
2301 sd_print_result(sdkp, "Read Capacity(16) failed", the_result);
2302 read_capacity_error(sdkp, sdp, &sshdr, sense_valid, the_result);
2306 sector_size = get_unaligned_be32(&buffer[8]);
2307 lba = get_unaligned_be64(&buffer[0]);
2309 if (sd_read_protection_type(sdkp, buffer) < 0) {
2314 if (!sd_addressable_capacity(lba, sector_size)) {
2315 sd_printk(KERN_ERR, sdkp, "Too big for this kernel. Use a "
2316 "kernel compiled with support for large block "
2322 /* Logical blocks per physical block exponent */
2323 sdkp->physical_block_size = (1 << (buffer[13] & 0xf)) * sector_size;
2326 sdkp->rc_basis = (buffer[12] >> 4) & 0x3;
2328 /* Lowest aligned logical block */
2329 alignment = ((buffer[14] & 0x3f) << 8 | buffer[15]) * sector_size;
2330 blk_queue_alignment_offset(sdp->request_queue, alignment);
2331 if (alignment && sdkp->first_scan)
2332 sd_printk(KERN_NOTICE, sdkp,
2333 "physical block alignment offset: %u\n", alignment);
2335 if (buffer[14] & 0x80) { /* LBPME */
2338 if (buffer[14] & 0x40) /* LBPRZ */
2341 sd_config_discard(sdkp, SD_LBP_WS16);
2344 sdkp->capacity = lba + 1;
2348 static int read_capacity_10(struct scsi_disk *sdkp, struct scsi_device *sdp,
2349 unsigned char *buffer)
2351 unsigned char cmd[16];
2352 struct scsi_sense_hdr sshdr;
2353 int sense_valid = 0;
2355 int retries = 3, reset_retries = READ_CAPACITY_RETRIES_ON_RESET;
2357 unsigned sector_size;
2360 cmd[0] = READ_CAPACITY;
2361 memset(&cmd[1], 0, 9);
2362 memset(buffer, 0, 8);
2364 the_result = scsi_execute_req(sdp, cmd, DMA_FROM_DEVICE,
2366 SD_TIMEOUT, SD_MAX_RETRIES, NULL);
2368 if (media_not_present(sdkp, &sshdr))
2372 sense_valid = scsi_sense_valid(&sshdr);
2374 sshdr.sense_key == UNIT_ATTENTION &&
2375 sshdr.asc == 0x29 && sshdr.ascq == 0x00)
2376 /* Device reset might occur several times,
2377 * give it one more chance */
2378 if (--reset_retries > 0)
2383 } while (the_result && retries);
2386 sd_print_result(sdkp, "Read Capacity(10) failed", the_result);
2387 read_capacity_error(sdkp, sdp, &sshdr, sense_valid, the_result);
2391 sector_size = get_unaligned_be32(&buffer[4]);
2392 lba = get_unaligned_be32(&buffer[0]);
2394 if (sdp->no_read_capacity_16 && (lba == 0xffffffff)) {
2395 /* Some buggy (usb cardreader) devices return an lba of
2396 0xffffffff when the want to report a size of 0 (with
2397 which they really mean no media is present) */
2399 sdkp->physical_block_size = sector_size;
2403 if (!sd_addressable_capacity(lba, sector_size)) {
2404 sd_printk(KERN_ERR, sdkp, "Too big for this kernel. Use a "
2405 "kernel compiled with support for large block "
2411 sdkp->capacity = lba + 1;
2412 sdkp->physical_block_size = sector_size;
2416 static int sd_try_rc16_first(struct scsi_device *sdp)
2418 if (sdp->host->max_cmd_len < 16)
2420 if (sdp->try_rc_10_first)
2422 if (sdp->scsi_level > SCSI_SPC_2)
2424 if (scsi_device_protection(sdp))
2430 * read disk capacity
2433 sd_read_capacity(struct scsi_disk *sdkp, unsigned char *buffer)
2436 struct scsi_device *sdp = sdkp->device;
2438 if (sd_try_rc16_first(sdp)) {
2439 sector_size = read_capacity_16(sdkp, sdp, buffer);
2440 if (sector_size == -EOVERFLOW)
2442 if (sector_size == -ENODEV)
2444 if (sector_size < 0)
2445 sector_size = read_capacity_10(sdkp, sdp, buffer);
2446 if (sector_size < 0)
2449 sector_size = read_capacity_10(sdkp, sdp, buffer);
2450 if (sector_size == -EOVERFLOW)
2452 if (sector_size < 0)
2454 if ((sizeof(sdkp->capacity) > 4) &&
2455 (sdkp->capacity > 0xffffffffULL)) {
2456 int old_sector_size = sector_size;
2457 sd_printk(KERN_NOTICE, sdkp, "Very big device. "
2458 "Trying to use READ CAPACITY(16).\n");
2459 sector_size = read_capacity_16(sdkp, sdp, buffer);
2460 if (sector_size < 0) {
2461 sd_printk(KERN_NOTICE, sdkp,
2462 "Using 0xffffffff as device size\n");
2463 sdkp->capacity = 1 + (sector_t) 0xffffffff;
2464 sector_size = old_sector_size;
2470 /* Some devices are known to return the total number of blocks,
2471 * not the highest block number. Some devices have versions
2472 * which do this and others which do not. Some devices we might
2473 * suspect of doing this but we don't know for certain.
2475 * If we know the reported capacity is wrong, decrement it. If
2476 * we can only guess, then assume the number of blocks is even
2477 * (usually true but not always) and err on the side of lowering
2480 if (sdp->fix_capacity ||
2481 (sdp->guess_capacity && (sdkp->capacity & 0x01))) {
2482 sd_printk(KERN_INFO, sdkp, "Adjusting the sector count "
2483 "from its reported value: %llu\n",
2484 (unsigned long long) sdkp->capacity);
2489 if (sector_size == 0) {
2491 sd_printk(KERN_NOTICE, sdkp, "Sector size 0 reported, "
2495 if (sector_size != 512 &&
2496 sector_size != 1024 &&
2497 sector_size != 2048 &&
2498 sector_size != 4096) {
2499 sd_printk(KERN_NOTICE, sdkp, "Unsupported sector size %d.\n",
2502 * The user might want to re-format the drive with
2503 * a supported sectorsize. Once this happens, it
2504 * would be relatively trivial to set the thing up.
2505 * For this reason, we leave the thing in the table.
2509 * set a bogus sector size so the normal read/write
2510 * logic in the block layer will eventually refuse any
2511 * request on this device without tripping over power
2512 * of two sector size assumptions
2516 blk_queue_logical_block_size(sdp->request_queue, sector_size);
2517 blk_queue_physical_block_size(sdp->request_queue,
2518 sdkp->physical_block_size);
2519 sdkp->device->sector_size = sector_size;
2521 if (sdkp->capacity > 0xffffffff)
2522 sdp->use_16_for_rw = 1;
2527 * Print disk capacity
2530 sd_print_capacity(struct scsi_disk *sdkp,
2531 sector_t old_capacity)
2533 int sector_size = sdkp->device->sector_size;
2534 char cap_str_2[10], cap_str_10[10];
2536 string_get_size(sdkp->capacity, sector_size,
2537 STRING_UNITS_2, cap_str_2, sizeof(cap_str_2));
2538 string_get_size(sdkp->capacity, sector_size,
2539 STRING_UNITS_10, cap_str_10,
2540 sizeof(cap_str_10));
2542 if (sdkp->first_scan || old_capacity != sdkp->capacity) {
2543 sd_printk(KERN_NOTICE, sdkp,
2544 "%llu %d-byte logical blocks: (%s/%s)\n",
2545 (unsigned long long)sdkp->capacity,
2546 sector_size, cap_str_10, cap_str_2);
2548 if (sdkp->physical_block_size != sector_size)
2549 sd_printk(KERN_NOTICE, sdkp,
2550 "%u-byte physical blocks\n",
2551 sdkp->physical_block_size);
2553 sd_zbc_print_zones(sdkp);
2557 /* called with buffer of length 512 */
2559 sd_do_mode_sense(struct scsi_device *sdp, int dbd, int modepage,
2560 unsigned char *buffer, int len, struct scsi_mode_data *data,
2561 struct scsi_sense_hdr *sshdr)
2563 return scsi_mode_sense(sdp, dbd, modepage, buffer, len,
2564 SD_TIMEOUT, SD_MAX_RETRIES, data,
2569 * read write protect setting, if possible - called only in sd_revalidate_disk()
2570 * called with buffer of length SD_BUF_SIZE
2573 sd_read_write_protect_flag(struct scsi_disk *sdkp, unsigned char *buffer)
2576 struct scsi_device *sdp = sdkp->device;
2577 struct scsi_mode_data data;
2578 int old_wp = sdkp->write_prot;
2580 set_disk_ro(sdkp->disk, 0);
2581 if (sdp->skip_ms_page_3f) {
2582 sd_first_printk(KERN_NOTICE, sdkp, "Assuming Write Enabled\n");
2586 if (sdp->use_192_bytes_for_3f) {
2587 res = sd_do_mode_sense(sdp, 0, 0x3F, buffer, 192, &data, NULL);
2590 * First attempt: ask for all pages (0x3F), but only 4 bytes.
2591 * We have to start carefully: some devices hang if we ask
2592 * for more than is available.
2594 res = sd_do_mode_sense(sdp, 0, 0x3F, buffer, 4, &data, NULL);
2597 * Second attempt: ask for page 0 When only page 0 is
2598 * implemented, a request for page 3F may return Sense Key
2599 * 5: Illegal Request, Sense Code 24: Invalid field in
2602 if (!scsi_status_is_good(res))
2603 res = sd_do_mode_sense(sdp, 0, 0, buffer, 4, &data, NULL);
2606 * Third attempt: ask 255 bytes, as we did earlier.
2608 if (!scsi_status_is_good(res))
2609 res = sd_do_mode_sense(sdp, 0, 0x3F, buffer, 255,
2613 if (!scsi_status_is_good(res)) {
2614 sd_first_printk(KERN_WARNING, sdkp,
2615 "Test WP failed, assume Write Enabled\n");
2617 sdkp->write_prot = ((data.device_specific & 0x80) != 0);
2618 set_disk_ro(sdkp->disk, sdkp->write_prot);
2619 if (sdkp->first_scan || old_wp != sdkp->write_prot) {
2620 sd_printk(KERN_NOTICE, sdkp, "Write Protect is %s\n",
2621 sdkp->write_prot ? "on" : "off");
2622 sd_printk(KERN_DEBUG, sdkp, "Mode Sense: %4ph\n", buffer);
2628 * sd_read_cache_type - called only from sd_revalidate_disk()
2629 * called with buffer of length SD_BUF_SIZE
2632 sd_read_cache_type(struct scsi_disk *sdkp, unsigned char *buffer)
2635 struct scsi_device *sdp = sdkp->device;
2640 struct scsi_mode_data data;
2641 struct scsi_sense_hdr sshdr;
2642 int old_wce = sdkp->WCE;
2643 int old_rcd = sdkp->RCD;
2644 int old_dpofua = sdkp->DPOFUA;
2647 if (sdkp->cache_override)
2651 if (sdp->skip_ms_page_8) {
2652 if (sdp->type == TYPE_RBC)
2655 if (sdp->skip_ms_page_3f)
2658 if (sdp->use_192_bytes_for_3f)
2662 } else if (sdp->type == TYPE_RBC) {
2670 /* cautiously ask */
2671 res = sd_do_mode_sense(sdp, dbd, modepage, buffer, first_len,
2674 if (!scsi_status_is_good(res))
2677 if (!data.header_length) {
2680 sd_first_printk(KERN_ERR, sdkp,
2681 "Missing header in MODE_SENSE response\n");
2684 /* that went OK, now ask for the proper length */
2688 * We're only interested in the first three bytes, actually.
2689 * But the data cache page is defined for the first 20.
2693 else if (len > SD_BUF_SIZE) {
2694 sd_first_printk(KERN_NOTICE, sdkp, "Truncating mode parameter "
2695 "data from %d to %d bytes\n", len, SD_BUF_SIZE);
2698 if (modepage == 0x3F && sdp->use_192_bytes_for_3f)
2702 if (len > first_len)
2703 res = sd_do_mode_sense(sdp, dbd, modepage, buffer, len,
2706 if (scsi_status_is_good(res)) {
2707 int offset = data.header_length + data.block_descriptor_length;
2709 while (offset < len) {
2710 u8 page_code = buffer[offset] & 0x3F;
2711 u8 spf = buffer[offset] & 0x40;
2713 if (page_code == 8 || page_code == 6) {
2714 /* We're interested only in the first 3 bytes.
2716 if (len - offset <= 2) {
2717 sd_first_printk(KERN_ERR, sdkp,
2718 "Incomplete mode parameter "
2722 modepage = page_code;
2726 /* Go to the next page */
2727 if (spf && len - offset > 3)
2728 offset += 4 + (buffer[offset+2] << 8) +
2730 else if (!spf && len - offset > 1)
2731 offset += 2 + buffer[offset+1];
2733 sd_first_printk(KERN_ERR, sdkp,
2735 "parameter data\n");
2741 sd_first_printk(KERN_ERR, sdkp, "No Caching mode page found\n");
2745 if (modepage == 8) {
2746 sdkp->WCE = ((buffer[offset + 2] & 0x04) != 0);
2747 sdkp->RCD = ((buffer[offset + 2] & 0x01) != 0);
2749 sdkp->WCE = ((buffer[offset + 2] & 0x01) == 0);
2753 sdkp->DPOFUA = (data.device_specific & 0x10) != 0;
2754 if (sdp->broken_fua) {
2755 sd_first_printk(KERN_NOTICE, sdkp, "Disabling FUA\n");
2757 } else if (sdkp->DPOFUA && !sdkp->device->use_10_for_rw &&
2758 !sdkp->device->use_16_for_rw) {
2759 sd_first_printk(KERN_NOTICE, sdkp,
2760 "Uses READ/WRITE(6), disabling FUA\n");
2764 /* No cache flush allowed for write protected devices */
2765 if (sdkp->WCE && sdkp->write_prot)
2768 if (sdkp->first_scan || old_wce != sdkp->WCE ||
2769 old_rcd != sdkp->RCD || old_dpofua != sdkp->DPOFUA)
2770 sd_printk(KERN_NOTICE, sdkp,
2771 "Write cache: %s, read cache: %s, %s\n",
2772 sdkp->WCE ? "enabled" : "disabled",
2773 sdkp->RCD ? "disabled" : "enabled",
2774 sdkp->DPOFUA ? "supports DPO and FUA"
2775 : "doesn't support DPO or FUA");
2781 if (scsi_sense_valid(&sshdr) &&
2782 sshdr.sense_key == ILLEGAL_REQUEST &&
2783 sshdr.asc == 0x24 && sshdr.ascq == 0x0)
2784 /* Invalid field in CDB */
2785 sd_first_printk(KERN_NOTICE, sdkp, "Cache data unavailable\n");
2787 sd_first_printk(KERN_ERR, sdkp,
2788 "Asking for cache data failed\n");
2791 if (sdp->wce_default_on) {
2792 sd_first_printk(KERN_NOTICE, sdkp,
2793 "Assuming drive cache: write back\n");
2796 sd_first_printk(KERN_ERR, sdkp,
2797 "Assuming drive cache: write through\n");
2805 * The ATO bit indicates whether the DIF application tag is available
2806 * for use by the operating system.
2808 static void sd_read_app_tag_own(struct scsi_disk *sdkp, unsigned char *buffer)
2811 struct scsi_device *sdp = sdkp->device;
2812 struct scsi_mode_data data;
2813 struct scsi_sense_hdr sshdr;
2815 if (sdp->type != TYPE_DISK && sdp->type != TYPE_ZBC)
2818 if (sdkp->protection_type == 0)
2821 res = scsi_mode_sense(sdp, 1, 0x0a, buffer, 36, SD_TIMEOUT,
2822 SD_MAX_RETRIES, &data, &sshdr);
2824 if (!scsi_status_is_good(res) || !data.header_length ||
2826 sd_first_printk(KERN_WARNING, sdkp,
2827 "getting Control mode page failed, assume no ATO\n");
2829 if (scsi_sense_valid(&sshdr))
2830 sd_print_sense_hdr(sdkp, &sshdr);
2835 offset = data.header_length + data.block_descriptor_length;
2837 if ((buffer[offset] & 0x3f) != 0x0a) {
2838 sd_first_printk(KERN_ERR, sdkp, "ATO Got wrong page\n");
2842 if ((buffer[offset + 5] & 0x80) == 0)
2851 * sd_read_block_limits - Query disk device for preferred I/O sizes.
2852 * @sdkp: disk to query
2854 static void sd_read_block_limits(struct scsi_disk *sdkp)
2856 unsigned int sector_sz = sdkp->device->sector_size;
2857 const int vpd_len = 64;
2858 unsigned char *buffer = kmalloc(vpd_len, GFP_KERNEL);
2861 /* Block Limits VPD */
2862 scsi_get_vpd_page(sdkp->device, 0xb0, buffer, vpd_len))
2865 blk_queue_io_min(sdkp->disk->queue,
2866 get_unaligned_be16(&buffer[6]) * sector_sz);
2868 sdkp->max_xfer_blocks = get_unaligned_be32(&buffer[8]);
2869 sdkp->opt_xfer_blocks = get_unaligned_be32(&buffer[12]);
2871 if (buffer[3] == 0x3c) {
2872 unsigned int lba_count, desc_count;
2874 sdkp->max_ws_blocks = (u32)get_unaligned_be64(&buffer[36]);
2879 lba_count = get_unaligned_be32(&buffer[20]);
2880 desc_count = get_unaligned_be32(&buffer[24]);
2882 if (lba_count && desc_count)
2883 sdkp->max_unmap_blocks = lba_count;
2885 sdkp->unmap_granularity = get_unaligned_be32(&buffer[28]);
2887 if (buffer[32] & 0x80)
2888 sdkp->unmap_alignment =
2889 get_unaligned_be32(&buffer[32]) & ~(1 << 31);
2891 if (!sdkp->lbpvpd) { /* LBP VPD page not provided */
2893 if (sdkp->max_unmap_blocks)
2894 sd_config_discard(sdkp, SD_LBP_UNMAP);
2896 sd_config_discard(sdkp, SD_LBP_WS16);
2898 } else { /* LBP VPD page tells us what to use */
2899 if (sdkp->lbpu && sdkp->max_unmap_blocks)
2900 sd_config_discard(sdkp, SD_LBP_UNMAP);
2901 else if (sdkp->lbpws)
2902 sd_config_discard(sdkp, SD_LBP_WS16);
2903 else if (sdkp->lbpws10)
2904 sd_config_discard(sdkp, SD_LBP_WS10);
2905 else if (sdkp->lbpu && sdkp->max_unmap_blocks)
2906 sd_config_discard(sdkp, SD_LBP_UNMAP);
2908 sd_config_discard(sdkp, SD_LBP_DISABLE);
2917 * sd_read_block_characteristics - Query block dev. characteristics
2918 * @sdkp: disk to query
2920 static void sd_read_block_characteristics(struct scsi_disk *sdkp)
2922 struct request_queue *q = sdkp->disk->queue;
2923 unsigned char *buffer;
2925 const int vpd_len = 64;
2927 buffer = kmalloc(vpd_len, GFP_KERNEL);
2930 /* Block Device Characteristics VPD */
2931 scsi_get_vpd_page(sdkp->device, 0xb1, buffer, vpd_len))
2934 rot = get_unaligned_be16(&buffer[4]);
2937 queue_flag_set_unlocked(QUEUE_FLAG_NONROT, q);
2938 queue_flag_clear_unlocked(QUEUE_FLAG_ADD_RANDOM, q);
2941 if (sdkp->device->type == TYPE_ZBC) {
2943 q->limits.zoned = BLK_ZONED_HM;
2945 sdkp->zoned = (buffer[8] >> 4) & 3;
2946 if (sdkp->zoned == 1)
2948 q->limits.zoned = BLK_ZONED_HA;
2951 * Treat drive-managed devices as
2952 * regular block devices.
2954 q->limits.zoned = BLK_ZONED_NONE;
2956 if (blk_queue_is_zoned(q) && sdkp->first_scan)
2957 sd_printk(KERN_NOTICE, sdkp, "Host-%s zoned block device\n",
2958 q->limits.zoned == BLK_ZONED_HM ? "managed" : "aware");
2965 * sd_read_block_provisioning - Query provisioning VPD page
2966 * @sdkp: disk to query
2968 static void sd_read_block_provisioning(struct scsi_disk *sdkp)
2970 unsigned char *buffer;
2971 const int vpd_len = 8;
2973 if (sdkp->lbpme == 0)
2976 buffer = kmalloc(vpd_len, GFP_KERNEL);
2978 if (!buffer || scsi_get_vpd_page(sdkp->device, 0xb2, buffer, vpd_len))
2982 sdkp->lbpu = (buffer[5] >> 7) & 1; /* UNMAP */
2983 sdkp->lbpws = (buffer[5] >> 6) & 1; /* WRITE SAME(16) with UNMAP */
2984 sdkp->lbpws10 = (buffer[5] >> 5) & 1; /* WRITE SAME(10) with UNMAP */
2990 static void sd_read_write_same(struct scsi_disk *sdkp, unsigned char *buffer)
2992 struct scsi_device *sdev = sdkp->device;
2994 if (sdev->host->no_write_same) {
2995 sdev->no_write_same = 1;
3000 if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE, INQUIRY) < 0) {
3001 /* too large values might cause issues with arcmsr */
3002 int vpd_buf_len = 64;
3004 sdev->no_report_opcodes = 1;
3006 /* Disable WRITE SAME if REPORT SUPPORTED OPERATION
3007 * CODES is unsupported and the device has an ATA
3008 * Information VPD page (SAT).
3010 if (!scsi_get_vpd_page(sdev, 0x89, buffer, vpd_buf_len))
3011 sdev->no_write_same = 1;
3014 if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE, WRITE_SAME_16) == 1)
3017 if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE, WRITE_SAME) == 1)
3021 static void sd_read_security(struct scsi_disk *sdkp, unsigned char *buffer)
3023 struct scsi_device *sdev = sdkp->device;
3025 if (!sdev->security_supported)
3028 if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE,
3029 SECURITY_PROTOCOL_IN) == 1 &&
3030 scsi_report_opcode(sdev, buffer, SD_BUF_SIZE,
3031 SECURITY_PROTOCOL_OUT) == 1)
3036 * sd_revalidate_disk - called the first time a new disk is seen,
3037 * performs disk spin up, read_capacity, etc.
3038 * @disk: struct gendisk we care about
3040 static int sd_revalidate_disk(struct gendisk *disk)
3042 struct scsi_disk *sdkp = scsi_disk(disk);
3043 struct scsi_device *sdp = sdkp->device;
3044 struct request_queue *q = sdkp->disk->queue;
3045 sector_t old_capacity = sdkp->capacity;
3046 unsigned char *buffer;
3047 unsigned int dev_max, rw_max;
3049 SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp,
3050 "sd_revalidate_disk\n"));
3053 * If the device is offline, don't try and read capacity or any
3054 * of the other niceties.
3056 if (!scsi_device_online(sdp))
3059 buffer = kmalloc(SD_BUF_SIZE, GFP_KERNEL);
3061 sd_printk(KERN_WARNING, sdkp, "sd_revalidate_disk: Memory "
3062 "allocation failure.\n");
3066 sd_spinup_disk(sdkp);
3069 * Without media there is no reason to ask; moreover, some devices
3070 * react badly if we do.
3072 if (sdkp->media_present) {
3073 sd_read_capacity(sdkp, buffer);
3075 if (scsi_device_supports_vpd(sdp)) {
3076 sd_read_block_provisioning(sdkp);
3077 sd_read_block_limits(sdkp);
3078 sd_read_block_characteristics(sdkp);
3079 sd_zbc_read_zones(sdkp, buffer);
3082 sd_print_capacity(sdkp, old_capacity);
3084 sd_read_write_protect_flag(sdkp, buffer);
3085 sd_read_cache_type(sdkp, buffer);
3086 sd_read_app_tag_own(sdkp, buffer);
3087 sd_read_write_same(sdkp, buffer);
3088 sd_read_security(sdkp, buffer);
3091 sdkp->first_scan = 0;
3094 * We now have all cache related info, determine how we deal
3095 * with flush requests.
3097 sd_set_flush_flag(sdkp);
3099 /* Initial block count limit based on CDB TRANSFER LENGTH field size. */
3100 dev_max = sdp->use_16_for_rw ? SD_MAX_XFER_BLOCKS : SD_DEF_XFER_BLOCKS;
3102 /* Some devices report a maximum block count for READ/WRITE requests. */
3103 dev_max = min_not_zero(dev_max, sdkp->max_xfer_blocks);
3104 q->limits.max_dev_sectors = logical_to_sectors(sdp, dev_max);
3107 * Use the device's preferred I/O size for reads and writes
3108 * unless the reported value is unreasonably small, large, or
3111 if (sdkp->opt_xfer_blocks &&
3112 sdkp->opt_xfer_blocks <= dev_max &&
3113 sdkp->opt_xfer_blocks <= SD_DEF_XFER_BLOCKS &&
3114 logical_to_bytes(sdp, sdkp->opt_xfer_blocks) >= PAGE_SIZE) {
3115 q->limits.io_opt = logical_to_bytes(sdp, sdkp->opt_xfer_blocks);
3116 rw_max = logical_to_sectors(sdp, sdkp->opt_xfer_blocks);
3118 rw_max = min_not_zero(logical_to_sectors(sdp, dev_max),
3119 (sector_t)BLK_DEF_MAX_SECTORS);
3121 /* Combine with controller limits */
3122 q->limits.max_sectors = min(rw_max, queue_max_hw_sectors(q));
3124 set_capacity(disk, logical_to_sectors(sdp, sdkp->capacity));
3125 sd_config_write_same(sdkp);
3133 * sd_unlock_native_capacity - unlock native capacity
3134 * @disk: struct gendisk to set capacity for
3136 * Block layer calls this function if it detects that partitions
3137 * on @disk reach beyond the end of the device. If the SCSI host
3138 * implements ->unlock_native_capacity() method, it's invoked to
3139 * give it a chance to adjust the device capacity.
3142 * Defined by block layer. Might sleep.
3144 static void sd_unlock_native_capacity(struct gendisk *disk)
3146 struct scsi_device *sdev = scsi_disk(disk)->device;
3148 if (sdev->host->hostt->unlock_native_capacity)
3149 sdev->host->hostt->unlock_native_capacity(sdev);
3153 * sd_format_disk_name - format disk name
3154 * @prefix: name prefix - ie. "sd" for SCSI disks
3155 * @index: index of the disk to format name for
3156 * @buf: output buffer
3157 * @buflen: length of the output buffer
3159 * SCSI disk names starts at sda. The 26th device is sdz and the
3160 * 27th is sdaa. The last one for two lettered suffix is sdzz
3161 * which is followed by sdaaa.
3163 * This is basically 26 base counting with one extra 'nil' entry
3164 * at the beginning from the second digit on and can be
3165 * determined using similar method as 26 base conversion with the
3166 * index shifted -1 after each digit is computed.
3172 * 0 on success, -errno on failure.
3174 static int sd_format_disk_name(char *prefix, int index, char *buf, int buflen)
3176 const int base = 'z' - 'a' + 1;
3177 char *begin = buf + strlen(prefix);
3178 char *end = buf + buflen;
3188 *--p = 'a' + (index % unit);
3189 index = (index / unit) - 1;
3190 } while (index >= 0);
3192 memmove(begin, p, end - p);
3193 memcpy(buf, prefix, strlen(prefix));
3199 * The asynchronous part of sd_probe
3201 static void sd_probe_async(void *data, async_cookie_t cookie)
3203 struct scsi_disk *sdkp = data;
3204 struct scsi_device *sdp;
3211 index = sdkp->index;
3212 dev = &sdp->sdev_gendev;
3214 gd->major = sd_major((index & 0xf0) >> 4);
3215 gd->first_minor = ((index & 0xf) << 4) | (index & 0xfff00);
3216 gd->minors = SD_MINORS;
3218 gd->fops = &sd_fops;
3219 gd->private_data = &sdkp->driver;
3220 gd->queue = sdkp->device->request_queue;
3222 /* defaults, until the device tells us otherwise */
3223 sdp->sector_size = 512;
3225 sdkp->media_present = 1;
3226 sdkp->write_prot = 0;
3227 sdkp->cache_override = 0;
3231 sdkp->first_scan = 1;
3232 sdkp->max_medium_access_timeouts = SD_MAX_MEDIUM_TIMEOUTS;
3234 sd_revalidate_disk(gd);
3236 gd->flags = GENHD_FL_EXT_DEVT;
3237 if (sdp->removable) {
3238 gd->flags |= GENHD_FL_REMOVABLE;
3239 gd->events |= DISK_EVENT_MEDIA_CHANGE;
3242 blk_pm_runtime_init(sdp->request_queue, dev);
3243 device_add_disk(dev, gd);
3245 sd_dif_config_host(sdkp);
3247 sd_revalidate_disk(gd);
3249 if (sdkp->security) {
3250 sdkp->opal_dev = init_opal_dev(sdp, &sd_sec_submit);
3252 sd_printk(KERN_NOTICE, sdkp, "supports TCG Opal\n");
3255 sd_printk(KERN_NOTICE, sdkp, "Attached SCSI %sdisk\n",
3256 sdp->removable ? "removable " : "");
3257 scsi_autopm_put_device(sdp);
3258 put_device(&sdkp->dev);
3262 * sd_probe - called during driver initialization and whenever a
3263 * new scsi device is attached to the system. It is called once
3264 * for each scsi device (not just disks) present.
3265 * @dev: pointer to device object
3267 * Returns 0 if successful (or not interested in this scsi device
3268 * (e.g. scanner)); 1 when there is an error.
3270 * Note: this function is invoked from the scsi mid-level.
3271 * This function sets up the mapping between a given
3272 * <host,channel,id,lun> (found in sdp) and new device name
3273 * (e.g. /dev/sda). More precisely it is the block device major
3274 * and minor number that is chosen here.
3276 * Assume sd_probe is not re-entrant (for time being)
3277 * Also think about sd_probe() and sd_remove() running coincidentally.
3279 static int sd_probe(struct device *dev)
3281 struct scsi_device *sdp = to_scsi_device(dev);
3282 struct scsi_disk *sdkp;
3287 scsi_autopm_get_device(sdp);
3289 if (sdp->type != TYPE_DISK &&
3290 sdp->type != TYPE_ZBC &&
3291 sdp->type != TYPE_MOD &&
3292 sdp->type != TYPE_RBC)
3295 #ifndef CONFIG_BLK_DEV_ZONED
3296 if (sdp->type == TYPE_ZBC)
3299 SCSI_LOG_HLQUEUE(3, sdev_printk(KERN_INFO, sdp,
3303 sdkp = kzalloc(sizeof(*sdkp), GFP_KERNEL);
3307 gd = alloc_disk(SD_MINORS);
3312 if (!ida_pre_get(&sd_index_ida, GFP_KERNEL))
3315 spin_lock(&sd_index_lock);
3316 error = ida_get_new(&sd_index_ida, &index);
3317 spin_unlock(&sd_index_lock);
3318 } while (error == -EAGAIN);
3321 sdev_printk(KERN_WARNING, sdp, "sd_probe: memory exhausted.\n");
3325 error = sd_format_disk_name("sd", index, gd->disk_name, DISK_NAME_LEN);
3327 sdev_printk(KERN_WARNING, sdp, "SCSI disk (sd) name length exceeded.\n");
3328 goto out_free_index;
3332 sdkp->driver = &sd_template;
3334 sdkp->index = index;
3335 atomic_set(&sdkp->openers, 0);
3336 atomic_set(&sdkp->device->ioerr_cnt, 0);
3338 if (!sdp->request_queue->rq_timeout) {
3339 if (sdp->type != TYPE_MOD)
3340 blk_queue_rq_timeout(sdp->request_queue, SD_TIMEOUT);
3342 blk_queue_rq_timeout(sdp->request_queue,
3346 device_initialize(&sdkp->dev);
3347 sdkp->dev.parent = dev;
3348 sdkp->dev.class = &sd_disk_class;
3349 dev_set_name(&sdkp->dev, "%s", dev_name(dev));
3351 error = device_add(&sdkp->dev);
3353 goto out_free_index;
3356 dev_set_drvdata(dev, sdkp);
3358 get_device(&sdkp->dev); /* prevent release before async_schedule */
3359 async_schedule_domain(sd_probe_async, sdkp, &scsi_sd_probe_domain);
3364 spin_lock(&sd_index_lock);
3365 ida_remove(&sd_index_ida, index);
3366 spin_unlock(&sd_index_lock);
3372 scsi_autopm_put_device(sdp);
3377 * sd_remove - called whenever a scsi disk (previously recognized by
3378 * sd_probe) is detached from the system. It is called (potentially
3379 * multiple times) during sd module unload.
3380 * @dev: pointer to device object
3382 * Note: this function is invoked from the scsi mid-level.
3383 * This function potentially frees up a device name (e.g. /dev/sdc)
3384 * that could be re-used by a subsequent sd_probe().
3385 * This function is not called when the built-in sd driver is "exit-ed".
3387 static int sd_remove(struct device *dev)
3389 struct scsi_disk *sdkp;
3392 sdkp = dev_get_drvdata(dev);
3393 devt = disk_devt(sdkp->disk);
3394 scsi_autopm_get_device(sdkp->device);
3396 async_synchronize_full_domain(&scsi_sd_pm_domain);
3397 async_synchronize_full_domain(&scsi_sd_probe_domain);
3398 device_del(&sdkp->dev);
3399 del_gendisk(sdkp->disk);
3402 sd_zbc_remove(sdkp);
3404 free_opal_dev(sdkp->opal_dev);
3406 blk_register_region(devt, SD_MINORS, NULL,
3407 sd_default_probe, NULL, NULL);
3409 mutex_lock(&sd_ref_mutex);
3410 dev_set_drvdata(dev, NULL);
3411 put_device(&sdkp->dev);
3412 mutex_unlock(&sd_ref_mutex);
3418 * scsi_disk_release - Called to free the scsi_disk structure
3419 * @dev: pointer to embedded class device
3421 * sd_ref_mutex must be held entering this routine. Because it is
3422 * called on last put, you should always use the scsi_disk_get()
3423 * scsi_disk_put() helpers which manipulate the semaphore directly
3424 * and never do a direct put_device.
3426 static void scsi_disk_release(struct device *dev)
3428 struct scsi_disk *sdkp = to_scsi_disk(dev);
3429 struct gendisk *disk = sdkp->disk;
3431 spin_lock(&sd_index_lock);
3432 ida_remove(&sd_index_ida, sdkp->index);
3433 spin_unlock(&sd_index_lock);
3435 disk->private_data = NULL;
3437 put_device(&sdkp->device->sdev_gendev);
3442 static int sd_start_stop_device(struct scsi_disk *sdkp, int start)
3444 unsigned char cmd[6] = { START_STOP }; /* START_VALID */
3445 struct scsi_sense_hdr sshdr;
3446 struct scsi_device *sdp = sdkp->device;
3450 cmd[4] |= 1; /* START */
3452 if (sdp->start_stop_pwr_cond)
3453 cmd[4] |= start ? 1 << 4 : 3 << 4; /* Active or Standby */
3455 if (!scsi_device_online(sdp))
3458 res = scsi_execute(sdp, cmd, DMA_NONE, NULL, 0, NULL, &sshdr,
3459 SD_TIMEOUT, SD_MAX_RETRIES, 0, RQF_PM, NULL);
3461 sd_print_result(sdkp, "Start/Stop Unit failed", res);
3462 if (driver_byte(res) & DRIVER_SENSE)
3463 sd_print_sense_hdr(sdkp, &sshdr);
3464 if (scsi_sense_valid(&sshdr) &&
3465 /* 0x3a is medium not present */
3470 /* SCSI error codes must not go to the generic layer */
3478 * Send a SYNCHRONIZE CACHE instruction down to the device through
3479 * the normal SCSI command structure. Wait for the command to
3482 static void sd_shutdown(struct device *dev)
3484 struct scsi_disk *sdkp = dev_get_drvdata(dev);
3487 return; /* this can happen */
3489 if (pm_runtime_suspended(dev))
3492 if (sdkp->WCE && sdkp->media_present) {
3493 sd_printk(KERN_NOTICE, sdkp, "Synchronizing SCSI cache\n");
3494 sd_sync_cache(sdkp);
3497 if (system_state != SYSTEM_RESTART && sdkp->device->manage_start_stop) {
3498 sd_printk(KERN_NOTICE, sdkp, "Stopping disk\n");
3499 sd_start_stop_device(sdkp, 0);
3503 static int sd_suspend_common(struct device *dev, bool ignore_stop_errors)
3505 struct scsi_disk *sdkp = dev_get_drvdata(dev);
3508 if (!sdkp) /* E.g.: runtime suspend following sd_remove() */
3511 if (sdkp->WCE && sdkp->media_present) {
3512 sd_printk(KERN_NOTICE, sdkp, "Synchronizing SCSI cache\n");
3513 ret = sd_sync_cache(sdkp);
3515 /* ignore OFFLINE device */
3522 if (sdkp->device->manage_start_stop) {
3523 sd_printk(KERN_NOTICE, sdkp, "Stopping disk\n");
3524 /* an error is not worth aborting a system sleep */
3525 ret = sd_start_stop_device(sdkp, 0);
3526 if (ignore_stop_errors)
3534 static int sd_suspend_system(struct device *dev)
3536 return sd_suspend_common(dev, true);
3539 static int sd_suspend_runtime(struct device *dev)
3541 return sd_suspend_common(dev, false);
3544 static int sd_resume(struct device *dev)
3546 struct scsi_disk *sdkp = dev_get_drvdata(dev);
3549 if (!sdkp) /* E.g.: runtime resume at the start of sd_probe() */
3552 if (!sdkp->device->manage_start_stop)
3555 sd_printk(KERN_NOTICE, sdkp, "Starting disk\n");
3556 ret = sd_start_stop_device(sdkp, 1);
3558 opal_unlock_from_suspend(sdkp->opal_dev);
3563 * init_sd - entry point for this driver (both when built in or when
3566 * Note: this function registers this driver with the scsi mid-level.
3568 static int __init init_sd(void)
3570 int majors = 0, i, err;
3572 SCSI_LOG_HLQUEUE(3, printk("init_sd: sd driver entry point\n"));
3574 for (i = 0; i < SD_MAJORS; i++) {
3575 if (register_blkdev(sd_major(i), "sd") != 0)
3578 blk_register_region(sd_major(i), SD_MINORS, NULL,
3579 sd_default_probe, NULL, NULL);
3585 err = class_register(&sd_disk_class);
3589 sd_cdb_cache = kmem_cache_create("sd_ext_cdb", SD_EXT_CDB_SIZE,
3591 if (!sd_cdb_cache) {
3592 printk(KERN_ERR "sd: can't init extended cdb cache\n");
3597 sd_cdb_pool = mempool_create_slab_pool(SD_MEMPOOL_SIZE, sd_cdb_cache);
3599 printk(KERN_ERR "sd: can't init extended cdb pool\n");
3604 err = scsi_register_driver(&sd_template.gendrv);
3606 goto err_out_driver;
3611 mempool_destroy(sd_cdb_pool);
3614 kmem_cache_destroy(sd_cdb_cache);
3617 class_unregister(&sd_disk_class);
3619 for (i = 0; i < SD_MAJORS; i++)
3620 unregister_blkdev(sd_major(i), "sd");
3625 * exit_sd - exit point for this driver (when it is a module).
3627 * Note: this function unregisters this driver from the scsi mid-level.
3629 static void __exit exit_sd(void)
3633 SCSI_LOG_HLQUEUE(3, printk("exit_sd: exiting sd driver\n"));
3635 scsi_unregister_driver(&sd_template.gendrv);
3636 mempool_destroy(sd_cdb_pool);
3637 kmem_cache_destroy(sd_cdb_cache);
3639 class_unregister(&sd_disk_class);
3641 for (i = 0; i < SD_MAJORS; i++) {
3642 blk_unregister_region(sd_major(i), SD_MINORS);
3643 unregister_blkdev(sd_major(i), "sd");
3647 module_init(init_sd);
3648 module_exit(exit_sd);
3650 static void sd_print_sense_hdr(struct scsi_disk *sdkp,
3651 struct scsi_sense_hdr *sshdr)
3653 scsi_print_sense_hdr(sdkp->device,
3654 sdkp->disk ? sdkp->disk->disk_name : NULL, sshdr);
3657 static void sd_print_result(const struct scsi_disk *sdkp, const char *msg,
3660 const char *hb_string = scsi_hostbyte_string(result);
3661 const char *db_string = scsi_driverbyte_string(result);
3663 if (hb_string || db_string)
3664 sd_printk(KERN_INFO, sdkp,
3665 "%s: Result: hostbyte=%s driverbyte=%s\n", msg,
3666 hb_string ? hb_string : "invalid",
3667 db_string ? db_string : "invalid");
3669 sd_printk(KERN_INFO, sdkp,
3670 "%s: Result: hostbyte=0x%02x driverbyte=0x%02x\n",
3671 msg, host_byte(result), driver_byte(result));