BKL: remove extraneous #include <smp_lock.h>
[linux-2.6/linux-acpi-2.6/ibm-acpi-2.6.git] / drivers / scsi / sd.c
blob956496182c805d8dd7f0aa58cc3ddb7c12c91b8b
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 <asm/uaccess.h>
54 #include <asm/unaligned.h>
56 #include <scsi/scsi.h>
57 #include <scsi/scsi_cmnd.h>
58 #include <scsi/scsi_dbg.h>
59 #include <scsi/scsi_device.h>
60 #include <scsi/scsi_driver.h>
61 #include <scsi/scsi_eh.h>
62 #include <scsi/scsi_host.h>
63 #include <scsi/scsi_ioctl.h>
64 #include <scsi/scsicam.h>
66 #include "sd.h"
67 #include "scsi_logging.h"
69 MODULE_AUTHOR("Eric Youngdale");
70 MODULE_DESCRIPTION("SCSI disk (sd) driver");
71 MODULE_LICENSE("GPL");
73 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK0_MAJOR);
74 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK1_MAJOR);
75 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK2_MAJOR);
76 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK3_MAJOR);
77 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK4_MAJOR);
78 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK5_MAJOR);
79 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK6_MAJOR);
80 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK7_MAJOR);
81 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK8_MAJOR);
82 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK9_MAJOR);
83 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK10_MAJOR);
84 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK11_MAJOR);
85 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK12_MAJOR);
86 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK13_MAJOR);
87 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK14_MAJOR);
88 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK15_MAJOR);
89 MODULE_ALIAS_SCSI_DEVICE(TYPE_DISK);
90 MODULE_ALIAS_SCSI_DEVICE(TYPE_MOD);
91 MODULE_ALIAS_SCSI_DEVICE(TYPE_RBC);
93 #if !defined(CONFIG_DEBUG_BLOCK_EXT_DEVT)
94 #define SD_MINORS 16
95 #else
96 #define SD_MINORS 0
97 #endif
99 static int sd_revalidate_disk(struct gendisk *);
100 static void sd_unlock_native_capacity(struct gendisk *disk);
101 static int sd_probe(struct device *);
102 static int sd_remove(struct device *);
103 static void sd_shutdown(struct device *);
104 static int sd_suspend(struct device *, pm_message_t state);
105 static int sd_resume(struct device *);
106 static void sd_rescan(struct device *);
107 static int sd_done(struct scsi_cmnd *);
108 static void sd_read_capacity(struct scsi_disk *sdkp, unsigned char *buffer);
109 static void scsi_disk_release(struct device *cdev);
110 static void sd_print_sense_hdr(struct scsi_disk *, struct scsi_sense_hdr *);
111 static void sd_print_result(struct scsi_disk *, int);
113 static DEFINE_SPINLOCK(sd_index_lock);
114 static DEFINE_IDA(sd_index_ida);
116 /* This semaphore is used to mediate the 0->1 reference get in the
117 * face of object destruction (i.e. we can't allow a get on an
118 * object after last put) */
119 static DEFINE_MUTEX(sd_ref_mutex);
121 static struct kmem_cache *sd_cdb_cache;
122 static mempool_t *sd_cdb_pool;
124 static const char *sd_cache_types[] = {
125 "write through", "none", "write back",
126 "write back, no read (daft)"
129 static ssize_t
130 sd_store_cache_type(struct device *dev, struct device_attribute *attr,
131 const char *buf, size_t count)
133 int i, ct = -1, rcd, wce, sp;
134 struct scsi_disk *sdkp = to_scsi_disk(dev);
135 struct scsi_device *sdp = sdkp->device;
136 char buffer[64];
137 char *buffer_data;
138 struct scsi_mode_data data;
139 struct scsi_sense_hdr sshdr;
140 int len;
142 if (sdp->type != TYPE_DISK)
143 /* no cache control on RBC devices; theoretically they
144 * can do it, but there's probably so many exceptions
145 * it's not worth the risk */
146 return -EINVAL;
148 for (i = 0; i < ARRAY_SIZE(sd_cache_types); i++) {
149 len = strlen(sd_cache_types[i]);
150 if (strncmp(sd_cache_types[i], buf, len) == 0 &&
151 buf[len] == '\n') {
152 ct = i;
153 break;
156 if (ct < 0)
157 return -EINVAL;
158 rcd = ct & 0x01 ? 1 : 0;
159 wce = ct & 0x02 ? 1 : 0;
160 if (scsi_mode_sense(sdp, 0x08, 8, buffer, sizeof(buffer), SD_TIMEOUT,
161 SD_MAX_RETRIES, &data, NULL))
162 return -EINVAL;
163 len = min_t(size_t, sizeof(buffer), data.length - data.header_length -
164 data.block_descriptor_length);
165 buffer_data = buffer + data.header_length +
166 data.block_descriptor_length;
167 buffer_data[2] &= ~0x05;
168 buffer_data[2] |= wce << 2 | rcd;
169 sp = buffer_data[0] & 0x80 ? 1 : 0;
171 if (scsi_mode_select(sdp, 1, sp, 8, buffer_data, len, SD_TIMEOUT,
172 SD_MAX_RETRIES, &data, &sshdr)) {
173 if (scsi_sense_valid(&sshdr))
174 sd_print_sense_hdr(sdkp, &sshdr);
175 return -EINVAL;
177 revalidate_disk(sdkp->disk);
178 return count;
181 static ssize_t
182 sd_store_manage_start_stop(struct device *dev, struct device_attribute *attr,
183 const char *buf, size_t count)
185 struct scsi_disk *sdkp = to_scsi_disk(dev);
186 struct scsi_device *sdp = sdkp->device;
188 if (!capable(CAP_SYS_ADMIN))
189 return -EACCES;
191 sdp->manage_start_stop = simple_strtoul(buf, NULL, 10);
193 return count;
196 static ssize_t
197 sd_store_allow_restart(struct device *dev, struct device_attribute *attr,
198 const char *buf, size_t count)
200 struct scsi_disk *sdkp = to_scsi_disk(dev);
201 struct scsi_device *sdp = sdkp->device;
203 if (!capable(CAP_SYS_ADMIN))
204 return -EACCES;
206 if (sdp->type != TYPE_DISK)
207 return -EINVAL;
209 sdp->allow_restart = simple_strtoul(buf, NULL, 10);
211 return count;
214 static ssize_t
215 sd_show_cache_type(struct device *dev, struct device_attribute *attr,
216 char *buf)
218 struct scsi_disk *sdkp = to_scsi_disk(dev);
219 int ct = sdkp->RCD + 2*sdkp->WCE;
221 return snprintf(buf, 40, "%s\n", sd_cache_types[ct]);
224 static ssize_t
225 sd_show_fua(struct device *dev, struct device_attribute *attr, char *buf)
227 struct scsi_disk *sdkp = to_scsi_disk(dev);
229 return snprintf(buf, 20, "%u\n", sdkp->DPOFUA);
232 static ssize_t
233 sd_show_manage_start_stop(struct device *dev, struct device_attribute *attr,
234 char *buf)
236 struct scsi_disk *sdkp = to_scsi_disk(dev);
237 struct scsi_device *sdp = sdkp->device;
239 return snprintf(buf, 20, "%u\n", sdp->manage_start_stop);
242 static ssize_t
243 sd_show_allow_restart(struct device *dev, struct device_attribute *attr,
244 char *buf)
246 struct scsi_disk *sdkp = to_scsi_disk(dev);
248 return snprintf(buf, 40, "%d\n", sdkp->device->allow_restart);
251 static ssize_t
252 sd_show_protection_type(struct device *dev, struct device_attribute *attr,
253 char *buf)
255 struct scsi_disk *sdkp = to_scsi_disk(dev);
257 return snprintf(buf, 20, "%u\n", sdkp->protection_type);
260 static ssize_t
261 sd_show_protection_mode(struct device *dev, struct device_attribute *attr,
262 char *buf)
264 struct scsi_disk *sdkp = to_scsi_disk(dev);
265 struct scsi_device *sdp = sdkp->device;
266 unsigned int dif, dix;
268 dif = scsi_host_dif_capable(sdp->host, sdkp->protection_type);
269 dix = scsi_host_dix_capable(sdp->host, sdkp->protection_type);
271 if (!dix && scsi_host_dix_capable(sdp->host, SD_DIF_TYPE0_PROTECTION)) {
272 dif = 0;
273 dix = 1;
276 if (!dif && !dix)
277 return snprintf(buf, 20, "none\n");
279 return snprintf(buf, 20, "%s%u\n", dix ? "dix" : "dif", dif);
282 static ssize_t
283 sd_show_app_tag_own(struct device *dev, struct device_attribute *attr,
284 char *buf)
286 struct scsi_disk *sdkp = to_scsi_disk(dev);
288 return snprintf(buf, 20, "%u\n", sdkp->ATO);
291 static ssize_t
292 sd_show_thin_provisioning(struct device *dev, struct device_attribute *attr,
293 char *buf)
295 struct scsi_disk *sdkp = to_scsi_disk(dev);
297 return snprintf(buf, 20, "%u\n", sdkp->thin_provisioning);
300 static struct device_attribute sd_disk_attrs[] = {
301 __ATTR(cache_type, S_IRUGO|S_IWUSR, sd_show_cache_type,
302 sd_store_cache_type),
303 __ATTR(FUA, S_IRUGO, sd_show_fua, NULL),
304 __ATTR(allow_restart, S_IRUGO|S_IWUSR, sd_show_allow_restart,
305 sd_store_allow_restart),
306 __ATTR(manage_start_stop, S_IRUGO|S_IWUSR, sd_show_manage_start_stop,
307 sd_store_manage_start_stop),
308 __ATTR(protection_type, S_IRUGO, sd_show_protection_type, NULL),
309 __ATTR(protection_mode, S_IRUGO, sd_show_protection_mode, NULL),
310 __ATTR(app_tag_own, S_IRUGO, sd_show_app_tag_own, NULL),
311 __ATTR(thin_provisioning, S_IRUGO, sd_show_thin_provisioning, NULL),
312 __ATTR_NULL,
315 static struct class sd_disk_class = {
316 .name = "scsi_disk",
317 .owner = THIS_MODULE,
318 .dev_release = scsi_disk_release,
319 .dev_attrs = sd_disk_attrs,
322 static struct scsi_driver sd_template = {
323 .owner = THIS_MODULE,
324 .gendrv = {
325 .name = "sd",
326 .probe = sd_probe,
327 .remove = sd_remove,
328 .suspend = sd_suspend,
329 .resume = sd_resume,
330 .shutdown = sd_shutdown,
332 .rescan = sd_rescan,
333 .done = sd_done,
337 * Device no to disk mapping:
339 * major disc2 disc p1
340 * |............|.............|....|....| <- dev_t
341 * 31 20 19 8 7 4 3 0
343 * Inside a major, we have 16k disks, however mapped non-
344 * contiguously. The first 16 disks are for major0, the next
345 * ones with major1, ... Disk 256 is for major0 again, disk 272
346 * for major1, ...
347 * As we stay compatible with our numbering scheme, we can reuse
348 * the well-know SCSI majors 8, 65--71, 136--143.
350 static int sd_major(int major_idx)
352 switch (major_idx) {
353 case 0:
354 return SCSI_DISK0_MAJOR;
355 case 1 ... 7:
356 return SCSI_DISK1_MAJOR + major_idx - 1;
357 case 8 ... 15:
358 return SCSI_DISK8_MAJOR + major_idx - 8;
359 default:
360 BUG();
361 return 0; /* shut up gcc */
365 static struct scsi_disk *__scsi_disk_get(struct gendisk *disk)
367 struct scsi_disk *sdkp = NULL;
369 if (disk->private_data) {
370 sdkp = scsi_disk(disk);
371 if (scsi_device_get(sdkp->device) == 0)
372 get_device(&sdkp->dev);
373 else
374 sdkp = NULL;
376 return sdkp;
379 static struct scsi_disk *scsi_disk_get(struct gendisk *disk)
381 struct scsi_disk *sdkp;
383 mutex_lock(&sd_ref_mutex);
384 sdkp = __scsi_disk_get(disk);
385 mutex_unlock(&sd_ref_mutex);
386 return sdkp;
389 static struct scsi_disk *scsi_disk_get_from_dev(struct device *dev)
391 struct scsi_disk *sdkp;
393 mutex_lock(&sd_ref_mutex);
394 sdkp = dev_get_drvdata(dev);
395 if (sdkp)
396 sdkp = __scsi_disk_get(sdkp->disk);
397 mutex_unlock(&sd_ref_mutex);
398 return sdkp;
401 static void scsi_disk_put(struct scsi_disk *sdkp)
403 struct scsi_device *sdev = sdkp->device;
405 mutex_lock(&sd_ref_mutex);
406 put_device(&sdkp->dev);
407 scsi_device_put(sdev);
408 mutex_unlock(&sd_ref_mutex);
411 static void sd_prot_op(struct scsi_cmnd *scmd, unsigned int dif)
413 unsigned int prot_op = SCSI_PROT_NORMAL;
414 unsigned int dix = scsi_prot_sg_count(scmd);
416 if (scmd->sc_data_direction == DMA_FROM_DEVICE) {
417 if (dif && dix)
418 prot_op = SCSI_PROT_READ_PASS;
419 else if (dif && !dix)
420 prot_op = SCSI_PROT_READ_STRIP;
421 else if (!dif && dix)
422 prot_op = SCSI_PROT_READ_INSERT;
423 } else {
424 if (dif && dix)
425 prot_op = SCSI_PROT_WRITE_PASS;
426 else if (dif && !dix)
427 prot_op = SCSI_PROT_WRITE_INSERT;
428 else if (!dif && dix)
429 prot_op = SCSI_PROT_WRITE_STRIP;
432 scsi_set_prot_op(scmd, prot_op);
433 scsi_set_prot_type(scmd, dif);
437 * scsi_setup_discard_cmnd - unmap blocks on thinly provisioned device
438 * @sdp: scsi device to operate one
439 * @rq: Request to prepare
441 * Will issue either UNMAP or WRITE SAME(16) depending on preference
442 * indicated by target device.
444 static int scsi_setup_discard_cmnd(struct scsi_device *sdp, struct request *rq)
446 struct scsi_disk *sdkp = scsi_disk(rq->rq_disk);
447 struct bio *bio = rq->bio;
448 sector_t sector = bio->bi_sector;
449 unsigned int nr_sectors = bio_sectors(bio);
450 unsigned int len;
451 int ret;
452 struct page *page;
454 if (sdkp->device->sector_size == 4096) {
455 sector >>= 3;
456 nr_sectors >>= 3;
459 rq->timeout = SD_TIMEOUT;
461 memset(rq->cmd, 0, rq->cmd_len);
463 page = alloc_page(GFP_ATOMIC | __GFP_ZERO);
464 if (!page)
465 return BLKPREP_DEFER;
467 if (sdkp->unmap) {
468 char *buf = page_address(page);
470 rq->cmd_len = 10;
471 rq->cmd[0] = UNMAP;
472 rq->cmd[8] = 24;
474 put_unaligned_be16(6 + 16, &buf[0]);
475 put_unaligned_be16(16, &buf[2]);
476 put_unaligned_be64(sector, &buf[8]);
477 put_unaligned_be32(nr_sectors, &buf[16]);
479 len = 24;
480 } else {
481 rq->cmd_len = 16;
482 rq->cmd[0] = WRITE_SAME_16;
483 rq->cmd[1] = 0x8; /* UNMAP */
484 put_unaligned_be64(sector, &rq->cmd[2]);
485 put_unaligned_be32(nr_sectors, &rq->cmd[10]);
487 len = sdkp->device->sector_size;
490 blk_add_request_payload(rq, page, len);
491 ret = scsi_setup_blk_pc_cmnd(sdp, rq);
492 rq->buffer = page_address(page);
493 if (ret != BLKPREP_OK) {
494 __free_page(page);
495 rq->buffer = NULL;
497 return ret;
500 static int scsi_setup_flush_cmnd(struct scsi_device *sdp, struct request *rq)
502 rq->timeout = SD_FLUSH_TIMEOUT;
503 rq->retries = SD_MAX_RETRIES;
504 rq->cmd[0] = SYNCHRONIZE_CACHE;
505 rq->cmd_len = 10;
507 return scsi_setup_blk_pc_cmnd(sdp, rq);
510 static void sd_unprep_fn(struct request_queue *q, struct request *rq)
512 if (rq->cmd_flags & REQ_DISCARD) {
513 free_page((unsigned long)rq->buffer);
514 rq->buffer = NULL;
519 * sd_init_command - build a scsi (read or write) command from
520 * information in the request structure.
521 * @SCpnt: pointer to mid-level's per scsi command structure that
522 * contains request and into which the scsi command is written
524 * Returns 1 if successful and 0 if error (or cannot be done now).
526 static int sd_prep_fn(struct request_queue *q, struct request *rq)
528 struct scsi_cmnd *SCpnt;
529 struct scsi_device *sdp = q->queuedata;
530 struct gendisk *disk = rq->rq_disk;
531 struct scsi_disk *sdkp;
532 sector_t block = blk_rq_pos(rq);
533 sector_t threshold;
534 unsigned int this_count = blk_rq_sectors(rq);
535 int ret, host_dif;
536 unsigned char protect;
539 * Discard request come in as REQ_TYPE_FS but we turn them into
540 * block PC requests to make life easier.
542 if (rq->cmd_flags & REQ_DISCARD) {
543 ret = scsi_setup_discard_cmnd(sdp, rq);
544 goto out;
545 } else if (rq->cmd_flags & REQ_FLUSH) {
546 ret = scsi_setup_flush_cmnd(sdp, rq);
547 goto out;
548 } else if (rq->cmd_type == REQ_TYPE_BLOCK_PC) {
549 ret = scsi_setup_blk_pc_cmnd(sdp, rq);
550 goto out;
551 } else if (rq->cmd_type != REQ_TYPE_FS) {
552 ret = BLKPREP_KILL;
553 goto out;
555 ret = scsi_setup_fs_cmnd(sdp, rq);
556 if (ret != BLKPREP_OK)
557 goto out;
558 SCpnt = rq->special;
559 sdkp = scsi_disk(disk);
561 /* from here on until we're complete, any goto out
562 * is used for a killable error condition */
563 ret = BLKPREP_KILL;
565 SCSI_LOG_HLQUEUE(1, scmd_printk(KERN_INFO, SCpnt,
566 "sd_init_command: block=%llu, "
567 "count=%d\n",
568 (unsigned long long)block,
569 this_count));
571 if (!sdp || !scsi_device_online(sdp) ||
572 block + blk_rq_sectors(rq) > get_capacity(disk)) {
573 SCSI_LOG_HLQUEUE(2, scmd_printk(KERN_INFO, SCpnt,
574 "Finishing %u sectors\n",
575 blk_rq_sectors(rq)));
576 SCSI_LOG_HLQUEUE(2, scmd_printk(KERN_INFO, SCpnt,
577 "Retry with 0x%p\n", SCpnt));
578 goto out;
581 if (sdp->changed) {
583 * quietly refuse to do anything to a changed disc until
584 * the changed bit has been reset
586 /* printk("SCSI disk has been changed. Prohibiting further I/O.\n"); */
587 goto out;
591 * Some SD card readers can't handle multi-sector accesses which touch
592 * the last one or two hardware sectors. Split accesses as needed.
594 threshold = get_capacity(disk) - SD_LAST_BUGGY_SECTORS *
595 (sdp->sector_size / 512);
597 if (unlikely(sdp->last_sector_bug && block + this_count > threshold)) {
598 if (block < threshold) {
599 /* Access up to the threshold but not beyond */
600 this_count = threshold - block;
601 } else {
602 /* Access only a single hardware sector */
603 this_count = sdp->sector_size / 512;
607 SCSI_LOG_HLQUEUE(2, scmd_printk(KERN_INFO, SCpnt, "block=%llu\n",
608 (unsigned long long)block));
611 * If we have a 1K hardware sectorsize, prevent access to single
612 * 512 byte sectors. In theory we could handle this - in fact
613 * the scsi cdrom driver must be able to handle this because
614 * we typically use 1K blocksizes, and cdroms typically have
615 * 2K hardware sectorsizes. Of course, things are simpler
616 * with the cdrom, since it is read-only. For performance
617 * reasons, the filesystems should be able to handle this
618 * and not force the scsi disk driver to use bounce buffers
619 * for this.
621 if (sdp->sector_size == 1024) {
622 if ((block & 1) || (blk_rq_sectors(rq) & 1)) {
623 scmd_printk(KERN_ERR, SCpnt,
624 "Bad block number requested\n");
625 goto out;
626 } else {
627 block = block >> 1;
628 this_count = this_count >> 1;
631 if (sdp->sector_size == 2048) {
632 if ((block & 3) || (blk_rq_sectors(rq) & 3)) {
633 scmd_printk(KERN_ERR, SCpnt,
634 "Bad block number requested\n");
635 goto out;
636 } else {
637 block = block >> 2;
638 this_count = this_count >> 2;
641 if (sdp->sector_size == 4096) {
642 if ((block & 7) || (blk_rq_sectors(rq) & 7)) {
643 scmd_printk(KERN_ERR, SCpnt,
644 "Bad block number requested\n");
645 goto out;
646 } else {
647 block = block >> 3;
648 this_count = this_count >> 3;
651 if (rq_data_dir(rq) == WRITE) {
652 if (!sdp->writeable) {
653 goto out;
655 SCpnt->cmnd[0] = WRITE_6;
656 SCpnt->sc_data_direction = DMA_TO_DEVICE;
658 if (blk_integrity_rq(rq) &&
659 sd_dif_prepare(rq, block, sdp->sector_size) == -EIO)
660 goto out;
662 } else if (rq_data_dir(rq) == READ) {
663 SCpnt->cmnd[0] = READ_6;
664 SCpnt->sc_data_direction = DMA_FROM_DEVICE;
665 } else {
666 scmd_printk(KERN_ERR, SCpnt, "Unknown command %x\n", rq->cmd_flags);
667 goto out;
670 SCSI_LOG_HLQUEUE(2, scmd_printk(KERN_INFO, SCpnt,
671 "%s %d/%u 512 byte blocks.\n",
672 (rq_data_dir(rq) == WRITE) ?
673 "writing" : "reading", this_count,
674 blk_rq_sectors(rq)));
676 /* Set RDPROTECT/WRPROTECT if disk is formatted with DIF */
677 host_dif = scsi_host_dif_capable(sdp->host, sdkp->protection_type);
678 if (host_dif)
679 protect = 1 << 5;
680 else
681 protect = 0;
683 if (host_dif == SD_DIF_TYPE2_PROTECTION) {
684 SCpnt->cmnd = mempool_alloc(sd_cdb_pool, GFP_ATOMIC);
686 if (unlikely(SCpnt->cmnd == NULL)) {
687 ret = BLKPREP_DEFER;
688 goto out;
691 SCpnt->cmd_len = SD_EXT_CDB_SIZE;
692 memset(SCpnt->cmnd, 0, SCpnt->cmd_len);
693 SCpnt->cmnd[0] = VARIABLE_LENGTH_CMD;
694 SCpnt->cmnd[7] = 0x18;
695 SCpnt->cmnd[9] = (rq_data_dir(rq) == READ) ? READ_32 : WRITE_32;
696 SCpnt->cmnd[10] = protect | ((rq->cmd_flags & REQ_FUA) ? 0x8 : 0);
698 /* LBA */
699 SCpnt->cmnd[12] = sizeof(block) > 4 ? (unsigned char) (block >> 56) & 0xff : 0;
700 SCpnt->cmnd[13] = sizeof(block) > 4 ? (unsigned char) (block >> 48) & 0xff : 0;
701 SCpnt->cmnd[14] = sizeof(block) > 4 ? (unsigned char) (block >> 40) & 0xff : 0;
702 SCpnt->cmnd[15] = sizeof(block) > 4 ? (unsigned char) (block >> 32) & 0xff : 0;
703 SCpnt->cmnd[16] = (unsigned char) (block >> 24) & 0xff;
704 SCpnt->cmnd[17] = (unsigned char) (block >> 16) & 0xff;
705 SCpnt->cmnd[18] = (unsigned char) (block >> 8) & 0xff;
706 SCpnt->cmnd[19] = (unsigned char) block & 0xff;
708 /* Expected Indirect LBA */
709 SCpnt->cmnd[20] = (unsigned char) (block >> 24) & 0xff;
710 SCpnt->cmnd[21] = (unsigned char) (block >> 16) & 0xff;
711 SCpnt->cmnd[22] = (unsigned char) (block >> 8) & 0xff;
712 SCpnt->cmnd[23] = (unsigned char) block & 0xff;
714 /* Transfer length */
715 SCpnt->cmnd[28] = (unsigned char) (this_count >> 24) & 0xff;
716 SCpnt->cmnd[29] = (unsigned char) (this_count >> 16) & 0xff;
717 SCpnt->cmnd[30] = (unsigned char) (this_count >> 8) & 0xff;
718 SCpnt->cmnd[31] = (unsigned char) this_count & 0xff;
719 } else if (block > 0xffffffff) {
720 SCpnt->cmnd[0] += READ_16 - READ_6;
721 SCpnt->cmnd[1] = protect | ((rq->cmd_flags & REQ_FUA) ? 0x8 : 0);
722 SCpnt->cmnd[2] = sizeof(block) > 4 ? (unsigned char) (block >> 56) & 0xff : 0;
723 SCpnt->cmnd[3] = sizeof(block) > 4 ? (unsigned char) (block >> 48) & 0xff : 0;
724 SCpnt->cmnd[4] = sizeof(block) > 4 ? (unsigned char) (block >> 40) & 0xff : 0;
725 SCpnt->cmnd[5] = sizeof(block) > 4 ? (unsigned char) (block >> 32) & 0xff : 0;
726 SCpnt->cmnd[6] = (unsigned char) (block >> 24) & 0xff;
727 SCpnt->cmnd[7] = (unsigned char) (block >> 16) & 0xff;
728 SCpnt->cmnd[8] = (unsigned char) (block >> 8) & 0xff;
729 SCpnt->cmnd[9] = (unsigned char) block & 0xff;
730 SCpnt->cmnd[10] = (unsigned char) (this_count >> 24) & 0xff;
731 SCpnt->cmnd[11] = (unsigned char) (this_count >> 16) & 0xff;
732 SCpnt->cmnd[12] = (unsigned char) (this_count >> 8) & 0xff;
733 SCpnt->cmnd[13] = (unsigned char) this_count & 0xff;
734 SCpnt->cmnd[14] = SCpnt->cmnd[15] = 0;
735 } else if ((this_count > 0xff) || (block > 0x1fffff) ||
736 scsi_device_protection(SCpnt->device) ||
737 SCpnt->device->use_10_for_rw) {
738 if (this_count > 0xffff)
739 this_count = 0xffff;
741 SCpnt->cmnd[0] += READ_10 - READ_6;
742 SCpnt->cmnd[1] = protect | ((rq->cmd_flags & REQ_FUA) ? 0x8 : 0);
743 SCpnt->cmnd[2] = (unsigned char) (block >> 24) & 0xff;
744 SCpnt->cmnd[3] = (unsigned char) (block >> 16) & 0xff;
745 SCpnt->cmnd[4] = (unsigned char) (block >> 8) & 0xff;
746 SCpnt->cmnd[5] = (unsigned char) block & 0xff;
747 SCpnt->cmnd[6] = SCpnt->cmnd[9] = 0;
748 SCpnt->cmnd[7] = (unsigned char) (this_count >> 8) & 0xff;
749 SCpnt->cmnd[8] = (unsigned char) this_count & 0xff;
750 } else {
751 if (unlikely(rq->cmd_flags & REQ_FUA)) {
753 * This happens only if this drive failed
754 * 10byte rw command with ILLEGAL_REQUEST
755 * during operation and thus turned off
756 * use_10_for_rw.
758 scmd_printk(KERN_ERR, SCpnt,
759 "FUA write on READ/WRITE(6) drive\n");
760 goto out;
763 SCpnt->cmnd[1] |= (unsigned char) ((block >> 16) & 0x1f);
764 SCpnt->cmnd[2] = (unsigned char) ((block >> 8) & 0xff);
765 SCpnt->cmnd[3] = (unsigned char) block & 0xff;
766 SCpnt->cmnd[4] = (unsigned char) this_count;
767 SCpnt->cmnd[5] = 0;
769 SCpnt->sdb.length = this_count * sdp->sector_size;
771 /* If DIF or DIX is enabled, tell HBA how to handle request */
772 if (host_dif || scsi_prot_sg_count(SCpnt))
773 sd_prot_op(SCpnt, host_dif);
776 * We shouldn't disconnect in the middle of a sector, so with a dumb
777 * host adapter, it's safe to assume that we can at least transfer
778 * this many bytes between each connect / disconnect.
780 SCpnt->transfersize = sdp->sector_size;
781 SCpnt->underflow = this_count << 9;
782 SCpnt->allowed = SD_MAX_RETRIES;
785 * This indicates that the command is ready from our end to be
786 * queued.
788 ret = BLKPREP_OK;
789 out:
790 return scsi_prep_return(q, rq, ret);
794 * sd_open - open a scsi disk device
795 * @inode: only i_rdev member may be used
796 * @filp: only f_mode and f_flags may be used
798 * Returns 0 if successful. Returns a negated errno value in case
799 * of error.
801 * Note: This can be called from a user context (e.g. fsck(1) )
802 * or from within the kernel (e.g. as a result of a mount(1) ).
803 * In the latter case @inode and @filp carry an abridged amount
804 * of information as noted above.
806 * Locking: called with bdev->bd_mutex held.
808 static int sd_open(struct block_device *bdev, fmode_t mode)
810 struct scsi_disk *sdkp = scsi_disk_get(bdev->bd_disk);
811 struct scsi_device *sdev;
812 int retval;
814 if (!sdkp)
815 return -ENXIO;
817 SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp, "sd_open\n"));
819 sdev = sdkp->device;
821 retval = scsi_autopm_get_device(sdev);
822 if (retval)
823 goto error_autopm;
826 * If the device is in error recovery, wait until it is done.
827 * If the device is offline, then disallow any access to it.
829 retval = -ENXIO;
830 if (!scsi_block_when_processing_errors(sdev))
831 goto error_out;
833 if (sdev->removable || sdkp->write_prot)
834 check_disk_change(bdev);
837 * If the drive is empty, just let the open fail.
839 retval = -ENOMEDIUM;
840 if (sdev->removable && !sdkp->media_present && !(mode & FMODE_NDELAY))
841 goto error_out;
844 * If the device has the write protect tab set, have the open fail
845 * if the user expects to be able to write to the thing.
847 retval = -EROFS;
848 if (sdkp->write_prot && (mode & FMODE_WRITE))
849 goto error_out;
852 * It is possible that the disk changing stuff resulted in
853 * the device being taken offline. If this is the case,
854 * report this to the user, and don't pretend that the
855 * open actually succeeded.
857 retval = -ENXIO;
858 if (!scsi_device_online(sdev))
859 goto error_out;
861 if ((atomic_inc_return(&sdkp->openers) == 1) && sdev->removable) {
862 if (scsi_block_when_processing_errors(sdev))
863 scsi_set_medium_removal(sdev, SCSI_REMOVAL_PREVENT);
866 return 0;
868 error_out:
869 scsi_autopm_put_device(sdev);
870 error_autopm:
871 scsi_disk_put(sdkp);
872 return retval;
876 * sd_release - invoked when the (last) close(2) is called on this
877 * scsi disk.
878 * @inode: only i_rdev member may be used
879 * @filp: only f_mode and f_flags may be used
881 * Returns 0.
883 * Note: may block (uninterruptible) if error recovery is underway
884 * on this disk.
886 * Locking: called with bdev->bd_mutex held.
888 static int sd_release(struct gendisk *disk, fmode_t mode)
890 struct scsi_disk *sdkp = scsi_disk(disk);
891 struct scsi_device *sdev = sdkp->device;
893 SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp, "sd_release\n"));
895 if (atomic_dec_return(&sdkp->openers) == 0 && sdev->removable) {
896 if (scsi_block_when_processing_errors(sdev))
897 scsi_set_medium_removal(sdev, SCSI_REMOVAL_ALLOW);
901 * XXX and what if there are packets in flight and this close()
902 * XXX is followed by a "rmmod sd_mod"?
905 scsi_autopm_put_device(sdev);
906 scsi_disk_put(sdkp);
907 return 0;
910 static int sd_getgeo(struct block_device *bdev, struct hd_geometry *geo)
912 struct scsi_disk *sdkp = scsi_disk(bdev->bd_disk);
913 struct scsi_device *sdp = sdkp->device;
914 struct Scsi_Host *host = sdp->host;
915 int diskinfo[4];
917 /* default to most commonly used values */
918 diskinfo[0] = 0x40; /* 1 << 6 */
919 diskinfo[1] = 0x20; /* 1 << 5 */
920 diskinfo[2] = sdkp->capacity >> 11;
922 /* override with calculated, extended default, or driver values */
923 if (host->hostt->bios_param)
924 host->hostt->bios_param(sdp, bdev, sdkp->capacity, diskinfo);
925 else
926 scsicam_bios_param(bdev, sdkp->capacity, diskinfo);
928 geo->heads = diskinfo[0];
929 geo->sectors = diskinfo[1];
930 geo->cylinders = diskinfo[2];
931 return 0;
935 * sd_ioctl - process an ioctl
936 * @inode: only i_rdev/i_bdev members may be used
937 * @filp: only f_mode and f_flags may be used
938 * @cmd: ioctl command number
939 * @arg: this is third argument given to ioctl(2) system call.
940 * Often contains a pointer.
942 * Returns 0 if successful (some ioctls return postive numbers on
943 * success as well). Returns a negated errno value in case of error.
945 * Note: most ioctls are forward onto the block subsystem or further
946 * down in the scsi subsystem.
948 static int sd_ioctl(struct block_device *bdev, fmode_t mode,
949 unsigned int cmd, unsigned long arg)
951 struct gendisk *disk = bdev->bd_disk;
952 struct scsi_device *sdp = scsi_disk(disk)->device;
953 void __user *p = (void __user *)arg;
954 int error;
956 SCSI_LOG_IOCTL(1, printk("sd_ioctl: disk=%s, cmd=0x%x\n",
957 disk->disk_name, cmd));
960 * If we are in the middle of error recovery, don't let anyone
961 * else try and use this device. Also, if error recovery fails, it
962 * may try and take the device offline, in which case all further
963 * access to the device is prohibited.
965 error = scsi_nonblockable_ioctl(sdp, cmd, p,
966 (mode & FMODE_NDELAY) != 0);
967 if (!scsi_block_when_processing_errors(sdp) || !error)
968 goto out;
971 * Send SCSI addressing ioctls directly to mid level, send other
972 * ioctls to block level and then onto mid level if they can't be
973 * resolved.
975 switch (cmd) {
976 case SCSI_IOCTL_GET_IDLUN:
977 case SCSI_IOCTL_GET_BUS_NUMBER:
978 error = scsi_ioctl(sdp, cmd, p);
979 break;
980 default:
981 error = scsi_cmd_ioctl(disk->queue, disk, mode, cmd, p);
982 if (error != -ENOTTY)
983 break;
984 error = scsi_ioctl(sdp, cmd, p);
985 break;
987 out:
988 return error;
991 static void set_media_not_present(struct scsi_disk *sdkp)
993 sdkp->media_present = 0;
994 sdkp->capacity = 0;
995 sdkp->device->changed = 1;
999 * sd_media_changed - check if our medium changed
1000 * @disk: kernel device descriptor
1002 * Returns 0 if not applicable or no change; 1 if change
1004 * Note: this function is invoked from the block subsystem.
1006 static int sd_media_changed(struct gendisk *disk)
1008 struct scsi_disk *sdkp = scsi_disk(disk);
1009 struct scsi_device *sdp = sdkp->device;
1010 struct scsi_sense_hdr *sshdr = NULL;
1011 int retval;
1013 SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp, "sd_media_changed\n"));
1015 if (!sdp->removable)
1016 return 0;
1019 * If the device is offline, don't send any commands - just pretend as
1020 * if the command failed. If the device ever comes back online, we
1021 * can deal with it then. It is only because of unrecoverable errors
1022 * that we would ever take a device offline in the first place.
1024 if (!scsi_device_online(sdp)) {
1025 set_media_not_present(sdkp);
1026 retval = 1;
1027 goto out;
1031 * Using TEST_UNIT_READY enables differentiation between drive with
1032 * no cartridge loaded - NOT READY, drive with changed cartridge -
1033 * UNIT ATTENTION, or with same cartridge - GOOD STATUS.
1035 * Drives that auto spin down. eg iomega jaz 1G, will be started
1036 * by sd_spinup_disk() from sd_revalidate_disk(), which happens whenever
1037 * sd_revalidate() is called.
1039 retval = -ENODEV;
1041 if (scsi_block_when_processing_errors(sdp)) {
1042 sshdr = kzalloc(sizeof(*sshdr), GFP_KERNEL);
1043 retval = scsi_test_unit_ready(sdp, SD_TIMEOUT, SD_MAX_RETRIES,
1044 sshdr);
1048 * Unable to test, unit probably not ready. This usually
1049 * means there is no disc in the drive. Mark as changed,
1050 * and we will figure it out later once the drive is
1051 * available again.
1053 if (retval || (scsi_sense_valid(sshdr) &&
1054 /* 0x3a is medium not present */
1055 sshdr->asc == 0x3a)) {
1056 set_media_not_present(sdkp);
1057 retval = 1;
1058 goto out;
1062 * For removable scsi disk we have to recognise the presence
1063 * of a disk in the drive. This is kept in the struct scsi_disk
1064 * struct and tested at open ! Daniel Roche (dan@lectra.fr)
1066 sdkp->media_present = 1;
1068 retval = sdp->changed;
1069 sdp->changed = 0;
1070 out:
1071 if (retval != sdkp->previous_state)
1072 sdev_evt_send_simple(sdp, SDEV_EVT_MEDIA_CHANGE, GFP_KERNEL);
1073 sdkp->previous_state = retval;
1074 kfree(sshdr);
1075 return retval;
1078 static int sd_sync_cache(struct scsi_disk *sdkp)
1080 int retries, res;
1081 struct scsi_device *sdp = sdkp->device;
1082 struct scsi_sense_hdr sshdr;
1084 if (!scsi_device_online(sdp))
1085 return -ENODEV;
1088 for (retries = 3; retries > 0; --retries) {
1089 unsigned char cmd[10] = { 0 };
1091 cmd[0] = SYNCHRONIZE_CACHE;
1093 * Leave the rest of the command zero to indicate
1094 * flush everything.
1096 res = scsi_execute_req(sdp, cmd, DMA_NONE, NULL, 0, &sshdr,
1097 SD_FLUSH_TIMEOUT, SD_MAX_RETRIES, NULL);
1098 if (res == 0)
1099 break;
1102 if (res) {
1103 sd_print_result(sdkp, res);
1104 if (driver_byte(res) & DRIVER_SENSE)
1105 sd_print_sense_hdr(sdkp, &sshdr);
1108 if (res)
1109 return -EIO;
1110 return 0;
1113 static void sd_rescan(struct device *dev)
1115 struct scsi_disk *sdkp = scsi_disk_get_from_dev(dev);
1117 if (sdkp) {
1118 revalidate_disk(sdkp->disk);
1119 scsi_disk_put(sdkp);
1124 #ifdef CONFIG_COMPAT
1126 * This gets directly called from VFS. When the ioctl
1127 * is not recognized we go back to the other translation paths.
1129 static int sd_compat_ioctl(struct block_device *bdev, fmode_t mode,
1130 unsigned int cmd, unsigned long arg)
1132 struct scsi_device *sdev = scsi_disk(bdev->bd_disk)->device;
1135 * If we are in the middle of error recovery, don't let anyone
1136 * else try and use this device. Also, if error recovery fails, it
1137 * may try and take the device offline, in which case all further
1138 * access to the device is prohibited.
1140 if (!scsi_block_when_processing_errors(sdev))
1141 return -ENODEV;
1143 if (sdev->host->hostt->compat_ioctl) {
1144 int ret;
1146 ret = sdev->host->hostt->compat_ioctl(sdev, cmd, (void __user *)arg);
1148 return ret;
1152 * Let the static ioctl translation table take care of it.
1154 return -ENOIOCTLCMD;
1156 #endif
1158 static const struct block_device_operations sd_fops = {
1159 .owner = THIS_MODULE,
1160 .open = sd_open,
1161 .release = sd_release,
1162 .ioctl = sd_ioctl,
1163 .getgeo = sd_getgeo,
1164 #ifdef CONFIG_COMPAT
1165 .compat_ioctl = sd_compat_ioctl,
1166 #endif
1167 .media_changed = sd_media_changed,
1168 .revalidate_disk = sd_revalidate_disk,
1169 .unlock_native_capacity = sd_unlock_native_capacity,
1172 static unsigned int sd_completed_bytes(struct scsi_cmnd *scmd)
1174 u64 start_lba = blk_rq_pos(scmd->request);
1175 u64 end_lba = blk_rq_pos(scmd->request) + (scsi_bufflen(scmd) / 512);
1176 u64 bad_lba;
1177 int info_valid;
1179 if (scmd->request->cmd_type != REQ_TYPE_FS)
1180 return 0;
1182 info_valid = scsi_get_sense_info_fld(scmd->sense_buffer,
1183 SCSI_SENSE_BUFFERSIZE,
1184 &bad_lba);
1185 if (!info_valid)
1186 return 0;
1188 if (scsi_bufflen(scmd) <= scmd->device->sector_size)
1189 return 0;
1191 if (scmd->device->sector_size < 512) {
1192 /* only legitimate sector_size here is 256 */
1193 start_lba <<= 1;
1194 end_lba <<= 1;
1195 } else {
1196 /* be careful ... don't want any overflows */
1197 u64 factor = scmd->device->sector_size / 512;
1198 do_div(start_lba, factor);
1199 do_div(end_lba, factor);
1202 /* The bad lba was reported incorrectly, we have no idea where
1203 * the error is.
1205 if (bad_lba < start_lba || bad_lba >= end_lba)
1206 return 0;
1208 /* This computation should always be done in terms of
1209 * the resolution of the device's medium.
1211 return (bad_lba - start_lba) * scmd->device->sector_size;
1215 * sd_done - bottom half handler: called when the lower level
1216 * driver has completed (successfully or otherwise) a scsi command.
1217 * @SCpnt: mid-level's per command structure.
1219 * Note: potentially run from within an ISR. Must not block.
1221 static int sd_done(struct scsi_cmnd *SCpnt)
1223 int result = SCpnt->result;
1224 unsigned int good_bytes = result ? 0 : scsi_bufflen(SCpnt);
1225 struct scsi_sense_hdr sshdr;
1226 struct scsi_disk *sdkp = scsi_disk(SCpnt->request->rq_disk);
1227 int sense_valid = 0;
1228 int sense_deferred = 0;
1230 if (SCpnt->request->cmd_flags & REQ_DISCARD) {
1231 if (!result)
1232 scsi_set_resid(SCpnt, 0);
1233 return good_bytes;
1236 if (result) {
1237 sense_valid = scsi_command_normalize_sense(SCpnt, &sshdr);
1238 if (sense_valid)
1239 sense_deferred = scsi_sense_is_deferred(&sshdr);
1241 #ifdef CONFIG_SCSI_LOGGING
1242 SCSI_LOG_HLCOMPLETE(1, scsi_print_result(SCpnt));
1243 if (sense_valid) {
1244 SCSI_LOG_HLCOMPLETE(1, scmd_printk(KERN_INFO, SCpnt,
1245 "sd_done: sb[respc,sk,asc,"
1246 "ascq]=%x,%x,%x,%x\n",
1247 sshdr.response_code,
1248 sshdr.sense_key, sshdr.asc,
1249 sshdr.ascq));
1251 #endif
1252 if (driver_byte(result) != DRIVER_SENSE &&
1253 (!sense_valid || sense_deferred))
1254 goto out;
1256 switch (sshdr.sense_key) {
1257 case HARDWARE_ERROR:
1258 case MEDIUM_ERROR:
1259 good_bytes = sd_completed_bytes(SCpnt);
1260 break;
1261 case RECOVERED_ERROR:
1262 good_bytes = scsi_bufflen(SCpnt);
1263 break;
1264 case NO_SENSE:
1265 /* This indicates a false check condition, so ignore it. An
1266 * unknown amount of data was transferred so treat it as an
1267 * error.
1269 scsi_print_sense("sd", SCpnt);
1270 SCpnt->result = 0;
1271 memset(SCpnt->sense_buffer, 0, SCSI_SENSE_BUFFERSIZE);
1272 break;
1273 case ABORTED_COMMAND: /* DIF: Target detected corruption */
1274 case ILLEGAL_REQUEST: /* DIX: Host detected corruption */
1275 if (sshdr.asc == 0x10)
1276 good_bytes = sd_completed_bytes(SCpnt);
1277 break;
1278 default:
1279 break;
1281 out:
1282 if (rq_data_dir(SCpnt->request) == READ && scsi_prot_sg_count(SCpnt))
1283 sd_dif_complete(SCpnt, good_bytes);
1285 if (scsi_host_dif_capable(sdkp->device->host, sdkp->protection_type)
1286 == SD_DIF_TYPE2_PROTECTION && SCpnt->cmnd != SCpnt->request->cmd) {
1288 /* We have to print a failed command here as the
1289 * extended CDB gets freed before scsi_io_completion()
1290 * is called.
1292 if (result)
1293 scsi_print_command(SCpnt);
1295 mempool_free(SCpnt->cmnd, sd_cdb_pool);
1296 SCpnt->cmnd = NULL;
1297 SCpnt->cmd_len = 0;
1300 return good_bytes;
1303 static int media_not_present(struct scsi_disk *sdkp,
1304 struct scsi_sense_hdr *sshdr)
1307 if (!scsi_sense_valid(sshdr))
1308 return 0;
1309 /* not invoked for commands that could return deferred errors */
1310 if (sshdr->sense_key != NOT_READY &&
1311 sshdr->sense_key != UNIT_ATTENTION)
1312 return 0;
1313 if (sshdr->asc != 0x3A) /* medium not present */
1314 return 0;
1316 set_media_not_present(sdkp);
1317 return 1;
1321 * spinup disk - called only in sd_revalidate_disk()
1323 static void
1324 sd_spinup_disk(struct scsi_disk *sdkp)
1326 unsigned char cmd[10];
1327 unsigned long spintime_expire = 0;
1328 int retries, spintime;
1329 unsigned int the_result;
1330 struct scsi_sense_hdr sshdr;
1331 int sense_valid = 0;
1333 spintime = 0;
1335 /* Spin up drives, as required. Only do this at boot time */
1336 /* Spinup needs to be done for module loads too. */
1337 do {
1338 retries = 0;
1340 do {
1341 cmd[0] = TEST_UNIT_READY;
1342 memset((void *) &cmd[1], 0, 9);
1344 the_result = scsi_execute_req(sdkp->device, cmd,
1345 DMA_NONE, NULL, 0,
1346 &sshdr, SD_TIMEOUT,
1347 SD_MAX_RETRIES, NULL);
1350 * If the drive has indicated to us that it
1351 * doesn't have any media in it, don't bother
1352 * with any more polling.
1354 if (media_not_present(sdkp, &sshdr))
1355 return;
1357 if (the_result)
1358 sense_valid = scsi_sense_valid(&sshdr);
1359 retries++;
1360 } while (retries < 3 &&
1361 (!scsi_status_is_good(the_result) ||
1362 ((driver_byte(the_result) & DRIVER_SENSE) &&
1363 sense_valid && sshdr.sense_key == UNIT_ATTENTION)));
1365 if ((driver_byte(the_result) & DRIVER_SENSE) == 0) {
1366 /* no sense, TUR either succeeded or failed
1367 * with a status error */
1368 if(!spintime && !scsi_status_is_good(the_result)) {
1369 sd_printk(KERN_NOTICE, sdkp, "Unit Not Ready\n");
1370 sd_print_result(sdkp, the_result);
1372 break;
1376 * The device does not want the automatic start to be issued.
1378 if (sdkp->device->no_start_on_add)
1379 break;
1381 if (sense_valid && sshdr.sense_key == NOT_READY) {
1382 if (sshdr.asc == 4 && sshdr.ascq == 3)
1383 break; /* manual intervention required */
1384 if (sshdr.asc == 4 && sshdr.ascq == 0xb)
1385 break; /* standby */
1386 if (sshdr.asc == 4 && sshdr.ascq == 0xc)
1387 break; /* unavailable */
1389 * Issue command to spin up drive when not ready
1391 if (!spintime) {
1392 sd_printk(KERN_NOTICE, sdkp, "Spinning up disk...");
1393 cmd[0] = START_STOP;
1394 cmd[1] = 1; /* Return immediately */
1395 memset((void *) &cmd[2], 0, 8);
1396 cmd[4] = 1; /* Start spin cycle */
1397 if (sdkp->device->start_stop_pwr_cond)
1398 cmd[4] |= 1 << 4;
1399 scsi_execute_req(sdkp->device, cmd, DMA_NONE,
1400 NULL, 0, &sshdr,
1401 SD_TIMEOUT, SD_MAX_RETRIES,
1402 NULL);
1403 spintime_expire = jiffies + 100 * HZ;
1404 spintime = 1;
1406 /* Wait 1 second for next try */
1407 msleep(1000);
1408 printk(".");
1411 * Wait for USB flash devices with slow firmware.
1412 * Yes, this sense key/ASC combination shouldn't
1413 * occur here. It's characteristic of these devices.
1415 } else if (sense_valid &&
1416 sshdr.sense_key == UNIT_ATTENTION &&
1417 sshdr.asc == 0x28) {
1418 if (!spintime) {
1419 spintime_expire = jiffies + 5 * HZ;
1420 spintime = 1;
1422 /* Wait 1 second for next try */
1423 msleep(1000);
1424 } else {
1425 /* we don't understand the sense code, so it's
1426 * probably pointless to loop */
1427 if(!spintime) {
1428 sd_printk(KERN_NOTICE, sdkp, "Unit Not Ready\n");
1429 sd_print_sense_hdr(sdkp, &sshdr);
1431 break;
1434 } while (spintime && time_before_eq(jiffies, spintime_expire));
1436 if (spintime) {
1437 if (scsi_status_is_good(the_result))
1438 printk("ready\n");
1439 else
1440 printk("not responding...\n");
1446 * Determine whether disk supports Data Integrity Field.
1448 static void sd_read_protection_type(struct scsi_disk *sdkp, unsigned char *buffer)
1450 struct scsi_device *sdp = sdkp->device;
1451 u8 type;
1453 if (scsi_device_protection(sdp) == 0 || (buffer[12] & 1) == 0)
1454 return;
1456 type = ((buffer[12] >> 1) & 7) + 1; /* P_TYPE 0 = Type 1 */
1458 if (type == sdkp->protection_type || !sdkp->first_scan)
1459 return;
1461 sdkp->protection_type = type;
1463 if (type > SD_DIF_TYPE3_PROTECTION) {
1464 sd_printk(KERN_ERR, sdkp, "formatted with unsupported " \
1465 "protection type %u. Disabling disk!\n", type);
1466 sdkp->capacity = 0;
1467 return;
1470 if (scsi_host_dif_capable(sdp->host, type))
1471 sd_printk(KERN_NOTICE, sdkp,
1472 "Enabling DIF Type %u protection\n", type);
1473 else
1474 sd_printk(KERN_NOTICE, sdkp,
1475 "Disabling DIF Type %u protection\n", type);
1478 static void read_capacity_error(struct scsi_disk *sdkp, struct scsi_device *sdp,
1479 struct scsi_sense_hdr *sshdr, int sense_valid,
1480 int the_result)
1482 sd_print_result(sdkp, the_result);
1483 if (driver_byte(the_result) & DRIVER_SENSE)
1484 sd_print_sense_hdr(sdkp, sshdr);
1485 else
1486 sd_printk(KERN_NOTICE, sdkp, "Sense not available.\n");
1489 * Set dirty bit for removable devices if not ready -
1490 * sometimes drives will not report this properly.
1492 if (sdp->removable &&
1493 sense_valid && sshdr->sense_key == NOT_READY)
1494 sdp->changed = 1;
1497 * We used to set media_present to 0 here to indicate no media
1498 * in the drive, but some drives fail read capacity even with
1499 * media present, so we can't do that.
1501 sdkp->capacity = 0; /* unknown mapped to zero - as usual */
1504 #define RC16_LEN 32
1505 #if RC16_LEN > SD_BUF_SIZE
1506 #error RC16_LEN must not be more than SD_BUF_SIZE
1507 #endif
1509 #define READ_CAPACITY_RETRIES_ON_RESET 10
1511 static int read_capacity_16(struct scsi_disk *sdkp, struct scsi_device *sdp,
1512 unsigned char *buffer)
1514 unsigned char cmd[16];
1515 struct scsi_sense_hdr sshdr;
1516 int sense_valid = 0;
1517 int the_result;
1518 int retries = 3, reset_retries = READ_CAPACITY_RETRIES_ON_RESET;
1519 unsigned int alignment;
1520 unsigned long long lba;
1521 unsigned sector_size;
1523 if (sdp->no_read_capacity_16)
1524 return -EINVAL;
1526 do {
1527 memset(cmd, 0, 16);
1528 cmd[0] = SERVICE_ACTION_IN;
1529 cmd[1] = SAI_READ_CAPACITY_16;
1530 cmd[13] = RC16_LEN;
1531 memset(buffer, 0, RC16_LEN);
1533 the_result = scsi_execute_req(sdp, cmd, DMA_FROM_DEVICE,
1534 buffer, RC16_LEN, &sshdr,
1535 SD_TIMEOUT, SD_MAX_RETRIES, NULL);
1537 if (media_not_present(sdkp, &sshdr))
1538 return -ENODEV;
1540 if (the_result) {
1541 sense_valid = scsi_sense_valid(&sshdr);
1542 if (sense_valid &&
1543 sshdr.sense_key == ILLEGAL_REQUEST &&
1544 (sshdr.asc == 0x20 || sshdr.asc == 0x24) &&
1545 sshdr.ascq == 0x00)
1546 /* Invalid Command Operation Code or
1547 * Invalid Field in CDB, just retry
1548 * silently with RC10 */
1549 return -EINVAL;
1550 if (sense_valid &&
1551 sshdr.sense_key == UNIT_ATTENTION &&
1552 sshdr.asc == 0x29 && sshdr.ascq == 0x00)
1553 /* Device reset might occur several times,
1554 * give it one more chance */
1555 if (--reset_retries > 0)
1556 continue;
1558 retries--;
1560 } while (the_result && retries);
1562 if (the_result) {
1563 sd_printk(KERN_NOTICE, sdkp, "READ CAPACITY(16) failed\n");
1564 read_capacity_error(sdkp, sdp, &sshdr, sense_valid, the_result);
1565 return -EINVAL;
1568 sector_size = get_unaligned_be32(&buffer[8]);
1569 lba = get_unaligned_be64(&buffer[0]);
1571 sd_read_protection_type(sdkp, buffer);
1573 if ((sizeof(sdkp->capacity) == 4) && (lba >= 0xffffffffULL)) {
1574 sd_printk(KERN_ERR, sdkp, "Too big for this kernel. Use a "
1575 "kernel compiled with support for large block "
1576 "devices.\n");
1577 sdkp->capacity = 0;
1578 return -EOVERFLOW;
1581 /* Logical blocks per physical block exponent */
1582 sdkp->physical_block_size = (1 << (buffer[13] & 0xf)) * sector_size;
1584 /* Lowest aligned logical block */
1585 alignment = ((buffer[14] & 0x3f) << 8 | buffer[15]) * sector_size;
1586 blk_queue_alignment_offset(sdp->request_queue, alignment);
1587 if (alignment && sdkp->first_scan)
1588 sd_printk(KERN_NOTICE, sdkp,
1589 "physical block alignment offset: %u\n", alignment);
1591 if (buffer[14] & 0x80) { /* TPE */
1592 struct request_queue *q = sdp->request_queue;
1594 sdkp->thin_provisioning = 1;
1595 q->limits.discard_granularity = sdkp->physical_block_size;
1596 q->limits.max_discard_sectors = 0xffffffff;
1598 if (buffer[14] & 0x40) /* TPRZ */
1599 q->limits.discard_zeroes_data = 1;
1601 queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, q);
1604 sdkp->capacity = lba + 1;
1605 return sector_size;
1608 static int read_capacity_10(struct scsi_disk *sdkp, struct scsi_device *sdp,
1609 unsigned char *buffer)
1611 unsigned char cmd[16];
1612 struct scsi_sense_hdr sshdr;
1613 int sense_valid = 0;
1614 int the_result;
1615 int retries = 3, reset_retries = READ_CAPACITY_RETRIES_ON_RESET;
1616 sector_t lba;
1617 unsigned sector_size;
1619 do {
1620 cmd[0] = READ_CAPACITY;
1621 memset(&cmd[1], 0, 9);
1622 memset(buffer, 0, 8);
1624 the_result = scsi_execute_req(sdp, cmd, DMA_FROM_DEVICE,
1625 buffer, 8, &sshdr,
1626 SD_TIMEOUT, SD_MAX_RETRIES, NULL);
1628 if (media_not_present(sdkp, &sshdr))
1629 return -ENODEV;
1631 if (the_result) {
1632 sense_valid = scsi_sense_valid(&sshdr);
1633 if (sense_valid &&
1634 sshdr.sense_key == UNIT_ATTENTION &&
1635 sshdr.asc == 0x29 && sshdr.ascq == 0x00)
1636 /* Device reset might occur several times,
1637 * give it one more chance */
1638 if (--reset_retries > 0)
1639 continue;
1641 retries--;
1643 } while (the_result && retries);
1645 if (the_result) {
1646 sd_printk(KERN_NOTICE, sdkp, "READ CAPACITY failed\n");
1647 read_capacity_error(sdkp, sdp, &sshdr, sense_valid, the_result);
1648 return -EINVAL;
1651 sector_size = get_unaligned_be32(&buffer[4]);
1652 lba = get_unaligned_be32(&buffer[0]);
1654 if (sdp->no_read_capacity_16 && (lba == 0xffffffff)) {
1655 /* Some buggy (usb cardreader) devices return an lba of
1656 0xffffffff when the want to report a size of 0 (with
1657 which they really mean no media is present) */
1658 sdkp->capacity = 0;
1659 sdkp->physical_block_size = sector_size;
1660 return sector_size;
1663 if ((sizeof(sdkp->capacity) == 4) && (lba == 0xffffffff)) {
1664 sd_printk(KERN_ERR, sdkp, "Too big for this kernel. Use a "
1665 "kernel compiled with support for large block "
1666 "devices.\n");
1667 sdkp->capacity = 0;
1668 return -EOVERFLOW;
1671 sdkp->capacity = lba + 1;
1672 sdkp->physical_block_size = sector_size;
1673 return sector_size;
1676 static int sd_try_rc16_first(struct scsi_device *sdp)
1678 if (sdp->host->max_cmd_len < 16)
1679 return 0;
1680 if (sdp->scsi_level > SCSI_SPC_2)
1681 return 1;
1682 if (scsi_device_protection(sdp))
1683 return 1;
1684 return 0;
1688 * read disk capacity
1690 static void
1691 sd_read_capacity(struct scsi_disk *sdkp, unsigned char *buffer)
1693 int sector_size;
1694 struct scsi_device *sdp = sdkp->device;
1695 sector_t old_capacity = sdkp->capacity;
1697 if (sd_try_rc16_first(sdp)) {
1698 sector_size = read_capacity_16(sdkp, sdp, buffer);
1699 if (sector_size == -EOVERFLOW)
1700 goto got_data;
1701 if (sector_size == -ENODEV)
1702 return;
1703 if (sector_size < 0)
1704 sector_size = read_capacity_10(sdkp, sdp, buffer);
1705 if (sector_size < 0)
1706 return;
1707 } else {
1708 sector_size = read_capacity_10(sdkp, sdp, buffer);
1709 if (sector_size == -EOVERFLOW)
1710 goto got_data;
1711 if (sector_size < 0)
1712 return;
1713 if ((sizeof(sdkp->capacity) > 4) &&
1714 (sdkp->capacity > 0xffffffffULL)) {
1715 int old_sector_size = sector_size;
1716 sd_printk(KERN_NOTICE, sdkp, "Very big device. "
1717 "Trying to use READ CAPACITY(16).\n");
1718 sector_size = read_capacity_16(sdkp, sdp, buffer);
1719 if (sector_size < 0) {
1720 sd_printk(KERN_NOTICE, sdkp,
1721 "Using 0xffffffff as device size\n");
1722 sdkp->capacity = 1 + (sector_t) 0xffffffff;
1723 sector_size = old_sector_size;
1724 goto got_data;
1729 /* Some devices are known to return the total number of blocks,
1730 * not the highest block number. Some devices have versions
1731 * which do this and others which do not. Some devices we might
1732 * suspect of doing this but we don't know for certain.
1734 * If we know the reported capacity is wrong, decrement it. If
1735 * we can only guess, then assume the number of blocks is even
1736 * (usually true but not always) and err on the side of lowering
1737 * the capacity.
1739 if (sdp->fix_capacity ||
1740 (sdp->guess_capacity && (sdkp->capacity & 0x01))) {
1741 sd_printk(KERN_INFO, sdkp, "Adjusting the sector count "
1742 "from its reported value: %llu\n",
1743 (unsigned long long) sdkp->capacity);
1744 --sdkp->capacity;
1747 got_data:
1748 if (sector_size == 0) {
1749 sector_size = 512;
1750 sd_printk(KERN_NOTICE, sdkp, "Sector size 0 reported, "
1751 "assuming 512.\n");
1754 if (sector_size != 512 &&
1755 sector_size != 1024 &&
1756 sector_size != 2048 &&
1757 sector_size != 4096 &&
1758 sector_size != 256) {
1759 sd_printk(KERN_NOTICE, sdkp, "Unsupported sector size %d.\n",
1760 sector_size);
1762 * The user might want to re-format the drive with
1763 * a supported sectorsize. Once this happens, it
1764 * would be relatively trivial to set the thing up.
1765 * For this reason, we leave the thing in the table.
1767 sdkp->capacity = 0;
1769 * set a bogus sector size so the normal read/write
1770 * logic in the block layer will eventually refuse any
1771 * request on this device without tripping over power
1772 * of two sector size assumptions
1774 sector_size = 512;
1776 blk_queue_logical_block_size(sdp->request_queue, sector_size);
1779 char cap_str_2[10], cap_str_10[10];
1780 u64 sz = (u64)sdkp->capacity << ilog2(sector_size);
1782 string_get_size(sz, STRING_UNITS_2, cap_str_2,
1783 sizeof(cap_str_2));
1784 string_get_size(sz, STRING_UNITS_10, cap_str_10,
1785 sizeof(cap_str_10));
1787 if (sdkp->first_scan || old_capacity != sdkp->capacity) {
1788 sd_printk(KERN_NOTICE, sdkp,
1789 "%llu %d-byte logical blocks: (%s/%s)\n",
1790 (unsigned long long)sdkp->capacity,
1791 sector_size, cap_str_10, cap_str_2);
1793 if (sdkp->physical_block_size != sector_size)
1794 sd_printk(KERN_NOTICE, sdkp,
1795 "%u-byte physical blocks\n",
1796 sdkp->physical_block_size);
1800 /* Rescale capacity to 512-byte units */
1801 if (sector_size == 4096)
1802 sdkp->capacity <<= 3;
1803 else if (sector_size == 2048)
1804 sdkp->capacity <<= 2;
1805 else if (sector_size == 1024)
1806 sdkp->capacity <<= 1;
1807 else if (sector_size == 256)
1808 sdkp->capacity >>= 1;
1810 blk_queue_physical_block_size(sdp->request_queue,
1811 sdkp->physical_block_size);
1812 sdkp->device->sector_size = sector_size;
1815 /* called with buffer of length 512 */
1816 static inline int
1817 sd_do_mode_sense(struct scsi_device *sdp, int dbd, int modepage,
1818 unsigned char *buffer, int len, struct scsi_mode_data *data,
1819 struct scsi_sense_hdr *sshdr)
1821 return scsi_mode_sense(sdp, dbd, modepage, buffer, len,
1822 SD_TIMEOUT, SD_MAX_RETRIES, data,
1823 sshdr);
1827 * read write protect setting, if possible - called only in sd_revalidate_disk()
1828 * called with buffer of length SD_BUF_SIZE
1830 static void
1831 sd_read_write_protect_flag(struct scsi_disk *sdkp, unsigned char *buffer)
1833 int res;
1834 struct scsi_device *sdp = sdkp->device;
1835 struct scsi_mode_data data;
1836 int old_wp = sdkp->write_prot;
1838 set_disk_ro(sdkp->disk, 0);
1839 if (sdp->skip_ms_page_3f) {
1840 sd_printk(KERN_NOTICE, sdkp, "Assuming Write Enabled\n");
1841 return;
1844 if (sdp->use_192_bytes_for_3f) {
1845 res = sd_do_mode_sense(sdp, 0, 0x3F, buffer, 192, &data, NULL);
1846 } else {
1848 * First attempt: ask for all pages (0x3F), but only 4 bytes.
1849 * We have to start carefully: some devices hang if we ask
1850 * for more than is available.
1852 res = sd_do_mode_sense(sdp, 0, 0x3F, buffer, 4, &data, NULL);
1855 * Second attempt: ask for page 0 When only page 0 is
1856 * implemented, a request for page 3F may return Sense Key
1857 * 5: Illegal Request, Sense Code 24: Invalid field in
1858 * CDB.
1860 if (!scsi_status_is_good(res))
1861 res = sd_do_mode_sense(sdp, 0, 0, buffer, 4, &data, NULL);
1864 * Third attempt: ask 255 bytes, as we did earlier.
1866 if (!scsi_status_is_good(res))
1867 res = sd_do_mode_sense(sdp, 0, 0x3F, buffer, 255,
1868 &data, NULL);
1871 if (!scsi_status_is_good(res)) {
1872 sd_printk(KERN_WARNING, sdkp,
1873 "Test WP failed, assume Write Enabled\n");
1874 } else {
1875 sdkp->write_prot = ((data.device_specific & 0x80) != 0);
1876 set_disk_ro(sdkp->disk, sdkp->write_prot);
1877 if (sdkp->first_scan || old_wp != sdkp->write_prot) {
1878 sd_printk(KERN_NOTICE, sdkp, "Write Protect is %s\n",
1879 sdkp->write_prot ? "on" : "off");
1880 sd_printk(KERN_DEBUG, sdkp,
1881 "Mode Sense: %02x %02x %02x %02x\n",
1882 buffer[0], buffer[1], buffer[2], buffer[3]);
1888 * sd_read_cache_type - called only from sd_revalidate_disk()
1889 * called with buffer of length SD_BUF_SIZE
1891 static void
1892 sd_read_cache_type(struct scsi_disk *sdkp, unsigned char *buffer)
1894 int len = 0, res;
1895 struct scsi_device *sdp = sdkp->device;
1897 int dbd;
1898 int modepage;
1899 struct scsi_mode_data data;
1900 struct scsi_sense_hdr sshdr;
1901 int old_wce = sdkp->WCE;
1902 int old_rcd = sdkp->RCD;
1903 int old_dpofua = sdkp->DPOFUA;
1905 if (sdp->skip_ms_page_8)
1906 goto defaults;
1908 if (sdp->type == TYPE_RBC) {
1909 modepage = 6;
1910 dbd = 8;
1911 } else {
1912 modepage = 8;
1913 dbd = 0;
1916 /* cautiously ask */
1917 res = sd_do_mode_sense(sdp, dbd, modepage, buffer, 4, &data, &sshdr);
1919 if (!scsi_status_is_good(res))
1920 goto bad_sense;
1922 if (!data.header_length) {
1923 modepage = 6;
1924 sd_printk(KERN_ERR, sdkp, "Missing header in MODE_SENSE response\n");
1927 /* that went OK, now ask for the proper length */
1928 len = data.length;
1931 * We're only interested in the first three bytes, actually.
1932 * But the data cache page is defined for the first 20.
1934 if (len < 3)
1935 goto bad_sense;
1936 if (len > 20)
1937 len = 20;
1939 /* Take headers and block descriptors into account */
1940 len += data.header_length + data.block_descriptor_length;
1941 if (len > SD_BUF_SIZE)
1942 goto bad_sense;
1944 /* Get the data */
1945 res = sd_do_mode_sense(sdp, dbd, modepage, buffer, len, &data, &sshdr);
1947 if (scsi_status_is_good(res)) {
1948 int offset = data.header_length + data.block_descriptor_length;
1950 if (offset >= SD_BUF_SIZE - 2) {
1951 sd_printk(KERN_ERR, sdkp, "Malformed MODE SENSE response\n");
1952 goto defaults;
1955 if ((buffer[offset] & 0x3f) != modepage) {
1956 sd_printk(KERN_ERR, sdkp, "Got wrong page\n");
1957 goto defaults;
1960 if (modepage == 8) {
1961 sdkp->WCE = ((buffer[offset + 2] & 0x04) != 0);
1962 sdkp->RCD = ((buffer[offset + 2] & 0x01) != 0);
1963 } else {
1964 sdkp->WCE = ((buffer[offset + 2] & 0x01) == 0);
1965 sdkp->RCD = 0;
1968 sdkp->DPOFUA = (data.device_specific & 0x10) != 0;
1969 if (sdkp->DPOFUA && !sdkp->device->use_10_for_rw) {
1970 sd_printk(KERN_NOTICE, sdkp,
1971 "Uses READ/WRITE(6), disabling FUA\n");
1972 sdkp->DPOFUA = 0;
1975 if (sdkp->first_scan || old_wce != sdkp->WCE ||
1976 old_rcd != sdkp->RCD || old_dpofua != sdkp->DPOFUA)
1977 sd_printk(KERN_NOTICE, sdkp,
1978 "Write cache: %s, read cache: %s, %s\n",
1979 sdkp->WCE ? "enabled" : "disabled",
1980 sdkp->RCD ? "disabled" : "enabled",
1981 sdkp->DPOFUA ? "supports DPO and FUA"
1982 : "doesn't support DPO or FUA");
1984 return;
1987 bad_sense:
1988 if (scsi_sense_valid(&sshdr) &&
1989 sshdr.sense_key == ILLEGAL_REQUEST &&
1990 sshdr.asc == 0x24 && sshdr.ascq == 0x0)
1991 /* Invalid field in CDB */
1992 sd_printk(KERN_NOTICE, sdkp, "Cache data unavailable\n");
1993 else
1994 sd_printk(KERN_ERR, sdkp, "Asking for cache data failed\n");
1996 defaults:
1997 sd_printk(KERN_ERR, sdkp, "Assuming drive cache: write through\n");
1998 sdkp->WCE = 0;
1999 sdkp->RCD = 0;
2000 sdkp->DPOFUA = 0;
2004 * The ATO bit indicates whether the DIF application tag is available
2005 * for use by the operating system.
2007 static void sd_read_app_tag_own(struct scsi_disk *sdkp, unsigned char *buffer)
2009 int res, offset;
2010 struct scsi_device *sdp = sdkp->device;
2011 struct scsi_mode_data data;
2012 struct scsi_sense_hdr sshdr;
2014 if (sdp->type != TYPE_DISK)
2015 return;
2017 if (sdkp->protection_type == 0)
2018 return;
2020 res = scsi_mode_sense(sdp, 1, 0x0a, buffer, 36, SD_TIMEOUT,
2021 SD_MAX_RETRIES, &data, &sshdr);
2023 if (!scsi_status_is_good(res) || !data.header_length ||
2024 data.length < 6) {
2025 sd_printk(KERN_WARNING, sdkp,
2026 "getting Control mode page failed, assume no ATO\n");
2028 if (scsi_sense_valid(&sshdr))
2029 sd_print_sense_hdr(sdkp, &sshdr);
2031 return;
2034 offset = data.header_length + data.block_descriptor_length;
2036 if ((buffer[offset] & 0x3f) != 0x0a) {
2037 sd_printk(KERN_ERR, sdkp, "ATO Got wrong page\n");
2038 return;
2041 if ((buffer[offset + 5] & 0x80) == 0)
2042 return;
2044 sdkp->ATO = 1;
2046 return;
2050 * sd_read_block_limits - Query disk device for preferred I/O sizes.
2051 * @disk: disk to query
2053 static void sd_read_block_limits(struct scsi_disk *sdkp)
2055 struct request_queue *q = sdkp->disk->queue;
2056 unsigned int sector_sz = sdkp->device->sector_size;
2057 const int vpd_len = 64;
2058 unsigned char *buffer = kmalloc(vpd_len, GFP_KERNEL);
2060 if (!buffer ||
2061 /* Block Limits VPD */
2062 scsi_get_vpd_page(sdkp->device, 0xb0, buffer, vpd_len))
2063 goto out;
2065 blk_queue_io_min(sdkp->disk->queue,
2066 get_unaligned_be16(&buffer[6]) * sector_sz);
2067 blk_queue_io_opt(sdkp->disk->queue,
2068 get_unaligned_be32(&buffer[12]) * sector_sz);
2070 /* Thin provisioning enabled and page length indicates TP support */
2071 if (sdkp->thin_provisioning && buffer[3] == 0x3c) {
2072 unsigned int lba_count, desc_count, granularity;
2074 lba_count = get_unaligned_be32(&buffer[20]);
2075 desc_count = get_unaligned_be32(&buffer[24]);
2077 if (lba_count && desc_count) {
2078 if (sdkp->tpvpd && !sdkp->tpu)
2079 sdkp->unmap = 0;
2080 else
2081 sdkp->unmap = 1;
2084 if (sdkp->tpvpd && !sdkp->tpu && !sdkp->tpws) {
2085 sd_printk(KERN_ERR, sdkp, "Thin provisioning is " \
2086 "enabled but neither TPU, nor TPWS are " \
2087 "set. Disabling discard!\n");
2088 goto out;
2091 if (lba_count)
2092 q->limits.max_discard_sectors =
2093 lba_count * sector_sz >> 9;
2095 granularity = get_unaligned_be32(&buffer[28]);
2097 if (granularity)
2098 q->limits.discard_granularity = granularity * sector_sz;
2100 if (buffer[32] & 0x80)
2101 q->limits.discard_alignment =
2102 get_unaligned_be32(&buffer[32]) & ~(1 << 31);
2105 out:
2106 kfree(buffer);
2110 * sd_read_block_characteristics - Query block dev. characteristics
2111 * @disk: disk to query
2113 static void sd_read_block_characteristics(struct scsi_disk *sdkp)
2115 unsigned char *buffer;
2116 u16 rot;
2117 const int vpd_len = 64;
2119 buffer = kmalloc(vpd_len, GFP_KERNEL);
2121 if (!buffer ||
2122 /* Block Device Characteristics VPD */
2123 scsi_get_vpd_page(sdkp->device, 0xb1, buffer, vpd_len))
2124 goto out;
2126 rot = get_unaligned_be16(&buffer[4]);
2128 if (rot == 1)
2129 queue_flag_set_unlocked(QUEUE_FLAG_NONROT, sdkp->disk->queue);
2131 out:
2132 kfree(buffer);
2136 * sd_read_thin_provisioning - Query thin provisioning VPD page
2137 * @disk: disk to query
2139 static void sd_read_thin_provisioning(struct scsi_disk *sdkp)
2141 unsigned char *buffer;
2142 const int vpd_len = 8;
2144 if (sdkp->thin_provisioning == 0)
2145 return;
2147 buffer = kmalloc(vpd_len, GFP_KERNEL);
2149 if (!buffer || scsi_get_vpd_page(sdkp->device, 0xb2, buffer, vpd_len))
2150 goto out;
2152 sdkp->tpvpd = 1;
2153 sdkp->tpu = (buffer[5] >> 7) & 1; /* UNMAP */
2154 sdkp->tpws = (buffer[5] >> 6) & 1; /* WRITE SAME(16) with UNMAP */
2156 out:
2157 kfree(buffer);
2160 static int sd_try_extended_inquiry(struct scsi_device *sdp)
2163 * Although VPD inquiries can go to SCSI-2 type devices,
2164 * some USB ones crash on receiving them, and the pages
2165 * we currently ask for are for SPC-3 and beyond
2167 if (sdp->scsi_level > SCSI_SPC_2)
2168 return 1;
2169 return 0;
2173 * sd_revalidate_disk - called the first time a new disk is seen,
2174 * performs disk spin up, read_capacity, etc.
2175 * @disk: struct gendisk we care about
2177 static int sd_revalidate_disk(struct gendisk *disk)
2179 struct scsi_disk *sdkp = scsi_disk(disk);
2180 struct scsi_device *sdp = sdkp->device;
2181 unsigned char *buffer;
2182 unsigned flush = 0;
2184 SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp,
2185 "sd_revalidate_disk\n"));
2188 * If the device is offline, don't try and read capacity or any
2189 * of the other niceties.
2191 if (!scsi_device_online(sdp))
2192 goto out;
2194 buffer = kmalloc(SD_BUF_SIZE, GFP_KERNEL);
2195 if (!buffer) {
2196 sd_printk(KERN_WARNING, sdkp, "sd_revalidate_disk: Memory "
2197 "allocation failure.\n");
2198 goto out;
2201 sd_spinup_disk(sdkp);
2204 * Without media there is no reason to ask; moreover, some devices
2205 * react badly if we do.
2207 if (sdkp->media_present) {
2208 sd_read_capacity(sdkp, buffer);
2210 if (sd_try_extended_inquiry(sdp)) {
2211 sd_read_thin_provisioning(sdkp);
2212 sd_read_block_limits(sdkp);
2213 sd_read_block_characteristics(sdkp);
2216 sd_read_write_protect_flag(sdkp, buffer);
2217 sd_read_cache_type(sdkp, buffer);
2218 sd_read_app_tag_own(sdkp, buffer);
2221 sdkp->first_scan = 0;
2224 * We now have all cache related info, determine how we deal
2225 * with flush requests.
2227 if (sdkp->WCE) {
2228 flush |= REQ_FLUSH;
2229 if (sdkp->DPOFUA)
2230 flush |= REQ_FUA;
2233 blk_queue_flush(sdkp->disk->queue, flush);
2235 set_capacity(disk, sdkp->capacity);
2236 kfree(buffer);
2238 out:
2239 return 0;
2243 * sd_unlock_native_capacity - unlock native capacity
2244 * @disk: struct gendisk to set capacity for
2246 * Block layer calls this function if it detects that partitions
2247 * on @disk reach beyond the end of the device. If the SCSI host
2248 * implements ->unlock_native_capacity() method, it's invoked to
2249 * give it a chance to adjust the device capacity.
2251 * CONTEXT:
2252 * Defined by block layer. Might sleep.
2254 static void sd_unlock_native_capacity(struct gendisk *disk)
2256 struct scsi_device *sdev = scsi_disk(disk)->device;
2258 if (sdev->host->hostt->unlock_native_capacity)
2259 sdev->host->hostt->unlock_native_capacity(sdev);
2263 * sd_format_disk_name - format disk name
2264 * @prefix: name prefix - ie. "sd" for SCSI disks
2265 * @index: index of the disk to format name for
2266 * @buf: output buffer
2267 * @buflen: length of the output buffer
2269 * SCSI disk names starts at sda. The 26th device is sdz and the
2270 * 27th is sdaa. The last one for two lettered suffix is sdzz
2271 * which is followed by sdaaa.
2273 * This is basically 26 base counting with one extra 'nil' entry
2274 * at the beginning from the second digit on and can be
2275 * determined using similar method as 26 base conversion with the
2276 * index shifted -1 after each digit is computed.
2278 * CONTEXT:
2279 * Don't care.
2281 * RETURNS:
2282 * 0 on success, -errno on failure.
2284 static int sd_format_disk_name(char *prefix, int index, char *buf, int buflen)
2286 const int base = 'z' - 'a' + 1;
2287 char *begin = buf + strlen(prefix);
2288 char *end = buf + buflen;
2289 char *p;
2290 int unit;
2292 p = end - 1;
2293 *p = '\0';
2294 unit = base;
2295 do {
2296 if (p == begin)
2297 return -EINVAL;
2298 *--p = 'a' + (index % unit);
2299 index = (index / unit) - 1;
2300 } while (index >= 0);
2302 memmove(begin, p, end - p);
2303 memcpy(buf, prefix, strlen(prefix));
2305 return 0;
2309 * The asynchronous part of sd_probe
2311 static void sd_probe_async(void *data, async_cookie_t cookie)
2313 struct scsi_disk *sdkp = data;
2314 struct scsi_device *sdp;
2315 struct gendisk *gd;
2316 u32 index;
2317 struct device *dev;
2319 sdp = sdkp->device;
2320 gd = sdkp->disk;
2321 index = sdkp->index;
2322 dev = &sdp->sdev_gendev;
2324 gd->major = sd_major((index & 0xf0) >> 4);
2325 gd->first_minor = ((index & 0xf) << 4) | (index & 0xfff00);
2326 gd->minors = SD_MINORS;
2328 gd->fops = &sd_fops;
2329 gd->private_data = &sdkp->driver;
2330 gd->queue = sdkp->device->request_queue;
2332 /* defaults, until the device tells us otherwise */
2333 sdp->sector_size = 512;
2334 sdkp->capacity = 0;
2335 sdkp->media_present = 1;
2336 sdkp->write_prot = 0;
2337 sdkp->WCE = 0;
2338 sdkp->RCD = 0;
2339 sdkp->ATO = 0;
2340 sdkp->first_scan = 1;
2342 sd_revalidate_disk(gd);
2344 blk_queue_prep_rq(sdp->request_queue, sd_prep_fn);
2345 blk_queue_unprep_rq(sdp->request_queue, sd_unprep_fn);
2347 gd->driverfs_dev = &sdp->sdev_gendev;
2348 gd->flags = GENHD_FL_EXT_DEVT;
2349 if (sdp->removable)
2350 gd->flags |= GENHD_FL_REMOVABLE;
2352 add_disk(gd);
2353 sd_dif_config_host(sdkp);
2355 sd_revalidate_disk(gd);
2357 sd_printk(KERN_NOTICE, sdkp, "Attached SCSI %sdisk\n",
2358 sdp->removable ? "removable " : "");
2359 scsi_autopm_put_device(sdp);
2360 put_device(&sdkp->dev);
2364 * sd_probe - called during driver initialization and whenever a
2365 * new scsi device is attached to the system. It is called once
2366 * for each scsi device (not just disks) present.
2367 * @dev: pointer to device object
2369 * Returns 0 if successful (or not interested in this scsi device
2370 * (e.g. scanner)); 1 when there is an error.
2372 * Note: this function is invoked from the scsi mid-level.
2373 * This function sets up the mapping between a given
2374 * <host,channel,id,lun> (found in sdp) and new device name
2375 * (e.g. /dev/sda). More precisely it is the block device major
2376 * and minor number that is chosen here.
2378 * Assume sd_attach is not re-entrant (for time being)
2379 * Also think about sd_attach() and sd_remove() running coincidentally.
2381 static int sd_probe(struct device *dev)
2383 struct scsi_device *sdp = to_scsi_device(dev);
2384 struct scsi_disk *sdkp;
2385 struct gendisk *gd;
2386 int index;
2387 int error;
2389 error = -ENODEV;
2390 if (sdp->type != TYPE_DISK && sdp->type != TYPE_MOD && sdp->type != TYPE_RBC)
2391 goto out;
2393 SCSI_LOG_HLQUEUE(3, sdev_printk(KERN_INFO, sdp,
2394 "sd_attach\n"));
2396 error = -ENOMEM;
2397 sdkp = kzalloc(sizeof(*sdkp), GFP_KERNEL);
2398 if (!sdkp)
2399 goto out;
2401 gd = alloc_disk(SD_MINORS);
2402 if (!gd)
2403 goto out_free;
2405 do {
2406 if (!ida_pre_get(&sd_index_ida, GFP_KERNEL))
2407 goto out_put;
2409 spin_lock(&sd_index_lock);
2410 error = ida_get_new(&sd_index_ida, &index);
2411 spin_unlock(&sd_index_lock);
2412 } while (error == -EAGAIN);
2414 if (error)
2415 goto out_put;
2417 if (index >= SD_MAX_DISKS) {
2418 error = -ENODEV;
2419 sdev_printk(KERN_WARNING, sdp, "SCSI disk (sd) name space exhausted.\n");
2420 goto out_free_index;
2423 error = sd_format_disk_name("sd", index, gd->disk_name, DISK_NAME_LEN);
2424 if (error)
2425 goto out_free_index;
2427 sdkp->device = sdp;
2428 sdkp->driver = &sd_template;
2429 sdkp->disk = gd;
2430 sdkp->index = index;
2431 atomic_set(&sdkp->openers, 0);
2432 sdkp->previous_state = 1;
2434 if (!sdp->request_queue->rq_timeout) {
2435 if (sdp->type != TYPE_MOD)
2436 blk_queue_rq_timeout(sdp->request_queue, SD_TIMEOUT);
2437 else
2438 blk_queue_rq_timeout(sdp->request_queue,
2439 SD_MOD_TIMEOUT);
2442 device_initialize(&sdkp->dev);
2443 sdkp->dev.parent = dev;
2444 sdkp->dev.class = &sd_disk_class;
2445 dev_set_name(&sdkp->dev, dev_name(dev));
2447 if (device_add(&sdkp->dev))
2448 goto out_free_index;
2450 get_device(dev);
2451 dev_set_drvdata(dev, sdkp);
2453 get_device(&sdkp->dev); /* prevent release before async_schedule */
2454 async_schedule(sd_probe_async, sdkp);
2456 return 0;
2458 out_free_index:
2459 spin_lock(&sd_index_lock);
2460 ida_remove(&sd_index_ida, index);
2461 spin_unlock(&sd_index_lock);
2462 out_put:
2463 put_disk(gd);
2464 out_free:
2465 kfree(sdkp);
2466 out:
2467 return error;
2471 * sd_remove - called whenever a scsi disk (previously recognized by
2472 * sd_probe) is detached from the system. It is called (potentially
2473 * multiple times) during sd module unload.
2474 * @sdp: pointer to mid level scsi device object
2476 * Note: this function is invoked from the scsi mid-level.
2477 * This function potentially frees up a device name (e.g. /dev/sdc)
2478 * that could be re-used by a subsequent sd_probe().
2479 * This function is not called when the built-in sd driver is "exit-ed".
2481 static int sd_remove(struct device *dev)
2483 struct scsi_disk *sdkp;
2485 sdkp = dev_get_drvdata(dev);
2486 scsi_autopm_get_device(sdkp->device);
2488 async_synchronize_full();
2489 blk_queue_prep_rq(sdkp->device->request_queue, scsi_prep_fn);
2490 blk_queue_unprep_rq(sdkp->device->request_queue, NULL);
2491 device_del(&sdkp->dev);
2492 del_gendisk(sdkp->disk);
2493 sd_shutdown(dev);
2495 mutex_lock(&sd_ref_mutex);
2496 dev_set_drvdata(dev, NULL);
2497 put_device(&sdkp->dev);
2498 mutex_unlock(&sd_ref_mutex);
2500 return 0;
2504 * scsi_disk_release - Called to free the scsi_disk structure
2505 * @dev: pointer to embedded class device
2507 * sd_ref_mutex must be held entering this routine. Because it is
2508 * called on last put, you should always use the scsi_disk_get()
2509 * scsi_disk_put() helpers which manipulate the semaphore directly
2510 * and never do a direct put_device.
2512 static void scsi_disk_release(struct device *dev)
2514 struct scsi_disk *sdkp = to_scsi_disk(dev);
2515 struct gendisk *disk = sdkp->disk;
2517 spin_lock(&sd_index_lock);
2518 ida_remove(&sd_index_ida, sdkp->index);
2519 spin_unlock(&sd_index_lock);
2521 disk->private_data = NULL;
2522 put_disk(disk);
2523 put_device(&sdkp->device->sdev_gendev);
2525 kfree(sdkp);
2528 static int sd_start_stop_device(struct scsi_disk *sdkp, int start)
2530 unsigned char cmd[6] = { START_STOP }; /* START_VALID */
2531 struct scsi_sense_hdr sshdr;
2532 struct scsi_device *sdp = sdkp->device;
2533 int res;
2535 if (start)
2536 cmd[4] |= 1; /* START */
2538 if (sdp->start_stop_pwr_cond)
2539 cmd[4] |= start ? 1 << 4 : 3 << 4; /* Active or Standby */
2541 if (!scsi_device_online(sdp))
2542 return -ENODEV;
2544 res = scsi_execute_req(sdp, cmd, DMA_NONE, NULL, 0, &sshdr,
2545 SD_TIMEOUT, SD_MAX_RETRIES, NULL);
2546 if (res) {
2547 sd_printk(KERN_WARNING, sdkp, "START_STOP FAILED\n");
2548 sd_print_result(sdkp, res);
2549 if (driver_byte(res) & DRIVER_SENSE)
2550 sd_print_sense_hdr(sdkp, &sshdr);
2553 return res;
2557 * Send a SYNCHRONIZE CACHE instruction down to the device through
2558 * the normal SCSI command structure. Wait for the command to
2559 * complete.
2561 static void sd_shutdown(struct device *dev)
2563 struct scsi_disk *sdkp = scsi_disk_get_from_dev(dev);
2565 if (!sdkp)
2566 return; /* this can happen */
2568 if (sdkp->WCE) {
2569 sd_printk(KERN_NOTICE, sdkp, "Synchronizing SCSI cache\n");
2570 sd_sync_cache(sdkp);
2573 if (system_state != SYSTEM_RESTART && sdkp->device->manage_start_stop) {
2574 sd_printk(KERN_NOTICE, sdkp, "Stopping disk\n");
2575 sd_start_stop_device(sdkp, 0);
2578 scsi_disk_put(sdkp);
2581 static int sd_suspend(struct device *dev, pm_message_t mesg)
2583 struct scsi_disk *sdkp = scsi_disk_get_from_dev(dev);
2584 int ret = 0;
2586 if (!sdkp)
2587 return 0; /* this can happen */
2589 if (sdkp->WCE) {
2590 sd_printk(KERN_NOTICE, sdkp, "Synchronizing SCSI cache\n");
2591 ret = sd_sync_cache(sdkp);
2592 if (ret)
2593 goto done;
2596 if ((mesg.event & PM_EVENT_SLEEP) && sdkp->device->manage_start_stop) {
2597 sd_printk(KERN_NOTICE, sdkp, "Stopping disk\n");
2598 ret = sd_start_stop_device(sdkp, 0);
2601 done:
2602 scsi_disk_put(sdkp);
2603 return ret;
2606 static int sd_resume(struct device *dev)
2608 struct scsi_disk *sdkp = scsi_disk_get_from_dev(dev);
2609 int ret = 0;
2611 if (!sdkp->device->manage_start_stop)
2612 goto done;
2614 sd_printk(KERN_NOTICE, sdkp, "Starting disk\n");
2615 ret = sd_start_stop_device(sdkp, 1);
2617 done:
2618 scsi_disk_put(sdkp);
2619 return ret;
2623 * init_sd - entry point for this driver (both when built in or when
2624 * a module).
2626 * Note: this function registers this driver with the scsi mid-level.
2628 static int __init init_sd(void)
2630 int majors = 0, i, err;
2632 SCSI_LOG_HLQUEUE(3, printk("init_sd: sd driver entry point\n"));
2634 for (i = 0; i < SD_MAJORS; i++)
2635 if (register_blkdev(sd_major(i), "sd") == 0)
2636 majors++;
2638 if (!majors)
2639 return -ENODEV;
2641 err = class_register(&sd_disk_class);
2642 if (err)
2643 goto err_out;
2645 err = scsi_register_driver(&sd_template.gendrv);
2646 if (err)
2647 goto err_out_class;
2649 sd_cdb_cache = kmem_cache_create("sd_ext_cdb", SD_EXT_CDB_SIZE,
2650 0, 0, NULL);
2651 if (!sd_cdb_cache) {
2652 printk(KERN_ERR "sd: can't init extended cdb cache\n");
2653 goto err_out_class;
2656 sd_cdb_pool = mempool_create_slab_pool(SD_MEMPOOL_SIZE, sd_cdb_cache);
2657 if (!sd_cdb_pool) {
2658 printk(KERN_ERR "sd: can't init extended cdb pool\n");
2659 goto err_out_cache;
2662 return 0;
2664 err_out_cache:
2665 kmem_cache_destroy(sd_cdb_cache);
2667 err_out_class:
2668 class_unregister(&sd_disk_class);
2669 err_out:
2670 for (i = 0; i < SD_MAJORS; i++)
2671 unregister_blkdev(sd_major(i), "sd");
2672 return err;
2676 * exit_sd - exit point for this driver (when it is a module).
2678 * Note: this function unregisters this driver from the scsi mid-level.
2680 static void __exit exit_sd(void)
2682 int i;
2684 SCSI_LOG_HLQUEUE(3, printk("exit_sd: exiting sd driver\n"));
2686 mempool_destroy(sd_cdb_pool);
2687 kmem_cache_destroy(sd_cdb_cache);
2689 scsi_unregister_driver(&sd_template.gendrv);
2690 class_unregister(&sd_disk_class);
2692 for (i = 0; i < SD_MAJORS; i++)
2693 unregister_blkdev(sd_major(i), "sd");
2696 module_init(init_sd);
2697 module_exit(exit_sd);
2699 static void sd_print_sense_hdr(struct scsi_disk *sdkp,
2700 struct scsi_sense_hdr *sshdr)
2702 sd_printk(KERN_INFO, sdkp, " ");
2703 scsi_show_sense_hdr(sshdr);
2704 sd_printk(KERN_INFO, sdkp, " ");
2705 scsi_show_extd_sense(sshdr->asc, sshdr->ascq);
2708 static void sd_print_result(struct scsi_disk *sdkp, int result)
2710 sd_printk(KERN_INFO, sdkp, " ");
2711 scsi_show_result(result);