qla2xxx: Fix missing device login retries.
[linux/fpc-iii.git] / drivers / scsi / sd.c
blob3b2fcb4fada0491c4500b42555fdef542b4399d2
1 /*
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>
36 #include <linux/fs.h>
37 #include <linux/kernel.h>
38 #include <linux/mm.h>
39 #include <linux/bio.h>
40 #include <linux/genhd.h>
41 #include <linux/hdreg.h>
42 #include <linux/errno.h>
43 #include <linux/idr.h>
44 #include <linux/interrupt.h>
45 #include <linux/init.h>
46 #include <linux/blkdev.h>
47 #include <linux/blkpg.h>
48 #include <linux/delay.h>
49 #include <linux/mutex.h>
50 #include <linux/string_helpers.h>
51 #include <linux/async.h>
52 #include <linux/slab.h>
53 #include <linux/pm_runtime.h>
54 #include <asm/uaccess.h>
55 #include <asm/unaligned.h>
57 #include <scsi/scsi.h>
58 #include <scsi/scsi_cmnd.h>
59 #include <scsi/scsi_dbg.h>
60 #include <scsi/scsi_device.h>
61 #include <scsi/scsi_driver.h>
62 #include <scsi/scsi_eh.h>
63 #include <scsi/scsi_host.h>
64 #include <scsi/scsi_ioctl.h>
65 #include <scsi/scsicam.h>
67 #include "sd.h"
68 #include "scsi_priv.h"
69 #include "scsi_logging.h"
71 MODULE_AUTHOR("Eric Youngdale");
72 MODULE_DESCRIPTION("SCSI disk (sd) driver");
73 MODULE_LICENSE("GPL");
75 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK0_MAJOR);
76 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK1_MAJOR);
77 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK2_MAJOR);
78 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK3_MAJOR);
79 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK4_MAJOR);
80 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK5_MAJOR);
81 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK6_MAJOR);
82 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK7_MAJOR);
83 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK8_MAJOR);
84 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK9_MAJOR);
85 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK10_MAJOR);
86 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK11_MAJOR);
87 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK12_MAJOR);
88 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK13_MAJOR);
89 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK14_MAJOR);
90 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK15_MAJOR);
91 MODULE_ALIAS_SCSI_DEVICE(TYPE_DISK);
92 MODULE_ALIAS_SCSI_DEVICE(TYPE_MOD);
93 MODULE_ALIAS_SCSI_DEVICE(TYPE_RBC);
95 #if !defined(CONFIG_DEBUG_BLOCK_EXT_DEVT)
96 #define SD_MINORS 16
97 #else
98 #define SD_MINORS 0
99 #endif
101 static void sd_config_discard(struct scsi_disk *, unsigned int);
102 static void sd_config_write_same(struct scsi_disk *);
103 static int sd_revalidate_disk(struct gendisk *);
104 static void sd_unlock_native_capacity(struct gendisk *disk);
105 static int sd_probe(struct device *);
106 static int sd_remove(struct device *);
107 static void sd_shutdown(struct device *);
108 static int sd_suspend_system(struct device *);
109 static int sd_suspend_runtime(struct device *);
110 static int sd_resume(struct device *);
111 static void sd_rescan(struct device *);
112 static int sd_init_command(struct scsi_cmnd *SCpnt);
113 static void sd_uninit_command(struct scsi_cmnd *SCpnt);
114 static int sd_done(struct scsi_cmnd *);
115 static int sd_eh_action(struct scsi_cmnd *, int);
116 static void sd_read_capacity(struct scsi_disk *sdkp, unsigned char *buffer);
117 static void scsi_disk_release(struct device *cdev);
118 static void sd_print_sense_hdr(struct scsi_disk *, struct scsi_sense_hdr *);
119 static void sd_print_result(const struct scsi_disk *, const char *, int);
121 static DEFINE_SPINLOCK(sd_index_lock);
122 static DEFINE_IDA(sd_index_ida);
124 /* This semaphore is used to mediate the 0->1 reference get in the
125 * face of object destruction (i.e. we can't allow a get on an
126 * object after last put) */
127 static DEFINE_MUTEX(sd_ref_mutex);
129 static struct kmem_cache *sd_cdb_cache;
130 static mempool_t *sd_cdb_pool;
132 static const char *sd_cache_types[] = {
133 "write through", "none", "write back",
134 "write back, no read (daft)"
137 static void sd_set_flush_flag(struct scsi_disk *sdkp)
139 unsigned flush = 0;
141 if (sdkp->WCE) {
142 flush |= REQ_FLUSH;
143 if (sdkp->DPOFUA)
144 flush |= REQ_FUA;
147 blk_queue_flush(sdkp->disk->queue, flush);
150 static ssize_t
151 cache_type_store(struct device *dev, struct device_attribute *attr,
152 const char *buf, size_t count)
154 int i, ct = -1, rcd, wce, sp;
155 struct scsi_disk *sdkp = to_scsi_disk(dev);
156 struct scsi_device *sdp = sdkp->device;
157 char buffer[64];
158 char *buffer_data;
159 struct scsi_mode_data data;
160 struct scsi_sense_hdr sshdr;
161 static const char temp[] = "temporary ";
162 int len;
164 if (sdp->type != TYPE_DISK)
165 /* no cache control on RBC devices; theoretically they
166 * can do it, but there's probably so many exceptions
167 * it's not worth the risk */
168 return -EINVAL;
170 if (strncmp(buf, temp, sizeof(temp) - 1) == 0) {
171 buf += sizeof(temp) - 1;
172 sdkp->cache_override = 1;
173 } else {
174 sdkp->cache_override = 0;
177 for (i = 0; i < ARRAY_SIZE(sd_cache_types); i++) {
178 len = strlen(sd_cache_types[i]);
179 if (strncmp(sd_cache_types[i], buf, len) == 0 &&
180 buf[len] == '\n') {
181 ct = i;
182 break;
185 if (ct < 0)
186 return -EINVAL;
187 rcd = ct & 0x01 ? 1 : 0;
188 wce = (ct & 0x02) && !sdkp->write_prot ? 1 : 0;
190 if (sdkp->cache_override) {
191 sdkp->WCE = wce;
192 sdkp->RCD = rcd;
193 sd_set_flush_flag(sdkp);
194 return count;
197 if (scsi_mode_sense(sdp, 0x08, 8, buffer, sizeof(buffer), SD_TIMEOUT,
198 SD_MAX_RETRIES, &data, NULL))
199 return -EINVAL;
200 len = min_t(size_t, sizeof(buffer), data.length - data.header_length -
201 data.block_descriptor_length);
202 buffer_data = buffer + data.header_length +
203 data.block_descriptor_length;
204 buffer_data[2] &= ~0x05;
205 buffer_data[2] |= wce << 2 | rcd;
206 sp = buffer_data[0] & 0x80 ? 1 : 0;
208 if (scsi_mode_select(sdp, 1, sp, 8, buffer_data, len, SD_TIMEOUT,
209 SD_MAX_RETRIES, &data, &sshdr)) {
210 if (scsi_sense_valid(&sshdr))
211 sd_print_sense_hdr(sdkp, &sshdr);
212 return -EINVAL;
214 revalidate_disk(sdkp->disk);
215 return count;
218 static ssize_t
219 manage_start_stop_show(struct device *dev, struct device_attribute *attr,
220 char *buf)
222 struct scsi_disk *sdkp = to_scsi_disk(dev);
223 struct scsi_device *sdp = sdkp->device;
225 return snprintf(buf, 20, "%u\n", sdp->manage_start_stop);
228 static ssize_t
229 manage_start_stop_store(struct device *dev, struct device_attribute *attr,
230 const char *buf, size_t count)
232 struct scsi_disk *sdkp = to_scsi_disk(dev);
233 struct scsi_device *sdp = sdkp->device;
235 if (!capable(CAP_SYS_ADMIN))
236 return -EACCES;
238 sdp->manage_start_stop = simple_strtoul(buf, NULL, 10);
240 return count;
242 static DEVICE_ATTR_RW(manage_start_stop);
244 static ssize_t
245 allow_restart_show(struct device *dev, struct device_attribute *attr, char *buf)
247 struct scsi_disk *sdkp = to_scsi_disk(dev);
249 return snprintf(buf, 40, "%d\n", sdkp->device->allow_restart);
252 static ssize_t
253 allow_restart_store(struct device *dev, struct device_attribute *attr,
254 const char *buf, size_t count)
256 struct scsi_disk *sdkp = to_scsi_disk(dev);
257 struct scsi_device *sdp = sdkp->device;
259 if (!capable(CAP_SYS_ADMIN))
260 return -EACCES;
262 if (sdp->type != TYPE_DISK)
263 return -EINVAL;
265 sdp->allow_restart = simple_strtoul(buf, NULL, 10);
267 return count;
269 static DEVICE_ATTR_RW(allow_restart);
271 static ssize_t
272 cache_type_show(struct device *dev, struct device_attribute *attr, char *buf)
274 struct scsi_disk *sdkp = to_scsi_disk(dev);
275 int ct = sdkp->RCD + 2*sdkp->WCE;
277 return snprintf(buf, 40, "%s\n", sd_cache_types[ct]);
279 static DEVICE_ATTR_RW(cache_type);
281 static ssize_t
282 FUA_show(struct device *dev, struct device_attribute *attr, char *buf)
284 struct scsi_disk *sdkp = to_scsi_disk(dev);
286 return snprintf(buf, 20, "%u\n", sdkp->DPOFUA);
288 static DEVICE_ATTR_RO(FUA);
290 static ssize_t
291 protection_type_show(struct device *dev, struct device_attribute *attr,
292 char *buf)
294 struct scsi_disk *sdkp = to_scsi_disk(dev);
296 return snprintf(buf, 20, "%u\n", sdkp->protection_type);
299 static ssize_t
300 protection_type_store(struct device *dev, struct device_attribute *attr,
301 const char *buf, size_t count)
303 struct scsi_disk *sdkp = to_scsi_disk(dev);
304 unsigned int val;
305 int err;
307 if (!capable(CAP_SYS_ADMIN))
308 return -EACCES;
310 err = kstrtouint(buf, 10, &val);
312 if (err)
313 return err;
315 if (val >= 0 && val <= SD_DIF_TYPE3_PROTECTION)
316 sdkp->protection_type = val;
318 return count;
320 static DEVICE_ATTR_RW(protection_type);
322 static ssize_t
323 protection_mode_show(struct device *dev, struct device_attribute *attr,
324 char *buf)
326 struct scsi_disk *sdkp = to_scsi_disk(dev);
327 struct scsi_device *sdp = sdkp->device;
328 unsigned int dif, dix;
330 dif = scsi_host_dif_capable(sdp->host, sdkp->protection_type);
331 dix = scsi_host_dix_capable(sdp->host, sdkp->protection_type);
333 if (!dix && scsi_host_dix_capable(sdp->host, SD_DIF_TYPE0_PROTECTION)) {
334 dif = 0;
335 dix = 1;
338 if (!dif && !dix)
339 return snprintf(buf, 20, "none\n");
341 return snprintf(buf, 20, "%s%u\n", dix ? "dix" : "dif", dif);
343 static DEVICE_ATTR_RO(protection_mode);
345 static ssize_t
346 app_tag_own_show(struct device *dev, struct device_attribute *attr, char *buf)
348 struct scsi_disk *sdkp = to_scsi_disk(dev);
350 return snprintf(buf, 20, "%u\n", sdkp->ATO);
352 static DEVICE_ATTR_RO(app_tag_own);
354 static ssize_t
355 thin_provisioning_show(struct device *dev, struct device_attribute *attr,
356 char *buf)
358 struct scsi_disk *sdkp = to_scsi_disk(dev);
360 return snprintf(buf, 20, "%u\n", sdkp->lbpme);
362 static DEVICE_ATTR_RO(thin_provisioning);
364 static const char *lbp_mode[] = {
365 [SD_LBP_FULL] = "full",
366 [SD_LBP_UNMAP] = "unmap",
367 [SD_LBP_WS16] = "writesame_16",
368 [SD_LBP_WS10] = "writesame_10",
369 [SD_LBP_ZERO] = "writesame_zero",
370 [SD_LBP_DISABLE] = "disabled",
373 static ssize_t
374 provisioning_mode_show(struct device *dev, struct device_attribute *attr,
375 char *buf)
377 struct scsi_disk *sdkp = to_scsi_disk(dev);
379 return snprintf(buf, 20, "%s\n", lbp_mode[sdkp->provisioning_mode]);
382 static ssize_t
383 provisioning_mode_store(struct device *dev, struct device_attribute *attr,
384 const char *buf, size_t count)
386 struct scsi_disk *sdkp = to_scsi_disk(dev);
387 struct scsi_device *sdp = sdkp->device;
389 if (!capable(CAP_SYS_ADMIN))
390 return -EACCES;
392 if (sdp->type != TYPE_DISK)
393 return -EINVAL;
395 if (!strncmp(buf, lbp_mode[SD_LBP_UNMAP], 20))
396 sd_config_discard(sdkp, SD_LBP_UNMAP);
397 else if (!strncmp(buf, lbp_mode[SD_LBP_WS16], 20))
398 sd_config_discard(sdkp, SD_LBP_WS16);
399 else if (!strncmp(buf, lbp_mode[SD_LBP_WS10], 20))
400 sd_config_discard(sdkp, SD_LBP_WS10);
401 else if (!strncmp(buf, lbp_mode[SD_LBP_ZERO], 20))
402 sd_config_discard(sdkp, SD_LBP_ZERO);
403 else if (!strncmp(buf, lbp_mode[SD_LBP_DISABLE], 20))
404 sd_config_discard(sdkp, SD_LBP_DISABLE);
405 else
406 return -EINVAL;
408 return count;
410 static DEVICE_ATTR_RW(provisioning_mode);
412 static ssize_t
413 max_medium_access_timeouts_show(struct device *dev,
414 struct device_attribute *attr, char *buf)
416 struct scsi_disk *sdkp = to_scsi_disk(dev);
418 return snprintf(buf, 20, "%u\n", sdkp->max_medium_access_timeouts);
421 static ssize_t
422 max_medium_access_timeouts_store(struct device *dev,
423 struct device_attribute *attr, const char *buf,
424 size_t count)
426 struct scsi_disk *sdkp = to_scsi_disk(dev);
427 int err;
429 if (!capable(CAP_SYS_ADMIN))
430 return -EACCES;
432 err = kstrtouint(buf, 10, &sdkp->max_medium_access_timeouts);
434 return err ? err : count;
436 static DEVICE_ATTR_RW(max_medium_access_timeouts);
438 static ssize_t
439 max_write_same_blocks_show(struct device *dev, struct device_attribute *attr,
440 char *buf)
442 struct scsi_disk *sdkp = to_scsi_disk(dev);
444 return snprintf(buf, 20, "%u\n", sdkp->max_ws_blocks);
447 static ssize_t
448 max_write_same_blocks_store(struct device *dev, struct device_attribute *attr,
449 const char *buf, size_t count)
451 struct scsi_disk *sdkp = to_scsi_disk(dev);
452 struct scsi_device *sdp = sdkp->device;
453 unsigned long max;
454 int err;
456 if (!capable(CAP_SYS_ADMIN))
457 return -EACCES;
459 if (sdp->type != TYPE_DISK)
460 return -EINVAL;
462 err = kstrtoul(buf, 10, &max);
464 if (err)
465 return err;
467 if (max == 0)
468 sdp->no_write_same = 1;
469 else if (max <= SD_MAX_WS16_BLOCKS) {
470 sdp->no_write_same = 0;
471 sdkp->max_ws_blocks = max;
474 sd_config_write_same(sdkp);
476 return count;
478 static DEVICE_ATTR_RW(max_write_same_blocks);
480 static struct attribute *sd_disk_attrs[] = {
481 &dev_attr_cache_type.attr,
482 &dev_attr_FUA.attr,
483 &dev_attr_allow_restart.attr,
484 &dev_attr_manage_start_stop.attr,
485 &dev_attr_protection_type.attr,
486 &dev_attr_protection_mode.attr,
487 &dev_attr_app_tag_own.attr,
488 &dev_attr_thin_provisioning.attr,
489 &dev_attr_provisioning_mode.attr,
490 &dev_attr_max_write_same_blocks.attr,
491 &dev_attr_max_medium_access_timeouts.attr,
492 NULL,
494 ATTRIBUTE_GROUPS(sd_disk);
496 static struct class sd_disk_class = {
497 .name = "scsi_disk",
498 .owner = THIS_MODULE,
499 .dev_release = scsi_disk_release,
500 .dev_groups = sd_disk_groups,
503 static const struct dev_pm_ops sd_pm_ops = {
504 .suspend = sd_suspend_system,
505 .resume = sd_resume,
506 .poweroff = sd_suspend_system,
507 .restore = sd_resume,
508 .runtime_suspend = sd_suspend_runtime,
509 .runtime_resume = sd_resume,
512 static struct scsi_driver sd_template = {
513 .gendrv = {
514 .name = "sd",
515 .owner = THIS_MODULE,
516 .probe = sd_probe,
517 .remove = sd_remove,
518 .shutdown = sd_shutdown,
519 .pm = &sd_pm_ops,
521 .rescan = sd_rescan,
522 .init_command = sd_init_command,
523 .uninit_command = sd_uninit_command,
524 .done = sd_done,
525 .eh_action = sd_eh_action,
529 * Dummy kobj_map->probe function.
530 * The default ->probe function will call modprobe, which is
531 * pointless as this module is already loaded.
533 static struct kobject *sd_default_probe(dev_t devt, int *partno, void *data)
535 return NULL;
539 * Device no to disk mapping:
541 * major disc2 disc p1
542 * |............|.............|....|....| <- dev_t
543 * 31 20 19 8 7 4 3 0
545 * Inside a major, we have 16k disks, however mapped non-
546 * contiguously. The first 16 disks are for major0, the next
547 * ones with major1, ... Disk 256 is for major0 again, disk 272
548 * for major1, ...
549 * As we stay compatible with our numbering scheme, we can reuse
550 * the well-know SCSI majors 8, 65--71, 136--143.
552 static int sd_major(int major_idx)
554 switch (major_idx) {
555 case 0:
556 return SCSI_DISK0_MAJOR;
557 case 1 ... 7:
558 return SCSI_DISK1_MAJOR + major_idx - 1;
559 case 8 ... 15:
560 return SCSI_DISK8_MAJOR + major_idx - 8;
561 default:
562 BUG();
563 return 0; /* shut up gcc */
567 static struct scsi_disk *scsi_disk_get(struct gendisk *disk)
569 struct scsi_disk *sdkp = NULL;
571 mutex_lock(&sd_ref_mutex);
573 if (disk->private_data) {
574 sdkp = scsi_disk(disk);
575 if (scsi_device_get(sdkp->device) == 0)
576 get_device(&sdkp->dev);
577 else
578 sdkp = NULL;
580 mutex_unlock(&sd_ref_mutex);
581 return sdkp;
584 static void scsi_disk_put(struct scsi_disk *sdkp)
586 struct scsi_device *sdev = sdkp->device;
588 mutex_lock(&sd_ref_mutex);
589 put_device(&sdkp->dev);
590 scsi_device_put(sdev);
591 mutex_unlock(&sd_ref_mutex);
594 static unsigned char sd_setup_protect_cmnd(struct scsi_cmnd *scmd,
595 unsigned int dix, unsigned int dif)
597 struct bio *bio = scmd->request->bio;
598 unsigned int prot_op = sd_prot_op(rq_data_dir(scmd->request), dix, dif);
599 unsigned int protect = 0;
601 if (dix) { /* DIX Type 0, 1, 2, 3 */
602 if (bio_integrity_flagged(bio, BIP_IP_CHECKSUM))
603 scmd->prot_flags |= SCSI_PROT_IP_CHECKSUM;
605 if (bio_integrity_flagged(bio, BIP_CTRL_NOCHECK) == false)
606 scmd->prot_flags |= SCSI_PROT_GUARD_CHECK;
609 if (dif != SD_DIF_TYPE3_PROTECTION) { /* DIX/DIF Type 0, 1, 2 */
610 scmd->prot_flags |= SCSI_PROT_REF_INCREMENT;
612 if (bio_integrity_flagged(bio, BIP_CTRL_NOCHECK) == false)
613 scmd->prot_flags |= SCSI_PROT_REF_CHECK;
616 if (dif) { /* DIX/DIF Type 1, 2, 3 */
617 scmd->prot_flags |= SCSI_PROT_TRANSFER_PI;
619 if (bio_integrity_flagged(bio, BIP_DISK_NOCHECK))
620 protect = 3 << 5; /* Disable target PI checking */
621 else
622 protect = 1 << 5; /* Enable target PI checking */
625 scsi_set_prot_op(scmd, prot_op);
626 scsi_set_prot_type(scmd, dif);
627 scmd->prot_flags &= sd_prot_flag_mask(prot_op);
629 return protect;
632 static void sd_config_discard(struct scsi_disk *sdkp, unsigned int mode)
634 struct request_queue *q = sdkp->disk->queue;
635 unsigned int logical_block_size = sdkp->device->sector_size;
636 unsigned int max_blocks = 0;
638 q->limits.discard_zeroes_data = 0;
639 q->limits.discard_alignment = sdkp->unmap_alignment *
640 logical_block_size;
641 q->limits.discard_granularity =
642 max(sdkp->physical_block_size,
643 sdkp->unmap_granularity * logical_block_size);
645 sdkp->provisioning_mode = mode;
647 switch (mode) {
649 case SD_LBP_DISABLE:
650 q->limits.max_discard_sectors = 0;
651 queue_flag_clear_unlocked(QUEUE_FLAG_DISCARD, q);
652 return;
654 case SD_LBP_UNMAP:
655 max_blocks = min_not_zero(sdkp->max_unmap_blocks,
656 (u32)SD_MAX_WS16_BLOCKS);
657 break;
659 case SD_LBP_WS16:
660 max_blocks = min_not_zero(sdkp->max_ws_blocks,
661 (u32)SD_MAX_WS16_BLOCKS);
662 q->limits.discard_zeroes_data = sdkp->lbprz;
663 break;
665 case SD_LBP_WS10:
666 max_blocks = min_not_zero(sdkp->max_ws_blocks,
667 (u32)SD_MAX_WS10_BLOCKS);
668 q->limits.discard_zeroes_data = sdkp->lbprz;
669 break;
671 case SD_LBP_ZERO:
672 max_blocks = min_not_zero(sdkp->max_ws_blocks,
673 (u32)SD_MAX_WS10_BLOCKS);
674 q->limits.discard_zeroes_data = 1;
675 break;
678 q->limits.max_discard_sectors = max_blocks * (logical_block_size >> 9);
679 queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, q);
683 * sd_setup_discard_cmnd - unmap blocks on thinly provisioned device
684 * @sdp: scsi device to operate one
685 * @rq: Request to prepare
687 * Will issue either UNMAP or WRITE SAME(16) depending on preference
688 * indicated by target device.
690 static int sd_setup_discard_cmnd(struct scsi_cmnd *cmd)
692 struct request *rq = cmd->request;
693 struct scsi_device *sdp = cmd->device;
694 struct scsi_disk *sdkp = scsi_disk(rq->rq_disk);
695 sector_t sector = blk_rq_pos(rq);
696 unsigned int nr_sectors = blk_rq_sectors(rq);
697 unsigned int nr_bytes = blk_rq_bytes(rq);
698 unsigned int len;
699 int ret;
700 char *buf;
701 struct page *page;
703 sector >>= ilog2(sdp->sector_size) - 9;
704 nr_sectors >>= ilog2(sdp->sector_size) - 9;
706 page = alloc_page(GFP_ATOMIC | __GFP_ZERO);
707 if (!page)
708 return BLKPREP_DEFER;
710 switch (sdkp->provisioning_mode) {
711 case SD_LBP_UNMAP:
712 buf = page_address(page);
714 cmd->cmd_len = 10;
715 cmd->cmnd[0] = UNMAP;
716 cmd->cmnd[8] = 24;
718 put_unaligned_be16(6 + 16, &buf[0]);
719 put_unaligned_be16(16, &buf[2]);
720 put_unaligned_be64(sector, &buf[8]);
721 put_unaligned_be32(nr_sectors, &buf[16]);
723 len = 24;
724 break;
726 case SD_LBP_WS16:
727 cmd->cmd_len = 16;
728 cmd->cmnd[0] = WRITE_SAME_16;
729 cmd->cmnd[1] = 0x8; /* UNMAP */
730 put_unaligned_be64(sector, &cmd->cmnd[2]);
731 put_unaligned_be32(nr_sectors, &cmd->cmnd[10]);
733 len = sdkp->device->sector_size;
734 break;
736 case SD_LBP_WS10:
737 case SD_LBP_ZERO:
738 cmd->cmd_len = 10;
739 cmd->cmnd[0] = WRITE_SAME;
740 if (sdkp->provisioning_mode == SD_LBP_WS10)
741 cmd->cmnd[1] = 0x8; /* UNMAP */
742 put_unaligned_be32(sector, &cmd->cmnd[2]);
743 put_unaligned_be16(nr_sectors, &cmd->cmnd[7]);
745 len = sdkp->device->sector_size;
746 break;
748 default:
749 ret = BLKPREP_KILL;
750 goto out;
753 rq->completion_data = page;
754 rq->timeout = SD_TIMEOUT;
756 cmd->transfersize = len;
757 cmd->allowed = SD_MAX_RETRIES;
760 * Initially __data_len is set to the amount of data that needs to be
761 * transferred to the target. This amount depends on whether WRITE SAME
762 * or UNMAP is being used. After the scatterlist has been mapped by
763 * scsi_init_io() we set __data_len to the size of the area to be
764 * discarded on disk. This allows us to report completion on the full
765 * amount of blocks described by the request.
767 blk_add_request_payload(rq, page, len);
768 ret = scsi_init_io(cmd);
769 rq->__data_len = nr_bytes;
771 out:
772 if (ret != BLKPREP_OK)
773 __free_page(page);
774 return ret;
777 static void sd_config_write_same(struct scsi_disk *sdkp)
779 struct request_queue *q = sdkp->disk->queue;
780 unsigned int logical_block_size = sdkp->device->sector_size;
782 if (sdkp->device->no_write_same) {
783 sdkp->max_ws_blocks = 0;
784 goto out;
787 /* Some devices can not handle block counts above 0xffff despite
788 * supporting WRITE SAME(16). Consequently we default to 64k
789 * blocks per I/O unless the device explicitly advertises a
790 * bigger limit.
792 if (sdkp->max_ws_blocks > SD_MAX_WS10_BLOCKS)
793 sdkp->max_ws_blocks = min_not_zero(sdkp->max_ws_blocks,
794 (u32)SD_MAX_WS16_BLOCKS);
795 else if (sdkp->ws16 || sdkp->ws10 || sdkp->device->no_report_opcodes)
796 sdkp->max_ws_blocks = min_not_zero(sdkp->max_ws_blocks,
797 (u32)SD_MAX_WS10_BLOCKS);
798 else {
799 sdkp->device->no_write_same = 1;
800 sdkp->max_ws_blocks = 0;
803 out:
804 blk_queue_max_write_same_sectors(q, sdkp->max_ws_blocks *
805 (logical_block_size >> 9));
809 * sd_setup_write_same_cmnd - write the same data to multiple blocks
810 * @cmd: command to prepare
812 * Will issue either WRITE SAME(10) or WRITE SAME(16) depending on
813 * preference indicated by target device.
815 static int sd_setup_write_same_cmnd(struct scsi_cmnd *cmd)
817 struct request *rq = cmd->request;
818 struct scsi_device *sdp = cmd->device;
819 struct scsi_disk *sdkp = scsi_disk(rq->rq_disk);
820 struct bio *bio = rq->bio;
821 sector_t sector = blk_rq_pos(rq);
822 unsigned int nr_sectors = blk_rq_sectors(rq);
823 unsigned int nr_bytes = blk_rq_bytes(rq);
824 int ret;
826 if (sdkp->device->no_write_same)
827 return BLKPREP_KILL;
829 BUG_ON(bio_offset(bio) || bio_iovec(bio).bv_len != sdp->sector_size);
831 sector >>= ilog2(sdp->sector_size) - 9;
832 nr_sectors >>= ilog2(sdp->sector_size) - 9;
834 rq->timeout = SD_WRITE_SAME_TIMEOUT;
836 if (sdkp->ws16 || sector > 0xffffffff || nr_sectors > 0xffff) {
837 cmd->cmd_len = 16;
838 cmd->cmnd[0] = WRITE_SAME_16;
839 put_unaligned_be64(sector, &cmd->cmnd[2]);
840 put_unaligned_be32(nr_sectors, &cmd->cmnd[10]);
841 } else {
842 cmd->cmd_len = 10;
843 cmd->cmnd[0] = WRITE_SAME;
844 put_unaligned_be32(sector, &cmd->cmnd[2]);
845 put_unaligned_be16(nr_sectors, &cmd->cmnd[7]);
848 cmd->transfersize = sdp->sector_size;
849 cmd->allowed = SD_MAX_RETRIES;
852 * For WRITE_SAME the data transferred in the DATA IN buffer is
853 * different from the amount of data actually written to the target.
855 * We set up __data_len to the amount of data transferred from the
856 * DATA IN buffer so that blk_rq_map_sg set up the proper S/G list
857 * to transfer a single sector of data first, but then reset it to
858 * the amount of data to be written right after so that the I/O path
859 * knows how much to actually write.
861 rq->__data_len = sdp->sector_size;
862 ret = scsi_init_io(cmd);
863 rq->__data_len = nr_bytes;
864 return ret;
867 static int sd_setup_flush_cmnd(struct scsi_cmnd *cmd)
869 struct request *rq = cmd->request;
871 /* flush requests don't perform I/O, zero the S/G table */
872 memset(&cmd->sdb, 0, sizeof(cmd->sdb));
874 cmd->cmnd[0] = SYNCHRONIZE_CACHE;
875 cmd->cmd_len = 10;
876 cmd->transfersize = 0;
877 cmd->allowed = SD_MAX_RETRIES;
879 rq->timeout = rq->q->rq_timeout * SD_FLUSH_TIMEOUT_MULTIPLIER;
880 return BLKPREP_OK;
883 static int sd_setup_read_write_cmnd(struct scsi_cmnd *SCpnt)
885 struct request *rq = SCpnt->request;
886 struct scsi_device *sdp = SCpnt->device;
887 struct gendisk *disk = rq->rq_disk;
888 struct scsi_disk *sdkp;
889 sector_t block = blk_rq_pos(rq);
890 sector_t threshold;
891 unsigned int this_count = blk_rq_sectors(rq);
892 unsigned int dif, dix;
893 int ret;
894 unsigned char protect;
896 ret = scsi_init_io(SCpnt);
897 if (ret != BLKPREP_OK)
898 goto out;
899 SCpnt = rq->special;
900 sdkp = scsi_disk(disk);
902 /* from here on until we're complete, any goto out
903 * is used for a killable error condition */
904 ret = BLKPREP_KILL;
906 SCSI_LOG_HLQUEUE(1,
907 scmd_printk(KERN_INFO, SCpnt,
908 "%s: block=%llu, count=%d\n",
909 __func__, (unsigned long long)block, this_count));
911 if (!sdp || !scsi_device_online(sdp) ||
912 block + blk_rq_sectors(rq) > get_capacity(disk)) {
913 SCSI_LOG_HLQUEUE(2, scmd_printk(KERN_INFO, SCpnt,
914 "Finishing %u sectors\n",
915 blk_rq_sectors(rq)));
916 SCSI_LOG_HLQUEUE(2, scmd_printk(KERN_INFO, SCpnt,
917 "Retry with 0x%p\n", SCpnt));
918 goto out;
921 if (sdp->changed) {
923 * quietly refuse to do anything to a changed disc until
924 * the changed bit has been reset
926 /* printk("SCSI disk has been changed or is not present. Prohibiting further I/O.\n"); */
927 goto out;
931 * Some SD card readers can't handle multi-sector accesses which touch
932 * the last one or two hardware sectors. Split accesses as needed.
934 threshold = get_capacity(disk) - SD_LAST_BUGGY_SECTORS *
935 (sdp->sector_size / 512);
937 if (unlikely(sdp->last_sector_bug && block + this_count > threshold)) {
938 if (block < threshold) {
939 /* Access up to the threshold but not beyond */
940 this_count = threshold - block;
941 } else {
942 /* Access only a single hardware sector */
943 this_count = sdp->sector_size / 512;
947 SCSI_LOG_HLQUEUE(2, scmd_printk(KERN_INFO, SCpnt, "block=%llu\n",
948 (unsigned long long)block));
951 * If we have a 1K hardware sectorsize, prevent access to single
952 * 512 byte sectors. In theory we could handle this - in fact
953 * the scsi cdrom driver must be able to handle this because
954 * we typically use 1K blocksizes, and cdroms typically have
955 * 2K hardware sectorsizes. Of course, things are simpler
956 * with the cdrom, since it is read-only. For performance
957 * reasons, the filesystems should be able to handle this
958 * and not force the scsi disk driver to use bounce buffers
959 * for this.
961 if (sdp->sector_size == 1024) {
962 if ((block & 1) || (blk_rq_sectors(rq) & 1)) {
963 scmd_printk(KERN_ERR, SCpnt,
964 "Bad block number requested\n");
965 goto out;
966 } else {
967 block = block >> 1;
968 this_count = this_count >> 1;
971 if (sdp->sector_size == 2048) {
972 if ((block & 3) || (blk_rq_sectors(rq) & 3)) {
973 scmd_printk(KERN_ERR, SCpnt,
974 "Bad block number requested\n");
975 goto out;
976 } else {
977 block = block >> 2;
978 this_count = this_count >> 2;
981 if (sdp->sector_size == 4096) {
982 if ((block & 7) || (blk_rq_sectors(rq) & 7)) {
983 scmd_printk(KERN_ERR, SCpnt,
984 "Bad block number requested\n");
985 goto out;
986 } else {
987 block = block >> 3;
988 this_count = this_count >> 3;
991 if (rq_data_dir(rq) == WRITE) {
992 SCpnt->cmnd[0] = WRITE_6;
994 if (blk_integrity_rq(rq))
995 sd_dif_prepare(SCpnt);
997 } else if (rq_data_dir(rq) == READ) {
998 SCpnt->cmnd[0] = READ_6;
999 } else {
1000 scmd_printk(KERN_ERR, SCpnt, "Unknown command %llx\n", (unsigned long long) rq->cmd_flags);
1001 goto out;
1004 SCSI_LOG_HLQUEUE(2, scmd_printk(KERN_INFO, SCpnt,
1005 "%s %d/%u 512 byte blocks.\n",
1006 (rq_data_dir(rq) == WRITE) ?
1007 "writing" : "reading", this_count,
1008 blk_rq_sectors(rq)));
1010 dix = scsi_prot_sg_count(SCpnt);
1011 dif = scsi_host_dif_capable(SCpnt->device->host, sdkp->protection_type);
1013 if (dif || dix)
1014 protect = sd_setup_protect_cmnd(SCpnt, dix, dif);
1015 else
1016 protect = 0;
1018 if (protect && sdkp->protection_type == SD_DIF_TYPE2_PROTECTION) {
1019 SCpnt->cmnd = mempool_alloc(sd_cdb_pool, GFP_ATOMIC);
1021 if (unlikely(SCpnt->cmnd == NULL)) {
1022 ret = BLKPREP_DEFER;
1023 goto out;
1026 SCpnt->cmd_len = SD_EXT_CDB_SIZE;
1027 memset(SCpnt->cmnd, 0, SCpnt->cmd_len);
1028 SCpnt->cmnd[0] = VARIABLE_LENGTH_CMD;
1029 SCpnt->cmnd[7] = 0x18;
1030 SCpnt->cmnd[9] = (rq_data_dir(rq) == READ) ? READ_32 : WRITE_32;
1031 SCpnt->cmnd[10] = protect | ((rq->cmd_flags & REQ_FUA) ? 0x8 : 0);
1033 /* LBA */
1034 SCpnt->cmnd[12] = sizeof(block) > 4 ? (unsigned char) (block >> 56) & 0xff : 0;
1035 SCpnt->cmnd[13] = sizeof(block) > 4 ? (unsigned char) (block >> 48) & 0xff : 0;
1036 SCpnt->cmnd[14] = sizeof(block) > 4 ? (unsigned char) (block >> 40) & 0xff : 0;
1037 SCpnt->cmnd[15] = sizeof(block) > 4 ? (unsigned char) (block >> 32) & 0xff : 0;
1038 SCpnt->cmnd[16] = (unsigned char) (block >> 24) & 0xff;
1039 SCpnt->cmnd[17] = (unsigned char) (block >> 16) & 0xff;
1040 SCpnt->cmnd[18] = (unsigned char) (block >> 8) & 0xff;
1041 SCpnt->cmnd[19] = (unsigned char) block & 0xff;
1043 /* Expected Indirect LBA */
1044 SCpnt->cmnd[20] = (unsigned char) (block >> 24) & 0xff;
1045 SCpnt->cmnd[21] = (unsigned char) (block >> 16) & 0xff;
1046 SCpnt->cmnd[22] = (unsigned char) (block >> 8) & 0xff;
1047 SCpnt->cmnd[23] = (unsigned char) block & 0xff;
1049 /* Transfer length */
1050 SCpnt->cmnd[28] = (unsigned char) (this_count >> 24) & 0xff;
1051 SCpnt->cmnd[29] = (unsigned char) (this_count >> 16) & 0xff;
1052 SCpnt->cmnd[30] = (unsigned char) (this_count >> 8) & 0xff;
1053 SCpnt->cmnd[31] = (unsigned char) this_count & 0xff;
1054 } else if (sdp->use_16_for_rw || (this_count > 0xffff)) {
1055 SCpnt->cmnd[0] += READ_16 - READ_6;
1056 SCpnt->cmnd[1] = protect | ((rq->cmd_flags & REQ_FUA) ? 0x8 : 0);
1057 SCpnt->cmnd[2] = sizeof(block) > 4 ? (unsigned char) (block >> 56) & 0xff : 0;
1058 SCpnt->cmnd[3] = sizeof(block) > 4 ? (unsigned char) (block >> 48) & 0xff : 0;
1059 SCpnt->cmnd[4] = sizeof(block) > 4 ? (unsigned char) (block >> 40) & 0xff : 0;
1060 SCpnt->cmnd[5] = sizeof(block) > 4 ? (unsigned char) (block >> 32) & 0xff : 0;
1061 SCpnt->cmnd[6] = (unsigned char) (block >> 24) & 0xff;
1062 SCpnt->cmnd[7] = (unsigned char) (block >> 16) & 0xff;
1063 SCpnt->cmnd[8] = (unsigned char) (block >> 8) & 0xff;
1064 SCpnt->cmnd[9] = (unsigned char) block & 0xff;
1065 SCpnt->cmnd[10] = (unsigned char) (this_count >> 24) & 0xff;
1066 SCpnt->cmnd[11] = (unsigned char) (this_count >> 16) & 0xff;
1067 SCpnt->cmnd[12] = (unsigned char) (this_count >> 8) & 0xff;
1068 SCpnt->cmnd[13] = (unsigned char) this_count & 0xff;
1069 SCpnt->cmnd[14] = SCpnt->cmnd[15] = 0;
1070 } else if ((this_count > 0xff) || (block > 0x1fffff) ||
1071 scsi_device_protection(SCpnt->device) ||
1072 SCpnt->device->use_10_for_rw) {
1073 SCpnt->cmnd[0] += READ_10 - READ_6;
1074 SCpnt->cmnd[1] = protect | ((rq->cmd_flags & REQ_FUA) ? 0x8 : 0);
1075 SCpnt->cmnd[2] = (unsigned char) (block >> 24) & 0xff;
1076 SCpnt->cmnd[3] = (unsigned char) (block >> 16) & 0xff;
1077 SCpnt->cmnd[4] = (unsigned char) (block >> 8) & 0xff;
1078 SCpnt->cmnd[5] = (unsigned char) block & 0xff;
1079 SCpnt->cmnd[6] = SCpnt->cmnd[9] = 0;
1080 SCpnt->cmnd[7] = (unsigned char) (this_count >> 8) & 0xff;
1081 SCpnt->cmnd[8] = (unsigned char) this_count & 0xff;
1082 } else {
1083 if (unlikely(rq->cmd_flags & REQ_FUA)) {
1085 * This happens only if this drive failed
1086 * 10byte rw command with ILLEGAL_REQUEST
1087 * during operation and thus turned off
1088 * use_10_for_rw.
1090 scmd_printk(KERN_ERR, SCpnt,
1091 "FUA write on READ/WRITE(6) drive\n");
1092 goto out;
1095 SCpnt->cmnd[1] |= (unsigned char) ((block >> 16) & 0x1f);
1096 SCpnt->cmnd[2] = (unsigned char) ((block >> 8) & 0xff);
1097 SCpnt->cmnd[3] = (unsigned char) block & 0xff;
1098 SCpnt->cmnd[4] = (unsigned char) this_count;
1099 SCpnt->cmnd[5] = 0;
1101 SCpnt->sdb.length = this_count * sdp->sector_size;
1104 * We shouldn't disconnect in the middle of a sector, so with a dumb
1105 * host adapter, it's safe to assume that we can at least transfer
1106 * this many bytes between each connect / disconnect.
1108 SCpnt->transfersize = sdp->sector_size;
1109 SCpnt->underflow = this_count << 9;
1110 SCpnt->allowed = SD_MAX_RETRIES;
1113 * This indicates that the command is ready from our end to be
1114 * queued.
1116 ret = BLKPREP_OK;
1117 out:
1118 return ret;
1121 static int sd_init_command(struct scsi_cmnd *cmd)
1123 struct request *rq = cmd->request;
1125 if (rq->cmd_flags & REQ_DISCARD)
1126 return sd_setup_discard_cmnd(cmd);
1127 else if (rq->cmd_flags & REQ_WRITE_SAME)
1128 return sd_setup_write_same_cmnd(cmd);
1129 else if (rq->cmd_flags & REQ_FLUSH)
1130 return sd_setup_flush_cmnd(cmd);
1131 else
1132 return sd_setup_read_write_cmnd(cmd);
1135 static void sd_uninit_command(struct scsi_cmnd *SCpnt)
1137 struct request *rq = SCpnt->request;
1139 if (rq->cmd_flags & REQ_DISCARD)
1140 __free_page(rq->completion_data);
1142 if (SCpnt->cmnd != rq->cmd) {
1143 mempool_free(SCpnt->cmnd, sd_cdb_pool);
1144 SCpnt->cmnd = NULL;
1145 SCpnt->cmd_len = 0;
1150 * sd_open - open a scsi disk device
1151 * @inode: only i_rdev member may be used
1152 * @filp: only f_mode and f_flags may be used
1154 * Returns 0 if successful. Returns a negated errno value in case
1155 * of error.
1157 * Note: This can be called from a user context (e.g. fsck(1) )
1158 * or from within the kernel (e.g. as a result of a mount(1) ).
1159 * In the latter case @inode and @filp carry an abridged amount
1160 * of information as noted above.
1162 * Locking: called with bdev->bd_mutex held.
1164 static int sd_open(struct block_device *bdev, fmode_t mode)
1166 struct scsi_disk *sdkp = scsi_disk_get(bdev->bd_disk);
1167 struct scsi_device *sdev;
1168 int retval;
1170 if (!sdkp)
1171 return -ENXIO;
1173 SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp, "sd_open\n"));
1175 sdev = sdkp->device;
1178 * If the device is in error recovery, wait until it is done.
1179 * If the device is offline, then disallow any access to it.
1181 retval = -ENXIO;
1182 if (!scsi_block_when_processing_errors(sdev))
1183 goto error_out;
1185 if (sdev->removable || sdkp->write_prot)
1186 check_disk_change(bdev);
1189 * If the drive is empty, just let the open fail.
1191 retval = -ENOMEDIUM;
1192 if (sdev->removable && !sdkp->media_present && !(mode & FMODE_NDELAY))
1193 goto error_out;
1196 * If the device has the write protect tab set, have the open fail
1197 * if the user expects to be able to write to the thing.
1199 retval = -EROFS;
1200 if (sdkp->write_prot && (mode & FMODE_WRITE))
1201 goto error_out;
1204 * It is possible that the disk changing stuff resulted in
1205 * the device being taken offline. If this is the case,
1206 * report this to the user, and don't pretend that the
1207 * open actually succeeded.
1209 retval = -ENXIO;
1210 if (!scsi_device_online(sdev))
1211 goto error_out;
1213 if ((atomic_inc_return(&sdkp->openers) == 1) && sdev->removable) {
1214 if (scsi_block_when_processing_errors(sdev))
1215 scsi_set_medium_removal(sdev, SCSI_REMOVAL_PREVENT);
1218 return 0;
1220 error_out:
1221 scsi_disk_put(sdkp);
1222 return retval;
1226 * sd_release - invoked when the (last) close(2) is called on this
1227 * scsi disk.
1228 * @inode: only i_rdev member may be used
1229 * @filp: only f_mode and f_flags may be used
1231 * Returns 0.
1233 * Note: may block (uninterruptible) if error recovery is underway
1234 * on this disk.
1236 * Locking: called with bdev->bd_mutex held.
1238 static void sd_release(struct gendisk *disk, fmode_t mode)
1240 struct scsi_disk *sdkp = scsi_disk(disk);
1241 struct scsi_device *sdev = sdkp->device;
1243 SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp, "sd_release\n"));
1245 if (atomic_dec_return(&sdkp->openers) == 0 && sdev->removable) {
1246 if (scsi_block_when_processing_errors(sdev))
1247 scsi_set_medium_removal(sdev, SCSI_REMOVAL_ALLOW);
1251 * XXX and what if there are packets in flight and this close()
1252 * XXX is followed by a "rmmod sd_mod"?
1255 scsi_disk_put(sdkp);
1258 static int sd_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1260 struct scsi_disk *sdkp = scsi_disk(bdev->bd_disk);
1261 struct scsi_device *sdp = sdkp->device;
1262 struct Scsi_Host *host = sdp->host;
1263 int diskinfo[4];
1265 /* default to most commonly used values */
1266 diskinfo[0] = 0x40; /* 1 << 6 */
1267 diskinfo[1] = 0x20; /* 1 << 5 */
1268 diskinfo[2] = sdkp->capacity >> 11;
1270 /* override with calculated, extended default, or driver values */
1271 if (host->hostt->bios_param)
1272 host->hostt->bios_param(sdp, bdev, sdkp->capacity, diskinfo);
1273 else
1274 scsicam_bios_param(bdev, sdkp->capacity, diskinfo);
1276 geo->heads = diskinfo[0];
1277 geo->sectors = diskinfo[1];
1278 geo->cylinders = diskinfo[2];
1279 return 0;
1283 * sd_ioctl - process an ioctl
1284 * @inode: only i_rdev/i_bdev members may be used
1285 * @filp: only f_mode and f_flags may be used
1286 * @cmd: ioctl command number
1287 * @arg: this is third argument given to ioctl(2) system call.
1288 * Often contains a pointer.
1290 * Returns 0 if successful (some ioctls return positive numbers on
1291 * success as well). Returns a negated errno value in case of error.
1293 * Note: most ioctls are forward onto the block subsystem or further
1294 * down in the scsi subsystem.
1296 static int sd_ioctl(struct block_device *bdev, fmode_t mode,
1297 unsigned int cmd, unsigned long arg)
1299 struct gendisk *disk = bdev->bd_disk;
1300 struct scsi_disk *sdkp = scsi_disk(disk);
1301 struct scsi_device *sdp = sdkp->device;
1302 void __user *p = (void __user *)arg;
1303 int error;
1305 SCSI_LOG_IOCTL(1, sd_printk(KERN_INFO, sdkp, "sd_ioctl: disk=%s, "
1306 "cmd=0x%x\n", disk->disk_name, cmd));
1308 error = scsi_verify_blk_ioctl(bdev, cmd);
1309 if (error < 0)
1310 return error;
1313 * If we are in the middle of error recovery, don't let anyone
1314 * else try and use this device. Also, if error recovery fails, it
1315 * may try and take the device offline, in which case all further
1316 * access to the device is prohibited.
1318 error = scsi_ioctl_block_when_processing_errors(sdp, cmd,
1319 (mode & FMODE_NDELAY) != 0);
1320 if (error)
1321 goto out;
1324 * Send SCSI addressing ioctls directly to mid level, send other
1325 * ioctls to block level and then onto mid level if they can't be
1326 * resolved.
1328 switch (cmd) {
1329 case SCSI_IOCTL_GET_IDLUN:
1330 case SCSI_IOCTL_GET_BUS_NUMBER:
1331 error = scsi_ioctl(sdp, cmd, p);
1332 break;
1333 default:
1334 error = scsi_cmd_blk_ioctl(bdev, mode, cmd, p);
1335 if (error != -ENOTTY)
1336 break;
1337 error = scsi_ioctl(sdp, cmd, p);
1338 break;
1340 out:
1341 return error;
1344 static void set_media_not_present(struct scsi_disk *sdkp)
1346 if (sdkp->media_present)
1347 sdkp->device->changed = 1;
1349 if (sdkp->device->removable) {
1350 sdkp->media_present = 0;
1351 sdkp->capacity = 0;
1355 static int media_not_present(struct scsi_disk *sdkp,
1356 struct scsi_sense_hdr *sshdr)
1358 if (!scsi_sense_valid(sshdr))
1359 return 0;
1361 /* not invoked for commands that could return deferred errors */
1362 switch (sshdr->sense_key) {
1363 case UNIT_ATTENTION:
1364 case NOT_READY:
1365 /* medium not present */
1366 if (sshdr->asc == 0x3A) {
1367 set_media_not_present(sdkp);
1368 return 1;
1371 return 0;
1375 * sd_check_events - check media events
1376 * @disk: kernel device descriptor
1377 * @clearing: disk events currently being cleared
1379 * Returns mask of DISK_EVENT_*.
1381 * Note: this function is invoked from the block subsystem.
1383 static unsigned int sd_check_events(struct gendisk *disk, unsigned int clearing)
1385 struct scsi_disk *sdkp = scsi_disk(disk);
1386 struct scsi_device *sdp = sdkp->device;
1387 struct scsi_sense_hdr *sshdr = NULL;
1388 int retval;
1390 SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp, "sd_check_events\n"));
1393 * If the device is offline, don't send any commands - just pretend as
1394 * if the command failed. If the device ever comes back online, we
1395 * can deal with it then. It is only because of unrecoverable errors
1396 * that we would ever take a device offline in the first place.
1398 if (!scsi_device_online(sdp)) {
1399 set_media_not_present(sdkp);
1400 goto out;
1404 * Using TEST_UNIT_READY enables differentiation between drive with
1405 * no cartridge loaded - NOT READY, drive with changed cartridge -
1406 * UNIT ATTENTION, or with same cartridge - GOOD STATUS.
1408 * Drives that auto spin down. eg iomega jaz 1G, will be started
1409 * by sd_spinup_disk() from sd_revalidate_disk(), which happens whenever
1410 * sd_revalidate() is called.
1412 retval = -ENODEV;
1414 if (scsi_block_when_processing_errors(sdp)) {
1415 sshdr = kzalloc(sizeof(*sshdr), GFP_KERNEL);
1416 retval = scsi_test_unit_ready(sdp, SD_TIMEOUT, SD_MAX_RETRIES,
1417 sshdr);
1420 /* failed to execute TUR, assume media not present */
1421 if (host_byte(retval)) {
1422 set_media_not_present(sdkp);
1423 goto out;
1426 if (media_not_present(sdkp, sshdr))
1427 goto out;
1430 * For removable scsi disk we have to recognise the presence
1431 * of a disk in the drive.
1433 if (!sdkp->media_present)
1434 sdp->changed = 1;
1435 sdkp->media_present = 1;
1436 out:
1438 * sdp->changed is set under the following conditions:
1440 * Medium present state has changed in either direction.
1441 * Device has indicated UNIT_ATTENTION.
1443 kfree(sshdr);
1444 retval = sdp->changed ? DISK_EVENT_MEDIA_CHANGE : 0;
1445 sdp->changed = 0;
1446 return retval;
1449 static int sd_sync_cache(struct scsi_disk *sdkp)
1451 int retries, res;
1452 struct scsi_device *sdp = sdkp->device;
1453 const int timeout = sdp->request_queue->rq_timeout
1454 * SD_FLUSH_TIMEOUT_MULTIPLIER;
1455 struct scsi_sense_hdr sshdr;
1457 if (!scsi_device_online(sdp))
1458 return -ENODEV;
1460 for (retries = 3; retries > 0; --retries) {
1461 unsigned char cmd[10] = { 0 };
1463 cmd[0] = SYNCHRONIZE_CACHE;
1465 * Leave the rest of the command zero to indicate
1466 * flush everything.
1468 res = scsi_execute_req_flags(sdp, cmd, DMA_NONE, NULL, 0,
1469 &sshdr, timeout, SD_MAX_RETRIES,
1470 NULL, REQ_PM);
1471 if (res == 0)
1472 break;
1475 if (res) {
1476 sd_print_result(sdkp, "Synchronize Cache(10) failed", res);
1478 if (driver_byte(res) & DRIVER_SENSE)
1479 sd_print_sense_hdr(sdkp, &sshdr);
1480 /* we need to evaluate the error return */
1481 if (scsi_sense_valid(&sshdr) &&
1482 (sshdr.asc == 0x3a || /* medium not present */
1483 sshdr.asc == 0x20)) /* invalid command */
1484 /* this is no error here */
1485 return 0;
1487 switch (host_byte(res)) {
1488 /* ignore errors due to racing a disconnection */
1489 case DID_BAD_TARGET:
1490 case DID_NO_CONNECT:
1491 return 0;
1492 /* signal the upper layer it might try again */
1493 case DID_BUS_BUSY:
1494 case DID_IMM_RETRY:
1495 case DID_REQUEUE:
1496 case DID_SOFT_ERROR:
1497 return -EBUSY;
1498 default:
1499 return -EIO;
1502 return 0;
1505 static void sd_rescan(struct device *dev)
1507 struct scsi_disk *sdkp = dev_get_drvdata(dev);
1509 revalidate_disk(sdkp->disk);
1513 #ifdef CONFIG_COMPAT
1515 * This gets directly called from VFS. When the ioctl
1516 * is not recognized we go back to the other translation paths.
1518 static int sd_compat_ioctl(struct block_device *bdev, fmode_t mode,
1519 unsigned int cmd, unsigned long arg)
1521 struct scsi_device *sdev = scsi_disk(bdev->bd_disk)->device;
1522 int error;
1524 error = scsi_ioctl_block_when_processing_errors(sdev, cmd,
1525 (mode & FMODE_NDELAY) != 0);
1526 if (error)
1527 return error;
1530 * Let the static ioctl translation table take care of it.
1532 if (!sdev->host->hostt->compat_ioctl)
1533 return -ENOIOCTLCMD;
1534 return sdev->host->hostt->compat_ioctl(sdev, cmd, (void __user *)arg);
1536 #endif
1538 static const struct block_device_operations sd_fops = {
1539 .owner = THIS_MODULE,
1540 .open = sd_open,
1541 .release = sd_release,
1542 .ioctl = sd_ioctl,
1543 .getgeo = sd_getgeo,
1544 #ifdef CONFIG_COMPAT
1545 .compat_ioctl = sd_compat_ioctl,
1546 #endif
1547 .check_events = sd_check_events,
1548 .revalidate_disk = sd_revalidate_disk,
1549 .unlock_native_capacity = sd_unlock_native_capacity,
1553 * sd_eh_action - error handling callback
1554 * @scmd: sd-issued command that has failed
1555 * @eh_disp: The recovery disposition suggested by the midlayer
1557 * This function is called by the SCSI midlayer upon completion of an
1558 * error test command (currently TEST UNIT READY). The result of sending
1559 * the eh command is passed in eh_disp. We're looking for devices that
1560 * fail medium access commands but are OK with non access commands like
1561 * test unit ready (so wrongly see the device as having a successful
1562 * recovery)
1564 static int sd_eh_action(struct scsi_cmnd *scmd, int eh_disp)
1566 struct scsi_disk *sdkp = scsi_disk(scmd->request->rq_disk);
1568 if (!scsi_device_online(scmd->device) ||
1569 !scsi_medium_access_command(scmd) ||
1570 host_byte(scmd->result) != DID_TIME_OUT ||
1571 eh_disp != SUCCESS)
1572 return eh_disp;
1575 * The device has timed out executing a medium access command.
1576 * However, the TEST UNIT READY command sent during error
1577 * handling completed successfully. Either the device is in the
1578 * process of recovering or has it suffered an internal failure
1579 * that prevents access to the storage medium.
1581 sdkp->medium_access_timed_out++;
1584 * If the device keeps failing read/write commands but TEST UNIT
1585 * READY always completes successfully we assume that medium
1586 * access is no longer possible and take the device offline.
1588 if (sdkp->medium_access_timed_out >= sdkp->max_medium_access_timeouts) {
1589 scmd_printk(KERN_ERR, scmd,
1590 "Medium access timeout failure. Offlining disk!\n");
1591 scsi_device_set_state(scmd->device, SDEV_OFFLINE);
1593 return FAILED;
1596 return eh_disp;
1599 static unsigned int sd_completed_bytes(struct scsi_cmnd *scmd)
1601 u64 start_lba = blk_rq_pos(scmd->request);
1602 u64 end_lba = blk_rq_pos(scmd->request) + (scsi_bufflen(scmd) / 512);
1603 u64 factor = scmd->device->sector_size / 512;
1604 u64 bad_lba;
1605 int info_valid;
1607 * resid is optional but mostly filled in. When it's unused,
1608 * its value is zero, so we assume the whole buffer transferred
1610 unsigned int transferred = scsi_bufflen(scmd) - scsi_get_resid(scmd);
1611 unsigned int good_bytes;
1613 if (scmd->request->cmd_type != REQ_TYPE_FS)
1614 return 0;
1616 info_valid = scsi_get_sense_info_fld(scmd->sense_buffer,
1617 SCSI_SENSE_BUFFERSIZE,
1618 &bad_lba);
1619 if (!info_valid)
1620 return 0;
1622 if (scsi_bufflen(scmd) <= scmd->device->sector_size)
1623 return 0;
1625 /* be careful ... don't want any overflows */
1626 do_div(start_lba, factor);
1627 do_div(end_lba, factor);
1629 /* The bad lba was reported incorrectly, we have no idea where
1630 * the error is.
1632 if (bad_lba < start_lba || bad_lba >= end_lba)
1633 return 0;
1635 /* This computation should always be done in terms of
1636 * the resolution of the device's medium.
1638 good_bytes = (bad_lba - start_lba) * scmd->device->sector_size;
1639 return min(good_bytes, transferred);
1643 * sd_done - bottom half handler: called when the lower level
1644 * driver has completed (successfully or otherwise) a scsi command.
1645 * @SCpnt: mid-level's per command structure.
1647 * Note: potentially run from within an ISR. Must not block.
1649 static int sd_done(struct scsi_cmnd *SCpnt)
1651 int result = SCpnt->result;
1652 unsigned int good_bytes = result ? 0 : scsi_bufflen(SCpnt);
1653 struct scsi_sense_hdr sshdr;
1654 struct scsi_disk *sdkp = scsi_disk(SCpnt->request->rq_disk);
1655 struct request *req = SCpnt->request;
1656 int sense_valid = 0;
1657 int sense_deferred = 0;
1658 unsigned char op = SCpnt->cmnd[0];
1659 unsigned char unmap = SCpnt->cmnd[1] & 8;
1661 if (req->cmd_flags & REQ_DISCARD || req->cmd_flags & REQ_WRITE_SAME) {
1662 if (!result) {
1663 good_bytes = blk_rq_bytes(req);
1664 scsi_set_resid(SCpnt, 0);
1665 } else {
1666 good_bytes = 0;
1667 scsi_set_resid(SCpnt, blk_rq_bytes(req));
1671 if (result) {
1672 sense_valid = scsi_command_normalize_sense(SCpnt, &sshdr);
1673 if (sense_valid)
1674 sense_deferred = scsi_sense_is_deferred(&sshdr);
1676 sdkp->medium_access_timed_out = 0;
1678 if (driver_byte(result) != DRIVER_SENSE &&
1679 (!sense_valid || sense_deferred))
1680 goto out;
1682 switch (sshdr.sense_key) {
1683 case HARDWARE_ERROR:
1684 case MEDIUM_ERROR:
1685 good_bytes = sd_completed_bytes(SCpnt);
1686 break;
1687 case RECOVERED_ERROR:
1688 good_bytes = scsi_bufflen(SCpnt);
1689 break;
1690 case NO_SENSE:
1691 /* This indicates a false check condition, so ignore it. An
1692 * unknown amount of data was transferred so treat it as an
1693 * error.
1695 SCpnt->result = 0;
1696 memset(SCpnt->sense_buffer, 0, SCSI_SENSE_BUFFERSIZE);
1697 break;
1698 case ABORTED_COMMAND:
1699 if (sshdr.asc == 0x10) /* DIF: Target detected corruption */
1700 good_bytes = sd_completed_bytes(SCpnt);
1701 break;
1702 case ILLEGAL_REQUEST:
1703 if (sshdr.asc == 0x10) /* DIX: Host detected corruption */
1704 good_bytes = sd_completed_bytes(SCpnt);
1705 /* INVALID COMMAND OPCODE or INVALID FIELD IN CDB */
1706 if (sshdr.asc == 0x20 || sshdr.asc == 0x24) {
1707 switch (op) {
1708 case UNMAP:
1709 sd_config_discard(sdkp, SD_LBP_DISABLE);
1710 break;
1711 case WRITE_SAME_16:
1712 case WRITE_SAME:
1713 if (unmap)
1714 sd_config_discard(sdkp, SD_LBP_DISABLE);
1715 else {
1716 sdkp->device->no_write_same = 1;
1717 sd_config_write_same(sdkp);
1719 good_bytes = 0;
1720 req->__data_len = blk_rq_bytes(req);
1721 req->cmd_flags |= REQ_QUIET;
1725 break;
1726 default:
1727 break;
1729 out:
1730 SCSI_LOG_HLCOMPLETE(1, scmd_printk(KERN_INFO, SCpnt,
1731 "sd_done: completed %d of %d bytes\n",
1732 good_bytes, scsi_bufflen(SCpnt)));
1734 if (rq_data_dir(SCpnt->request) == READ && scsi_prot_sg_count(SCpnt))
1735 sd_dif_complete(SCpnt, good_bytes);
1737 return good_bytes;
1741 * spinup disk - called only in sd_revalidate_disk()
1743 static void
1744 sd_spinup_disk(struct scsi_disk *sdkp)
1746 unsigned char cmd[10];
1747 unsigned long spintime_expire = 0;
1748 int retries, spintime;
1749 unsigned int the_result;
1750 struct scsi_sense_hdr sshdr;
1751 int sense_valid = 0;
1753 spintime = 0;
1755 /* Spin up drives, as required. Only do this at boot time */
1756 /* Spinup needs to be done for module loads too. */
1757 do {
1758 retries = 0;
1760 do {
1761 cmd[0] = TEST_UNIT_READY;
1762 memset((void *) &cmd[1], 0, 9);
1764 the_result = scsi_execute_req(sdkp->device, cmd,
1765 DMA_NONE, NULL, 0,
1766 &sshdr, SD_TIMEOUT,
1767 SD_MAX_RETRIES, NULL);
1770 * If the drive has indicated to us that it
1771 * doesn't have any media in it, don't bother
1772 * with any more polling.
1774 if (media_not_present(sdkp, &sshdr))
1775 return;
1777 if (the_result)
1778 sense_valid = scsi_sense_valid(&sshdr);
1779 retries++;
1780 } while (retries < 3 &&
1781 (!scsi_status_is_good(the_result) ||
1782 ((driver_byte(the_result) & DRIVER_SENSE) &&
1783 sense_valid && sshdr.sense_key == UNIT_ATTENTION)));
1785 if ((driver_byte(the_result) & DRIVER_SENSE) == 0) {
1786 /* no sense, TUR either succeeded or failed
1787 * with a status error */
1788 if(!spintime && !scsi_status_is_good(the_result)) {
1789 sd_print_result(sdkp, "Test Unit Ready failed",
1790 the_result);
1792 break;
1796 * The device does not want the automatic start to be issued.
1798 if (sdkp->device->no_start_on_add)
1799 break;
1801 if (sense_valid && sshdr.sense_key == NOT_READY) {
1802 if (sshdr.asc == 4 && sshdr.ascq == 3)
1803 break; /* manual intervention required */
1804 if (sshdr.asc == 4 && sshdr.ascq == 0xb)
1805 break; /* standby */
1806 if (sshdr.asc == 4 && sshdr.ascq == 0xc)
1807 break; /* unavailable */
1809 * Issue command to spin up drive when not ready
1811 if (!spintime) {
1812 sd_printk(KERN_NOTICE, sdkp, "Spinning up disk...");
1813 cmd[0] = START_STOP;
1814 cmd[1] = 1; /* Return immediately */
1815 memset((void *) &cmd[2], 0, 8);
1816 cmd[4] = 1; /* Start spin cycle */
1817 if (sdkp->device->start_stop_pwr_cond)
1818 cmd[4] |= 1 << 4;
1819 scsi_execute_req(sdkp->device, cmd, DMA_NONE,
1820 NULL, 0, &sshdr,
1821 SD_TIMEOUT, SD_MAX_RETRIES,
1822 NULL);
1823 spintime_expire = jiffies + 100 * HZ;
1824 spintime = 1;
1826 /* Wait 1 second for next try */
1827 msleep(1000);
1828 printk(".");
1831 * Wait for USB flash devices with slow firmware.
1832 * Yes, this sense key/ASC combination shouldn't
1833 * occur here. It's characteristic of these devices.
1835 } else if (sense_valid &&
1836 sshdr.sense_key == UNIT_ATTENTION &&
1837 sshdr.asc == 0x28) {
1838 if (!spintime) {
1839 spintime_expire = jiffies + 5 * HZ;
1840 spintime = 1;
1842 /* Wait 1 second for next try */
1843 msleep(1000);
1844 } else {
1845 /* we don't understand the sense code, so it's
1846 * probably pointless to loop */
1847 if(!spintime) {
1848 sd_printk(KERN_NOTICE, sdkp, "Unit Not Ready\n");
1849 sd_print_sense_hdr(sdkp, &sshdr);
1851 break;
1854 } while (spintime && time_before_eq(jiffies, spintime_expire));
1856 if (spintime) {
1857 if (scsi_status_is_good(the_result))
1858 printk("ready\n");
1859 else
1860 printk("not responding...\n");
1866 * Determine whether disk supports Data Integrity Field.
1868 static int sd_read_protection_type(struct scsi_disk *sdkp, unsigned char *buffer)
1870 struct scsi_device *sdp = sdkp->device;
1871 u8 type;
1872 int ret = 0;
1874 if (scsi_device_protection(sdp) == 0 || (buffer[12] & 1) == 0)
1875 return ret;
1877 type = ((buffer[12] >> 1) & 7) + 1; /* P_TYPE 0 = Type 1 */
1879 if (type > SD_DIF_TYPE3_PROTECTION)
1880 ret = -ENODEV;
1881 else if (scsi_host_dif_capable(sdp->host, type))
1882 ret = 1;
1884 if (sdkp->first_scan || type != sdkp->protection_type)
1885 switch (ret) {
1886 case -ENODEV:
1887 sd_printk(KERN_ERR, sdkp, "formatted with unsupported" \
1888 " protection type %u. Disabling disk!\n",
1889 type);
1890 break;
1891 case 1:
1892 sd_printk(KERN_NOTICE, sdkp,
1893 "Enabling DIF Type %u protection\n", type);
1894 break;
1895 case 0:
1896 sd_printk(KERN_NOTICE, sdkp,
1897 "Disabling DIF Type %u protection\n", type);
1898 break;
1901 sdkp->protection_type = type;
1903 return ret;
1906 static void read_capacity_error(struct scsi_disk *sdkp, struct scsi_device *sdp,
1907 struct scsi_sense_hdr *sshdr, int sense_valid,
1908 int the_result)
1910 if (driver_byte(the_result) & DRIVER_SENSE)
1911 sd_print_sense_hdr(sdkp, sshdr);
1912 else
1913 sd_printk(KERN_NOTICE, sdkp, "Sense not available.\n");
1916 * Set dirty bit for removable devices if not ready -
1917 * sometimes drives will not report this properly.
1919 if (sdp->removable &&
1920 sense_valid && sshdr->sense_key == NOT_READY)
1921 set_media_not_present(sdkp);
1924 * We used to set media_present to 0 here to indicate no media
1925 * in the drive, but some drives fail read capacity even with
1926 * media present, so we can't do that.
1928 sdkp->capacity = 0; /* unknown mapped to zero - as usual */
1931 #define RC16_LEN 32
1932 #if RC16_LEN > SD_BUF_SIZE
1933 #error RC16_LEN must not be more than SD_BUF_SIZE
1934 #endif
1936 #define READ_CAPACITY_RETRIES_ON_RESET 10
1938 static int read_capacity_16(struct scsi_disk *sdkp, struct scsi_device *sdp,
1939 unsigned char *buffer)
1941 unsigned char cmd[16];
1942 struct scsi_sense_hdr sshdr;
1943 int sense_valid = 0;
1944 int the_result;
1945 int retries = 3, reset_retries = READ_CAPACITY_RETRIES_ON_RESET;
1946 unsigned int alignment;
1947 unsigned long long lba;
1948 unsigned sector_size;
1950 if (sdp->no_read_capacity_16)
1951 return -EINVAL;
1953 do {
1954 memset(cmd, 0, 16);
1955 cmd[0] = SERVICE_ACTION_IN_16;
1956 cmd[1] = SAI_READ_CAPACITY_16;
1957 cmd[13] = RC16_LEN;
1958 memset(buffer, 0, RC16_LEN);
1960 the_result = scsi_execute_req(sdp, cmd, DMA_FROM_DEVICE,
1961 buffer, RC16_LEN, &sshdr,
1962 SD_TIMEOUT, SD_MAX_RETRIES, NULL);
1964 if (media_not_present(sdkp, &sshdr))
1965 return -ENODEV;
1967 if (the_result) {
1968 sense_valid = scsi_sense_valid(&sshdr);
1969 if (sense_valid &&
1970 sshdr.sense_key == ILLEGAL_REQUEST &&
1971 (sshdr.asc == 0x20 || sshdr.asc == 0x24) &&
1972 sshdr.ascq == 0x00)
1973 /* Invalid Command Operation Code or
1974 * Invalid Field in CDB, just retry
1975 * silently with RC10 */
1976 return -EINVAL;
1977 if (sense_valid &&
1978 sshdr.sense_key == UNIT_ATTENTION &&
1979 sshdr.asc == 0x29 && sshdr.ascq == 0x00)
1980 /* Device reset might occur several times,
1981 * give it one more chance */
1982 if (--reset_retries > 0)
1983 continue;
1985 retries--;
1987 } while (the_result && retries);
1989 if (the_result) {
1990 sd_print_result(sdkp, "Read Capacity(16) failed", the_result);
1991 read_capacity_error(sdkp, sdp, &sshdr, sense_valid, the_result);
1992 return -EINVAL;
1995 sector_size = get_unaligned_be32(&buffer[8]);
1996 lba = get_unaligned_be64(&buffer[0]);
1998 if (sd_read_protection_type(sdkp, buffer) < 0) {
1999 sdkp->capacity = 0;
2000 return -ENODEV;
2003 if ((sizeof(sdkp->capacity) == 4) && (lba >= 0xffffffffULL)) {
2004 sd_printk(KERN_ERR, sdkp, "Too big for this kernel. Use a "
2005 "kernel compiled with support for large block "
2006 "devices.\n");
2007 sdkp->capacity = 0;
2008 return -EOVERFLOW;
2011 /* Logical blocks per physical block exponent */
2012 sdkp->physical_block_size = (1 << (buffer[13] & 0xf)) * sector_size;
2014 /* Lowest aligned logical block */
2015 alignment = ((buffer[14] & 0x3f) << 8 | buffer[15]) * sector_size;
2016 blk_queue_alignment_offset(sdp->request_queue, alignment);
2017 if (alignment && sdkp->first_scan)
2018 sd_printk(KERN_NOTICE, sdkp,
2019 "physical block alignment offset: %u\n", alignment);
2021 if (buffer[14] & 0x80) { /* LBPME */
2022 sdkp->lbpme = 1;
2024 if (buffer[14] & 0x40) /* LBPRZ */
2025 sdkp->lbprz = 1;
2027 sd_config_discard(sdkp, SD_LBP_WS16);
2030 sdkp->capacity = lba + 1;
2031 return sector_size;
2034 static int read_capacity_10(struct scsi_disk *sdkp, struct scsi_device *sdp,
2035 unsigned char *buffer)
2037 unsigned char cmd[16];
2038 struct scsi_sense_hdr sshdr;
2039 int sense_valid = 0;
2040 int the_result;
2041 int retries = 3, reset_retries = READ_CAPACITY_RETRIES_ON_RESET;
2042 sector_t lba;
2043 unsigned sector_size;
2045 do {
2046 cmd[0] = READ_CAPACITY;
2047 memset(&cmd[1], 0, 9);
2048 memset(buffer, 0, 8);
2050 the_result = scsi_execute_req(sdp, cmd, DMA_FROM_DEVICE,
2051 buffer, 8, &sshdr,
2052 SD_TIMEOUT, SD_MAX_RETRIES, NULL);
2054 if (media_not_present(sdkp, &sshdr))
2055 return -ENODEV;
2057 if (the_result) {
2058 sense_valid = scsi_sense_valid(&sshdr);
2059 if (sense_valid &&
2060 sshdr.sense_key == UNIT_ATTENTION &&
2061 sshdr.asc == 0x29 && sshdr.ascq == 0x00)
2062 /* Device reset might occur several times,
2063 * give it one more chance */
2064 if (--reset_retries > 0)
2065 continue;
2067 retries--;
2069 } while (the_result && retries);
2071 if (the_result) {
2072 sd_print_result(sdkp, "Read Capacity(10) failed", the_result);
2073 read_capacity_error(sdkp, sdp, &sshdr, sense_valid, the_result);
2074 return -EINVAL;
2077 sector_size = get_unaligned_be32(&buffer[4]);
2078 lba = get_unaligned_be32(&buffer[0]);
2080 if (sdp->no_read_capacity_16 && (lba == 0xffffffff)) {
2081 /* Some buggy (usb cardreader) devices return an lba of
2082 0xffffffff when the want to report a size of 0 (with
2083 which they really mean no media is present) */
2084 sdkp->capacity = 0;
2085 sdkp->physical_block_size = sector_size;
2086 return sector_size;
2089 if ((sizeof(sdkp->capacity) == 4) && (lba == 0xffffffff)) {
2090 sd_printk(KERN_ERR, sdkp, "Too big for this kernel. Use a "
2091 "kernel compiled with support for large block "
2092 "devices.\n");
2093 sdkp->capacity = 0;
2094 return -EOVERFLOW;
2097 sdkp->capacity = lba + 1;
2098 sdkp->physical_block_size = sector_size;
2099 return sector_size;
2102 static int sd_try_rc16_first(struct scsi_device *sdp)
2104 if (sdp->host->max_cmd_len < 16)
2105 return 0;
2106 if (sdp->try_rc_10_first)
2107 return 0;
2108 if (sdp->scsi_level > SCSI_SPC_2)
2109 return 1;
2110 if (scsi_device_protection(sdp))
2111 return 1;
2112 return 0;
2116 * read disk capacity
2118 static void
2119 sd_read_capacity(struct scsi_disk *sdkp, unsigned char *buffer)
2121 int sector_size;
2122 struct scsi_device *sdp = sdkp->device;
2123 sector_t old_capacity = sdkp->capacity;
2125 if (sd_try_rc16_first(sdp)) {
2126 sector_size = read_capacity_16(sdkp, sdp, buffer);
2127 if (sector_size == -EOVERFLOW)
2128 goto got_data;
2129 if (sector_size == -ENODEV)
2130 return;
2131 if (sector_size < 0)
2132 sector_size = read_capacity_10(sdkp, sdp, buffer);
2133 if (sector_size < 0)
2134 return;
2135 } else {
2136 sector_size = read_capacity_10(sdkp, sdp, buffer);
2137 if (sector_size == -EOVERFLOW)
2138 goto got_data;
2139 if (sector_size < 0)
2140 return;
2141 if ((sizeof(sdkp->capacity) > 4) &&
2142 (sdkp->capacity > 0xffffffffULL)) {
2143 int old_sector_size = sector_size;
2144 sd_printk(KERN_NOTICE, sdkp, "Very big device. "
2145 "Trying to use READ CAPACITY(16).\n");
2146 sector_size = read_capacity_16(sdkp, sdp, buffer);
2147 if (sector_size < 0) {
2148 sd_printk(KERN_NOTICE, sdkp,
2149 "Using 0xffffffff as device size\n");
2150 sdkp->capacity = 1 + (sector_t) 0xffffffff;
2151 sector_size = old_sector_size;
2152 goto got_data;
2157 /* Some devices are known to return the total number of blocks,
2158 * not the highest block number. Some devices have versions
2159 * which do this and others which do not. Some devices we might
2160 * suspect of doing this but we don't know for certain.
2162 * If we know the reported capacity is wrong, decrement it. If
2163 * we can only guess, then assume the number of blocks is even
2164 * (usually true but not always) and err on the side of lowering
2165 * the capacity.
2167 if (sdp->fix_capacity ||
2168 (sdp->guess_capacity && (sdkp->capacity & 0x01))) {
2169 sd_printk(KERN_INFO, sdkp, "Adjusting the sector count "
2170 "from its reported value: %llu\n",
2171 (unsigned long long) sdkp->capacity);
2172 --sdkp->capacity;
2175 got_data:
2176 if (sector_size == 0) {
2177 sector_size = 512;
2178 sd_printk(KERN_NOTICE, sdkp, "Sector size 0 reported, "
2179 "assuming 512.\n");
2182 if (sector_size != 512 &&
2183 sector_size != 1024 &&
2184 sector_size != 2048 &&
2185 sector_size != 4096) {
2186 sd_printk(KERN_NOTICE, sdkp, "Unsupported sector size %d.\n",
2187 sector_size);
2189 * The user might want to re-format the drive with
2190 * a supported sectorsize. Once this happens, it
2191 * would be relatively trivial to set the thing up.
2192 * For this reason, we leave the thing in the table.
2194 sdkp->capacity = 0;
2196 * set a bogus sector size so the normal read/write
2197 * logic in the block layer will eventually refuse any
2198 * request on this device without tripping over power
2199 * of two sector size assumptions
2201 sector_size = 512;
2203 blk_queue_logical_block_size(sdp->request_queue, sector_size);
2206 char cap_str_2[10], cap_str_10[10];
2208 string_get_size(sdkp->capacity, sector_size,
2209 STRING_UNITS_2, cap_str_2, sizeof(cap_str_2));
2210 string_get_size(sdkp->capacity, sector_size,
2211 STRING_UNITS_10, cap_str_10,
2212 sizeof(cap_str_10));
2214 if (sdkp->first_scan || old_capacity != sdkp->capacity) {
2215 sd_printk(KERN_NOTICE, sdkp,
2216 "%llu %d-byte logical blocks: (%s/%s)\n",
2217 (unsigned long long)sdkp->capacity,
2218 sector_size, cap_str_10, cap_str_2);
2220 if (sdkp->physical_block_size != sector_size)
2221 sd_printk(KERN_NOTICE, sdkp,
2222 "%u-byte physical blocks\n",
2223 sdkp->physical_block_size);
2227 if (sdkp->capacity > 0xffffffff) {
2228 sdp->use_16_for_rw = 1;
2229 sdkp->max_xfer_blocks = SD_MAX_XFER_BLOCKS;
2230 } else
2231 sdkp->max_xfer_blocks = SD_DEF_XFER_BLOCKS;
2233 /* Rescale capacity to 512-byte units */
2234 if (sector_size == 4096)
2235 sdkp->capacity <<= 3;
2236 else if (sector_size == 2048)
2237 sdkp->capacity <<= 2;
2238 else if (sector_size == 1024)
2239 sdkp->capacity <<= 1;
2241 blk_queue_physical_block_size(sdp->request_queue,
2242 sdkp->physical_block_size);
2243 sdkp->device->sector_size = sector_size;
2246 /* called with buffer of length 512 */
2247 static inline int
2248 sd_do_mode_sense(struct scsi_device *sdp, int dbd, int modepage,
2249 unsigned char *buffer, int len, struct scsi_mode_data *data,
2250 struct scsi_sense_hdr *sshdr)
2252 return scsi_mode_sense(sdp, dbd, modepage, buffer, len,
2253 SD_TIMEOUT, SD_MAX_RETRIES, data,
2254 sshdr);
2258 * read write protect setting, if possible - called only in sd_revalidate_disk()
2259 * called with buffer of length SD_BUF_SIZE
2261 static void
2262 sd_read_write_protect_flag(struct scsi_disk *sdkp, unsigned char *buffer)
2264 int res;
2265 struct scsi_device *sdp = sdkp->device;
2266 struct scsi_mode_data data;
2267 int old_wp = sdkp->write_prot;
2269 set_disk_ro(sdkp->disk, 0);
2270 if (sdp->skip_ms_page_3f) {
2271 sd_first_printk(KERN_NOTICE, sdkp, "Assuming Write Enabled\n");
2272 return;
2275 if (sdp->use_192_bytes_for_3f) {
2276 res = sd_do_mode_sense(sdp, 0, 0x3F, buffer, 192, &data, NULL);
2277 } else {
2279 * First attempt: ask for all pages (0x3F), but only 4 bytes.
2280 * We have to start carefully: some devices hang if we ask
2281 * for more than is available.
2283 res = sd_do_mode_sense(sdp, 0, 0x3F, buffer, 4, &data, NULL);
2286 * Second attempt: ask for page 0 When only page 0 is
2287 * implemented, a request for page 3F may return Sense Key
2288 * 5: Illegal Request, Sense Code 24: Invalid field in
2289 * CDB.
2291 if (!scsi_status_is_good(res))
2292 res = sd_do_mode_sense(sdp, 0, 0, buffer, 4, &data, NULL);
2295 * Third attempt: ask 255 bytes, as we did earlier.
2297 if (!scsi_status_is_good(res))
2298 res = sd_do_mode_sense(sdp, 0, 0x3F, buffer, 255,
2299 &data, NULL);
2302 if (!scsi_status_is_good(res)) {
2303 sd_first_printk(KERN_WARNING, sdkp,
2304 "Test WP failed, assume Write Enabled\n");
2305 } else {
2306 sdkp->write_prot = ((data.device_specific & 0x80) != 0);
2307 set_disk_ro(sdkp->disk, sdkp->write_prot);
2308 if (sdkp->first_scan || old_wp != sdkp->write_prot) {
2309 sd_printk(KERN_NOTICE, sdkp, "Write Protect is %s\n",
2310 sdkp->write_prot ? "on" : "off");
2311 sd_printk(KERN_DEBUG, sdkp,
2312 "Mode Sense: %02x %02x %02x %02x\n",
2313 buffer[0], buffer[1], buffer[2], buffer[3]);
2319 * sd_read_cache_type - called only from sd_revalidate_disk()
2320 * called with buffer of length SD_BUF_SIZE
2322 static void
2323 sd_read_cache_type(struct scsi_disk *sdkp, unsigned char *buffer)
2325 int len = 0, res;
2326 struct scsi_device *sdp = sdkp->device;
2328 int dbd;
2329 int modepage;
2330 int first_len;
2331 struct scsi_mode_data data;
2332 struct scsi_sense_hdr sshdr;
2333 int old_wce = sdkp->WCE;
2334 int old_rcd = sdkp->RCD;
2335 int old_dpofua = sdkp->DPOFUA;
2338 if (sdkp->cache_override)
2339 return;
2341 first_len = 4;
2342 if (sdp->skip_ms_page_8) {
2343 if (sdp->type == TYPE_RBC)
2344 goto defaults;
2345 else {
2346 if (sdp->skip_ms_page_3f)
2347 goto defaults;
2348 modepage = 0x3F;
2349 if (sdp->use_192_bytes_for_3f)
2350 first_len = 192;
2351 dbd = 0;
2353 } else if (sdp->type == TYPE_RBC) {
2354 modepage = 6;
2355 dbd = 8;
2356 } else {
2357 modepage = 8;
2358 dbd = 0;
2361 /* cautiously ask */
2362 res = sd_do_mode_sense(sdp, dbd, modepage, buffer, first_len,
2363 &data, &sshdr);
2365 if (!scsi_status_is_good(res))
2366 goto bad_sense;
2368 if (!data.header_length) {
2369 modepage = 6;
2370 first_len = 0;
2371 sd_first_printk(KERN_ERR, sdkp,
2372 "Missing header in MODE_SENSE response\n");
2375 /* that went OK, now ask for the proper length */
2376 len = data.length;
2379 * We're only interested in the first three bytes, actually.
2380 * But the data cache page is defined for the first 20.
2382 if (len < 3)
2383 goto bad_sense;
2384 else if (len > SD_BUF_SIZE) {
2385 sd_first_printk(KERN_NOTICE, sdkp, "Truncating mode parameter "
2386 "data from %d to %d bytes\n", len, SD_BUF_SIZE);
2387 len = SD_BUF_SIZE;
2389 if (modepage == 0x3F && sdp->use_192_bytes_for_3f)
2390 len = 192;
2392 /* Get the data */
2393 if (len > first_len)
2394 res = sd_do_mode_sense(sdp, dbd, modepage, buffer, len,
2395 &data, &sshdr);
2397 if (scsi_status_is_good(res)) {
2398 int offset = data.header_length + data.block_descriptor_length;
2400 while (offset < len) {
2401 u8 page_code = buffer[offset] & 0x3F;
2402 u8 spf = buffer[offset] & 0x40;
2404 if (page_code == 8 || page_code == 6) {
2405 /* We're interested only in the first 3 bytes.
2407 if (len - offset <= 2) {
2408 sd_first_printk(KERN_ERR, sdkp,
2409 "Incomplete mode parameter "
2410 "data\n");
2411 goto defaults;
2412 } else {
2413 modepage = page_code;
2414 goto Page_found;
2416 } else {
2417 /* Go to the next page */
2418 if (spf && len - offset > 3)
2419 offset += 4 + (buffer[offset+2] << 8) +
2420 buffer[offset+3];
2421 else if (!spf && len - offset > 1)
2422 offset += 2 + buffer[offset+1];
2423 else {
2424 sd_first_printk(KERN_ERR, sdkp,
2425 "Incomplete mode "
2426 "parameter data\n");
2427 goto defaults;
2432 sd_first_printk(KERN_ERR, sdkp, "No Caching mode page found\n");
2433 goto defaults;
2435 Page_found:
2436 if (modepage == 8) {
2437 sdkp->WCE = ((buffer[offset + 2] & 0x04) != 0);
2438 sdkp->RCD = ((buffer[offset + 2] & 0x01) != 0);
2439 } else {
2440 sdkp->WCE = ((buffer[offset + 2] & 0x01) == 0);
2441 sdkp->RCD = 0;
2444 sdkp->DPOFUA = (data.device_specific & 0x10) != 0;
2445 if (sdp->broken_fua) {
2446 sd_first_printk(KERN_NOTICE, sdkp, "Disabling FUA\n");
2447 sdkp->DPOFUA = 0;
2448 } else if (sdkp->DPOFUA && !sdkp->device->use_10_for_rw) {
2449 sd_first_printk(KERN_NOTICE, sdkp,
2450 "Uses READ/WRITE(6), disabling FUA\n");
2451 sdkp->DPOFUA = 0;
2454 /* No cache flush allowed for write protected devices */
2455 if (sdkp->WCE && sdkp->write_prot)
2456 sdkp->WCE = 0;
2458 if (sdkp->first_scan || old_wce != sdkp->WCE ||
2459 old_rcd != sdkp->RCD || old_dpofua != sdkp->DPOFUA)
2460 sd_printk(KERN_NOTICE, sdkp,
2461 "Write cache: %s, read cache: %s, %s\n",
2462 sdkp->WCE ? "enabled" : "disabled",
2463 sdkp->RCD ? "disabled" : "enabled",
2464 sdkp->DPOFUA ? "supports DPO and FUA"
2465 : "doesn't support DPO or FUA");
2467 return;
2470 bad_sense:
2471 if (scsi_sense_valid(&sshdr) &&
2472 sshdr.sense_key == ILLEGAL_REQUEST &&
2473 sshdr.asc == 0x24 && sshdr.ascq == 0x0)
2474 /* Invalid field in CDB */
2475 sd_first_printk(KERN_NOTICE, sdkp, "Cache data unavailable\n");
2476 else
2477 sd_first_printk(KERN_ERR, sdkp,
2478 "Asking for cache data failed\n");
2480 defaults:
2481 if (sdp->wce_default_on) {
2482 sd_first_printk(KERN_NOTICE, sdkp,
2483 "Assuming drive cache: write back\n");
2484 sdkp->WCE = 1;
2485 } else {
2486 sd_first_printk(KERN_ERR, sdkp,
2487 "Assuming drive cache: write through\n");
2488 sdkp->WCE = 0;
2490 sdkp->RCD = 0;
2491 sdkp->DPOFUA = 0;
2495 * The ATO bit indicates whether the DIF application tag is available
2496 * for use by the operating system.
2498 static void sd_read_app_tag_own(struct scsi_disk *sdkp, unsigned char *buffer)
2500 int res, offset;
2501 struct scsi_device *sdp = sdkp->device;
2502 struct scsi_mode_data data;
2503 struct scsi_sense_hdr sshdr;
2505 if (sdp->type != TYPE_DISK)
2506 return;
2508 if (sdkp->protection_type == 0)
2509 return;
2511 res = scsi_mode_sense(sdp, 1, 0x0a, buffer, 36, SD_TIMEOUT,
2512 SD_MAX_RETRIES, &data, &sshdr);
2514 if (!scsi_status_is_good(res) || !data.header_length ||
2515 data.length < 6) {
2516 sd_first_printk(KERN_WARNING, sdkp,
2517 "getting Control mode page failed, assume no ATO\n");
2519 if (scsi_sense_valid(&sshdr))
2520 sd_print_sense_hdr(sdkp, &sshdr);
2522 return;
2525 offset = data.header_length + data.block_descriptor_length;
2527 if ((buffer[offset] & 0x3f) != 0x0a) {
2528 sd_first_printk(KERN_ERR, sdkp, "ATO Got wrong page\n");
2529 return;
2532 if ((buffer[offset + 5] & 0x80) == 0)
2533 return;
2535 sdkp->ATO = 1;
2537 return;
2541 * sd_read_block_limits - Query disk device for preferred I/O sizes.
2542 * @disk: disk to query
2544 static void sd_read_block_limits(struct scsi_disk *sdkp)
2546 unsigned int sector_sz = sdkp->device->sector_size;
2547 const int vpd_len = 64;
2548 u32 max_xfer_length;
2549 unsigned char *buffer = kmalloc(vpd_len, GFP_KERNEL);
2551 if (!buffer ||
2552 /* Block Limits VPD */
2553 scsi_get_vpd_page(sdkp->device, 0xb0, buffer, vpd_len))
2554 goto out;
2556 max_xfer_length = get_unaligned_be32(&buffer[8]);
2557 if (max_xfer_length)
2558 sdkp->max_xfer_blocks = max_xfer_length;
2560 blk_queue_io_min(sdkp->disk->queue,
2561 get_unaligned_be16(&buffer[6]) * sector_sz);
2562 blk_queue_io_opt(sdkp->disk->queue,
2563 get_unaligned_be32(&buffer[12]) * sector_sz);
2565 if (buffer[3] == 0x3c) {
2566 unsigned int lba_count, desc_count;
2568 sdkp->max_ws_blocks = (u32)get_unaligned_be64(&buffer[36]);
2570 if (!sdkp->lbpme)
2571 goto out;
2573 lba_count = get_unaligned_be32(&buffer[20]);
2574 desc_count = get_unaligned_be32(&buffer[24]);
2576 if (lba_count && desc_count)
2577 sdkp->max_unmap_blocks = lba_count;
2579 sdkp->unmap_granularity = get_unaligned_be32(&buffer[28]);
2581 if (buffer[32] & 0x80)
2582 sdkp->unmap_alignment =
2583 get_unaligned_be32(&buffer[32]) & ~(1 << 31);
2585 if (!sdkp->lbpvpd) { /* LBP VPD page not provided */
2587 if (sdkp->max_unmap_blocks)
2588 sd_config_discard(sdkp, SD_LBP_UNMAP);
2589 else
2590 sd_config_discard(sdkp, SD_LBP_WS16);
2592 } else { /* LBP VPD page tells us what to use */
2593 if (sdkp->lbpu && sdkp->max_unmap_blocks && !sdkp->lbprz)
2594 sd_config_discard(sdkp, SD_LBP_UNMAP);
2595 else if (sdkp->lbpws)
2596 sd_config_discard(sdkp, SD_LBP_WS16);
2597 else if (sdkp->lbpws10)
2598 sd_config_discard(sdkp, SD_LBP_WS10);
2599 else if (sdkp->lbpu && sdkp->max_unmap_blocks)
2600 sd_config_discard(sdkp, SD_LBP_UNMAP);
2601 else
2602 sd_config_discard(sdkp, SD_LBP_DISABLE);
2606 out:
2607 kfree(buffer);
2611 * sd_read_block_characteristics - Query block dev. characteristics
2612 * @disk: disk to query
2614 static void sd_read_block_characteristics(struct scsi_disk *sdkp)
2616 unsigned char *buffer;
2617 u16 rot;
2618 const int vpd_len = 64;
2620 buffer = kmalloc(vpd_len, GFP_KERNEL);
2622 if (!buffer ||
2623 /* Block Device Characteristics VPD */
2624 scsi_get_vpd_page(sdkp->device, 0xb1, buffer, vpd_len))
2625 goto out;
2627 rot = get_unaligned_be16(&buffer[4]);
2629 if (rot == 1) {
2630 queue_flag_set_unlocked(QUEUE_FLAG_NONROT, sdkp->disk->queue);
2631 queue_flag_clear_unlocked(QUEUE_FLAG_ADD_RANDOM, sdkp->disk->queue);
2634 out:
2635 kfree(buffer);
2639 * sd_read_block_provisioning - Query provisioning VPD page
2640 * @disk: disk to query
2642 static void sd_read_block_provisioning(struct scsi_disk *sdkp)
2644 unsigned char *buffer;
2645 const int vpd_len = 8;
2647 if (sdkp->lbpme == 0)
2648 return;
2650 buffer = kmalloc(vpd_len, GFP_KERNEL);
2652 if (!buffer || scsi_get_vpd_page(sdkp->device, 0xb2, buffer, vpd_len))
2653 goto out;
2655 sdkp->lbpvpd = 1;
2656 sdkp->lbpu = (buffer[5] >> 7) & 1; /* UNMAP */
2657 sdkp->lbpws = (buffer[5] >> 6) & 1; /* WRITE SAME(16) with UNMAP */
2658 sdkp->lbpws10 = (buffer[5] >> 5) & 1; /* WRITE SAME(10) with UNMAP */
2660 out:
2661 kfree(buffer);
2664 static void sd_read_write_same(struct scsi_disk *sdkp, unsigned char *buffer)
2666 struct scsi_device *sdev = sdkp->device;
2668 if (sdev->host->no_write_same) {
2669 sdev->no_write_same = 1;
2671 return;
2674 if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE, INQUIRY) < 0) {
2675 /* too large values might cause issues with arcmsr */
2676 int vpd_buf_len = 64;
2678 sdev->no_report_opcodes = 1;
2680 /* Disable WRITE SAME if REPORT SUPPORTED OPERATION
2681 * CODES is unsupported and the device has an ATA
2682 * Information VPD page (SAT).
2684 if (!scsi_get_vpd_page(sdev, 0x89, buffer, vpd_buf_len))
2685 sdev->no_write_same = 1;
2688 if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE, WRITE_SAME_16) == 1)
2689 sdkp->ws16 = 1;
2691 if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE, WRITE_SAME) == 1)
2692 sdkp->ws10 = 1;
2695 static int sd_try_extended_inquiry(struct scsi_device *sdp)
2697 /* Attempt VPD inquiry if the device blacklist explicitly calls
2698 * for it.
2700 if (sdp->try_vpd_pages)
2701 return 1;
2703 * Although VPD inquiries can go to SCSI-2 type devices,
2704 * some USB ones crash on receiving them, and the pages
2705 * we currently ask for are for SPC-3 and beyond
2707 if (sdp->scsi_level > SCSI_SPC_2 && !sdp->skip_vpd_pages)
2708 return 1;
2709 return 0;
2713 * sd_revalidate_disk - called the first time a new disk is seen,
2714 * performs disk spin up, read_capacity, etc.
2715 * @disk: struct gendisk we care about
2717 static int sd_revalidate_disk(struct gendisk *disk)
2719 struct scsi_disk *sdkp = scsi_disk(disk);
2720 struct scsi_device *sdp = sdkp->device;
2721 unsigned char *buffer;
2722 unsigned int max_xfer;
2724 SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp,
2725 "sd_revalidate_disk\n"));
2728 * If the device is offline, don't try and read capacity or any
2729 * of the other niceties.
2731 if (!scsi_device_online(sdp))
2732 goto out;
2734 buffer = kmalloc(SD_BUF_SIZE, GFP_KERNEL);
2735 if (!buffer) {
2736 sd_printk(KERN_WARNING, sdkp, "sd_revalidate_disk: Memory "
2737 "allocation failure.\n");
2738 goto out;
2741 sd_spinup_disk(sdkp);
2744 * Without media there is no reason to ask; moreover, some devices
2745 * react badly if we do.
2747 if (sdkp->media_present) {
2748 sd_read_capacity(sdkp, buffer);
2750 if (sd_try_extended_inquiry(sdp)) {
2751 sd_read_block_provisioning(sdkp);
2752 sd_read_block_limits(sdkp);
2753 sd_read_block_characteristics(sdkp);
2756 sd_read_write_protect_flag(sdkp, buffer);
2757 sd_read_cache_type(sdkp, buffer);
2758 sd_read_app_tag_own(sdkp, buffer);
2759 sd_read_write_same(sdkp, buffer);
2762 sdkp->first_scan = 0;
2765 * We now have all cache related info, determine how we deal
2766 * with flush requests.
2768 sd_set_flush_flag(sdkp);
2770 max_xfer = sdkp->max_xfer_blocks;
2771 max_xfer <<= ilog2(sdp->sector_size) - 9;
2773 max_xfer = min_not_zero(queue_max_hw_sectors(sdkp->disk->queue),
2774 max_xfer);
2775 blk_queue_max_hw_sectors(sdkp->disk->queue, max_xfer);
2776 set_capacity(disk, sdkp->capacity);
2777 sd_config_write_same(sdkp);
2778 kfree(buffer);
2780 out:
2781 return 0;
2785 * sd_unlock_native_capacity - unlock native capacity
2786 * @disk: struct gendisk to set capacity for
2788 * Block layer calls this function if it detects that partitions
2789 * on @disk reach beyond the end of the device. If the SCSI host
2790 * implements ->unlock_native_capacity() method, it's invoked to
2791 * give it a chance to adjust the device capacity.
2793 * CONTEXT:
2794 * Defined by block layer. Might sleep.
2796 static void sd_unlock_native_capacity(struct gendisk *disk)
2798 struct scsi_device *sdev = scsi_disk(disk)->device;
2800 if (sdev->host->hostt->unlock_native_capacity)
2801 sdev->host->hostt->unlock_native_capacity(sdev);
2805 * sd_format_disk_name - format disk name
2806 * @prefix: name prefix - ie. "sd" for SCSI disks
2807 * @index: index of the disk to format name for
2808 * @buf: output buffer
2809 * @buflen: length of the output buffer
2811 * SCSI disk names starts at sda. The 26th device is sdz and the
2812 * 27th is sdaa. The last one for two lettered suffix is sdzz
2813 * which is followed by sdaaa.
2815 * This is basically 26 base counting with one extra 'nil' entry
2816 * at the beginning from the second digit on and can be
2817 * determined using similar method as 26 base conversion with the
2818 * index shifted -1 after each digit is computed.
2820 * CONTEXT:
2821 * Don't care.
2823 * RETURNS:
2824 * 0 on success, -errno on failure.
2826 static int sd_format_disk_name(char *prefix, int index, char *buf, int buflen)
2828 const int base = 'z' - 'a' + 1;
2829 char *begin = buf + strlen(prefix);
2830 char *end = buf + buflen;
2831 char *p;
2832 int unit;
2834 p = end - 1;
2835 *p = '\0';
2836 unit = base;
2837 do {
2838 if (p == begin)
2839 return -EINVAL;
2840 *--p = 'a' + (index % unit);
2841 index = (index / unit) - 1;
2842 } while (index >= 0);
2844 memmove(begin, p, end - p);
2845 memcpy(buf, prefix, strlen(prefix));
2847 return 0;
2851 * The asynchronous part of sd_probe
2853 static void sd_probe_async(void *data, async_cookie_t cookie)
2855 struct scsi_disk *sdkp = data;
2856 struct scsi_device *sdp;
2857 struct gendisk *gd;
2858 u32 index;
2859 struct device *dev;
2861 sdp = sdkp->device;
2862 gd = sdkp->disk;
2863 index = sdkp->index;
2864 dev = &sdp->sdev_gendev;
2866 gd->major = sd_major((index & 0xf0) >> 4);
2867 gd->first_minor = ((index & 0xf) << 4) | (index & 0xfff00);
2868 gd->minors = SD_MINORS;
2870 gd->fops = &sd_fops;
2871 gd->private_data = &sdkp->driver;
2872 gd->queue = sdkp->device->request_queue;
2874 /* defaults, until the device tells us otherwise */
2875 sdp->sector_size = 512;
2876 sdkp->capacity = 0;
2877 sdkp->media_present = 1;
2878 sdkp->write_prot = 0;
2879 sdkp->cache_override = 0;
2880 sdkp->WCE = 0;
2881 sdkp->RCD = 0;
2882 sdkp->ATO = 0;
2883 sdkp->first_scan = 1;
2884 sdkp->max_medium_access_timeouts = SD_MAX_MEDIUM_TIMEOUTS;
2886 sd_revalidate_disk(gd);
2888 gd->driverfs_dev = &sdp->sdev_gendev;
2889 gd->flags = GENHD_FL_EXT_DEVT;
2890 if (sdp->removable) {
2891 gd->flags |= GENHD_FL_REMOVABLE;
2892 gd->events |= DISK_EVENT_MEDIA_CHANGE;
2895 blk_pm_runtime_init(sdp->request_queue, dev);
2896 add_disk(gd);
2897 if (sdkp->capacity)
2898 sd_dif_config_host(sdkp);
2900 sd_revalidate_disk(gd);
2902 sd_printk(KERN_NOTICE, sdkp, "Attached SCSI %sdisk\n",
2903 sdp->removable ? "removable " : "");
2904 scsi_autopm_put_device(sdp);
2905 put_device(&sdkp->dev);
2909 * sd_probe - called during driver initialization and whenever a
2910 * new scsi device is attached to the system. It is called once
2911 * for each scsi device (not just disks) present.
2912 * @dev: pointer to device object
2914 * Returns 0 if successful (or not interested in this scsi device
2915 * (e.g. scanner)); 1 when there is an error.
2917 * Note: this function is invoked from the scsi mid-level.
2918 * This function sets up the mapping between a given
2919 * <host,channel,id,lun> (found in sdp) and new device name
2920 * (e.g. /dev/sda). More precisely it is the block device major
2921 * and minor number that is chosen here.
2923 * Assume sd_probe is not re-entrant (for time being)
2924 * Also think about sd_probe() and sd_remove() running coincidentally.
2926 static int sd_probe(struct device *dev)
2928 struct scsi_device *sdp = to_scsi_device(dev);
2929 struct scsi_disk *sdkp;
2930 struct gendisk *gd;
2931 int index;
2932 int error;
2934 scsi_autopm_get_device(sdp);
2935 error = -ENODEV;
2936 if (sdp->type != TYPE_DISK && sdp->type != TYPE_MOD && sdp->type != TYPE_RBC)
2937 goto out;
2939 SCSI_LOG_HLQUEUE(3, sdev_printk(KERN_INFO, sdp,
2940 "sd_probe\n"));
2942 error = -ENOMEM;
2943 sdkp = kzalloc(sizeof(*sdkp), GFP_KERNEL);
2944 if (!sdkp)
2945 goto out;
2947 gd = alloc_disk(SD_MINORS);
2948 if (!gd)
2949 goto out_free;
2951 do {
2952 if (!ida_pre_get(&sd_index_ida, GFP_KERNEL))
2953 goto out_put;
2955 spin_lock(&sd_index_lock);
2956 error = ida_get_new(&sd_index_ida, &index);
2957 spin_unlock(&sd_index_lock);
2958 } while (error == -EAGAIN);
2960 if (error) {
2961 sdev_printk(KERN_WARNING, sdp, "sd_probe: memory exhausted.\n");
2962 goto out_put;
2965 error = sd_format_disk_name("sd", index, gd->disk_name, DISK_NAME_LEN);
2966 if (error) {
2967 sdev_printk(KERN_WARNING, sdp, "SCSI disk (sd) name length exceeded.\n");
2968 goto out_free_index;
2971 sdkp->device = sdp;
2972 sdkp->driver = &sd_template;
2973 sdkp->disk = gd;
2974 sdkp->index = index;
2975 atomic_set(&sdkp->openers, 0);
2976 atomic_set(&sdkp->device->ioerr_cnt, 0);
2978 if (!sdp->request_queue->rq_timeout) {
2979 if (sdp->type != TYPE_MOD)
2980 blk_queue_rq_timeout(sdp->request_queue, SD_TIMEOUT);
2981 else
2982 blk_queue_rq_timeout(sdp->request_queue,
2983 SD_MOD_TIMEOUT);
2986 device_initialize(&sdkp->dev);
2987 sdkp->dev.parent = dev;
2988 sdkp->dev.class = &sd_disk_class;
2989 dev_set_name(&sdkp->dev, "%s", dev_name(dev));
2991 error = device_add(&sdkp->dev);
2992 if (error)
2993 goto out_free_index;
2995 get_device(dev);
2996 dev_set_drvdata(dev, sdkp);
2998 get_device(&sdkp->dev); /* prevent release before async_schedule */
2999 async_schedule_domain(sd_probe_async, sdkp, &scsi_sd_probe_domain);
3001 return 0;
3003 out_free_index:
3004 spin_lock(&sd_index_lock);
3005 ida_remove(&sd_index_ida, index);
3006 spin_unlock(&sd_index_lock);
3007 out_put:
3008 put_disk(gd);
3009 out_free:
3010 kfree(sdkp);
3011 out:
3012 scsi_autopm_put_device(sdp);
3013 return error;
3017 * sd_remove - called whenever a scsi disk (previously recognized by
3018 * sd_probe) is detached from the system. It is called (potentially
3019 * multiple times) during sd module unload.
3020 * @sdp: pointer to mid level scsi device object
3022 * Note: this function is invoked from the scsi mid-level.
3023 * This function potentially frees up a device name (e.g. /dev/sdc)
3024 * that could be re-used by a subsequent sd_probe().
3025 * This function is not called when the built-in sd driver is "exit-ed".
3027 static int sd_remove(struct device *dev)
3029 struct scsi_disk *sdkp;
3030 dev_t devt;
3032 sdkp = dev_get_drvdata(dev);
3033 devt = disk_devt(sdkp->disk);
3034 scsi_autopm_get_device(sdkp->device);
3036 async_synchronize_full_domain(&scsi_sd_pm_domain);
3037 async_synchronize_full_domain(&scsi_sd_probe_domain);
3038 device_del(&sdkp->dev);
3039 del_gendisk(sdkp->disk);
3040 sd_shutdown(dev);
3042 blk_register_region(devt, SD_MINORS, NULL,
3043 sd_default_probe, NULL, NULL);
3045 mutex_lock(&sd_ref_mutex);
3046 dev_set_drvdata(dev, NULL);
3047 put_device(&sdkp->dev);
3048 mutex_unlock(&sd_ref_mutex);
3050 return 0;
3054 * scsi_disk_release - Called to free the scsi_disk structure
3055 * @dev: pointer to embedded class device
3057 * sd_ref_mutex must be held entering this routine. Because it is
3058 * called on last put, you should always use the scsi_disk_get()
3059 * scsi_disk_put() helpers which manipulate the semaphore directly
3060 * and never do a direct put_device.
3062 static void scsi_disk_release(struct device *dev)
3064 struct scsi_disk *sdkp = to_scsi_disk(dev);
3065 struct gendisk *disk = sdkp->disk;
3067 spin_lock(&sd_index_lock);
3068 ida_remove(&sd_index_ida, sdkp->index);
3069 spin_unlock(&sd_index_lock);
3071 blk_integrity_unregister(disk);
3072 disk->private_data = NULL;
3073 put_disk(disk);
3074 put_device(&sdkp->device->sdev_gendev);
3076 kfree(sdkp);
3079 static int sd_start_stop_device(struct scsi_disk *sdkp, int start)
3081 unsigned char cmd[6] = { START_STOP }; /* START_VALID */
3082 struct scsi_sense_hdr sshdr;
3083 struct scsi_device *sdp = sdkp->device;
3084 int res;
3086 if (start)
3087 cmd[4] |= 1; /* START */
3089 if (sdp->start_stop_pwr_cond)
3090 cmd[4] |= start ? 1 << 4 : 3 << 4; /* Active or Standby */
3092 if (!scsi_device_online(sdp))
3093 return -ENODEV;
3095 res = scsi_execute_req_flags(sdp, cmd, DMA_NONE, NULL, 0, &sshdr,
3096 SD_TIMEOUT, SD_MAX_RETRIES, NULL, REQ_PM);
3097 if (res) {
3098 sd_print_result(sdkp, "Start/Stop Unit failed", res);
3099 if (driver_byte(res) & DRIVER_SENSE)
3100 sd_print_sense_hdr(sdkp, &sshdr);
3101 if (scsi_sense_valid(&sshdr) &&
3102 /* 0x3a is medium not present */
3103 sshdr.asc == 0x3a)
3104 res = 0;
3107 /* SCSI error codes must not go to the generic layer */
3108 if (res)
3109 return -EIO;
3111 return 0;
3115 * Send a SYNCHRONIZE CACHE instruction down to the device through
3116 * the normal SCSI command structure. Wait for the command to
3117 * complete.
3119 static void sd_shutdown(struct device *dev)
3121 struct scsi_disk *sdkp = dev_get_drvdata(dev);
3123 if (!sdkp)
3124 return; /* this can happen */
3126 if (pm_runtime_suspended(dev))
3127 return;
3129 if (sdkp->WCE && sdkp->media_present) {
3130 sd_printk(KERN_NOTICE, sdkp, "Synchronizing SCSI cache\n");
3131 sd_sync_cache(sdkp);
3134 if (system_state != SYSTEM_RESTART && sdkp->device->manage_start_stop) {
3135 sd_printk(KERN_NOTICE, sdkp, "Stopping disk\n");
3136 sd_start_stop_device(sdkp, 0);
3140 static int sd_suspend_common(struct device *dev, bool ignore_stop_errors)
3142 struct scsi_disk *sdkp = dev_get_drvdata(dev);
3143 int ret = 0;
3145 if (!sdkp)
3146 return 0; /* this can happen */
3148 if (sdkp->WCE && sdkp->media_present) {
3149 sd_printk(KERN_NOTICE, sdkp, "Synchronizing SCSI cache\n");
3150 ret = sd_sync_cache(sdkp);
3151 if (ret) {
3152 /* ignore OFFLINE device */
3153 if (ret == -ENODEV)
3154 ret = 0;
3155 goto done;
3159 if (sdkp->device->manage_start_stop) {
3160 sd_printk(KERN_NOTICE, sdkp, "Stopping disk\n");
3161 /* an error is not worth aborting a system sleep */
3162 ret = sd_start_stop_device(sdkp, 0);
3163 if (ignore_stop_errors)
3164 ret = 0;
3167 done:
3168 return ret;
3171 static int sd_suspend_system(struct device *dev)
3173 return sd_suspend_common(dev, true);
3176 static int sd_suspend_runtime(struct device *dev)
3178 return sd_suspend_common(dev, false);
3181 static int sd_resume(struct device *dev)
3183 struct scsi_disk *sdkp = dev_get_drvdata(dev);
3185 if (!sdkp->device->manage_start_stop)
3186 return 0;
3188 sd_printk(KERN_NOTICE, sdkp, "Starting disk\n");
3189 return sd_start_stop_device(sdkp, 1);
3193 * init_sd - entry point for this driver (both when built in or when
3194 * a module).
3196 * Note: this function registers this driver with the scsi mid-level.
3198 static int __init init_sd(void)
3200 int majors = 0, i, err;
3202 SCSI_LOG_HLQUEUE(3, printk("init_sd: sd driver entry point\n"));
3204 for (i = 0; i < SD_MAJORS; i++) {
3205 if (register_blkdev(sd_major(i), "sd") != 0)
3206 continue;
3207 majors++;
3208 blk_register_region(sd_major(i), SD_MINORS, NULL,
3209 sd_default_probe, NULL, NULL);
3212 if (!majors)
3213 return -ENODEV;
3215 err = class_register(&sd_disk_class);
3216 if (err)
3217 goto err_out;
3219 sd_cdb_cache = kmem_cache_create("sd_ext_cdb", SD_EXT_CDB_SIZE,
3220 0, 0, NULL);
3221 if (!sd_cdb_cache) {
3222 printk(KERN_ERR "sd: can't init extended cdb cache\n");
3223 err = -ENOMEM;
3224 goto err_out_class;
3227 sd_cdb_pool = mempool_create_slab_pool(SD_MEMPOOL_SIZE, sd_cdb_cache);
3228 if (!sd_cdb_pool) {
3229 printk(KERN_ERR "sd: can't init extended cdb pool\n");
3230 err = -ENOMEM;
3231 goto err_out_cache;
3234 err = scsi_register_driver(&sd_template.gendrv);
3235 if (err)
3236 goto err_out_driver;
3238 return 0;
3240 err_out_driver:
3241 mempool_destroy(sd_cdb_pool);
3243 err_out_cache:
3244 kmem_cache_destroy(sd_cdb_cache);
3246 err_out_class:
3247 class_unregister(&sd_disk_class);
3248 err_out:
3249 for (i = 0; i < SD_MAJORS; i++)
3250 unregister_blkdev(sd_major(i), "sd");
3251 return err;
3255 * exit_sd - exit point for this driver (when it is a module).
3257 * Note: this function unregisters this driver from the scsi mid-level.
3259 static void __exit exit_sd(void)
3261 int i;
3263 SCSI_LOG_HLQUEUE(3, printk("exit_sd: exiting sd driver\n"));
3265 scsi_unregister_driver(&sd_template.gendrv);
3266 mempool_destroy(sd_cdb_pool);
3267 kmem_cache_destroy(sd_cdb_cache);
3269 class_unregister(&sd_disk_class);
3271 for (i = 0; i < SD_MAJORS; i++) {
3272 blk_unregister_region(sd_major(i), SD_MINORS);
3273 unregister_blkdev(sd_major(i), "sd");
3277 module_init(init_sd);
3278 module_exit(exit_sd);
3280 static void sd_print_sense_hdr(struct scsi_disk *sdkp,
3281 struct scsi_sense_hdr *sshdr)
3283 scsi_print_sense_hdr(sdkp->device,
3284 sdkp->disk ? sdkp->disk->disk_name : NULL, sshdr);
3287 static void sd_print_result(const struct scsi_disk *sdkp, const char *msg,
3288 int result)
3290 const char *hb_string = scsi_hostbyte_string(result);
3291 const char *db_string = scsi_driverbyte_string(result);
3293 if (hb_string || db_string)
3294 sd_printk(KERN_INFO, sdkp,
3295 "%s: Result: hostbyte=%s driverbyte=%s\n", msg,
3296 hb_string ? hb_string : "invalid",
3297 db_string ? db_string : "invalid");
3298 else
3299 sd_printk(KERN_INFO, sdkp,
3300 "%s: Result: hostbyte=0x%02x driverbyte=0x%02x\n",
3301 msg, host_byte(result), driver_byte(result));