Loading...
1/*
2 * sd.c Copyright (C) 1992 Drew Eckhardt
3 * Copyright (C) 1993, 1994, 1995, 1999 Eric Youngdale
4 *
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.
25 *
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.
33 */
34
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 <linux/pr.h>
55#include <linux/t10-pi.h>
56#include <linux/uaccess.h>
57#include <asm/unaligned.h>
58
59#include <scsi/scsi.h>
60#include <scsi/scsi_cmnd.h>
61#include <scsi/scsi_dbg.h>
62#include <scsi/scsi_device.h>
63#include <scsi/scsi_driver.h>
64#include <scsi/scsi_eh.h>
65#include <scsi/scsi_host.h>
66#include <scsi/scsi_ioctl.h>
67#include <scsi/scsicam.h>
68
69#include "sd.h"
70#include "scsi_priv.h"
71#include "scsi_logging.h"
72
73MODULE_AUTHOR("Eric Youngdale");
74MODULE_DESCRIPTION("SCSI disk (sd) driver");
75MODULE_LICENSE("GPL");
76
77MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK0_MAJOR);
78MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK1_MAJOR);
79MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK2_MAJOR);
80MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK3_MAJOR);
81MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK4_MAJOR);
82MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK5_MAJOR);
83MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK6_MAJOR);
84MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK7_MAJOR);
85MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK8_MAJOR);
86MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK9_MAJOR);
87MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK10_MAJOR);
88MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK11_MAJOR);
89MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK12_MAJOR);
90MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK13_MAJOR);
91MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK14_MAJOR);
92MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK15_MAJOR);
93MODULE_ALIAS_SCSI_DEVICE(TYPE_DISK);
94MODULE_ALIAS_SCSI_DEVICE(TYPE_MOD);
95MODULE_ALIAS_SCSI_DEVICE(TYPE_RBC);
96MODULE_ALIAS_SCSI_DEVICE(TYPE_ZBC);
97
98#if !defined(CONFIG_DEBUG_BLOCK_EXT_DEVT)
99#define SD_MINORS 16
100#else
101#define SD_MINORS 0
102#endif
103
104static void sd_config_discard(struct scsi_disk *, unsigned int);
105static void sd_config_write_same(struct scsi_disk *);
106static int sd_revalidate_disk(struct gendisk *);
107static void sd_unlock_native_capacity(struct gendisk *disk);
108static int sd_probe(struct device *);
109static int sd_remove(struct device *);
110static void sd_shutdown(struct device *);
111static int sd_suspend_system(struct device *);
112static int sd_suspend_runtime(struct device *);
113static int sd_resume(struct device *);
114static void sd_rescan(struct device *);
115static int sd_init_command(struct scsi_cmnd *SCpnt);
116static void sd_uninit_command(struct scsi_cmnd *SCpnt);
117static int sd_done(struct scsi_cmnd *);
118static int sd_eh_action(struct scsi_cmnd *, int);
119static void sd_read_capacity(struct scsi_disk *sdkp, unsigned char *buffer);
120static void scsi_disk_release(struct device *cdev);
121static void sd_print_sense_hdr(struct scsi_disk *, struct scsi_sense_hdr *);
122static void sd_print_result(const struct scsi_disk *, const char *, int);
123
124static DEFINE_SPINLOCK(sd_index_lock);
125static DEFINE_IDA(sd_index_ida);
126
127/* This semaphore is used to mediate the 0->1 reference get in the
128 * face of object destruction (i.e. we can't allow a get on an
129 * object after last put) */
130static DEFINE_MUTEX(sd_ref_mutex);
131
132static struct kmem_cache *sd_cdb_cache;
133static mempool_t *sd_cdb_pool;
134
135static const char *sd_cache_types[] = {
136 "write through", "none", "write back",
137 "write back, no read (daft)"
138};
139
140static void sd_set_flush_flag(struct scsi_disk *sdkp)
141{
142 bool wc = false, fua = false;
143
144 if (sdkp->WCE) {
145 wc = true;
146 if (sdkp->DPOFUA)
147 fua = true;
148 }
149
150 blk_queue_write_cache(sdkp->disk->queue, wc, fua);
151}
152
153static ssize_t
154cache_type_store(struct device *dev, struct device_attribute *attr,
155 const char *buf, size_t count)
156{
157 int i, ct = -1, rcd, wce, sp;
158 struct scsi_disk *sdkp = to_scsi_disk(dev);
159 struct scsi_device *sdp = sdkp->device;
160 char buffer[64];
161 char *buffer_data;
162 struct scsi_mode_data data;
163 struct scsi_sense_hdr sshdr;
164 static const char temp[] = "temporary ";
165 int len;
166
167 if (sdp->type != TYPE_DISK && sdp->type != TYPE_ZBC)
168 /* no cache control on RBC devices; theoretically they
169 * can do it, but there's probably so many exceptions
170 * it's not worth the risk */
171 return -EINVAL;
172
173 if (strncmp(buf, temp, sizeof(temp) - 1) == 0) {
174 buf += sizeof(temp) - 1;
175 sdkp->cache_override = 1;
176 } else {
177 sdkp->cache_override = 0;
178 }
179
180 for (i = 0; i < ARRAY_SIZE(sd_cache_types); i++) {
181 len = strlen(sd_cache_types[i]);
182 if (strncmp(sd_cache_types[i], buf, len) == 0 &&
183 buf[len] == '\n') {
184 ct = i;
185 break;
186 }
187 }
188 if (ct < 0)
189 return -EINVAL;
190 rcd = ct & 0x01 ? 1 : 0;
191 wce = (ct & 0x02) && !sdkp->write_prot ? 1 : 0;
192
193 if (sdkp->cache_override) {
194 sdkp->WCE = wce;
195 sdkp->RCD = rcd;
196 sd_set_flush_flag(sdkp);
197 return count;
198 }
199
200 if (scsi_mode_sense(sdp, 0x08, 8, buffer, sizeof(buffer), SD_TIMEOUT,
201 SD_MAX_RETRIES, &data, NULL))
202 return -EINVAL;
203 len = min_t(size_t, sizeof(buffer), data.length - data.header_length -
204 data.block_descriptor_length);
205 buffer_data = buffer + data.header_length +
206 data.block_descriptor_length;
207 buffer_data[2] &= ~0x05;
208 buffer_data[2] |= wce << 2 | rcd;
209 sp = buffer_data[0] & 0x80 ? 1 : 0;
210 buffer_data[0] &= ~0x80;
211
212 if (scsi_mode_select(sdp, 1, sp, 8, buffer_data, len, SD_TIMEOUT,
213 SD_MAX_RETRIES, &data, &sshdr)) {
214 if (scsi_sense_valid(&sshdr))
215 sd_print_sense_hdr(sdkp, &sshdr);
216 return -EINVAL;
217 }
218 revalidate_disk(sdkp->disk);
219 return count;
220}
221
222static ssize_t
223manage_start_stop_show(struct device *dev, struct device_attribute *attr,
224 char *buf)
225{
226 struct scsi_disk *sdkp = to_scsi_disk(dev);
227 struct scsi_device *sdp = sdkp->device;
228
229 return snprintf(buf, 20, "%u\n", sdp->manage_start_stop);
230}
231
232static ssize_t
233manage_start_stop_store(struct device *dev, struct device_attribute *attr,
234 const char *buf, size_t count)
235{
236 struct scsi_disk *sdkp = to_scsi_disk(dev);
237 struct scsi_device *sdp = sdkp->device;
238
239 if (!capable(CAP_SYS_ADMIN))
240 return -EACCES;
241
242 sdp->manage_start_stop = simple_strtoul(buf, NULL, 10);
243
244 return count;
245}
246static DEVICE_ATTR_RW(manage_start_stop);
247
248static ssize_t
249allow_restart_show(struct device *dev, struct device_attribute *attr, char *buf)
250{
251 struct scsi_disk *sdkp = to_scsi_disk(dev);
252
253 return snprintf(buf, 40, "%d\n", sdkp->device->allow_restart);
254}
255
256static ssize_t
257allow_restart_store(struct device *dev, struct device_attribute *attr,
258 const char *buf, size_t count)
259{
260 struct scsi_disk *sdkp = to_scsi_disk(dev);
261 struct scsi_device *sdp = sdkp->device;
262
263 if (!capable(CAP_SYS_ADMIN))
264 return -EACCES;
265
266 if (sdp->type != TYPE_DISK && sdp->type != TYPE_ZBC)
267 return -EINVAL;
268
269 sdp->allow_restart = simple_strtoul(buf, NULL, 10);
270
271 return count;
272}
273static DEVICE_ATTR_RW(allow_restart);
274
275static ssize_t
276cache_type_show(struct device *dev, struct device_attribute *attr, char *buf)
277{
278 struct scsi_disk *sdkp = to_scsi_disk(dev);
279 int ct = sdkp->RCD + 2*sdkp->WCE;
280
281 return snprintf(buf, 40, "%s\n", sd_cache_types[ct]);
282}
283static DEVICE_ATTR_RW(cache_type);
284
285static ssize_t
286FUA_show(struct device *dev, struct device_attribute *attr, char *buf)
287{
288 struct scsi_disk *sdkp = to_scsi_disk(dev);
289
290 return snprintf(buf, 20, "%u\n", sdkp->DPOFUA);
291}
292static DEVICE_ATTR_RO(FUA);
293
294static ssize_t
295protection_type_show(struct device *dev, struct device_attribute *attr,
296 char *buf)
297{
298 struct scsi_disk *sdkp = to_scsi_disk(dev);
299
300 return snprintf(buf, 20, "%u\n", sdkp->protection_type);
301}
302
303static ssize_t
304protection_type_store(struct device *dev, struct device_attribute *attr,
305 const char *buf, size_t count)
306{
307 struct scsi_disk *sdkp = to_scsi_disk(dev);
308 unsigned int val;
309 int err;
310
311 if (!capable(CAP_SYS_ADMIN))
312 return -EACCES;
313
314 err = kstrtouint(buf, 10, &val);
315
316 if (err)
317 return err;
318
319 if (val >= 0 && val <= T10_PI_TYPE3_PROTECTION)
320 sdkp->protection_type = val;
321
322 return count;
323}
324static DEVICE_ATTR_RW(protection_type);
325
326static ssize_t
327protection_mode_show(struct device *dev, struct device_attribute *attr,
328 char *buf)
329{
330 struct scsi_disk *sdkp = to_scsi_disk(dev);
331 struct scsi_device *sdp = sdkp->device;
332 unsigned int dif, dix;
333
334 dif = scsi_host_dif_capable(sdp->host, sdkp->protection_type);
335 dix = scsi_host_dix_capable(sdp->host, sdkp->protection_type);
336
337 if (!dix && scsi_host_dix_capable(sdp->host, T10_PI_TYPE0_PROTECTION)) {
338 dif = 0;
339 dix = 1;
340 }
341
342 if (!dif && !dix)
343 return snprintf(buf, 20, "none\n");
344
345 return snprintf(buf, 20, "%s%u\n", dix ? "dix" : "dif", dif);
346}
347static DEVICE_ATTR_RO(protection_mode);
348
349static ssize_t
350app_tag_own_show(struct device *dev, struct device_attribute *attr, char *buf)
351{
352 struct scsi_disk *sdkp = to_scsi_disk(dev);
353
354 return snprintf(buf, 20, "%u\n", sdkp->ATO);
355}
356static DEVICE_ATTR_RO(app_tag_own);
357
358static ssize_t
359thin_provisioning_show(struct device *dev, struct device_attribute *attr,
360 char *buf)
361{
362 struct scsi_disk *sdkp = to_scsi_disk(dev);
363
364 return snprintf(buf, 20, "%u\n", sdkp->lbpme);
365}
366static DEVICE_ATTR_RO(thin_provisioning);
367
368static const char *lbp_mode[] = {
369 [SD_LBP_FULL] = "full",
370 [SD_LBP_UNMAP] = "unmap",
371 [SD_LBP_WS16] = "writesame_16",
372 [SD_LBP_WS10] = "writesame_10",
373 [SD_LBP_ZERO] = "writesame_zero",
374 [SD_LBP_DISABLE] = "disabled",
375};
376
377static ssize_t
378provisioning_mode_show(struct device *dev, struct device_attribute *attr,
379 char *buf)
380{
381 struct scsi_disk *sdkp = to_scsi_disk(dev);
382
383 return snprintf(buf, 20, "%s\n", lbp_mode[sdkp->provisioning_mode]);
384}
385
386static ssize_t
387provisioning_mode_store(struct device *dev, struct device_attribute *attr,
388 const char *buf, size_t count)
389{
390 struct scsi_disk *sdkp = to_scsi_disk(dev);
391 struct scsi_device *sdp = sdkp->device;
392
393 if (!capable(CAP_SYS_ADMIN))
394 return -EACCES;
395
396 if (sd_is_zoned(sdkp)) {
397 sd_config_discard(sdkp, SD_LBP_DISABLE);
398 return count;
399 }
400
401 if (sdp->type != TYPE_DISK)
402 return -EINVAL;
403
404 if (!strncmp(buf, lbp_mode[SD_LBP_UNMAP], 20))
405 sd_config_discard(sdkp, SD_LBP_UNMAP);
406 else if (!strncmp(buf, lbp_mode[SD_LBP_WS16], 20))
407 sd_config_discard(sdkp, SD_LBP_WS16);
408 else if (!strncmp(buf, lbp_mode[SD_LBP_WS10], 20))
409 sd_config_discard(sdkp, SD_LBP_WS10);
410 else if (!strncmp(buf, lbp_mode[SD_LBP_ZERO], 20))
411 sd_config_discard(sdkp, SD_LBP_ZERO);
412 else if (!strncmp(buf, lbp_mode[SD_LBP_DISABLE], 20))
413 sd_config_discard(sdkp, SD_LBP_DISABLE);
414 else
415 return -EINVAL;
416
417 return count;
418}
419static DEVICE_ATTR_RW(provisioning_mode);
420
421static ssize_t
422max_medium_access_timeouts_show(struct device *dev,
423 struct device_attribute *attr, char *buf)
424{
425 struct scsi_disk *sdkp = to_scsi_disk(dev);
426
427 return snprintf(buf, 20, "%u\n", sdkp->max_medium_access_timeouts);
428}
429
430static ssize_t
431max_medium_access_timeouts_store(struct device *dev,
432 struct device_attribute *attr, const char *buf,
433 size_t count)
434{
435 struct scsi_disk *sdkp = to_scsi_disk(dev);
436 int err;
437
438 if (!capable(CAP_SYS_ADMIN))
439 return -EACCES;
440
441 err = kstrtouint(buf, 10, &sdkp->max_medium_access_timeouts);
442
443 return err ? err : count;
444}
445static DEVICE_ATTR_RW(max_medium_access_timeouts);
446
447static ssize_t
448max_write_same_blocks_show(struct device *dev, struct device_attribute *attr,
449 char *buf)
450{
451 struct scsi_disk *sdkp = to_scsi_disk(dev);
452
453 return snprintf(buf, 20, "%u\n", sdkp->max_ws_blocks);
454}
455
456static ssize_t
457max_write_same_blocks_store(struct device *dev, struct device_attribute *attr,
458 const char *buf, size_t count)
459{
460 struct scsi_disk *sdkp = to_scsi_disk(dev);
461 struct scsi_device *sdp = sdkp->device;
462 unsigned long max;
463 int err;
464
465 if (!capable(CAP_SYS_ADMIN))
466 return -EACCES;
467
468 if (sdp->type != TYPE_DISK && sdp->type != TYPE_ZBC)
469 return -EINVAL;
470
471 err = kstrtoul(buf, 10, &max);
472
473 if (err)
474 return err;
475
476 if (max == 0)
477 sdp->no_write_same = 1;
478 else if (max <= SD_MAX_WS16_BLOCKS) {
479 sdp->no_write_same = 0;
480 sdkp->max_ws_blocks = max;
481 }
482
483 sd_config_write_same(sdkp);
484
485 return count;
486}
487static DEVICE_ATTR_RW(max_write_same_blocks);
488
489static struct attribute *sd_disk_attrs[] = {
490 &dev_attr_cache_type.attr,
491 &dev_attr_FUA.attr,
492 &dev_attr_allow_restart.attr,
493 &dev_attr_manage_start_stop.attr,
494 &dev_attr_protection_type.attr,
495 &dev_attr_protection_mode.attr,
496 &dev_attr_app_tag_own.attr,
497 &dev_attr_thin_provisioning.attr,
498 &dev_attr_provisioning_mode.attr,
499 &dev_attr_max_write_same_blocks.attr,
500 &dev_attr_max_medium_access_timeouts.attr,
501 NULL,
502};
503ATTRIBUTE_GROUPS(sd_disk);
504
505static struct class sd_disk_class = {
506 .name = "scsi_disk",
507 .owner = THIS_MODULE,
508 .dev_release = scsi_disk_release,
509 .dev_groups = sd_disk_groups,
510};
511
512static const struct dev_pm_ops sd_pm_ops = {
513 .suspend = sd_suspend_system,
514 .resume = sd_resume,
515 .poweroff = sd_suspend_system,
516 .restore = sd_resume,
517 .runtime_suspend = sd_suspend_runtime,
518 .runtime_resume = sd_resume,
519};
520
521static struct scsi_driver sd_template = {
522 .gendrv = {
523 .name = "sd",
524 .owner = THIS_MODULE,
525 .probe = sd_probe,
526 .remove = sd_remove,
527 .shutdown = sd_shutdown,
528 .pm = &sd_pm_ops,
529 },
530 .rescan = sd_rescan,
531 .init_command = sd_init_command,
532 .uninit_command = sd_uninit_command,
533 .done = sd_done,
534 .eh_action = sd_eh_action,
535};
536
537/*
538 * Dummy kobj_map->probe function.
539 * The default ->probe function will call modprobe, which is
540 * pointless as this module is already loaded.
541 */
542static struct kobject *sd_default_probe(dev_t devt, int *partno, void *data)
543{
544 return NULL;
545}
546
547/*
548 * Device no to disk mapping:
549 *
550 * major disc2 disc p1
551 * |............|.............|....|....| <- dev_t
552 * 31 20 19 8 7 4 3 0
553 *
554 * Inside a major, we have 16k disks, however mapped non-
555 * contiguously. The first 16 disks are for major0, the next
556 * ones with major1, ... Disk 256 is for major0 again, disk 272
557 * for major1, ...
558 * As we stay compatible with our numbering scheme, we can reuse
559 * the well-know SCSI majors 8, 65--71, 136--143.
560 */
561static int sd_major(int major_idx)
562{
563 switch (major_idx) {
564 case 0:
565 return SCSI_DISK0_MAJOR;
566 case 1 ... 7:
567 return SCSI_DISK1_MAJOR + major_idx - 1;
568 case 8 ... 15:
569 return SCSI_DISK8_MAJOR + major_idx - 8;
570 default:
571 BUG();
572 return 0; /* shut up gcc */
573 }
574}
575
576static struct scsi_disk *scsi_disk_get(struct gendisk *disk)
577{
578 struct scsi_disk *sdkp = NULL;
579
580 mutex_lock(&sd_ref_mutex);
581
582 if (disk->private_data) {
583 sdkp = scsi_disk(disk);
584 if (scsi_device_get(sdkp->device) == 0)
585 get_device(&sdkp->dev);
586 else
587 sdkp = NULL;
588 }
589 mutex_unlock(&sd_ref_mutex);
590 return sdkp;
591}
592
593static void scsi_disk_put(struct scsi_disk *sdkp)
594{
595 struct scsi_device *sdev = sdkp->device;
596
597 mutex_lock(&sd_ref_mutex);
598 put_device(&sdkp->dev);
599 scsi_device_put(sdev);
600 mutex_unlock(&sd_ref_mutex);
601}
602
603static unsigned char sd_setup_protect_cmnd(struct scsi_cmnd *scmd,
604 unsigned int dix, unsigned int dif)
605{
606 struct bio *bio = scmd->request->bio;
607 unsigned int prot_op = sd_prot_op(rq_data_dir(scmd->request), dix, dif);
608 unsigned int protect = 0;
609
610 if (dix) { /* DIX Type 0, 1, 2, 3 */
611 if (bio_integrity_flagged(bio, BIP_IP_CHECKSUM))
612 scmd->prot_flags |= SCSI_PROT_IP_CHECKSUM;
613
614 if (bio_integrity_flagged(bio, BIP_CTRL_NOCHECK) == false)
615 scmd->prot_flags |= SCSI_PROT_GUARD_CHECK;
616 }
617
618 if (dif != T10_PI_TYPE3_PROTECTION) { /* DIX/DIF Type 0, 1, 2 */
619 scmd->prot_flags |= SCSI_PROT_REF_INCREMENT;
620
621 if (bio_integrity_flagged(bio, BIP_CTRL_NOCHECK) == false)
622 scmd->prot_flags |= SCSI_PROT_REF_CHECK;
623 }
624
625 if (dif) { /* DIX/DIF Type 1, 2, 3 */
626 scmd->prot_flags |= SCSI_PROT_TRANSFER_PI;
627
628 if (bio_integrity_flagged(bio, BIP_DISK_NOCHECK))
629 protect = 3 << 5; /* Disable target PI checking */
630 else
631 protect = 1 << 5; /* Enable target PI checking */
632 }
633
634 scsi_set_prot_op(scmd, prot_op);
635 scsi_set_prot_type(scmd, dif);
636 scmd->prot_flags &= sd_prot_flag_mask(prot_op);
637
638 return protect;
639}
640
641static void sd_config_discard(struct scsi_disk *sdkp, unsigned int mode)
642{
643 struct request_queue *q = sdkp->disk->queue;
644 unsigned int logical_block_size = sdkp->device->sector_size;
645 unsigned int max_blocks = 0;
646
647 q->limits.discard_zeroes_data = 0;
648
649 /*
650 * When LBPRZ is reported, discard alignment and granularity
651 * must be fixed to the logical block size. Otherwise the block
652 * layer will drop misaligned portions of the request which can
653 * lead to data corruption. If LBPRZ is not set, we honor the
654 * device preference.
655 */
656 if (sdkp->lbprz) {
657 q->limits.discard_alignment = 0;
658 q->limits.discard_granularity = logical_block_size;
659 } else {
660 q->limits.discard_alignment = sdkp->unmap_alignment *
661 logical_block_size;
662 q->limits.discard_granularity =
663 max(sdkp->physical_block_size,
664 sdkp->unmap_granularity * logical_block_size);
665 }
666
667 sdkp->provisioning_mode = mode;
668
669 switch (mode) {
670
671 case SD_LBP_DISABLE:
672 blk_queue_max_discard_sectors(q, 0);
673 queue_flag_clear_unlocked(QUEUE_FLAG_DISCARD, q);
674 return;
675
676 case SD_LBP_UNMAP:
677 max_blocks = min_not_zero(sdkp->max_unmap_blocks,
678 (u32)SD_MAX_WS16_BLOCKS);
679 break;
680
681 case SD_LBP_WS16:
682 max_blocks = min_not_zero(sdkp->max_ws_blocks,
683 (u32)SD_MAX_WS16_BLOCKS);
684 q->limits.discard_zeroes_data = sdkp->lbprz;
685 break;
686
687 case SD_LBP_WS10:
688 max_blocks = min_not_zero(sdkp->max_ws_blocks,
689 (u32)SD_MAX_WS10_BLOCKS);
690 q->limits.discard_zeroes_data = sdkp->lbprz;
691 break;
692
693 case SD_LBP_ZERO:
694 max_blocks = min_not_zero(sdkp->max_ws_blocks,
695 (u32)SD_MAX_WS10_BLOCKS);
696 q->limits.discard_zeroes_data = 1;
697 break;
698 }
699
700 blk_queue_max_discard_sectors(q, max_blocks * (logical_block_size >> 9));
701 queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, q);
702}
703
704/**
705 * sd_setup_discard_cmnd - unmap blocks on thinly provisioned device
706 * @sdp: scsi device to operate one
707 * @rq: Request to prepare
708 *
709 * Will issue either UNMAP or WRITE SAME(16) depending on preference
710 * indicated by target device.
711 **/
712static int sd_setup_discard_cmnd(struct scsi_cmnd *cmd)
713{
714 struct request *rq = cmd->request;
715 struct scsi_device *sdp = cmd->device;
716 struct scsi_disk *sdkp = scsi_disk(rq->rq_disk);
717 sector_t sector = blk_rq_pos(rq);
718 unsigned int nr_sectors = blk_rq_sectors(rq);
719 unsigned int len;
720 int ret;
721 char *buf;
722 struct page *page;
723
724 sector >>= ilog2(sdp->sector_size) - 9;
725 nr_sectors >>= ilog2(sdp->sector_size) - 9;
726
727 page = alloc_page(GFP_ATOMIC | __GFP_ZERO);
728 if (!page)
729 return BLKPREP_DEFER;
730
731 switch (sdkp->provisioning_mode) {
732 case SD_LBP_UNMAP:
733 buf = page_address(page);
734
735 cmd->cmd_len = 10;
736 cmd->cmnd[0] = UNMAP;
737 cmd->cmnd[8] = 24;
738
739 put_unaligned_be16(6 + 16, &buf[0]);
740 put_unaligned_be16(16, &buf[2]);
741 put_unaligned_be64(sector, &buf[8]);
742 put_unaligned_be32(nr_sectors, &buf[16]);
743
744 len = 24;
745 break;
746
747 case SD_LBP_WS16:
748 cmd->cmd_len = 16;
749 cmd->cmnd[0] = WRITE_SAME_16;
750 cmd->cmnd[1] = 0x8; /* UNMAP */
751 put_unaligned_be64(sector, &cmd->cmnd[2]);
752 put_unaligned_be32(nr_sectors, &cmd->cmnd[10]);
753
754 len = sdkp->device->sector_size;
755 break;
756
757 case SD_LBP_WS10:
758 case SD_LBP_ZERO:
759 cmd->cmd_len = 10;
760 cmd->cmnd[0] = WRITE_SAME;
761 if (sdkp->provisioning_mode == SD_LBP_WS10)
762 cmd->cmnd[1] = 0x8; /* UNMAP */
763 put_unaligned_be32(sector, &cmd->cmnd[2]);
764 put_unaligned_be16(nr_sectors, &cmd->cmnd[7]);
765
766 len = sdkp->device->sector_size;
767 break;
768
769 default:
770 ret = BLKPREP_INVALID;
771 goto out;
772 }
773
774 rq->timeout = SD_TIMEOUT;
775
776 cmd->transfersize = len;
777 cmd->allowed = SD_MAX_RETRIES;
778
779 rq->special_vec.bv_page = page;
780 rq->special_vec.bv_offset = 0;
781 rq->special_vec.bv_len = len;
782
783 rq->rq_flags |= RQF_SPECIAL_PAYLOAD;
784 rq->resid_len = len;
785
786 ret = scsi_init_io(cmd);
787out:
788 if (ret != BLKPREP_OK)
789 __free_page(page);
790 return ret;
791}
792
793static void sd_config_write_same(struct scsi_disk *sdkp)
794{
795 struct request_queue *q = sdkp->disk->queue;
796 unsigned int logical_block_size = sdkp->device->sector_size;
797
798 if (sdkp->device->no_write_same) {
799 sdkp->max_ws_blocks = 0;
800 goto out;
801 }
802
803 /* Some devices can not handle block counts above 0xffff despite
804 * supporting WRITE SAME(16). Consequently we default to 64k
805 * blocks per I/O unless the device explicitly advertises a
806 * bigger limit.
807 */
808 if (sdkp->max_ws_blocks > SD_MAX_WS10_BLOCKS)
809 sdkp->max_ws_blocks = min_not_zero(sdkp->max_ws_blocks,
810 (u32)SD_MAX_WS16_BLOCKS);
811 else if (sdkp->ws16 || sdkp->ws10 || sdkp->device->no_report_opcodes)
812 sdkp->max_ws_blocks = min_not_zero(sdkp->max_ws_blocks,
813 (u32)SD_MAX_WS10_BLOCKS);
814 else {
815 sdkp->device->no_write_same = 1;
816 sdkp->max_ws_blocks = 0;
817 }
818
819out:
820 blk_queue_max_write_same_sectors(q, sdkp->max_ws_blocks *
821 (logical_block_size >> 9));
822}
823
824/**
825 * sd_setup_write_same_cmnd - write the same data to multiple blocks
826 * @cmd: command to prepare
827 *
828 * Will issue either WRITE SAME(10) or WRITE SAME(16) depending on
829 * preference indicated by target device.
830 **/
831static int sd_setup_write_same_cmnd(struct scsi_cmnd *cmd)
832{
833 struct request *rq = cmd->request;
834 struct scsi_device *sdp = cmd->device;
835 struct scsi_disk *sdkp = scsi_disk(rq->rq_disk);
836 struct bio *bio = rq->bio;
837 sector_t sector = blk_rq_pos(rq);
838 unsigned int nr_sectors = blk_rq_sectors(rq);
839 unsigned int nr_bytes = blk_rq_bytes(rq);
840 int ret;
841
842 if (sdkp->device->no_write_same)
843 return BLKPREP_INVALID;
844
845 BUG_ON(bio_offset(bio) || bio_iovec(bio).bv_len != sdp->sector_size);
846
847 if (sd_is_zoned(sdkp)) {
848 ret = sd_zbc_setup_write_cmnd(cmd);
849 if (ret != BLKPREP_OK)
850 return ret;
851 }
852
853 sector >>= ilog2(sdp->sector_size) - 9;
854 nr_sectors >>= ilog2(sdp->sector_size) - 9;
855
856 rq->timeout = SD_WRITE_SAME_TIMEOUT;
857
858 if (sdkp->ws16 || sector > 0xffffffff || nr_sectors > 0xffff) {
859 cmd->cmd_len = 16;
860 cmd->cmnd[0] = WRITE_SAME_16;
861 put_unaligned_be64(sector, &cmd->cmnd[2]);
862 put_unaligned_be32(nr_sectors, &cmd->cmnd[10]);
863 } else {
864 cmd->cmd_len = 10;
865 cmd->cmnd[0] = WRITE_SAME;
866 put_unaligned_be32(sector, &cmd->cmnd[2]);
867 put_unaligned_be16(nr_sectors, &cmd->cmnd[7]);
868 }
869
870 cmd->transfersize = sdp->sector_size;
871 cmd->allowed = SD_MAX_RETRIES;
872
873 /*
874 * For WRITE SAME the data transferred via the DATA OUT buffer is
875 * different from the amount of data actually written to the target.
876 *
877 * We set up __data_len to the amount of data transferred via the
878 * DATA OUT buffer so that blk_rq_map_sg sets up the proper S/G list
879 * to transfer a single sector of data first, but then reset it to
880 * the amount of data to be written right after so that the I/O path
881 * knows how much to actually write.
882 */
883 rq->__data_len = sdp->sector_size;
884 ret = scsi_init_io(cmd);
885 rq->__data_len = nr_bytes;
886 return ret;
887}
888
889static int sd_setup_flush_cmnd(struct scsi_cmnd *cmd)
890{
891 struct request *rq = cmd->request;
892
893 /* flush requests don't perform I/O, zero the S/G table */
894 memset(&cmd->sdb, 0, sizeof(cmd->sdb));
895
896 cmd->cmnd[0] = SYNCHRONIZE_CACHE;
897 cmd->cmd_len = 10;
898 cmd->transfersize = 0;
899 cmd->allowed = SD_MAX_RETRIES;
900
901 rq->timeout = rq->q->rq_timeout * SD_FLUSH_TIMEOUT_MULTIPLIER;
902 return BLKPREP_OK;
903}
904
905static int sd_setup_read_write_cmnd(struct scsi_cmnd *SCpnt)
906{
907 struct request *rq = SCpnt->request;
908 struct scsi_device *sdp = SCpnt->device;
909 struct gendisk *disk = rq->rq_disk;
910 struct scsi_disk *sdkp = scsi_disk(disk);
911 sector_t block = blk_rq_pos(rq);
912 sector_t threshold;
913 unsigned int this_count = blk_rq_sectors(rq);
914 unsigned int dif, dix;
915 bool zoned_write = sd_is_zoned(sdkp) && rq_data_dir(rq) == WRITE;
916 int ret;
917 unsigned char protect;
918
919 if (zoned_write) {
920 ret = sd_zbc_setup_write_cmnd(SCpnt);
921 if (ret != BLKPREP_OK)
922 return ret;
923 }
924
925 ret = scsi_init_io(SCpnt);
926 if (ret != BLKPREP_OK)
927 goto out;
928 SCpnt = rq->special;
929
930 /* from here on until we're complete, any goto out
931 * is used for a killable error condition */
932 ret = BLKPREP_KILL;
933
934 SCSI_LOG_HLQUEUE(1,
935 scmd_printk(KERN_INFO, SCpnt,
936 "%s: block=%llu, count=%d\n",
937 __func__, (unsigned long long)block, this_count));
938
939 if (!sdp || !scsi_device_online(sdp) ||
940 block + blk_rq_sectors(rq) > get_capacity(disk)) {
941 SCSI_LOG_HLQUEUE(2, scmd_printk(KERN_INFO, SCpnt,
942 "Finishing %u sectors\n",
943 blk_rq_sectors(rq)));
944 SCSI_LOG_HLQUEUE(2, scmd_printk(KERN_INFO, SCpnt,
945 "Retry with 0x%p\n", SCpnt));
946 goto out;
947 }
948
949 if (sdp->changed) {
950 /*
951 * quietly refuse to do anything to a changed disc until
952 * the changed bit has been reset
953 */
954 /* printk("SCSI disk has been changed or is not present. Prohibiting further I/O.\n"); */
955 goto out;
956 }
957
958 /*
959 * Some SD card readers can't handle multi-sector accesses which touch
960 * the last one or two hardware sectors. Split accesses as needed.
961 */
962 threshold = get_capacity(disk) - SD_LAST_BUGGY_SECTORS *
963 (sdp->sector_size / 512);
964
965 if (unlikely(sdp->last_sector_bug && block + this_count > threshold)) {
966 if (block < threshold) {
967 /* Access up to the threshold but not beyond */
968 this_count = threshold - block;
969 } else {
970 /* Access only a single hardware sector */
971 this_count = sdp->sector_size / 512;
972 }
973 }
974
975 SCSI_LOG_HLQUEUE(2, scmd_printk(KERN_INFO, SCpnt, "block=%llu\n",
976 (unsigned long long)block));
977
978 /*
979 * If we have a 1K hardware sectorsize, prevent access to single
980 * 512 byte sectors. In theory we could handle this - in fact
981 * the scsi cdrom driver must be able to handle this because
982 * we typically use 1K blocksizes, and cdroms typically have
983 * 2K hardware sectorsizes. Of course, things are simpler
984 * with the cdrom, since it is read-only. For performance
985 * reasons, the filesystems should be able to handle this
986 * and not force the scsi disk driver to use bounce buffers
987 * for this.
988 */
989 if (sdp->sector_size == 1024) {
990 if ((block & 1) || (blk_rq_sectors(rq) & 1)) {
991 scmd_printk(KERN_ERR, SCpnt,
992 "Bad block number requested\n");
993 goto out;
994 } else {
995 block = block >> 1;
996 this_count = this_count >> 1;
997 }
998 }
999 if (sdp->sector_size == 2048) {
1000 if ((block & 3) || (blk_rq_sectors(rq) & 3)) {
1001 scmd_printk(KERN_ERR, SCpnt,
1002 "Bad block number requested\n");
1003 goto out;
1004 } else {
1005 block = block >> 2;
1006 this_count = this_count >> 2;
1007 }
1008 }
1009 if (sdp->sector_size == 4096) {
1010 if ((block & 7) || (blk_rq_sectors(rq) & 7)) {
1011 scmd_printk(KERN_ERR, SCpnt,
1012 "Bad block number requested\n");
1013 goto out;
1014 } else {
1015 block = block >> 3;
1016 this_count = this_count >> 3;
1017 }
1018 }
1019 if (rq_data_dir(rq) == WRITE) {
1020 SCpnt->cmnd[0] = WRITE_6;
1021
1022 if (blk_integrity_rq(rq))
1023 sd_dif_prepare(SCpnt);
1024
1025 } else if (rq_data_dir(rq) == READ) {
1026 SCpnt->cmnd[0] = READ_6;
1027 } else {
1028 scmd_printk(KERN_ERR, SCpnt, "Unknown command %d\n", req_op(rq));
1029 goto out;
1030 }
1031
1032 SCSI_LOG_HLQUEUE(2, scmd_printk(KERN_INFO, SCpnt,
1033 "%s %d/%u 512 byte blocks.\n",
1034 (rq_data_dir(rq) == WRITE) ?
1035 "writing" : "reading", this_count,
1036 blk_rq_sectors(rq)));
1037
1038 dix = scsi_prot_sg_count(SCpnt);
1039 dif = scsi_host_dif_capable(SCpnt->device->host, sdkp->protection_type);
1040
1041 if (dif || dix)
1042 protect = sd_setup_protect_cmnd(SCpnt, dix, dif);
1043 else
1044 protect = 0;
1045
1046 if (protect && sdkp->protection_type == T10_PI_TYPE2_PROTECTION) {
1047 SCpnt->cmnd = mempool_alloc(sd_cdb_pool, GFP_ATOMIC);
1048
1049 if (unlikely(SCpnt->cmnd == NULL)) {
1050 ret = BLKPREP_DEFER;
1051 goto out;
1052 }
1053
1054 SCpnt->cmd_len = SD_EXT_CDB_SIZE;
1055 memset(SCpnt->cmnd, 0, SCpnt->cmd_len);
1056 SCpnt->cmnd[0] = VARIABLE_LENGTH_CMD;
1057 SCpnt->cmnd[7] = 0x18;
1058 SCpnt->cmnd[9] = (rq_data_dir(rq) == READ) ? READ_32 : WRITE_32;
1059 SCpnt->cmnd[10] = protect | ((rq->cmd_flags & REQ_FUA) ? 0x8 : 0);
1060
1061 /* LBA */
1062 SCpnt->cmnd[12] = sizeof(block) > 4 ? (unsigned char) (block >> 56) & 0xff : 0;
1063 SCpnt->cmnd[13] = sizeof(block) > 4 ? (unsigned char) (block >> 48) & 0xff : 0;
1064 SCpnt->cmnd[14] = sizeof(block) > 4 ? (unsigned char) (block >> 40) & 0xff : 0;
1065 SCpnt->cmnd[15] = sizeof(block) > 4 ? (unsigned char) (block >> 32) & 0xff : 0;
1066 SCpnt->cmnd[16] = (unsigned char) (block >> 24) & 0xff;
1067 SCpnt->cmnd[17] = (unsigned char) (block >> 16) & 0xff;
1068 SCpnt->cmnd[18] = (unsigned char) (block >> 8) & 0xff;
1069 SCpnt->cmnd[19] = (unsigned char) block & 0xff;
1070
1071 /* Expected Indirect LBA */
1072 SCpnt->cmnd[20] = (unsigned char) (block >> 24) & 0xff;
1073 SCpnt->cmnd[21] = (unsigned char) (block >> 16) & 0xff;
1074 SCpnt->cmnd[22] = (unsigned char) (block >> 8) & 0xff;
1075 SCpnt->cmnd[23] = (unsigned char) block & 0xff;
1076
1077 /* Transfer length */
1078 SCpnt->cmnd[28] = (unsigned char) (this_count >> 24) & 0xff;
1079 SCpnt->cmnd[29] = (unsigned char) (this_count >> 16) & 0xff;
1080 SCpnt->cmnd[30] = (unsigned char) (this_count >> 8) & 0xff;
1081 SCpnt->cmnd[31] = (unsigned char) this_count & 0xff;
1082 } else if (sdp->use_16_for_rw || (this_count > 0xffff)) {
1083 SCpnt->cmnd[0] += READ_16 - READ_6;
1084 SCpnt->cmnd[1] = protect | ((rq->cmd_flags & REQ_FUA) ? 0x8 : 0);
1085 SCpnt->cmnd[2] = sizeof(block) > 4 ? (unsigned char) (block >> 56) & 0xff : 0;
1086 SCpnt->cmnd[3] = sizeof(block) > 4 ? (unsigned char) (block >> 48) & 0xff : 0;
1087 SCpnt->cmnd[4] = sizeof(block) > 4 ? (unsigned char) (block >> 40) & 0xff : 0;
1088 SCpnt->cmnd[5] = sizeof(block) > 4 ? (unsigned char) (block >> 32) & 0xff : 0;
1089 SCpnt->cmnd[6] = (unsigned char) (block >> 24) & 0xff;
1090 SCpnt->cmnd[7] = (unsigned char) (block >> 16) & 0xff;
1091 SCpnt->cmnd[8] = (unsigned char) (block >> 8) & 0xff;
1092 SCpnt->cmnd[9] = (unsigned char) block & 0xff;
1093 SCpnt->cmnd[10] = (unsigned char) (this_count >> 24) & 0xff;
1094 SCpnt->cmnd[11] = (unsigned char) (this_count >> 16) & 0xff;
1095 SCpnt->cmnd[12] = (unsigned char) (this_count >> 8) & 0xff;
1096 SCpnt->cmnd[13] = (unsigned char) this_count & 0xff;
1097 SCpnt->cmnd[14] = SCpnt->cmnd[15] = 0;
1098 } else if ((this_count > 0xff) || (block > 0x1fffff) ||
1099 scsi_device_protection(SCpnt->device) ||
1100 SCpnt->device->use_10_for_rw) {
1101 SCpnt->cmnd[0] += READ_10 - READ_6;
1102 SCpnt->cmnd[1] = protect | ((rq->cmd_flags & REQ_FUA) ? 0x8 : 0);
1103 SCpnt->cmnd[2] = (unsigned char) (block >> 24) & 0xff;
1104 SCpnt->cmnd[3] = (unsigned char) (block >> 16) & 0xff;
1105 SCpnt->cmnd[4] = (unsigned char) (block >> 8) & 0xff;
1106 SCpnt->cmnd[5] = (unsigned char) block & 0xff;
1107 SCpnt->cmnd[6] = SCpnt->cmnd[9] = 0;
1108 SCpnt->cmnd[7] = (unsigned char) (this_count >> 8) & 0xff;
1109 SCpnt->cmnd[8] = (unsigned char) this_count & 0xff;
1110 } else {
1111 if (unlikely(rq->cmd_flags & REQ_FUA)) {
1112 /*
1113 * This happens only if this drive failed
1114 * 10byte rw command with ILLEGAL_REQUEST
1115 * during operation and thus turned off
1116 * use_10_for_rw.
1117 */
1118 scmd_printk(KERN_ERR, SCpnt,
1119 "FUA write on READ/WRITE(6) drive\n");
1120 goto out;
1121 }
1122
1123 SCpnt->cmnd[1] |= (unsigned char) ((block >> 16) & 0x1f);
1124 SCpnt->cmnd[2] = (unsigned char) ((block >> 8) & 0xff);
1125 SCpnt->cmnd[3] = (unsigned char) block & 0xff;
1126 SCpnt->cmnd[4] = (unsigned char) this_count;
1127 SCpnt->cmnd[5] = 0;
1128 }
1129 SCpnt->sdb.length = this_count * sdp->sector_size;
1130
1131 /*
1132 * We shouldn't disconnect in the middle of a sector, so with a dumb
1133 * host adapter, it's safe to assume that we can at least transfer
1134 * this many bytes between each connect / disconnect.
1135 */
1136 SCpnt->transfersize = sdp->sector_size;
1137 SCpnt->underflow = this_count << 9;
1138 SCpnt->allowed = SD_MAX_RETRIES;
1139
1140 /*
1141 * This indicates that the command is ready from our end to be
1142 * queued.
1143 */
1144 ret = BLKPREP_OK;
1145 out:
1146 if (zoned_write && ret != BLKPREP_OK)
1147 sd_zbc_cancel_write_cmnd(SCpnt);
1148
1149 return ret;
1150}
1151
1152static int sd_init_command(struct scsi_cmnd *cmd)
1153{
1154 struct request *rq = cmd->request;
1155
1156 switch (req_op(rq)) {
1157 case REQ_OP_DISCARD:
1158 return sd_setup_discard_cmnd(cmd);
1159 case REQ_OP_WRITE_SAME:
1160 return sd_setup_write_same_cmnd(cmd);
1161 case REQ_OP_FLUSH:
1162 return sd_setup_flush_cmnd(cmd);
1163 case REQ_OP_READ:
1164 case REQ_OP_WRITE:
1165 return sd_setup_read_write_cmnd(cmd);
1166 case REQ_OP_ZONE_REPORT:
1167 return sd_zbc_setup_report_cmnd(cmd);
1168 case REQ_OP_ZONE_RESET:
1169 return sd_zbc_setup_reset_cmnd(cmd);
1170 default:
1171 BUG();
1172 }
1173}
1174
1175static void sd_uninit_command(struct scsi_cmnd *SCpnt)
1176{
1177 struct request *rq = SCpnt->request;
1178
1179 if (rq->rq_flags & RQF_SPECIAL_PAYLOAD)
1180 __free_page(rq->special_vec.bv_page);
1181
1182 if (SCpnt->cmnd != rq->cmd) {
1183 mempool_free(SCpnt->cmnd, sd_cdb_pool);
1184 SCpnt->cmnd = NULL;
1185 SCpnt->cmd_len = 0;
1186 }
1187}
1188
1189/**
1190 * sd_open - open a scsi disk device
1191 * @inode: only i_rdev member may be used
1192 * @filp: only f_mode and f_flags may be used
1193 *
1194 * Returns 0 if successful. Returns a negated errno value in case
1195 * of error.
1196 *
1197 * Note: This can be called from a user context (e.g. fsck(1) )
1198 * or from within the kernel (e.g. as a result of a mount(1) ).
1199 * In the latter case @inode and @filp carry an abridged amount
1200 * of information as noted above.
1201 *
1202 * Locking: called with bdev->bd_mutex held.
1203 **/
1204static int sd_open(struct block_device *bdev, fmode_t mode)
1205{
1206 struct scsi_disk *sdkp = scsi_disk_get(bdev->bd_disk);
1207 struct scsi_device *sdev;
1208 int retval;
1209
1210 if (!sdkp)
1211 return -ENXIO;
1212
1213 SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp, "sd_open\n"));
1214
1215 sdev = sdkp->device;
1216
1217 /*
1218 * If the device is in error recovery, wait until it is done.
1219 * If the device is offline, then disallow any access to it.
1220 */
1221 retval = -ENXIO;
1222 if (!scsi_block_when_processing_errors(sdev))
1223 goto error_out;
1224
1225 if (sdev->removable || sdkp->write_prot)
1226 check_disk_change(bdev);
1227
1228 /*
1229 * If the drive is empty, just let the open fail.
1230 */
1231 retval = -ENOMEDIUM;
1232 if (sdev->removable && !sdkp->media_present && !(mode & FMODE_NDELAY))
1233 goto error_out;
1234
1235 /*
1236 * If the device has the write protect tab set, have the open fail
1237 * if the user expects to be able to write to the thing.
1238 */
1239 retval = -EROFS;
1240 if (sdkp->write_prot && (mode & FMODE_WRITE))
1241 goto error_out;
1242
1243 /*
1244 * It is possible that the disk changing stuff resulted in
1245 * the device being taken offline. If this is the case,
1246 * report this to the user, and don't pretend that the
1247 * open actually succeeded.
1248 */
1249 retval = -ENXIO;
1250 if (!scsi_device_online(sdev))
1251 goto error_out;
1252
1253 if ((atomic_inc_return(&sdkp->openers) == 1) && sdev->removable) {
1254 if (scsi_block_when_processing_errors(sdev))
1255 scsi_set_medium_removal(sdev, SCSI_REMOVAL_PREVENT);
1256 }
1257
1258 return 0;
1259
1260error_out:
1261 scsi_disk_put(sdkp);
1262 return retval;
1263}
1264
1265/**
1266 * sd_release - invoked when the (last) close(2) is called on this
1267 * scsi disk.
1268 * @inode: only i_rdev member may be used
1269 * @filp: only f_mode and f_flags may be used
1270 *
1271 * Returns 0.
1272 *
1273 * Note: may block (uninterruptible) if error recovery is underway
1274 * on this disk.
1275 *
1276 * Locking: called with bdev->bd_mutex held.
1277 **/
1278static void sd_release(struct gendisk *disk, fmode_t mode)
1279{
1280 struct scsi_disk *sdkp = scsi_disk(disk);
1281 struct scsi_device *sdev = sdkp->device;
1282
1283 SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp, "sd_release\n"));
1284
1285 if (atomic_dec_return(&sdkp->openers) == 0 && sdev->removable) {
1286 if (scsi_block_when_processing_errors(sdev))
1287 scsi_set_medium_removal(sdev, SCSI_REMOVAL_ALLOW);
1288 }
1289
1290 /*
1291 * XXX and what if there are packets in flight and this close()
1292 * XXX is followed by a "rmmod sd_mod"?
1293 */
1294
1295 scsi_disk_put(sdkp);
1296}
1297
1298static int sd_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1299{
1300 struct scsi_disk *sdkp = scsi_disk(bdev->bd_disk);
1301 struct scsi_device *sdp = sdkp->device;
1302 struct Scsi_Host *host = sdp->host;
1303 sector_t capacity = logical_to_sectors(sdp, sdkp->capacity);
1304 int diskinfo[4];
1305
1306 /* default to most commonly used values */
1307 diskinfo[0] = 0x40; /* 1 << 6 */
1308 diskinfo[1] = 0x20; /* 1 << 5 */
1309 diskinfo[2] = capacity >> 11;
1310
1311 /* override with calculated, extended default, or driver values */
1312 if (host->hostt->bios_param)
1313 host->hostt->bios_param(sdp, bdev, capacity, diskinfo);
1314 else
1315 scsicam_bios_param(bdev, capacity, diskinfo);
1316
1317 geo->heads = diskinfo[0];
1318 geo->sectors = diskinfo[1];
1319 geo->cylinders = diskinfo[2];
1320 return 0;
1321}
1322
1323/**
1324 * sd_ioctl - process an ioctl
1325 * @inode: only i_rdev/i_bdev members may be used
1326 * @filp: only f_mode and f_flags may be used
1327 * @cmd: ioctl command number
1328 * @arg: this is third argument given to ioctl(2) system call.
1329 * Often contains a pointer.
1330 *
1331 * Returns 0 if successful (some ioctls return positive numbers on
1332 * success as well). Returns a negated errno value in case of error.
1333 *
1334 * Note: most ioctls are forward onto the block subsystem or further
1335 * down in the scsi subsystem.
1336 **/
1337static int sd_ioctl(struct block_device *bdev, fmode_t mode,
1338 unsigned int cmd, unsigned long arg)
1339{
1340 struct gendisk *disk = bdev->bd_disk;
1341 struct scsi_disk *sdkp = scsi_disk(disk);
1342 struct scsi_device *sdp = sdkp->device;
1343 void __user *p = (void __user *)arg;
1344 int error;
1345
1346 SCSI_LOG_IOCTL(1, sd_printk(KERN_INFO, sdkp, "sd_ioctl: disk=%s, "
1347 "cmd=0x%x\n", disk->disk_name, cmd));
1348
1349 error = scsi_verify_blk_ioctl(bdev, cmd);
1350 if (error < 0)
1351 return error;
1352
1353 /*
1354 * If we are in the middle of error recovery, don't let anyone
1355 * else try and use this device. Also, if error recovery fails, it
1356 * may try and take the device offline, in which case all further
1357 * access to the device is prohibited.
1358 */
1359 error = scsi_ioctl_block_when_processing_errors(sdp, cmd,
1360 (mode & FMODE_NDELAY) != 0);
1361 if (error)
1362 goto out;
1363
1364 /*
1365 * Send SCSI addressing ioctls directly to mid level, send other
1366 * ioctls to block level and then onto mid level if they can't be
1367 * resolved.
1368 */
1369 switch (cmd) {
1370 case SCSI_IOCTL_GET_IDLUN:
1371 case SCSI_IOCTL_GET_BUS_NUMBER:
1372 error = scsi_ioctl(sdp, cmd, p);
1373 break;
1374 default:
1375 error = scsi_cmd_blk_ioctl(bdev, mode, cmd, p);
1376 if (error != -ENOTTY)
1377 break;
1378 error = scsi_ioctl(sdp, cmd, p);
1379 break;
1380 }
1381out:
1382 return error;
1383}
1384
1385static void set_media_not_present(struct scsi_disk *sdkp)
1386{
1387 if (sdkp->media_present)
1388 sdkp->device->changed = 1;
1389
1390 if (sdkp->device->removable) {
1391 sdkp->media_present = 0;
1392 sdkp->capacity = 0;
1393 }
1394}
1395
1396static int media_not_present(struct scsi_disk *sdkp,
1397 struct scsi_sense_hdr *sshdr)
1398{
1399 if (!scsi_sense_valid(sshdr))
1400 return 0;
1401
1402 /* not invoked for commands that could return deferred errors */
1403 switch (sshdr->sense_key) {
1404 case UNIT_ATTENTION:
1405 case NOT_READY:
1406 /* medium not present */
1407 if (sshdr->asc == 0x3A) {
1408 set_media_not_present(sdkp);
1409 return 1;
1410 }
1411 }
1412 return 0;
1413}
1414
1415/**
1416 * sd_check_events - check media events
1417 * @disk: kernel device descriptor
1418 * @clearing: disk events currently being cleared
1419 *
1420 * Returns mask of DISK_EVENT_*.
1421 *
1422 * Note: this function is invoked from the block subsystem.
1423 **/
1424static unsigned int sd_check_events(struct gendisk *disk, unsigned int clearing)
1425{
1426 struct scsi_disk *sdkp = scsi_disk_get(disk);
1427 struct scsi_device *sdp;
1428 struct scsi_sense_hdr *sshdr = NULL;
1429 int retval;
1430
1431 if (!sdkp)
1432 return 0;
1433
1434 sdp = sdkp->device;
1435 SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp, "sd_check_events\n"));
1436
1437 /*
1438 * If the device is offline, don't send any commands - just pretend as
1439 * if the command failed. If the device ever comes back online, we
1440 * can deal with it then. It is only because of unrecoverable errors
1441 * that we would ever take a device offline in the first place.
1442 */
1443 if (!scsi_device_online(sdp)) {
1444 set_media_not_present(sdkp);
1445 goto out;
1446 }
1447
1448 /*
1449 * Using TEST_UNIT_READY enables differentiation between drive with
1450 * no cartridge loaded - NOT READY, drive with changed cartridge -
1451 * UNIT ATTENTION, or with same cartridge - GOOD STATUS.
1452 *
1453 * Drives that auto spin down. eg iomega jaz 1G, will be started
1454 * by sd_spinup_disk() from sd_revalidate_disk(), which happens whenever
1455 * sd_revalidate() is called.
1456 */
1457 retval = -ENODEV;
1458
1459 if (scsi_block_when_processing_errors(sdp)) {
1460 sshdr = kzalloc(sizeof(*sshdr), GFP_KERNEL);
1461 retval = scsi_test_unit_ready(sdp, SD_TIMEOUT, SD_MAX_RETRIES,
1462 sshdr);
1463 }
1464
1465 /* failed to execute TUR, assume media not present */
1466 if (host_byte(retval)) {
1467 set_media_not_present(sdkp);
1468 goto out;
1469 }
1470
1471 if (media_not_present(sdkp, sshdr))
1472 goto out;
1473
1474 /*
1475 * For removable scsi disk we have to recognise the presence
1476 * of a disk in the drive.
1477 */
1478 if (!sdkp->media_present)
1479 sdp->changed = 1;
1480 sdkp->media_present = 1;
1481out:
1482 /*
1483 * sdp->changed is set under the following conditions:
1484 *
1485 * Medium present state has changed in either direction.
1486 * Device has indicated UNIT_ATTENTION.
1487 */
1488 kfree(sshdr);
1489 retval = sdp->changed ? DISK_EVENT_MEDIA_CHANGE : 0;
1490 sdp->changed = 0;
1491 scsi_disk_put(sdkp);
1492 return retval;
1493}
1494
1495static int sd_sync_cache(struct scsi_disk *sdkp)
1496{
1497 int retries, res;
1498 struct scsi_device *sdp = sdkp->device;
1499 const int timeout = sdp->request_queue->rq_timeout
1500 * SD_FLUSH_TIMEOUT_MULTIPLIER;
1501 struct scsi_sense_hdr sshdr;
1502
1503 if (!scsi_device_online(sdp))
1504 return -ENODEV;
1505
1506 for (retries = 3; retries > 0; --retries) {
1507 unsigned char cmd[10] = { 0 };
1508
1509 cmd[0] = SYNCHRONIZE_CACHE;
1510 /*
1511 * Leave the rest of the command zero to indicate
1512 * flush everything.
1513 */
1514 res = scsi_execute_req_flags(sdp, cmd, DMA_NONE, NULL, 0,
1515 &sshdr, timeout, SD_MAX_RETRIES,
1516 NULL, 0, RQF_PM);
1517 if (res == 0)
1518 break;
1519 }
1520
1521 if (res) {
1522 sd_print_result(sdkp, "Synchronize Cache(10) failed", res);
1523
1524 if (driver_byte(res) & DRIVER_SENSE)
1525 sd_print_sense_hdr(sdkp, &sshdr);
1526 /* we need to evaluate the error return */
1527 if (scsi_sense_valid(&sshdr) &&
1528 (sshdr.asc == 0x3a || /* medium not present */
1529 sshdr.asc == 0x20)) /* invalid command */
1530 /* this is no error here */
1531 return 0;
1532
1533 switch (host_byte(res)) {
1534 /* ignore errors due to racing a disconnection */
1535 case DID_BAD_TARGET:
1536 case DID_NO_CONNECT:
1537 return 0;
1538 /* signal the upper layer it might try again */
1539 case DID_BUS_BUSY:
1540 case DID_IMM_RETRY:
1541 case DID_REQUEUE:
1542 case DID_SOFT_ERROR:
1543 return -EBUSY;
1544 default:
1545 return -EIO;
1546 }
1547 }
1548 return 0;
1549}
1550
1551static void sd_rescan(struct device *dev)
1552{
1553 struct scsi_disk *sdkp = dev_get_drvdata(dev);
1554
1555 revalidate_disk(sdkp->disk);
1556}
1557
1558
1559#ifdef CONFIG_COMPAT
1560/*
1561 * This gets directly called from VFS. When the ioctl
1562 * is not recognized we go back to the other translation paths.
1563 */
1564static int sd_compat_ioctl(struct block_device *bdev, fmode_t mode,
1565 unsigned int cmd, unsigned long arg)
1566{
1567 struct scsi_device *sdev = scsi_disk(bdev->bd_disk)->device;
1568 int error;
1569
1570 error = scsi_ioctl_block_when_processing_errors(sdev, cmd,
1571 (mode & FMODE_NDELAY) != 0);
1572 if (error)
1573 return error;
1574
1575 /*
1576 * Let the static ioctl translation table take care of it.
1577 */
1578 if (!sdev->host->hostt->compat_ioctl)
1579 return -ENOIOCTLCMD;
1580 return sdev->host->hostt->compat_ioctl(sdev, cmd, (void __user *)arg);
1581}
1582#endif
1583
1584static char sd_pr_type(enum pr_type type)
1585{
1586 switch (type) {
1587 case PR_WRITE_EXCLUSIVE:
1588 return 0x01;
1589 case PR_EXCLUSIVE_ACCESS:
1590 return 0x03;
1591 case PR_WRITE_EXCLUSIVE_REG_ONLY:
1592 return 0x05;
1593 case PR_EXCLUSIVE_ACCESS_REG_ONLY:
1594 return 0x06;
1595 case PR_WRITE_EXCLUSIVE_ALL_REGS:
1596 return 0x07;
1597 case PR_EXCLUSIVE_ACCESS_ALL_REGS:
1598 return 0x08;
1599 default:
1600 return 0;
1601 }
1602};
1603
1604static int sd_pr_command(struct block_device *bdev, u8 sa,
1605 u64 key, u64 sa_key, u8 type, u8 flags)
1606{
1607 struct scsi_device *sdev = scsi_disk(bdev->bd_disk)->device;
1608 struct scsi_sense_hdr sshdr;
1609 int result;
1610 u8 cmd[16] = { 0, };
1611 u8 data[24] = { 0, };
1612
1613 cmd[0] = PERSISTENT_RESERVE_OUT;
1614 cmd[1] = sa;
1615 cmd[2] = type;
1616 put_unaligned_be32(sizeof(data), &cmd[5]);
1617
1618 put_unaligned_be64(key, &data[0]);
1619 put_unaligned_be64(sa_key, &data[8]);
1620 data[20] = flags;
1621
1622 result = scsi_execute_req(sdev, cmd, DMA_TO_DEVICE, &data, sizeof(data),
1623 &sshdr, SD_TIMEOUT, SD_MAX_RETRIES, NULL);
1624
1625 if ((driver_byte(result) & DRIVER_SENSE) &&
1626 (scsi_sense_valid(&sshdr))) {
1627 sdev_printk(KERN_INFO, sdev, "PR command failed: %d\n", result);
1628 scsi_print_sense_hdr(sdev, NULL, &sshdr);
1629 }
1630
1631 return result;
1632}
1633
1634static int sd_pr_register(struct block_device *bdev, u64 old_key, u64 new_key,
1635 u32 flags)
1636{
1637 if (flags & ~PR_FL_IGNORE_KEY)
1638 return -EOPNOTSUPP;
1639 return sd_pr_command(bdev, (flags & PR_FL_IGNORE_KEY) ? 0x06 : 0x00,
1640 old_key, new_key, 0,
1641 (1 << 0) /* APTPL */);
1642}
1643
1644static int sd_pr_reserve(struct block_device *bdev, u64 key, enum pr_type type,
1645 u32 flags)
1646{
1647 if (flags)
1648 return -EOPNOTSUPP;
1649 return sd_pr_command(bdev, 0x01, key, 0, sd_pr_type(type), 0);
1650}
1651
1652static int sd_pr_release(struct block_device *bdev, u64 key, enum pr_type type)
1653{
1654 return sd_pr_command(bdev, 0x02, key, 0, sd_pr_type(type), 0);
1655}
1656
1657static int sd_pr_preempt(struct block_device *bdev, u64 old_key, u64 new_key,
1658 enum pr_type type, bool abort)
1659{
1660 return sd_pr_command(bdev, abort ? 0x05 : 0x04, old_key, new_key,
1661 sd_pr_type(type), 0);
1662}
1663
1664static int sd_pr_clear(struct block_device *bdev, u64 key)
1665{
1666 return sd_pr_command(bdev, 0x03, key, 0, 0, 0);
1667}
1668
1669static const struct pr_ops sd_pr_ops = {
1670 .pr_register = sd_pr_register,
1671 .pr_reserve = sd_pr_reserve,
1672 .pr_release = sd_pr_release,
1673 .pr_preempt = sd_pr_preempt,
1674 .pr_clear = sd_pr_clear,
1675};
1676
1677static const struct block_device_operations sd_fops = {
1678 .owner = THIS_MODULE,
1679 .open = sd_open,
1680 .release = sd_release,
1681 .ioctl = sd_ioctl,
1682 .getgeo = sd_getgeo,
1683#ifdef CONFIG_COMPAT
1684 .compat_ioctl = sd_compat_ioctl,
1685#endif
1686 .check_events = sd_check_events,
1687 .revalidate_disk = sd_revalidate_disk,
1688 .unlock_native_capacity = sd_unlock_native_capacity,
1689 .pr_ops = &sd_pr_ops,
1690};
1691
1692/**
1693 * sd_eh_action - error handling callback
1694 * @scmd: sd-issued command that has failed
1695 * @eh_disp: The recovery disposition suggested by the midlayer
1696 *
1697 * This function is called by the SCSI midlayer upon completion of an
1698 * error test command (currently TEST UNIT READY). The result of sending
1699 * the eh command is passed in eh_disp. We're looking for devices that
1700 * fail medium access commands but are OK with non access commands like
1701 * test unit ready (so wrongly see the device as having a successful
1702 * recovery)
1703 **/
1704static int sd_eh_action(struct scsi_cmnd *scmd, int eh_disp)
1705{
1706 struct scsi_disk *sdkp = scsi_disk(scmd->request->rq_disk);
1707
1708 if (!scsi_device_online(scmd->device) ||
1709 !scsi_medium_access_command(scmd) ||
1710 host_byte(scmd->result) != DID_TIME_OUT ||
1711 eh_disp != SUCCESS)
1712 return eh_disp;
1713
1714 /*
1715 * The device has timed out executing a medium access command.
1716 * However, the TEST UNIT READY command sent during error
1717 * handling completed successfully. Either the device is in the
1718 * process of recovering or has it suffered an internal failure
1719 * that prevents access to the storage medium.
1720 */
1721 sdkp->medium_access_timed_out++;
1722
1723 /*
1724 * If the device keeps failing read/write commands but TEST UNIT
1725 * READY always completes successfully we assume that medium
1726 * access is no longer possible and take the device offline.
1727 */
1728 if (sdkp->medium_access_timed_out >= sdkp->max_medium_access_timeouts) {
1729 scmd_printk(KERN_ERR, scmd,
1730 "Medium access timeout failure. Offlining disk!\n");
1731 scsi_device_set_state(scmd->device, SDEV_OFFLINE);
1732
1733 return FAILED;
1734 }
1735
1736 return eh_disp;
1737}
1738
1739static unsigned int sd_completed_bytes(struct scsi_cmnd *scmd)
1740{
1741 u64 start_lba = blk_rq_pos(scmd->request);
1742 u64 end_lba = blk_rq_pos(scmd->request) + (scsi_bufflen(scmd) / 512);
1743 u64 factor = scmd->device->sector_size / 512;
1744 u64 bad_lba;
1745 int info_valid;
1746 /*
1747 * resid is optional but mostly filled in. When it's unused,
1748 * its value is zero, so we assume the whole buffer transferred
1749 */
1750 unsigned int transferred = scsi_bufflen(scmd) - scsi_get_resid(scmd);
1751 unsigned int good_bytes;
1752
1753 if (scmd->request->cmd_type != REQ_TYPE_FS)
1754 return 0;
1755
1756 info_valid = scsi_get_sense_info_fld(scmd->sense_buffer,
1757 SCSI_SENSE_BUFFERSIZE,
1758 &bad_lba);
1759 if (!info_valid)
1760 return 0;
1761
1762 if (scsi_bufflen(scmd) <= scmd->device->sector_size)
1763 return 0;
1764
1765 /* be careful ... don't want any overflows */
1766 do_div(start_lba, factor);
1767 do_div(end_lba, factor);
1768
1769 /* The bad lba was reported incorrectly, we have no idea where
1770 * the error is.
1771 */
1772 if (bad_lba < start_lba || bad_lba >= end_lba)
1773 return 0;
1774
1775 /* This computation should always be done in terms of
1776 * the resolution of the device's medium.
1777 */
1778 good_bytes = (bad_lba - start_lba) * scmd->device->sector_size;
1779 return min(good_bytes, transferred);
1780}
1781
1782/**
1783 * sd_done - bottom half handler: called when the lower level
1784 * driver has completed (successfully or otherwise) a scsi command.
1785 * @SCpnt: mid-level's per command structure.
1786 *
1787 * Note: potentially run from within an ISR. Must not block.
1788 **/
1789static int sd_done(struct scsi_cmnd *SCpnt)
1790{
1791 int result = SCpnt->result;
1792 unsigned int good_bytes = result ? 0 : scsi_bufflen(SCpnt);
1793 unsigned int sector_size = SCpnt->device->sector_size;
1794 unsigned int resid;
1795 struct scsi_sense_hdr sshdr;
1796 struct scsi_disk *sdkp = scsi_disk(SCpnt->request->rq_disk);
1797 struct request *req = SCpnt->request;
1798 int sense_valid = 0;
1799 int sense_deferred = 0;
1800 unsigned char op = SCpnt->cmnd[0];
1801 unsigned char unmap = SCpnt->cmnd[1] & 8;
1802
1803 switch (req_op(req)) {
1804 case REQ_OP_DISCARD:
1805 case REQ_OP_WRITE_SAME:
1806 case REQ_OP_ZONE_RESET:
1807 if (!result) {
1808 good_bytes = blk_rq_bytes(req);
1809 scsi_set_resid(SCpnt, 0);
1810 } else {
1811 good_bytes = 0;
1812 scsi_set_resid(SCpnt, blk_rq_bytes(req));
1813 }
1814 break;
1815 case REQ_OP_ZONE_REPORT:
1816 if (!result) {
1817 good_bytes = scsi_bufflen(SCpnt)
1818 - scsi_get_resid(SCpnt);
1819 scsi_set_resid(SCpnt, 0);
1820 } else {
1821 good_bytes = 0;
1822 scsi_set_resid(SCpnt, blk_rq_bytes(req));
1823 }
1824 break;
1825 default:
1826 /*
1827 * In case of bogus fw or device, we could end up having
1828 * an unaligned partial completion. Check this here and force
1829 * alignment.
1830 */
1831 resid = scsi_get_resid(SCpnt);
1832 if (resid & (sector_size - 1)) {
1833 sd_printk(KERN_INFO, sdkp,
1834 "Unaligned partial completion (resid=%u, sector_sz=%u)\n",
1835 resid, sector_size);
1836 resid = min(scsi_bufflen(SCpnt),
1837 round_up(resid, sector_size));
1838 scsi_set_resid(SCpnt, resid);
1839 }
1840 }
1841
1842 if (result) {
1843 sense_valid = scsi_command_normalize_sense(SCpnt, &sshdr);
1844 if (sense_valid)
1845 sense_deferred = scsi_sense_is_deferred(&sshdr);
1846 }
1847 sdkp->medium_access_timed_out = 0;
1848
1849 if (driver_byte(result) != DRIVER_SENSE &&
1850 (!sense_valid || sense_deferred))
1851 goto out;
1852
1853 switch (sshdr.sense_key) {
1854 case HARDWARE_ERROR:
1855 case MEDIUM_ERROR:
1856 good_bytes = sd_completed_bytes(SCpnt);
1857 break;
1858 case RECOVERED_ERROR:
1859 good_bytes = scsi_bufflen(SCpnt);
1860 break;
1861 case NO_SENSE:
1862 /* This indicates a false check condition, so ignore it. An
1863 * unknown amount of data was transferred so treat it as an
1864 * error.
1865 */
1866 SCpnt->result = 0;
1867 memset(SCpnt->sense_buffer, 0, SCSI_SENSE_BUFFERSIZE);
1868 break;
1869 case ABORTED_COMMAND:
1870 if (sshdr.asc == 0x10) /* DIF: Target detected corruption */
1871 good_bytes = sd_completed_bytes(SCpnt);
1872 break;
1873 case ILLEGAL_REQUEST:
1874 if (sshdr.asc == 0x10) /* DIX: Host detected corruption */
1875 good_bytes = sd_completed_bytes(SCpnt);
1876 /* INVALID COMMAND OPCODE or INVALID FIELD IN CDB */
1877 if (sshdr.asc == 0x20 || sshdr.asc == 0x24) {
1878 switch (op) {
1879 case UNMAP:
1880 sd_config_discard(sdkp, SD_LBP_DISABLE);
1881 break;
1882 case WRITE_SAME_16:
1883 case WRITE_SAME:
1884 if (unmap)
1885 sd_config_discard(sdkp, SD_LBP_DISABLE);
1886 else {
1887 sdkp->device->no_write_same = 1;
1888 sd_config_write_same(sdkp);
1889
1890 good_bytes = 0;
1891 req->__data_len = blk_rq_bytes(req);
1892 req->rq_flags |= RQF_QUIET;
1893 }
1894 }
1895 }
1896 break;
1897 default:
1898 break;
1899 }
1900
1901 out:
1902 if (sd_is_zoned(sdkp))
1903 sd_zbc_complete(SCpnt, good_bytes, &sshdr);
1904
1905 SCSI_LOG_HLCOMPLETE(1, scmd_printk(KERN_INFO, SCpnt,
1906 "sd_done: completed %d of %d bytes\n",
1907 good_bytes, scsi_bufflen(SCpnt)));
1908
1909 if (rq_data_dir(SCpnt->request) == READ && scsi_prot_sg_count(SCpnt))
1910 sd_dif_complete(SCpnt, good_bytes);
1911
1912 return good_bytes;
1913}
1914
1915/*
1916 * spinup disk - called only in sd_revalidate_disk()
1917 */
1918static void
1919sd_spinup_disk(struct scsi_disk *sdkp)
1920{
1921 unsigned char cmd[10];
1922 unsigned long spintime_expire = 0;
1923 int retries, spintime;
1924 unsigned int the_result;
1925 struct scsi_sense_hdr sshdr;
1926 int sense_valid = 0;
1927
1928 spintime = 0;
1929
1930 /* Spin up drives, as required. Only do this at boot time */
1931 /* Spinup needs to be done for module loads too. */
1932 do {
1933 retries = 0;
1934
1935 do {
1936 cmd[0] = TEST_UNIT_READY;
1937 memset((void *) &cmd[1], 0, 9);
1938
1939 the_result = scsi_execute_req(sdkp->device, cmd,
1940 DMA_NONE, NULL, 0,
1941 &sshdr, SD_TIMEOUT,
1942 SD_MAX_RETRIES, NULL);
1943
1944 /*
1945 * If the drive has indicated to us that it
1946 * doesn't have any media in it, don't bother
1947 * with any more polling.
1948 */
1949 if (media_not_present(sdkp, &sshdr))
1950 return;
1951
1952 if (the_result)
1953 sense_valid = scsi_sense_valid(&sshdr);
1954 retries++;
1955 } while (retries < 3 &&
1956 (!scsi_status_is_good(the_result) ||
1957 ((driver_byte(the_result) & DRIVER_SENSE) &&
1958 sense_valid && sshdr.sense_key == UNIT_ATTENTION)));
1959
1960 if ((driver_byte(the_result) & DRIVER_SENSE) == 0) {
1961 /* no sense, TUR either succeeded or failed
1962 * with a status error */
1963 if(!spintime && !scsi_status_is_good(the_result)) {
1964 sd_print_result(sdkp, "Test Unit Ready failed",
1965 the_result);
1966 }
1967 break;
1968 }
1969
1970 /*
1971 * The device does not want the automatic start to be issued.
1972 */
1973 if (sdkp->device->no_start_on_add)
1974 break;
1975
1976 if (sense_valid && sshdr.sense_key == NOT_READY) {
1977 if (sshdr.asc == 4 && sshdr.ascq == 3)
1978 break; /* manual intervention required */
1979 if (sshdr.asc == 4 && sshdr.ascq == 0xb)
1980 break; /* standby */
1981 if (sshdr.asc == 4 && sshdr.ascq == 0xc)
1982 break; /* unavailable */
1983 /*
1984 * Issue command to spin up drive when not ready
1985 */
1986 if (!spintime) {
1987 sd_printk(KERN_NOTICE, sdkp, "Spinning up disk...");
1988 cmd[0] = START_STOP;
1989 cmd[1] = 1; /* Return immediately */
1990 memset((void *) &cmd[2], 0, 8);
1991 cmd[4] = 1; /* Start spin cycle */
1992 if (sdkp->device->start_stop_pwr_cond)
1993 cmd[4] |= 1 << 4;
1994 scsi_execute_req(sdkp->device, cmd, DMA_NONE,
1995 NULL, 0, &sshdr,
1996 SD_TIMEOUT, SD_MAX_RETRIES,
1997 NULL);
1998 spintime_expire = jiffies + 100 * HZ;
1999 spintime = 1;
2000 }
2001 /* Wait 1 second for next try */
2002 msleep(1000);
2003 printk(".");
2004
2005 /*
2006 * Wait for USB flash devices with slow firmware.
2007 * Yes, this sense key/ASC combination shouldn't
2008 * occur here. It's characteristic of these devices.
2009 */
2010 } else if (sense_valid &&
2011 sshdr.sense_key == UNIT_ATTENTION &&
2012 sshdr.asc == 0x28) {
2013 if (!spintime) {
2014 spintime_expire = jiffies + 5 * HZ;
2015 spintime = 1;
2016 }
2017 /* Wait 1 second for next try */
2018 msleep(1000);
2019 } else {
2020 /* we don't understand the sense code, so it's
2021 * probably pointless to loop */
2022 if(!spintime) {
2023 sd_printk(KERN_NOTICE, sdkp, "Unit Not Ready\n");
2024 sd_print_sense_hdr(sdkp, &sshdr);
2025 }
2026 break;
2027 }
2028
2029 } while (spintime && time_before_eq(jiffies, spintime_expire));
2030
2031 if (spintime) {
2032 if (scsi_status_is_good(the_result))
2033 printk("ready\n");
2034 else
2035 printk("not responding...\n");
2036 }
2037}
2038
2039/*
2040 * Determine whether disk supports Data Integrity Field.
2041 */
2042static int sd_read_protection_type(struct scsi_disk *sdkp, unsigned char *buffer)
2043{
2044 struct scsi_device *sdp = sdkp->device;
2045 u8 type;
2046 int ret = 0;
2047
2048 if (scsi_device_protection(sdp) == 0 || (buffer[12] & 1) == 0)
2049 return ret;
2050
2051 type = ((buffer[12] >> 1) & 7) + 1; /* P_TYPE 0 = Type 1 */
2052
2053 if (type > T10_PI_TYPE3_PROTECTION)
2054 ret = -ENODEV;
2055 else if (scsi_host_dif_capable(sdp->host, type))
2056 ret = 1;
2057
2058 if (sdkp->first_scan || type != sdkp->protection_type)
2059 switch (ret) {
2060 case -ENODEV:
2061 sd_printk(KERN_ERR, sdkp, "formatted with unsupported" \
2062 " protection type %u. Disabling disk!\n",
2063 type);
2064 break;
2065 case 1:
2066 sd_printk(KERN_NOTICE, sdkp,
2067 "Enabling DIF Type %u protection\n", type);
2068 break;
2069 case 0:
2070 sd_printk(KERN_NOTICE, sdkp,
2071 "Disabling DIF Type %u protection\n", type);
2072 break;
2073 }
2074
2075 sdkp->protection_type = type;
2076
2077 return ret;
2078}
2079
2080static void read_capacity_error(struct scsi_disk *sdkp, struct scsi_device *sdp,
2081 struct scsi_sense_hdr *sshdr, int sense_valid,
2082 int the_result)
2083{
2084 if (driver_byte(the_result) & DRIVER_SENSE)
2085 sd_print_sense_hdr(sdkp, sshdr);
2086 else
2087 sd_printk(KERN_NOTICE, sdkp, "Sense not available.\n");
2088
2089 /*
2090 * Set dirty bit for removable devices if not ready -
2091 * sometimes drives will not report this properly.
2092 */
2093 if (sdp->removable &&
2094 sense_valid && sshdr->sense_key == NOT_READY)
2095 set_media_not_present(sdkp);
2096
2097 /*
2098 * We used to set media_present to 0 here to indicate no media
2099 * in the drive, but some drives fail read capacity even with
2100 * media present, so we can't do that.
2101 */
2102 sdkp->capacity = 0; /* unknown mapped to zero - as usual */
2103}
2104
2105#define RC16_LEN 32
2106#if RC16_LEN > SD_BUF_SIZE
2107#error RC16_LEN must not be more than SD_BUF_SIZE
2108#endif
2109
2110#define READ_CAPACITY_RETRIES_ON_RESET 10
2111
2112static int read_capacity_16(struct scsi_disk *sdkp, struct scsi_device *sdp,
2113 unsigned char *buffer)
2114{
2115 unsigned char cmd[16];
2116 struct scsi_sense_hdr sshdr;
2117 int sense_valid = 0;
2118 int the_result;
2119 int retries = 3, reset_retries = READ_CAPACITY_RETRIES_ON_RESET;
2120 unsigned int alignment;
2121 unsigned long long lba;
2122 unsigned sector_size;
2123
2124 if (sdp->no_read_capacity_16)
2125 return -EINVAL;
2126
2127 do {
2128 memset(cmd, 0, 16);
2129 cmd[0] = SERVICE_ACTION_IN_16;
2130 cmd[1] = SAI_READ_CAPACITY_16;
2131 cmd[13] = RC16_LEN;
2132 memset(buffer, 0, RC16_LEN);
2133
2134 the_result = scsi_execute_req(sdp, cmd, DMA_FROM_DEVICE,
2135 buffer, RC16_LEN, &sshdr,
2136 SD_TIMEOUT, SD_MAX_RETRIES, NULL);
2137
2138 if (media_not_present(sdkp, &sshdr))
2139 return -ENODEV;
2140
2141 if (the_result) {
2142 sense_valid = scsi_sense_valid(&sshdr);
2143 if (sense_valid &&
2144 sshdr.sense_key == ILLEGAL_REQUEST &&
2145 (sshdr.asc == 0x20 || sshdr.asc == 0x24) &&
2146 sshdr.ascq == 0x00)
2147 /* Invalid Command Operation Code or
2148 * Invalid Field in CDB, just retry
2149 * silently with RC10 */
2150 return -EINVAL;
2151 if (sense_valid &&
2152 sshdr.sense_key == UNIT_ATTENTION &&
2153 sshdr.asc == 0x29 && sshdr.ascq == 0x00)
2154 /* Device reset might occur several times,
2155 * give it one more chance */
2156 if (--reset_retries > 0)
2157 continue;
2158 }
2159 retries--;
2160
2161 } while (the_result && retries);
2162
2163 if (the_result) {
2164 sd_print_result(sdkp, "Read Capacity(16) failed", the_result);
2165 read_capacity_error(sdkp, sdp, &sshdr, sense_valid, the_result);
2166 return -EINVAL;
2167 }
2168
2169 sector_size = get_unaligned_be32(&buffer[8]);
2170 lba = get_unaligned_be64(&buffer[0]);
2171
2172 if (sd_read_protection_type(sdkp, buffer) < 0) {
2173 sdkp->capacity = 0;
2174 return -ENODEV;
2175 }
2176
2177 if ((sizeof(sdkp->capacity) == 4) && (lba >= 0xffffffffULL)) {
2178 sd_printk(KERN_ERR, sdkp, "Too big for this kernel. Use a "
2179 "kernel compiled with support for large block "
2180 "devices.\n");
2181 sdkp->capacity = 0;
2182 return -EOVERFLOW;
2183 }
2184
2185 /* Logical blocks per physical block exponent */
2186 sdkp->physical_block_size = (1 << (buffer[13] & 0xf)) * sector_size;
2187
2188 /* RC basis */
2189 sdkp->rc_basis = (buffer[12] >> 4) & 0x3;
2190
2191 /* Lowest aligned logical block */
2192 alignment = ((buffer[14] & 0x3f) << 8 | buffer[15]) * sector_size;
2193 blk_queue_alignment_offset(sdp->request_queue, alignment);
2194 if (alignment && sdkp->first_scan)
2195 sd_printk(KERN_NOTICE, sdkp,
2196 "physical block alignment offset: %u\n", alignment);
2197
2198 if (buffer[14] & 0x80) { /* LBPME */
2199 sdkp->lbpme = 1;
2200
2201 if (buffer[14] & 0x40) /* LBPRZ */
2202 sdkp->lbprz = 1;
2203
2204 sd_config_discard(sdkp, SD_LBP_WS16);
2205 }
2206
2207 sdkp->capacity = lba + 1;
2208 return sector_size;
2209}
2210
2211static int read_capacity_10(struct scsi_disk *sdkp, struct scsi_device *sdp,
2212 unsigned char *buffer)
2213{
2214 unsigned char cmd[16];
2215 struct scsi_sense_hdr sshdr;
2216 int sense_valid = 0;
2217 int the_result;
2218 int retries = 3, reset_retries = READ_CAPACITY_RETRIES_ON_RESET;
2219 sector_t lba;
2220 unsigned sector_size;
2221
2222 do {
2223 cmd[0] = READ_CAPACITY;
2224 memset(&cmd[1], 0, 9);
2225 memset(buffer, 0, 8);
2226
2227 the_result = scsi_execute_req(sdp, cmd, DMA_FROM_DEVICE,
2228 buffer, 8, &sshdr,
2229 SD_TIMEOUT, SD_MAX_RETRIES, NULL);
2230
2231 if (media_not_present(sdkp, &sshdr))
2232 return -ENODEV;
2233
2234 if (the_result) {
2235 sense_valid = scsi_sense_valid(&sshdr);
2236 if (sense_valid &&
2237 sshdr.sense_key == UNIT_ATTENTION &&
2238 sshdr.asc == 0x29 && sshdr.ascq == 0x00)
2239 /* Device reset might occur several times,
2240 * give it one more chance */
2241 if (--reset_retries > 0)
2242 continue;
2243 }
2244 retries--;
2245
2246 } while (the_result && retries);
2247
2248 if (the_result) {
2249 sd_print_result(sdkp, "Read Capacity(10) failed", the_result);
2250 read_capacity_error(sdkp, sdp, &sshdr, sense_valid, the_result);
2251 return -EINVAL;
2252 }
2253
2254 sector_size = get_unaligned_be32(&buffer[4]);
2255 lba = get_unaligned_be32(&buffer[0]);
2256
2257 if (sdp->no_read_capacity_16 && (lba == 0xffffffff)) {
2258 /* Some buggy (usb cardreader) devices return an lba of
2259 0xffffffff when the want to report a size of 0 (with
2260 which they really mean no media is present) */
2261 sdkp->capacity = 0;
2262 sdkp->physical_block_size = sector_size;
2263 return sector_size;
2264 }
2265
2266 if ((sizeof(sdkp->capacity) == 4) && (lba == 0xffffffff)) {
2267 sd_printk(KERN_ERR, sdkp, "Too big for this kernel. Use a "
2268 "kernel compiled with support for large block "
2269 "devices.\n");
2270 sdkp->capacity = 0;
2271 return -EOVERFLOW;
2272 }
2273
2274 sdkp->capacity = lba + 1;
2275 sdkp->physical_block_size = sector_size;
2276 return sector_size;
2277}
2278
2279static int sd_try_rc16_first(struct scsi_device *sdp)
2280{
2281 if (sdp->host->max_cmd_len < 16)
2282 return 0;
2283 if (sdp->try_rc_10_first)
2284 return 0;
2285 if (sdp->scsi_level > SCSI_SPC_2)
2286 return 1;
2287 if (scsi_device_protection(sdp))
2288 return 1;
2289 return 0;
2290}
2291
2292/*
2293 * read disk capacity
2294 */
2295static void
2296sd_read_capacity(struct scsi_disk *sdkp, unsigned char *buffer)
2297{
2298 int sector_size;
2299 struct scsi_device *sdp = sdkp->device;
2300
2301 if (sd_try_rc16_first(sdp)) {
2302 sector_size = read_capacity_16(sdkp, sdp, buffer);
2303 if (sector_size == -EOVERFLOW)
2304 goto got_data;
2305 if (sector_size == -ENODEV)
2306 return;
2307 if (sector_size < 0)
2308 sector_size = read_capacity_10(sdkp, sdp, buffer);
2309 if (sector_size < 0)
2310 return;
2311 } else {
2312 sector_size = read_capacity_10(sdkp, sdp, buffer);
2313 if (sector_size == -EOVERFLOW)
2314 goto got_data;
2315 if (sector_size < 0)
2316 return;
2317 if ((sizeof(sdkp->capacity) > 4) &&
2318 (sdkp->capacity > 0xffffffffULL)) {
2319 int old_sector_size = sector_size;
2320 sd_printk(KERN_NOTICE, sdkp, "Very big device. "
2321 "Trying to use READ CAPACITY(16).\n");
2322 sector_size = read_capacity_16(sdkp, sdp, buffer);
2323 if (sector_size < 0) {
2324 sd_printk(KERN_NOTICE, sdkp,
2325 "Using 0xffffffff as device size\n");
2326 sdkp->capacity = 1 + (sector_t) 0xffffffff;
2327 sector_size = old_sector_size;
2328 goto got_data;
2329 }
2330 }
2331 }
2332
2333 /* Some devices are known to return the total number of blocks,
2334 * not the highest block number. Some devices have versions
2335 * which do this and others which do not. Some devices we might
2336 * suspect of doing this but we don't know for certain.
2337 *
2338 * If we know the reported capacity is wrong, decrement it. If
2339 * we can only guess, then assume the number of blocks is even
2340 * (usually true but not always) and err on the side of lowering
2341 * the capacity.
2342 */
2343 if (sdp->fix_capacity ||
2344 (sdp->guess_capacity && (sdkp->capacity & 0x01))) {
2345 sd_printk(KERN_INFO, sdkp, "Adjusting the sector count "
2346 "from its reported value: %llu\n",
2347 (unsigned long long) sdkp->capacity);
2348 --sdkp->capacity;
2349 }
2350
2351got_data:
2352 if (sector_size == 0) {
2353 sector_size = 512;
2354 sd_printk(KERN_NOTICE, sdkp, "Sector size 0 reported, "
2355 "assuming 512.\n");
2356 }
2357
2358 if (sector_size != 512 &&
2359 sector_size != 1024 &&
2360 sector_size != 2048 &&
2361 sector_size != 4096) {
2362 sd_printk(KERN_NOTICE, sdkp, "Unsupported sector size %d.\n",
2363 sector_size);
2364 /*
2365 * The user might want to re-format the drive with
2366 * a supported sectorsize. Once this happens, it
2367 * would be relatively trivial to set the thing up.
2368 * For this reason, we leave the thing in the table.
2369 */
2370 sdkp->capacity = 0;
2371 /*
2372 * set a bogus sector size so the normal read/write
2373 * logic in the block layer will eventually refuse any
2374 * request on this device without tripping over power
2375 * of two sector size assumptions
2376 */
2377 sector_size = 512;
2378 }
2379 blk_queue_logical_block_size(sdp->request_queue, sector_size);
2380 blk_queue_physical_block_size(sdp->request_queue,
2381 sdkp->physical_block_size);
2382 sdkp->device->sector_size = sector_size;
2383
2384 if (sdkp->capacity > 0xffffffff)
2385 sdp->use_16_for_rw = 1;
2386
2387}
2388
2389/*
2390 * Print disk capacity
2391 */
2392static void
2393sd_print_capacity(struct scsi_disk *sdkp,
2394 sector_t old_capacity)
2395{
2396 int sector_size = sdkp->device->sector_size;
2397 char cap_str_2[10], cap_str_10[10];
2398
2399 string_get_size(sdkp->capacity, sector_size,
2400 STRING_UNITS_2, cap_str_2, sizeof(cap_str_2));
2401 string_get_size(sdkp->capacity, sector_size,
2402 STRING_UNITS_10, cap_str_10,
2403 sizeof(cap_str_10));
2404
2405 if (sdkp->first_scan || old_capacity != sdkp->capacity) {
2406 sd_printk(KERN_NOTICE, sdkp,
2407 "%llu %d-byte logical blocks: (%s/%s)\n",
2408 (unsigned long long)sdkp->capacity,
2409 sector_size, cap_str_10, cap_str_2);
2410
2411 if (sdkp->physical_block_size != sector_size)
2412 sd_printk(KERN_NOTICE, sdkp,
2413 "%u-byte physical blocks\n",
2414 sdkp->physical_block_size);
2415
2416 sd_zbc_print_zones(sdkp);
2417 }
2418}
2419
2420/* called with buffer of length 512 */
2421static inline int
2422sd_do_mode_sense(struct scsi_device *sdp, int dbd, int modepage,
2423 unsigned char *buffer, int len, struct scsi_mode_data *data,
2424 struct scsi_sense_hdr *sshdr)
2425{
2426 return scsi_mode_sense(sdp, dbd, modepage, buffer, len,
2427 SD_TIMEOUT, SD_MAX_RETRIES, data,
2428 sshdr);
2429}
2430
2431/*
2432 * read write protect setting, if possible - called only in sd_revalidate_disk()
2433 * called with buffer of length SD_BUF_SIZE
2434 */
2435static void
2436sd_read_write_protect_flag(struct scsi_disk *sdkp, unsigned char *buffer)
2437{
2438 int res;
2439 struct scsi_device *sdp = sdkp->device;
2440 struct scsi_mode_data data;
2441 int old_wp = sdkp->write_prot;
2442
2443 set_disk_ro(sdkp->disk, 0);
2444 if (sdp->skip_ms_page_3f) {
2445 sd_first_printk(KERN_NOTICE, sdkp, "Assuming Write Enabled\n");
2446 return;
2447 }
2448
2449 if (sdp->use_192_bytes_for_3f) {
2450 res = sd_do_mode_sense(sdp, 0, 0x3F, buffer, 192, &data, NULL);
2451 } else {
2452 /*
2453 * First attempt: ask for all pages (0x3F), but only 4 bytes.
2454 * We have to start carefully: some devices hang if we ask
2455 * for more than is available.
2456 */
2457 res = sd_do_mode_sense(sdp, 0, 0x3F, buffer, 4, &data, NULL);
2458
2459 /*
2460 * Second attempt: ask for page 0 When only page 0 is
2461 * implemented, a request for page 3F may return Sense Key
2462 * 5: Illegal Request, Sense Code 24: Invalid field in
2463 * CDB.
2464 */
2465 if (!scsi_status_is_good(res))
2466 res = sd_do_mode_sense(sdp, 0, 0, buffer, 4, &data, NULL);
2467
2468 /*
2469 * Third attempt: ask 255 bytes, as we did earlier.
2470 */
2471 if (!scsi_status_is_good(res))
2472 res = sd_do_mode_sense(sdp, 0, 0x3F, buffer, 255,
2473 &data, NULL);
2474 }
2475
2476 if (!scsi_status_is_good(res)) {
2477 sd_first_printk(KERN_WARNING, sdkp,
2478 "Test WP failed, assume Write Enabled\n");
2479 } else {
2480 sdkp->write_prot = ((data.device_specific & 0x80) != 0);
2481 set_disk_ro(sdkp->disk, sdkp->write_prot);
2482 if (sdkp->first_scan || old_wp != sdkp->write_prot) {
2483 sd_printk(KERN_NOTICE, sdkp, "Write Protect is %s\n",
2484 sdkp->write_prot ? "on" : "off");
2485 sd_printk(KERN_DEBUG, sdkp, "Mode Sense: %4ph\n", buffer);
2486 }
2487 }
2488}
2489
2490/*
2491 * sd_read_cache_type - called only from sd_revalidate_disk()
2492 * called with buffer of length SD_BUF_SIZE
2493 */
2494static void
2495sd_read_cache_type(struct scsi_disk *sdkp, unsigned char *buffer)
2496{
2497 int len = 0, res;
2498 struct scsi_device *sdp = sdkp->device;
2499
2500 int dbd;
2501 int modepage;
2502 int first_len;
2503 struct scsi_mode_data data;
2504 struct scsi_sense_hdr sshdr;
2505 int old_wce = sdkp->WCE;
2506 int old_rcd = sdkp->RCD;
2507 int old_dpofua = sdkp->DPOFUA;
2508
2509
2510 if (sdkp->cache_override)
2511 return;
2512
2513 first_len = 4;
2514 if (sdp->skip_ms_page_8) {
2515 if (sdp->type == TYPE_RBC)
2516 goto defaults;
2517 else {
2518 if (sdp->skip_ms_page_3f)
2519 goto defaults;
2520 modepage = 0x3F;
2521 if (sdp->use_192_bytes_for_3f)
2522 first_len = 192;
2523 dbd = 0;
2524 }
2525 } else if (sdp->type == TYPE_RBC) {
2526 modepage = 6;
2527 dbd = 8;
2528 } else {
2529 modepage = 8;
2530 dbd = 0;
2531 }
2532
2533 /* cautiously ask */
2534 res = sd_do_mode_sense(sdp, dbd, modepage, buffer, first_len,
2535 &data, &sshdr);
2536
2537 if (!scsi_status_is_good(res))
2538 goto bad_sense;
2539
2540 if (!data.header_length) {
2541 modepage = 6;
2542 first_len = 0;
2543 sd_first_printk(KERN_ERR, sdkp,
2544 "Missing header in MODE_SENSE response\n");
2545 }
2546
2547 /* that went OK, now ask for the proper length */
2548 len = data.length;
2549
2550 /*
2551 * We're only interested in the first three bytes, actually.
2552 * But the data cache page is defined for the first 20.
2553 */
2554 if (len < 3)
2555 goto bad_sense;
2556 else if (len > SD_BUF_SIZE) {
2557 sd_first_printk(KERN_NOTICE, sdkp, "Truncating mode parameter "
2558 "data from %d to %d bytes\n", len, SD_BUF_SIZE);
2559 len = SD_BUF_SIZE;
2560 }
2561 if (modepage == 0x3F && sdp->use_192_bytes_for_3f)
2562 len = 192;
2563
2564 /* Get the data */
2565 if (len > first_len)
2566 res = sd_do_mode_sense(sdp, dbd, modepage, buffer, len,
2567 &data, &sshdr);
2568
2569 if (scsi_status_is_good(res)) {
2570 int offset = data.header_length + data.block_descriptor_length;
2571
2572 while (offset < len) {
2573 u8 page_code = buffer[offset] & 0x3F;
2574 u8 spf = buffer[offset] & 0x40;
2575
2576 if (page_code == 8 || page_code == 6) {
2577 /* We're interested only in the first 3 bytes.
2578 */
2579 if (len - offset <= 2) {
2580 sd_first_printk(KERN_ERR, sdkp,
2581 "Incomplete mode parameter "
2582 "data\n");
2583 goto defaults;
2584 } else {
2585 modepage = page_code;
2586 goto Page_found;
2587 }
2588 } else {
2589 /* Go to the next page */
2590 if (spf && len - offset > 3)
2591 offset += 4 + (buffer[offset+2] << 8) +
2592 buffer[offset+3];
2593 else if (!spf && len - offset > 1)
2594 offset += 2 + buffer[offset+1];
2595 else {
2596 sd_first_printk(KERN_ERR, sdkp,
2597 "Incomplete mode "
2598 "parameter data\n");
2599 goto defaults;
2600 }
2601 }
2602 }
2603
2604 sd_first_printk(KERN_ERR, sdkp, "No Caching mode page found\n");
2605 goto defaults;
2606
2607 Page_found:
2608 if (modepage == 8) {
2609 sdkp->WCE = ((buffer[offset + 2] & 0x04) != 0);
2610 sdkp->RCD = ((buffer[offset + 2] & 0x01) != 0);
2611 } else {
2612 sdkp->WCE = ((buffer[offset + 2] & 0x01) == 0);
2613 sdkp->RCD = 0;
2614 }
2615
2616 sdkp->DPOFUA = (data.device_specific & 0x10) != 0;
2617 if (sdp->broken_fua) {
2618 sd_first_printk(KERN_NOTICE, sdkp, "Disabling FUA\n");
2619 sdkp->DPOFUA = 0;
2620 } else if (sdkp->DPOFUA && !sdkp->device->use_10_for_rw &&
2621 !sdkp->device->use_16_for_rw) {
2622 sd_first_printk(KERN_NOTICE, sdkp,
2623 "Uses READ/WRITE(6), disabling FUA\n");
2624 sdkp->DPOFUA = 0;
2625 }
2626
2627 /* No cache flush allowed for write protected devices */
2628 if (sdkp->WCE && sdkp->write_prot)
2629 sdkp->WCE = 0;
2630
2631 if (sdkp->first_scan || old_wce != sdkp->WCE ||
2632 old_rcd != sdkp->RCD || old_dpofua != sdkp->DPOFUA)
2633 sd_printk(KERN_NOTICE, sdkp,
2634 "Write cache: %s, read cache: %s, %s\n",
2635 sdkp->WCE ? "enabled" : "disabled",
2636 sdkp->RCD ? "disabled" : "enabled",
2637 sdkp->DPOFUA ? "supports DPO and FUA"
2638 : "doesn't support DPO or FUA");
2639
2640 return;
2641 }
2642
2643bad_sense:
2644 if (scsi_sense_valid(&sshdr) &&
2645 sshdr.sense_key == ILLEGAL_REQUEST &&
2646 sshdr.asc == 0x24 && sshdr.ascq == 0x0)
2647 /* Invalid field in CDB */
2648 sd_first_printk(KERN_NOTICE, sdkp, "Cache data unavailable\n");
2649 else
2650 sd_first_printk(KERN_ERR, sdkp,
2651 "Asking for cache data failed\n");
2652
2653defaults:
2654 if (sdp->wce_default_on) {
2655 sd_first_printk(KERN_NOTICE, sdkp,
2656 "Assuming drive cache: write back\n");
2657 sdkp->WCE = 1;
2658 } else {
2659 sd_first_printk(KERN_ERR, sdkp,
2660 "Assuming drive cache: write through\n");
2661 sdkp->WCE = 0;
2662 }
2663 sdkp->RCD = 0;
2664 sdkp->DPOFUA = 0;
2665}
2666
2667/*
2668 * The ATO bit indicates whether the DIF application tag is available
2669 * for use by the operating system.
2670 */
2671static void sd_read_app_tag_own(struct scsi_disk *sdkp, unsigned char *buffer)
2672{
2673 int res, offset;
2674 struct scsi_device *sdp = sdkp->device;
2675 struct scsi_mode_data data;
2676 struct scsi_sense_hdr sshdr;
2677
2678 if (sdp->type != TYPE_DISK && sdp->type != TYPE_ZBC)
2679 return;
2680
2681 if (sdkp->protection_type == 0)
2682 return;
2683
2684 res = scsi_mode_sense(sdp, 1, 0x0a, buffer, 36, SD_TIMEOUT,
2685 SD_MAX_RETRIES, &data, &sshdr);
2686
2687 if (!scsi_status_is_good(res) || !data.header_length ||
2688 data.length < 6) {
2689 sd_first_printk(KERN_WARNING, sdkp,
2690 "getting Control mode page failed, assume no ATO\n");
2691
2692 if (scsi_sense_valid(&sshdr))
2693 sd_print_sense_hdr(sdkp, &sshdr);
2694
2695 return;
2696 }
2697
2698 offset = data.header_length + data.block_descriptor_length;
2699
2700 if ((buffer[offset] & 0x3f) != 0x0a) {
2701 sd_first_printk(KERN_ERR, sdkp, "ATO Got wrong page\n");
2702 return;
2703 }
2704
2705 if ((buffer[offset + 5] & 0x80) == 0)
2706 return;
2707
2708 sdkp->ATO = 1;
2709
2710 return;
2711}
2712
2713/**
2714 * sd_read_block_limits - Query disk device for preferred I/O sizes.
2715 * @disk: disk to query
2716 */
2717static void sd_read_block_limits(struct scsi_disk *sdkp)
2718{
2719 unsigned int sector_sz = sdkp->device->sector_size;
2720 const int vpd_len = 64;
2721 unsigned char *buffer = kmalloc(vpd_len, GFP_KERNEL);
2722
2723 if (!buffer ||
2724 /* Block Limits VPD */
2725 scsi_get_vpd_page(sdkp->device, 0xb0, buffer, vpd_len))
2726 goto out;
2727
2728 blk_queue_io_min(sdkp->disk->queue,
2729 get_unaligned_be16(&buffer[6]) * sector_sz);
2730
2731 sdkp->max_xfer_blocks = get_unaligned_be32(&buffer[8]);
2732 sdkp->opt_xfer_blocks = get_unaligned_be32(&buffer[12]);
2733
2734 if (buffer[3] == 0x3c) {
2735 unsigned int lba_count, desc_count;
2736
2737 sdkp->max_ws_blocks = (u32)get_unaligned_be64(&buffer[36]);
2738
2739 if (!sdkp->lbpme)
2740 goto out;
2741
2742 lba_count = get_unaligned_be32(&buffer[20]);
2743 desc_count = get_unaligned_be32(&buffer[24]);
2744
2745 if (lba_count && desc_count)
2746 sdkp->max_unmap_blocks = lba_count;
2747
2748 sdkp->unmap_granularity = get_unaligned_be32(&buffer[28]);
2749
2750 if (buffer[32] & 0x80)
2751 sdkp->unmap_alignment =
2752 get_unaligned_be32(&buffer[32]) & ~(1 << 31);
2753
2754 if (!sdkp->lbpvpd) { /* LBP VPD page not provided */
2755
2756 if (sdkp->max_unmap_blocks)
2757 sd_config_discard(sdkp, SD_LBP_UNMAP);
2758 else
2759 sd_config_discard(sdkp, SD_LBP_WS16);
2760
2761 } else { /* LBP VPD page tells us what to use */
2762 if (sdkp->lbpu && sdkp->max_unmap_blocks && !sdkp->lbprz)
2763 sd_config_discard(sdkp, SD_LBP_UNMAP);
2764 else if (sdkp->lbpws)
2765 sd_config_discard(sdkp, SD_LBP_WS16);
2766 else if (sdkp->lbpws10)
2767 sd_config_discard(sdkp, SD_LBP_WS10);
2768 else if (sdkp->lbpu && sdkp->max_unmap_blocks)
2769 sd_config_discard(sdkp, SD_LBP_UNMAP);
2770 else
2771 sd_config_discard(sdkp, SD_LBP_DISABLE);
2772 }
2773 }
2774
2775 out:
2776 kfree(buffer);
2777}
2778
2779/**
2780 * sd_read_block_characteristics - Query block dev. characteristics
2781 * @disk: disk to query
2782 */
2783static void sd_read_block_characteristics(struct scsi_disk *sdkp)
2784{
2785 struct request_queue *q = sdkp->disk->queue;
2786 unsigned char *buffer;
2787 u16 rot;
2788 const int vpd_len = 64;
2789
2790 buffer = kmalloc(vpd_len, GFP_KERNEL);
2791
2792 if (!buffer ||
2793 /* Block Device Characteristics VPD */
2794 scsi_get_vpd_page(sdkp->device, 0xb1, buffer, vpd_len))
2795 goto out;
2796
2797 rot = get_unaligned_be16(&buffer[4]);
2798
2799 if (rot == 1) {
2800 queue_flag_set_unlocked(QUEUE_FLAG_NONROT, q);
2801 queue_flag_clear_unlocked(QUEUE_FLAG_ADD_RANDOM, q);
2802 }
2803
2804 if (sdkp->device->type == TYPE_ZBC) {
2805 /* Host-managed */
2806 q->limits.zoned = BLK_ZONED_HM;
2807 } else {
2808 sdkp->zoned = (buffer[8] >> 4) & 3;
2809 if (sdkp->zoned == 1)
2810 /* Host-aware */
2811 q->limits.zoned = BLK_ZONED_HA;
2812 else
2813 /*
2814 * Treat drive-managed devices as
2815 * regular block devices.
2816 */
2817 q->limits.zoned = BLK_ZONED_NONE;
2818 }
2819 if (blk_queue_is_zoned(q) && sdkp->first_scan)
2820 sd_printk(KERN_NOTICE, sdkp, "Host-%s zoned block device\n",
2821 q->limits.zoned == BLK_ZONED_HM ? "managed" : "aware");
2822
2823 out:
2824 kfree(buffer);
2825}
2826
2827/**
2828 * sd_read_block_provisioning - Query provisioning VPD page
2829 * @disk: disk to query
2830 */
2831static void sd_read_block_provisioning(struct scsi_disk *sdkp)
2832{
2833 unsigned char *buffer;
2834 const int vpd_len = 8;
2835
2836 if (sdkp->lbpme == 0)
2837 return;
2838
2839 buffer = kmalloc(vpd_len, GFP_KERNEL);
2840
2841 if (!buffer || scsi_get_vpd_page(sdkp->device, 0xb2, buffer, vpd_len))
2842 goto out;
2843
2844 sdkp->lbpvpd = 1;
2845 sdkp->lbpu = (buffer[5] >> 7) & 1; /* UNMAP */
2846 sdkp->lbpws = (buffer[5] >> 6) & 1; /* WRITE SAME(16) with UNMAP */
2847 sdkp->lbpws10 = (buffer[5] >> 5) & 1; /* WRITE SAME(10) with UNMAP */
2848
2849 out:
2850 kfree(buffer);
2851}
2852
2853static void sd_read_write_same(struct scsi_disk *sdkp, unsigned char *buffer)
2854{
2855 struct scsi_device *sdev = sdkp->device;
2856
2857 if (sdev->host->no_write_same) {
2858 sdev->no_write_same = 1;
2859
2860 return;
2861 }
2862
2863 if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE, INQUIRY) < 0) {
2864 /* too large values might cause issues with arcmsr */
2865 int vpd_buf_len = 64;
2866
2867 sdev->no_report_opcodes = 1;
2868
2869 /* Disable WRITE SAME if REPORT SUPPORTED OPERATION
2870 * CODES is unsupported and the device has an ATA
2871 * Information VPD page (SAT).
2872 */
2873 if (!scsi_get_vpd_page(sdev, 0x89, buffer, vpd_buf_len))
2874 sdev->no_write_same = 1;
2875 }
2876
2877 if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE, WRITE_SAME_16) == 1)
2878 sdkp->ws16 = 1;
2879
2880 if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE, WRITE_SAME) == 1)
2881 sdkp->ws10 = 1;
2882}
2883
2884/**
2885 * sd_revalidate_disk - called the first time a new disk is seen,
2886 * performs disk spin up, read_capacity, etc.
2887 * @disk: struct gendisk we care about
2888 **/
2889static int sd_revalidate_disk(struct gendisk *disk)
2890{
2891 struct scsi_disk *sdkp = scsi_disk(disk);
2892 struct scsi_device *sdp = sdkp->device;
2893 struct request_queue *q = sdkp->disk->queue;
2894 sector_t old_capacity = sdkp->capacity;
2895 unsigned char *buffer;
2896 unsigned int dev_max, rw_max;
2897
2898 SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp,
2899 "sd_revalidate_disk\n"));
2900
2901 /*
2902 * If the device is offline, don't try and read capacity or any
2903 * of the other niceties.
2904 */
2905 if (!scsi_device_online(sdp))
2906 goto out;
2907
2908 buffer = kmalloc(SD_BUF_SIZE, GFP_KERNEL);
2909 if (!buffer) {
2910 sd_printk(KERN_WARNING, sdkp, "sd_revalidate_disk: Memory "
2911 "allocation failure.\n");
2912 goto out;
2913 }
2914
2915 sd_spinup_disk(sdkp);
2916
2917 /*
2918 * Without media there is no reason to ask; moreover, some devices
2919 * react badly if we do.
2920 */
2921 if (sdkp->media_present) {
2922 sd_read_capacity(sdkp, buffer);
2923
2924 if (scsi_device_supports_vpd(sdp)) {
2925 sd_read_block_provisioning(sdkp);
2926 sd_read_block_limits(sdkp);
2927 sd_read_block_characteristics(sdkp);
2928 sd_zbc_read_zones(sdkp, buffer);
2929 }
2930
2931 sd_print_capacity(sdkp, old_capacity);
2932
2933 sd_read_write_protect_flag(sdkp, buffer);
2934 sd_read_cache_type(sdkp, buffer);
2935 sd_read_app_tag_own(sdkp, buffer);
2936 sd_read_write_same(sdkp, buffer);
2937 }
2938
2939 sdkp->first_scan = 0;
2940
2941 /*
2942 * We now have all cache related info, determine how we deal
2943 * with flush requests.
2944 */
2945 sd_set_flush_flag(sdkp);
2946
2947 /* Initial block count limit based on CDB TRANSFER LENGTH field size. */
2948 dev_max = sdp->use_16_for_rw ? SD_MAX_XFER_BLOCKS : SD_DEF_XFER_BLOCKS;
2949
2950 /* Some devices report a maximum block count for READ/WRITE requests. */
2951 dev_max = min_not_zero(dev_max, sdkp->max_xfer_blocks);
2952 q->limits.max_dev_sectors = logical_to_sectors(sdp, dev_max);
2953
2954 /*
2955 * Use the device's preferred I/O size for reads and writes
2956 * unless the reported value is unreasonably small, large, or
2957 * garbage.
2958 */
2959 if (sdkp->opt_xfer_blocks &&
2960 sdkp->opt_xfer_blocks <= dev_max &&
2961 sdkp->opt_xfer_blocks <= SD_DEF_XFER_BLOCKS &&
2962 logical_to_bytes(sdp, sdkp->opt_xfer_blocks) >= PAGE_SIZE) {
2963 q->limits.io_opt = logical_to_bytes(sdp, sdkp->opt_xfer_blocks);
2964 rw_max = logical_to_sectors(sdp, sdkp->opt_xfer_blocks);
2965 } else
2966 rw_max = BLK_DEF_MAX_SECTORS;
2967
2968 /* Combine with controller limits */
2969 q->limits.max_sectors = min(rw_max, queue_max_hw_sectors(q));
2970
2971 set_capacity(disk, logical_to_sectors(sdp, sdkp->capacity));
2972 sd_config_write_same(sdkp);
2973 kfree(buffer);
2974
2975 out:
2976 return 0;
2977}
2978
2979/**
2980 * sd_unlock_native_capacity - unlock native capacity
2981 * @disk: struct gendisk to set capacity for
2982 *
2983 * Block layer calls this function if it detects that partitions
2984 * on @disk reach beyond the end of the device. If the SCSI host
2985 * implements ->unlock_native_capacity() method, it's invoked to
2986 * give it a chance to adjust the device capacity.
2987 *
2988 * CONTEXT:
2989 * Defined by block layer. Might sleep.
2990 */
2991static void sd_unlock_native_capacity(struct gendisk *disk)
2992{
2993 struct scsi_device *sdev = scsi_disk(disk)->device;
2994
2995 if (sdev->host->hostt->unlock_native_capacity)
2996 sdev->host->hostt->unlock_native_capacity(sdev);
2997}
2998
2999/**
3000 * sd_format_disk_name - format disk name
3001 * @prefix: name prefix - ie. "sd" for SCSI disks
3002 * @index: index of the disk to format name for
3003 * @buf: output buffer
3004 * @buflen: length of the output buffer
3005 *
3006 * SCSI disk names starts at sda. The 26th device is sdz and the
3007 * 27th is sdaa. The last one for two lettered suffix is sdzz
3008 * which is followed by sdaaa.
3009 *
3010 * This is basically 26 base counting with one extra 'nil' entry
3011 * at the beginning from the second digit on and can be
3012 * determined using similar method as 26 base conversion with the
3013 * index shifted -1 after each digit is computed.
3014 *
3015 * CONTEXT:
3016 * Don't care.
3017 *
3018 * RETURNS:
3019 * 0 on success, -errno on failure.
3020 */
3021static int sd_format_disk_name(char *prefix, int index, char *buf, int buflen)
3022{
3023 const int base = 'z' - 'a' + 1;
3024 char *begin = buf + strlen(prefix);
3025 char *end = buf + buflen;
3026 char *p;
3027 int unit;
3028
3029 p = end - 1;
3030 *p = '\0';
3031 unit = base;
3032 do {
3033 if (p == begin)
3034 return -EINVAL;
3035 *--p = 'a' + (index % unit);
3036 index = (index / unit) - 1;
3037 } while (index >= 0);
3038
3039 memmove(begin, p, end - p);
3040 memcpy(buf, prefix, strlen(prefix));
3041
3042 return 0;
3043}
3044
3045/*
3046 * The asynchronous part of sd_probe
3047 */
3048static void sd_probe_async(void *data, async_cookie_t cookie)
3049{
3050 struct scsi_disk *sdkp = data;
3051 struct scsi_device *sdp;
3052 struct gendisk *gd;
3053 u32 index;
3054 struct device *dev;
3055
3056 sdp = sdkp->device;
3057 gd = sdkp->disk;
3058 index = sdkp->index;
3059 dev = &sdp->sdev_gendev;
3060
3061 gd->major = sd_major((index & 0xf0) >> 4);
3062 gd->first_minor = ((index & 0xf) << 4) | (index & 0xfff00);
3063 gd->minors = SD_MINORS;
3064
3065 gd->fops = &sd_fops;
3066 gd->private_data = &sdkp->driver;
3067 gd->queue = sdkp->device->request_queue;
3068
3069 /* defaults, until the device tells us otherwise */
3070 sdp->sector_size = 512;
3071 sdkp->capacity = 0;
3072 sdkp->media_present = 1;
3073 sdkp->write_prot = 0;
3074 sdkp->cache_override = 0;
3075 sdkp->WCE = 0;
3076 sdkp->RCD = 0;
3077 sdkp->ATO = 0;
3078 sdkp->first_scan = 1;
3079 sdkp->max_medium_access_timeouts = SD_MAX_MEDIUM_TIMEOUTS;
3080
3081 sd_revalidate_disk(gd);
3082
3083 gd->flags = GENHD_FL_EXT_DEVT;
3084 if (sdp->removable) {
3085 gd->flags |= GENHD_FL_REMOVABLE;
3086 gd->events |= DISK_EVENT_MEDIA_CHANGE;
3087 }
3088
3089 blk_pm_runtime_init(sdp->request_queue, dev);
3090 device_add_disk(dev, gd);
3091 if (sdkp->capacity)
3092 sd_dif_config_host(sdkp);
3093
3094 sd_revalidate_disk(gd);
3095
3096 sd_printk(KERN_NOTICE, sdkp, "Attached SCSI %sdisk\n",
3097 sdp->removable ? "removable " : "");
3098 scsi_autopm_put_device(sdp);
3099 put_device(&sdkp->dev);
3100}
3101
3102/**
3103 * sd_probe - called during driver initialization and whenever a
3104 * new scsi device is attached to the system. It is called once
3105 * for each scsi device (not just disks) present.
3106 * @dev: pointer to device object
3107 *
3108 * Returns 0 if successful (or not interested in this scsi device
3109 * (e.g. scanner)); 1 when there is an error.
3110 *
3111 * Note: this function is invoked from the scsi mid-level.
3112 * This function sets up the mapping between a given
3113 * <host,channel,id,lun> (found in sdp) and new device name
3114 * (e.g. /dev/sda). More precisely it is the block device major
3115 * and minor number that is chosen here.
3116 *
3117 * Assume sd_probe is not re-entrant (for time being)
3118 * Also think about sd_probe() and sd_remove() running coincidentally.
3119 **/
3120static int sd_probe(struct device *dev)
3121{
3122 struct scsi_device *sdp = to_scsi_device(dev);
3123 struct scsi_disk *sdkp;
3124 struct gendisk *gd;
3125 int index;
3126 int error;
3127
3128 scsi_autopm_get_device(sdp);
3129 error = -ENODEV;
3130 if (sdp->type != TYPE_DISK &&
3131 sdp->type != TYPE_ZBC &&
3132 sdp->type != TYPE_MOD &&
3133 sdp->type != TYPE_RBC)
3134 goto out;
3135
3136#ifndef CONFIG_BLK_DEV_ZONED
3137 if (sdp->type == TYPE_ZBC)
3138 goto out;
3139#endif
3140 SCSI_LOG_HLQUEUE(3, sdev_printk(KERN_INFO, sdp,
3141 "sd_probe\n"));
3142
3143 error = -ENOMEM;
3144 sdkp = kzalloc(sizeof(*sdkp), GFP_KERNEL);
3145 if (!sdkp)
3146 goto out;
3147
3148 gd = alloc_disk(SD_MINORS);
3149 if (!gd)
3150 goto out_free;
3151
3152 do {
3153 if (!ida_pre_get(&sd_index_ida, GFP_KERNEL))
3154 goto out_put;
3155
3156 spin_lock(&sd_index_lock);
3157 error = ida_get_new(&sd_index_ida, &index);
3158 spin_unlock(&sd_index_lock);
3159 } while (error == -EAGAIN);
3160
3161 if (error) {
3162 sdev_printk(KERN_WARNING, sdp, "sd_probe: memory exhausted.\n");
3163 goto out_put;
3164 }
3165
3166 error = sd_format_disk_name("sd", index, gd->disk_name, DISK_NAME_LEN);
3167 if (error) {
3168 sdev_printk(KERN_WARNING, sdp, "SCSI disk (sd) name length exceeded.\n");
3169 goto out_free_index;
3170 }
3171
3172 sdkp->device = sdp;
3173 sdkp->driver = &sd_template;
3174 sdkp->disk = gd;
3175 sdkp->index = index;
3176 atomic_set(&sdkp->openers, 0);
3177 atomic_set(&sdkp->device->ioerr_cnt, 0);
3178
3179 if (!sdp->request_queue->rq_timeout) {
3180 if (sdp->type != TYPE_MOD)
3181 blk_queue_rq_timeout(sdp->request_queue, SD_TIMEOUT);
3182 else
3183 blk_queue_rq_timeout(sdp->request_queue,
3184 SD_MOD_TIMEOUT);
3185 }
3186
3187 device_initialize(&sdkp->dev);
3188 sdkp->dev.parent = dev;
3189 sdkp->dev.class = &sd_disk_class;
3190 dev_set_name(&sdkp->dev, "%s", dev_name(dev));
3191
3192 error = device_add(&sdkp->dev);
3193 if (error)
3194 goto out_free_index;
3195
3196 get_device(dev);
3197 dev_set_drvdata(dev, sdkp);
3198
3199 get_device(&sdkp->dev); /* prevent release before async_schedule */
3200 async_schedule_domain(sd_probe_async, sdkp, &scsi_sd_probe_domain);
3201
3202 return 0;
3203
3204 out_free_index:
3205 spin_lock(&sd_index_lock);
3206 ida_remove(&sd_index_ida, index);
3207 spin_unlock(&sd_index_lock);
3208 out_put:
3209 put_disk(gd);
3210 out_free:
3211 kfree(sdkp);
3212 out:
3213 scsi_autopm_put_device(sdp);
3214 return error;
3215}
3216
3217/**
3218 * sd_remove - called whenever a scsi disk (previously recognized by
3219 * sd_probe) is detached from the system. It is called (potentially
3220 * multiple times) during sd module unload.
3221 * @sdp: pointer to mid level scsi device object
3222 *
3223 * Note: this function is invoked from the scsi mid-level.
3224 * This function potentially frees up a device name (e.g. /dev/sdc)
3225 * that could be re-used by a subsequent sd_probe().
3226 * This function is not called when the built-in sd driver is "exit-ed".
3227 **/
3228static int sd_remove(struct device *dev)
3229{
3230 struct scsi_disk *sdkp;
3231 dev_t devt;
3232
3233 sdkp = dev_get_drvdata(dev);
3234 devt = disk_devt(sdkp->disk);
3235 scsi_autopm_get_device(sdkp->device);
3236
3237 async_synchronize_full_domain(&scsi_sd_pm_domain);
3238 async_synchronize_full_domain(&scsi_sd_probe_domain);
3239 device_del(&sdkp->dev);
3240 del_gendisk(sdkp->disk);
3241 sd_shutdown(dev);
3242
3243 sd_zbc_remove(sdkp);
3244
3245 blk_register_region(devt, SD_MINORS, NULL,
3246 sd_default_probe, NULL, NULL);
3247
3248 mutex_lock(&sd_ref_mutex);
3249 dev_set_drvdata(dev, NULL);
3250 put_device(&sdkp->dev);
3251 mutex_unlock(&sd_ref_mutex);
3252
3253 return 0;
3254}
3255
3256/**
3257 * scsi_disk_release - Called to free the scsi_disk structure
3258 * @dev: pointer to embedded class device
3259 *
3260 * sd_ref_mutex must be held entering this routine. Because it is
3261 * called on last put, you should always use the scsi_disk_get()
3262 * scsi_disk_put() helpers which manipulate the semaphore directly
3263 * and never do a direct put_device.
3264 **/
3265static void scsi_disk_release(struct device *dev)
3266{
3267 struct scsi_disk *sdkp = to_scsi_disk(dev);
3268 struct gendisk *disk = sdkp->disk;
3269
3270 spin_lock(&sd_index_lock);
3271 ida_remove(&sd_index_ida, sdkp->index);
3272 spin_unlock(&sd_index_lock);
3273
3274 disk->private_data = NULL;
3275 put_disk(disk);
3276 put_device(&sdkp->device->sdev_gendev);
3277
3278 kfree(sdkp);
3279}
3280
3281static int sd_start_stop_device(struct scsi_disk *sdkp, int start)
3282{
3283 unsigned char cmd[6] = { START_STOP }; /* START_VALID */
3284 struct scsi_sense_hdr sshdr;
3285 struct scsi_device *sdp = sdkp->device;
3286 int res;
3287
3288 if (start)
3289 cmd[4] |= 1; /* START */
3290
3291 if (sdp->start_stop_pwr_cond)
3292 cmd[4] |= start ? 1 << 4 : 3 << 4; /* Active or Standby */
3293
3294 if (!scsi_device_online(sdp))
3295 return -ENODEV;
3296
3297 res = scsi_execute_req_flags(sdp, cmd, DMA_NONE, NULL, 0, &sshdr,
3298 SD_TIMEOUT, SD_MAX_RETRIES, NULL, 0, RQF_PM);
3299 if (res) {
3300 sd_print_result(sdkp, "Start/Stop Unit failed", res);
3301 if (driver_byte(res) & DRIVER_SENSE)
3302 sd_print_sense_hdr(sdkp, &sshdr);
3303 if (scsi_sense_valid(&sshdr) &&
3304 /* 0x3a is medium not present */
3305 sshdr.asc == 0x3a)
3306 res = 0;
3307 }
3308
3309 /* SCSI error codes must not go to the generic layer */
3310 if (res)
3311 return -EIO;
3312
3313 return 0;
3314}
3315
3316/*
3317 * Send a SYNCHRONIZE CACHE instruction down to the device through
3318 * the normal SCSI command structure. Wait for the command to
3319 * complete.
3320 */
3321static void sd_shutdown(struct device *dev)
3322{
3323 struct scsi_disk *sdkp = dev_get_drvdata(dev);
3324
3325 if (!sdkp)
3326 return; /* this can happen */
3327
3328 if (pm_runtime_suspended(dev))
3329 return;
3330
3331 if (sdkp->WCE && sdkp->media_present) {
3332 sd_printk(KERN_NOTICE, sdkp, "Synchronizing SCSI cache\n");
3333 sd_sync_cache(sdkp);
3334 }
3335
3336 if (system_state != SYSTEM_RESTART && sdkp->device->manage_start_stop) {
3337 sd_printk(KERN_NOTICE, sdkp, "Stopping disk\n");
3338 sd_start_stop_device(sdkp, 0);
3339 }
3340}
3341
3342static int sd_suspend_common(struct device *dev, bool ignore_stop_errors)
3343{
3344 struct scsi_disk *sdkp = dev_get_drvdata(dev);
3345 int ret = 0;
3346
3347 if (!sdkp) /* E.g.: runtime suspend following sd_remove() */
3348 return 0;
3349
3350 if (sdkp->WCE && sdkp->media_present) {
3351 sd_printk(KERN_NOTICE, sdkp, "Synchronizing SCSI cache\n");
3352 ret = sd_sync_cache(sdkp);
3353 if (ret) {
3354 /* ignore OFFLINE device */
3355 if (ret == -ENODEV)
3356 ret = 0;
3357 goto done;
3358 }
3359 }
3360
3361 if (sdkp->device->manage_start_stop) {
3362 sd_printk(KERN_NOTICE, sdkp, "Stopping disk\n");
3363 /* an error is not worth aborting a system sleep */
3364 ret = sd_start_stop_device(sdkp, 0);
3365 if (ignore_stop_errors)
3366 ret = 0;
3367 }
3368
3369done:
3370 return ret;
3371}
3372
3373static int sd_suspend_system(struct device *dev)
3374{
3375 return sd_suspend_common(dev, true);
3376}
3377
3378static int sd_suspend_runtime(struct device *dev)
3379{
3380 return sd_suspend_common(dev, false);
3381}
3382
3383static int sd_resume(struct device *dev)
3384{
3385 struct scsi_disk *sdkp = dev_get_drvdata(dev);
3386
3387 if (!sdkp) /* E.g.: runtime resume at the start of sd_probe() */
3388 return 0;
3389
3390 if (!sdkp->device->manage_start_stop)
3391 return 0;
3392
3393 sd_printk(KERN_NOTICE, sdkp, "Starting disk\n");
3394 return sd_start_stop_device(sdkp, 1);
3395}
3396
3397/**
3398 * init_sd - entry point for this driver (both when built in or when
3399 * a module).
3400 *
3401 * Note: this function registers this driver with the scsi mid-level.
3402 **/
3403static int __init init_sd(void)
3404{
3405 int majors = 0, i, err;
3406
3407 SCSI_LOG_HLQUEUE(3, printk("init_sd: sd driver entry point\n"));
3408
3409 for (i = 0; i < SD_MAJORS; i++) {
3410 if (register_blkdev(sd_major(i), "sd") != 0)
3411 continue;
3412 majors++;
3413 blk_register_region(sd_major(i), SD_MINORS, NULL,
3414 sd_default_probe, NULL, NULL);
3415 }
3416
3417 if (!majors)
3418 return -ENODEV;
3419
3420 err = class_register(&sd_disk_class);
3421 if (err)
3422 goto err_out;
3423
3424 sd_cdb_cache = kmem_cache_create("sd_ext_cdb", SD_EXT_CDB_SIZE,
3425 0, 0, NULL);
3426 if (!sd_cdb_cache) {
3427 printk(KERN_ERR "sd: can't init extended cdb cache\n");
3428 err = -ENOMEM;
3429 goto err_out_class;
3430 }
3431
3432 sd_cdb_pool = mempool_create_slab_pool(SD_MEMPOOL_SIZE, sd_cdb_cache);
3433 if (!sd_cdb_pool) {
3434 printk(KERN_ERR "sd: can't init extended cdb pool\n");
3435 err = -ENOMEM;
3436 goto err_out_cache;
3437 }
3438
3439 err = scsi_register_driver(&sd_template.gendrv);
3440 if (err)
3441 goto err_out_driver;
3442
3443 return 0;
3444
3445err_out_driver:
3446 mempool_destroy(sd_cdb_pool);
3447
3448err_out_cache:
3449 kmem_cache_destroy(sd_cdb_cache);
3450
3451err_out_class:
3452 class_unregister(&sd_disk_class);
3453err_out:
3454 for (i = 0; i < SD_MAJORS; i++)
3455 unregister_blkdev(sd_major(i), "sd");
3456 return err;
3457}
3458
3459/**
3460 * exit_sd - exit point for this driver (when it is a module).
3461 *
3462 * Note: this function unregisters this driver from the scsi mid-level.
3463 **/
3464static void __exit exit_sd(void)
3465{
3466 int i;
3467
3468 SCSI_LOG_HLQUEUE(3, printk("exit_sd: exiting sd driver\n"));
3469
3470 scsi_unregister_driver(&sd_template.gendrv);
3471 mempool_destroy(sd_cdb_pool);
3472 kmem_cache_destroy(sd_cdb_cache);
3473
3474 class_unregister(&sd_disk_class);
3475
3476 for (i = 0; i < SD_MAJORS; i++) {
3477 blk_unregister_region(sd_major(i), SD_MINORS);
3478 unregister_blkdev(sd_major(i), "sd");
3479 }
3480}
3481
3482module_init(init_sd);
3483module_exit(exit_sd);
3484
3485static void sd_print_sense_hdr(struct scsi_disk *sdkp,
3486 struct scsi_sense_hdr *sshdr)
3487{
3488 scsi_print_sense_hdr(sdkp->device,
3489 sdkp->disk ? sdkp->disk->disk_name : NULL, sshdr);
3490}
3491
3492static void sd_print_result(const struct scsi_disk *sdkp, const char *msg,
3493 int result)
3494{
3495 const char *hb_string = scsi_hostbyte_string(result);
3496 const char *db_string = scsi_driverbyte_string(result);
3497
3498 if (hb_string || db_string)
3499 sd_printk(KERN_INFO, sdkp,
3500 "%s: Result: hostbyte=%s driverbyte=%s\n", msg,
3501 hb_string ? hb_string : "invalid",
3502 db_string ? db_string : "invalid");
3503 else
3504 sd_printk(KERN_INFO, sdkp,
3505 "%s: Result: hostbyte=0x%02x driverbyte=0x%02x\n",
3506 msg, host_byte(result), driver_byte(result));
3507}
3508
1// SPDX-License-Identifier: GPL-2.0-only
2/*
3 * sd.c Copyright (C) 1992 Drew Eckhardt
4 * Copyright (C) 1993, 1994, 1995, 1999 Eric Youngdale
5 *
6 * Linux scsi disk driver
7 * Initial versions: Drew Eckhardt
8 * Subsequent revisions: Eric Youngdale
9 * Modification history:
10 * - Drew Eckhardt <drew@colorado.edu> original
11 * - Eric Youngdale <eric@andante.org> add scatter-gather, multiple
12 * outstanding request, and other enhancements.
13 * Support loadable low-level scsi drivers.
14 * - Jirka Hanika <geo@ff.cuni.cz> support more scsi disks using
15 * eight major numbers.
16 * - Richard Gooch <rgooch@atnf.csiro.au> support devfs.
17 * - Torben Mathiasen <tmm@image.dk> Resource allocation fixes in
18 * sd_init and cleanups.
19 * - Alex Davis <letmein@erols.com> Fix problem where partition info
20 * not being read in sd_open. Fix problem where removable media
21 * could be ejected after sd_open.
22 * - Douglas Gilbert <dgilbert@interlog.com> cleanup for lk 2.5.x
23 * - Badari Pulavarty <pbadari@us.ibm.com>, Matthew Wilcox
24 * <willy@debian.org>, Kurt Garloff <garloff@suse.de>:
25 * Support 32k/1M disks.
26 *
27 * Logging policy (needs CONFIG_SCSI_LOGGING defined):
28 * - setting up transfer: SCSI_LOG_HLQUEUE levels 1 and 2
29 * - end of transfer (bh + scsi_lib): SCSI_LOG_HLCOMPLETE level 1
30 * - entering sd_ioctl: SCSI_LOG_IOCTL level 1
31 * - entering other commands: SCSI_LOG_HLQUEUE level 3
32 * Note: when the logging level is set by the user, it must be greater
33 * than the level indicated above to trigger output.
34 */
35
36#include <linux/module.h>
37#include <linux/fs.h>
38#include <linux/kernel.h>
39#include <linux/mm.h>
40#include <linux/bio.h>
41#include <linux/genhd.h>
42#include <linux/hdreg.h>
43#include <linux/errno.h>
44#include <linux/idr.h>
45#include <linux/interrupt.h>
46#include <linux/init.h>
47#include <linux/blkdev.h>
48#include <linux/blkpg.h>
49#include <linux/blk-pm.h>
50#include <linux/delay.h>
51#include <linux/mutex.h>
52#include <linux/string_helpers.h>
53#include <linux/async.h>
54#include <linux/slab.h>
55#include <linux/sed-opal.h>
56#include <linux/pm_runtime.h>
57#include <linux/pr.h>
58#include <linux/t10-pi.h>
59#include <linux/uaccess.h>
60#include <asm/unaligned.h>
61
62#include <scsi/scsi.h>
63#include <scsi/scsi_cmnd.h>
64#include <scsi/scsi_dbg.h>
65#include <scsi/scsi_device.h>
66#include <scsi/scsi_driver.h>
67#include <scsi/scsi_eh.h>
68#include <scsi/scsi_host.h>
69#include <scsi/scsi_ioctl.h>
70#include <scsi/scsicam.h>
71
72#include "sd.h"
73#include "scsi_priv.h"
74#include "scsi_logging.h"
75
76MODULE_AUTHOR("Eric Youngdale");
77MODULE_DESCRIPTION("SCSI disk (sd) driver");
78MODULE_LICENSE("GPL");
79
80MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK0_MAJOR);
81MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK1_MAJOR);
82MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK2_MAJOR);
83MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK3_MAJOR);
84MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK4_MAJOR);
85MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK5_MAJOR);
86MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK6_MAJOR);
87MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK7_MAJOR);
88MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK8_MAJOR);
89MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK9_MAJOR);
90MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK10_MAJOR);
91MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK11_MAJOR);
92MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK12_MAJOR);
93MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK13_MAJOR);
94MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK14_MAJOR);
95MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK15_MAJOR);
96MODULE_ALIAS_SCSI_DEVICE(TYPE_DISK);
97MODULE_ALIAS_SCSI_DEVICE(TYPE_MOD);
98MODULE_ALIAS_SCSI_DEVICE(TYPE_RBC);
99MODULE_ALIAS_SCSI_DEVICE(TYPE_ZBC);
100
101#if !defined(CONFIG_DEBUG_BLOCK_EXT_DEVT)
102#define SD_MINORS 16
103#else
104#define SD_MINORS 0
105#endif
106
107static void sd_config_discard(struct scsi_disk *, unsigned int);
108static void sd_config_write_same(struct scsi_disk *);
109static int sd_revalidate_disk(struct gendisk *);
110static void sd_unlock_native_capacity(struct gendisk *disk);
111static int sd_probe(struct device *);
112static int sd_remove(struct device *);
113static void sd_shutdown(struct device *);
114static int sd_suspend_system(struct device *);
115static int sd_suspend_runtime(struct device *);
116static int sd_resume(struct device *);
117static void sd_rescan(struct device *);
118static blk_status_t sd_init_command(struct scsi_cmnd *SCpnt);
119static void sd_uninit_command(struct scsi_cmnd *SCpnt);
120static int sd_done(struct scsi_cmnd *);
121static void sd_eh_reset(struct scsi_cmnd *);
122static int sd_eh_action(struct scsi_cmnd *, int);
123static void sd_read_capacity(struct scsi_disk *sdkp, unsigned char *buffer);
124static void scsi_disk_release(struct device *cdev);
125static void sd_print_sense_hdr(struct scsi_disk *, struct scsi_sense_hdr *);
126static void sd_print_result(const struct scsi_disk *, const char *, int);
127
128static DEFINE_IDA(sd_index_ida);
129
130/* This semaphore is used to mediate the 0->1 reference get in the
131 * face of object destruction (i.e. we can't allow a get on an
132 * object after last put) */
133static DEFINE_MUTEX(sd_ref_mutex);
134
135static struct kmem_cache *sd_cdb_cache;
136static mempool_t *sd_cdb_pool;
137static mempool_t *sd_page_pool;
138
139static const char *sd_cache_types[] = {
140 "write through", "none", "write back",
141 "write back, no read (daft)"
142};
143
144static void sd_set_flush_flag(struct scsi_disk *sdkp)
145{
146 bool wc = false, fua = false;
147
148 if (sdkp->WCE) {
149 wc = true;
150 if (sdkp->DPOFUA)
151 fua = true;
152 }
153
154 blk_queue_write_cache(sdkp->disk->queue, wc, fua);
155}
156
157static ssize_t
158cache_type_store(struct device *dev, struct device_attribute *attr,
159 const char *buf, size_t count)
160{
161 int ct, rcd, wce, sp;
162 struct scsi_disk *sdkp = to_scsi_disk(dev);
163 struct scsi_device *sdp = sdkp->device;
164 char buffer[64];
165 char *buffer_data;
166 struct scsi_mode_data data;
167 struct scsi_sense_hdr sshdr;
168 static const char temp[] = "temporary ";
169 int len;
170
171 if (sdp->type != TYPE_DISK && sdp->type != TYPE_ZBC)
172 /* no cache control on RBC devices; theoretically they
173 * can do it, but there's probably so many exceptions
174 * it's not worth the risk */
175 return -EINVAL;
176
177 if (strncmp(buf, temp, sizeof(temp) - 1) == 0) {
178 buf += sizeof(temp) - 1;
179 sdkp->cache_override = 1;
180 } else {
181 sdkp->cache_override = 0;
182 }
183
184 ct = sysfs_match_string(sd_cache_types, buf);
185 if (ct < 0)
186 return -EINVAL;
187
188 rcd = ct & 0x01 ? 1 : 0;
189 wce = (ct & 0x02) && !sdkp->write_prot ? 1 : 0;
190
191 if (sdkp->cache_override) {
192 sdkp->WCE = wce;
193 sdkp->RCD = rcd;
194 sd_set_flush_flag(sdkp);
195 return count;
196 }
197
198 if (scsi_mode_sense(sdp, 0x08, 8, buffer, sizeof(buffer), SD_TIMEOUT,
199 SD_MAX_RETRIES, &data, NULL))
200 return -EINVAL;
201 len = min_t(size_t, sizeof(buffer), data.length - data.header_length -
202 data.block_descriptor_length);
203 buffer_data = buffer + data.header_length +
204 data.block_descriptor_length;
205 buffer_data[2] &= ~0x05;
206 buffer_data[2] |= wce << 2 | rcd;
207 sp = buffer_data[0] & 0x80 ? 1 : 0;
208 buffer_data[0] &= ~0x80;
209
210 /*
211 * Ensure WP, DPOFUA, and RESERVED fields are cleared in
212 * received mode parameter buffer before doing MODE SELECT.
213 */
214 data.device_specific = 0;
215
216 if (scsi_mode_select(sdp, 1, sp, 8, buffer_data, len, SD_TIMEOUT,
217 SD_MAX_RETRIES, &data, &sshdr)) {
218 if (scsi_sense_valid(&sshdr))
219 sd_print_sense_hdr(sdkp, &sshdr);
220 return -EINVAL;
221 }
222 revalidate_disk(sdkp->disk);
223 return count;
224}
225
226static ssize_t
227manage_start_stop_show(struct device *dev, struct device_attribute *attr,
228 char *buf)
229{
230 struct scsi_disk *sdkp = to_scsi_disk(dev);
231 struct scsi_device *sdp = sdkp->device;
232
233 return sprintf(buf, "%u\n", sdp->manage_start_stop);
234}
235
236static ssize_t
237manage_start_stop_store(struct device *dev, struct device_attribute *attr,
238 const char *buf, size_t count)
239{
240 struct scsi_disk *sdkp = to_scsi_disk(dev);
241 struct scsi_device *sdp = sdkp->device;
242 bool v;
243
244 if (!capable(CAP_SYS_ADMIN))
245 return -EACCES;
246
247 if (kstrtobool(buf, &v))
248 return -EINVAL;
249
250 sdp->manage_start_stop = v;
251
252 return count;
253}
254static DEVICE_ATTR_RW(manage_start_stop);
255
256static ssize_t
257allow_restart_show(struct device *dev, struct device_attribute *attr, char *buf)
258{
259 struct scsi_disk *sdkp = to_scsi_disk(dev);
260
261 return sprintf(buf, "%u\n", sdkp->device->allow_restart);
262}
263
264static ssize_t
265allow_restart_store(struct device *dev, struct device_attribute *attr,
266 const char *buf, size_t count)
267{
268 bool v;
269 struct scsi_disk *sdkp = to_scsi_disk(dev);
270 struct scsi_device *sdp = sdkp->device;
271
272 if (!capable(CAP_SYS_ADMIN))
273 return -EACCES;
274
275 if (sdp->type != TYPE_DISK && sdp->type != TYPE_ZBC)
276 return -EINVAL;
277
278 if (kstrtobool(buf, &v))
279 return -EINVAL;
280
281 sdp->allow_restart = v;
282
283 return count;
284}
285static DEVICE_ATTR_RW(allow_restart);
286
287static ssize_t
288cache_type_show(struct device *dev, struct device_attribute *attr, char *buf)
289{
290 struct scsi_disk *sdkp = to_scsi_disk(dev);
291 int ct = sdkp->RCD + 2*sdkp->WCE;
292
293 return sprintf(buf, "%s\n", sd_cache_types[ct]);
294}
295static DEVICE_ATTR_RW(cache_type);
296
297static ssize_t
298FUA_show(struct device *dev, struct device_attribute *attr, char *buf)
299{
300 struct scsi_disk *sdkp = to_scsi_disk(dev);
301
302 return sprintf(buf, "%u\n", sdkp->DPOFUA);
303}
304static DEVICE_ATTR_RO(FUA);
305
306static ssize_t
307protection_type_show(struct device *dev, struct device_attribute *attr,
308 char *buf)
309{
310 struct scsi_disk *sdkp = to_scsi_disk(dev);
311
312 return sprintf(buf, "%u\n", sdkp->protection_type);
313}
314
315static ssize_t
316protection_type_store(struct device *dev, struct device_attribute *attr,
317 const char *buf, size_t count)
318{
319 struct scsi_disk *sdkp = to_scsi_disk(dev);
320 unsigned int val;
321 int err;
322
323 if (!capable(CAP_SYS_ADMIN))
324 return -EACCES;
325
326 err = kstrtouint(buf, 10, &val);
327
328 if (err)
329 return err;
330
331 if (val <= T10_PI_TYPE3_PROTECTION)
332 sdkp->protection_type = val;
333
334 return count;
335}
336static DEVICE_ATTR_RW(protection_type);
337
338static ssize_t
339protection_mode_show(struct device *dev, struct device_attribute *attr,
340 char *buf)
341{
342 struct scsi_disk *sdkp = to_scsi_disk(dev);
343 struct scsi_device *sdp = sdkp->device;
344 unsigned int dif, dix;
345
346 dif = scsi_host_dif_capable(sdp->host, sdkp->protection_type);
347 dix = scsi_host_dix_capable(sdp->host, sdkp->protection_type);
348
349 if (!dix && scsi_host_dix_capable(sdp->host, T10_PI_TYPE0_PROTECTION)) {
350 dif = 0;
351 dix = 1;
352 }
353
354 if (!dif && !dix)
355 return sprintf(buf, "none\n");
356
357 return sprintf(buf, "%s%u\n", dix ? "dix" : "dif", dif);
358}
359static DEVICE_ATTR_RO(protection_mode);
360
361static ssize_t
362app_tag_own_show(struct device *dev, struct device_attribute *attr, char *buf)
363{
364 struct scsi_disk *sdkp = to_scsi_disk(dev);
365
366 return sprintf(buf, "%u\n", sdkp->ATO);
367}
368static DEVICE_ATTR_RO(app_tag_own);
369
370static ssize_t
371thin_provisioning_show(struct device *dev, struct device_attribute *attr,
372 char *buf)
373{
374 struct scsi_disk *sdkp = to_scsi_disk(dev);
375
376 return sprintf(buf, "%u\n", sdkp->lbpme);
377}
378static DEVICE_ATTR_RO(thin_provisioning);
379
380/* sysfs_match_string() requires dense arrays */
381static const char *lbp_mode[] = {
382 [SD_LBP_FULL] = "full",
383 [SD_LBP_UNMAP] = "unmap",
384 [SD_LBP_WS16] = "writesame_16",
385 [SD_LBP_WS10] = "writesame_10",
386 [SD_LBP_ZERO] = "writesame_zero",
387 [SD_LBP_DISABLE] = "disabled",
388};
389
390static ssize_t
391provisioning_mode_show(struct device *dev, struct device_attribute *attr,
392 char *buf)
393{
394 struct scsi_disk *sdkp = to_scsi_disk(dev);
395
396 return sprintf(buf, "%s\n", lbp_mode[sdkp->provisioning_mode]);
397}
398
399static ssize_t
400provisioning_mode_store(struct device *dev, struct device_attribute *attr,
401 const char *buf, size_t count)
402{
403 struct scsi_disk *sdkp = to_scsi_disk(dev);
404 struct scsi_device *sdp = sdkp->device;
405 int mode;
406
407 if (!capable(CAP_SYS_ADMIN))
408 return -EACCES;
409
410 if (sd_is_zoned(sdkp)) {
411 sd_config_discard(sdkp, SD_LBP_DISABLE);
412 return count;
413 }
414
415 if (sdp->type != TYPE_DISK)
416 return -EINVAL;
417
418 mode = sysfs_match_string(lbp_mode, buf);
419 if (mode < 0)
420 return -EINVAL;
421
422 sd_config_discard(sdkp, mode);
423
424 return count;
425}
426static DEVICE_ATTR_RW(provisioning_mode);
427
428/* sysfs_match_string() requires dense arrays */
429static const char *zeroing_mode[] = {
430 [SD_ZERO_WRITE] = "write",
431 [SD_ZERO_WS] = "writesame",
432 [SD_ZERO_WS16_UNMAP] = "writesame_16_unmap",
433 [SD_ZERO_WS10_UNMAP] = "writesame_10_unmap",
434};
435
436static ssize_t
437zeroing_mode_show(struct device *dev, struct device_attribute *attr,
438 char *buf)
439{
440 struct scsi_disk *sdkp = to_scsi_disk(dev);
441
442 return sprintf(buf, "%s\n", zeroing_mode[sdkp->zeroing_mode]);
443}
444
445static ssize_t
446zeroing_mode_store(struct device *dev, struct device_attribute *attr,
447 const char *buf, size_t count)
448{
449 struct scsi_disk *sdkp = to_scsi_disk(dev);
450 int mode;
451
452 if (!capable(CAP_SYS_ADMIN))
453 return -EACCES;
454
455 mode = sysfs_match_string(zeroing_mode, buf);
456 if (mode < 0)
457 return -EINVAL;
458
459 sdkp->zeroing_mode = mode;
460
461 return count;
462}
463static DEVICE_ATTR_RW(zeroing_mode);
464
465static ssize_t
466max_medium_access_timeouts_show(struct device *dev,
467 struct device_attribute *attr, char *buf)
468{
469 struct scsi_disk *sdkp = to_scsi_disk(dev);
470
471 return sprintf(buf, "%u\n", sdkp->max_medium_access_timeouts);
472}
473
474static ssize_t
475max_medium_access_timeouts_store(struct device *dev,
476 struct device_attribute *attr, const char *buf,
477 size_t count)
478{
479 struct scsi_disk *sdkp = to_scsi_disk(dev);
480 int err;
481
482 if (!capable(CAP_SYS_ADMIN))
483 return -EACCES;
484
485 err = kstrtouint(buf, 10, &sdkp->max_medium_access_timeouts);
486
487 return err ? err : count;
488}
489static DEVICE_ATTR_RW(max_medium_access_timeouts);
490
491static ssize_t
492max_write_same_blocks_show(struct device *dev, struct device_attribute *attr,
493 char *buf)
494{
495 struct scsi_disk *sdkp = to_scsi_disk(dev);
496
497 return sprintf(buf, "%u\n", sdkp->max_ws_blocks);
498}
499
500static ssize_t
501max_write_same_blocks_store(struct device *dev, struct device_attribute *attr,
502 const char *buf, size_t count)
503{
504 struct scsi_disk *sdkp = to_scsi_disk(dev);
505 struct scsi_device *sdp = sdkp->device;
506 unsigned long max;
507 int err;
508
509 if (!capable(CAP_SYS_ADMIN))
510 return -EACCES;
511
512 if (sdp->type != TYPE_DISK && sdp->type != TYPE_ZBC)
513 return -EINVAL;
514
515 err = kstrtoul(buf, 10, &max);
516
517 if (err)
518 return err;
519
520 if (max == 0)
521 sdp->no_write_same = 1;
522 else if (max <= SD_MAX_WS16_BLOCKS) {
523 sdp->no_write_same = 0;
524 sdkp->max_ws_blocks = max;
525 }
526
527 sd_config_write_same(sdkp);
528
529 return count;
530}
531static DEVICE_ATTR_RW(max_write_same_blocks);
532
533static struct attribute *sd_disk_attrs[] = {
534 &dev_attr_cache_type.attr,
535 &dev_attr_FUA.attr,
536 &dev_attr_allow_restart.attr,
537 &dev_attr_manage_start_stop.attr,
538 &dev_attr_protection_type.attr,
539 &dev_attr_protection_mode.attr,
540 &dev_attr_app_tag_own.attr,
541 &dev_attr_thin_provisioning.attr,
542 &dev_attr_provisioning_mode.attr,
543 &dev_attr_zeroing_mode.attr,
544 &dev_attr_max_write_same_blocks.attr,
545 &dev_attr_max_medium_access_timeouts.attr,
546 NULL,
547};
548ATTRIBUTE_GROUPS(sd_disk);
549
550static struct class sd_disk_class = {
551 .name = "scsi_disk",
552 .owner = THIS_MODULE,
553 .dev_release = scsi_disk_release,
554 .dev_groups = sd_disk_groups,
555};
556
557static const struct dev_pm_ops sd_pm_ops = {
558 .suspend = sd_suspend_system,
559 .resume = sd_resume,
560 .poweroff = sd_suspend_system,
561 .restore = sd_resume,
562 .runtime_suspend = sd_suspend_runtime,
563 .runtime_resume = sd_resume,
564};
565
566static struct scsi_driver sd_template = {
567 .gendrv = {
568 .name = "sd",
569 .owner = THIS_MODULE,
570 .probe = sd_probe,
571 .probe_type = PROBE_PREFER_ASYNCHRONOUS,
572 .remove = sd_remove,
573 .shutdown = sd_shutdown,
574 .pm = &sd_pm_ops,
575 },
576 .rescan = sd_rescan,
577 .init_command = sd_init_command,
578 .uninit_command = sd_uninit_command,
579 .done = sd_done,
580 .eh_action = sd_eh_action,
581 .eh_reset = sd_eh_reset,
582};
583
584/*
585 * Dummy kobj_map->probe function.
586 * The default ->probe function will call modprobe, which is
587 * pointless as this module is already loaded.
588 */
589static struct kobject *sd_default_probe(dev_t devt, int *partno, void *data)
590{
591 return NULL;
592}
593
594/*
595 * Device no to disk mapping:
596 *
597 * major disc2 disc p1
598 * |............|.............|....|....| <- dev_t
599 * 31 20 19 8 7 4 3 0
600 *
601 * Inside a major, we have 16k disks, however mapped non-
602 * contiguously. The first 16 disks are for major0, the next
603 * ones with major1, ... Disk 256 is for major0 again, disk 272
604 * for major1, ...
605 * As we stay compatible with our numbering scheme, we can reuse
606 * the well-know SCSI majors 8, 65--71, 136--143.
607 */
608static int sd_major(int major_idx)
609{
610 switch (major_idx) {
611 case 0:
612 return SCSI_DISK0_MAJOR;
613 case 1 ... 7:
614 return SCSI_DISK1_MAJOR + major_idx - 1;
615 case 8 ... 15:
616 return SCSI_DISK8_MAJOR + major_idx - 8;
617 default:
618 BUG();
619 return 0; /* shut up gcc */
620 }
621}
622
623static struct scsi_disk *scsi_disk_get(struct gendisk *disk)
624{
625 struct scsi_disk *sdkp = NULL;
626
627 mutex_lock(&sd_ref_mutex);
628
629 if (disk->private_data) {
630 sdkp = scsi_disk(disk);
631 if (scsi_device_get(sdkp->device) == 0)
632 get_device(&sdkp->dev);
633 else
634 sdkp = NULL;
635 }
636 mutex_unlock(&sd_ref_mutex);
637 return sdkp;
638}
639
640static void scsi_disk_put(struct scsi_disk *sdkp)
641{
642 struct scsi_device *sdev = sdkp->device;
643
644 mutex_lock(&sd_ref_mutex);
645 put_device(&sdkp->dev);
646 scsi_device_put(sdev);
647 mutex_unlock(&sd_ref_mutex);
648}
649
650#ifdef CONFIG_BLK_SED_OPAL
651static int sd_sec_submit(void *data, u16 spsp, u8 secp, void *buffer,
652 size_t len, bool send)
653{
654 struct scsi_device *sdev = data;
655 u8 cdb[12] = { 0, };
656 int ret;
657
658 cdb[0] = send ? SECURITY_PROTOCOL_OUT : SECURITY_PROTOCOL_IN;
659 cdb[1] = secp;
660 put_unaligned_be16(spsp, &cdb[2]);
661 put_unaligned_be32(len, &cdb[6]);
662
663 ret = scsi_execute_req(sdev, cdb,
664 send ? DMA_TO_DEVICE : DMA_FROM_DEVICE,
665 buffer, len, NULL, SD_TIMEOUT, SD_MAX_RETRIES, NULL);
666 return ret <= 0 ? ret : -EIO;
667}
668#endif /* CONFIG_BLK_SED_OPAL */
669
670/*
671 * Look up the DIX operation based on whether the command is read or
672 * write and whether dix and dif are enabled.
673 */
674static unsigned int sd_prot_op(bool write, bool dix, bool dif)
675{
676 /* Lookup table: bit 2 (write), bit 1 (dix), bit 0 (dif) */
677 static const unsigned int ops[] = { /* wrt dix dif */
678 SCSI_PROT_NORMAL, /* 0 0 0 */
679 SCSI_PROT_READ_STRIP, /* 0 0 1 */
680 SCSI_PROT_READ_INSERT, /* 0 1 0 */
681 SCSI_PROT_READ_PASS, /* 0 1 1 */
682 SCSI_PROT_NORMAL, /* 1 0 0 */
683 SCSI_PROT_WRITE_INSERT, /* 1 0 1 */
684 SCSI_PROT_WRITE_STRIP, /* 1 1 0 */
685 SCSI_PROT_WRITE_PASS, /* 1 1 1 */
686 };
687
688 return ops[write << 2 | dix << 1 | dif];
689}
690
691/*
692 * Returns a mask of the protection flags that are valid for a given DIX
693 * operation.
694 */
695static unsigned int sd_prot_flag_mask(unsigned int prot_op)
696{
697 static const unsigned int flag_mask[] = {
698 [SCSI_PROT_NORMAL] = 0,
699
700 [SCSI_PROT_READ_STRIP] = SCSI_PROT_TRANSFER_PI |
701 SCSI_PROT_GUARD_CHECK |
702 SCSI_PROT_REF_CHECK |
703 SCSI_PROT_REF_INCREMENT,
704
705 [SCSI_PROT_READ_INSERT] = SCSI_PROT_REF_INCREMENT |
706 SCSI_PROT_IP_CHECKSUM,
707
708 [SCSI_PROT_READ_PASS] = SCSI_PROT_TRANSFER_PI |
709 SCSI_PROT_GUARD_CHECK |
710 SCSI_PROT_REF_CHECK |
711 SCSI_PROT_REF_INCREMENT |
712 SCSI_PROT_IP_CHECKSUM,
713
714 [SCSI_PROT_WRITE_INSERT] = SCSI_PROT_TRANSFER_PI |
715 SCSI_PROT_REF_INCREMENT,
716
717 [SCSI_PROT_WRITE_STRIP] = SCSI_PROT_GUARD_CHECK |
718 SCSI_PROT_REF_CHECK |
719 SCSI_PROT_REF_INCREMENT |
720 SCSI_PROT_IP_CHECKSUM,
721
722 [SCSI_PROT_WRITE_PASS] = SCSI_PROT_TRANSFER_PI |
723 SCSI_PROT_GUARD_CHECK |
724 SCSI_PROT_REF_CHECK |
725 SCSI_PROT_REF_INCREMENT |
726 SCSI_PROT_IP_CHECKSUM,
727 };
728
729 return flag_mask[prot_op];
730}
731
732static unsigned char sd_setup_protect_cmnd(struct scsi_cmnd *scmd,
733 unsigned int dix, unsigned int dif)
734{
735 struct bio *bio = scmd->request->bio;
736 unsigned int prot_op = sd_prot_op(rq_data_dir(scmd->request), dix, dif);
737 unsigned int protect = 0;
738
739 if (dix) { /* DIX Type 0, 1, 2, 3 */
740 if (bio_integrity_flagged(bio, BIP_IP_CHECKSUM))
741 scmd->prot_flags |= SCSI_PROT_IP_CHECKSUM;
742
743 if (bio_integrity_flagged(bio, BIP_CTRL_NOCHECK) == false)
744 scmd->prot_flags |= SCSI_PROT_GUARD_CHECK;
745 }
746
747 if (dif != T10_PI_TYPE3_PROTECTION) { /* DIX/DIF Type 0, 1, 2 */
748 scmd->prot_flags |= SCSI_PROT_REF_INCREMENT;
749
750 if (bio_integrity_flagged(bio, BIP_CTRL_NOCHECK) == false)
751 scmd->prot_flags |= SCSI_PROT_REF_CHECK;
752 }
753
754 if (dif) { /* DIX/DIF Type 1, 2, 3 */
755 scmd->prot_flags |= SCSI_PROT_TRANSFER_PI;
756
757 if (bio_integrity_flagged(bio, BIP_DISK_NOCHECK))
758 protect = 3 << 5; /* Disable target PI checking */
759 else
760 protect = 1 << 5; /* Enable target PI checking */
761 }
762
763 scsi_set_prot_op(scmd, prot_op);
764 scsi_set_prot_type(scmd, dif);
765 scmd->prot_flags &= sd_prot_flag_mask(prot_op);
766
767 return protect;
768}
769
770static void sd_config_discard(struct scsi_disk *sdkp, unsigned int mode)
771{
772 struct request_queue *q = sdkp->disk->queue;
773 unsigned int logical_block_size = sdkp->device->sector_size;
774 unsigned int max_blocks = 0;
775
776 q->limits.discard_alignment =
777 sdkp->unmap_alignment * logical_block_size;
778 q->limits.discard_granularity =
779 max(sdkp->physical_block_size,
780 sdkp->unmap_granularity * logical_block_size);
781 sdkp->provisioning_mode = mode;
782
783 switch (mode) {
784
785 case SD_LBP_FULL:
786 case SD_LBP_DISABLE:
787 blk_queue_max_discard_sectors(q, 0);
788 blk_queue_flag_clear(QUEUE_FLAG_DISCARD, q);
789 return;
790
791 case SD_LBP_UNMAP:
792 max_blocks = min_not_zero(sdkp->max_unmap_blocks,
793 (u32)SD_MAX_WS16_BLOCKS);
794 break;
795
796 case SD_LBP_WS16:
797 if (sdkp->device->unmap_limit_for_ws)
798 max_blocks = sdkp->max_unmap_blocks;
799 else
800 max_blocks = sdkp->max_ws_blocks;
801
802 max_blocks = min_not_zero(max_blocks, (u32)SD_MAX_WS16_BLOCKS);
803 break;
804
805 case SD_LBP_WS10:
806 if (sdkp->device->unmap_limit_for_ws)
807 max_blocks = sdkp->max_unmap_blocks;
808 else
809 max_blocks = sdkp->max_ws_blocks;
810
811 max_blocks = min_not_zero(max_blocks, (u32)SD_MAX_WS10_BLOCKS);
812 break;
813
814 case SD_LBP_ZERO:
815 max_blocks = min_not_zero(sdkp->max_ws_blocks,
816 (u32)SD_MAX_WS10_BLOCKS);
817 break;
818 }
819
820 blk_queue_max_discard_sectors(q, max_blocks * (logical_block_size >> 9));
821 blk_queue_flag_set(QUEUE_FLAG_DISCARD, q);
822}
823
824static blk_status_t sd_setup_unmap_cmnd(struct scsi_cmnd *cmd)
825{
826 struct scsi_device *sdp = cmd->device;
827 struct request *rq = cmd->request;
828 u64 lba = sectors_to_logical(sdp, blk_rq_pos(rq));
829 u32 nr_blocks = sectors_to_logical(sdp, blk_rq_sectors(rq));
830 unsigned int data_len = 24;
831 char *buf;
832
833 rq->special_vec.bv_page = mempool_alloc(sd_page_pool, GFP_ATOMIC);
834 if (!rq->special_vec.bv_page)
835 return BLK_STS_RESOURCE;
836 clear_highpage(rq->special_vec.bv_page);
837 rq->special_vec.bv_offset = 0;
838 rq->special_vec.bv_len = data_len;
839 rq->rq_flags |= RQF_SPECIAL_PAYLOAD;
840
841 cmd->cmd_len = 10;
842 cmd->cmnd[0] = UNMAP;
843 cmd->cmnd[8] = 24;
844
845 buf = page_address(rq->special_vec.bv_page);
846 put_unaligned_be16(6 + 16, &buf[0]);
847 put_unaligned_be16(16, &buf[2]);
848 put_unaligned_be64(lba, &buf[8]);
849 put_unaligned_be32(nr_blocks, &buf[16]);
850
851 cmd->allowed = SD_MAX_RETRIES;
852 cmd->transfersize = data_len;
853 rq->timeout = SD_TIMEOUT;
854
855 return scsi_init_io(cmd);
856}
857
858static blk_status_t sd_setup_write_same16_cmnd(struct scsi_cmnd *cmd,
859 bool unmap)
860{
861 struct scsi_device *sdp = cmd->device;
862 struct request *rq = cmd->request;
863 u64 lba = sectors_to_logical(sdp, blk_rq_pos(rq));
864 u32 nr_blocks = sectors_to_logical(sdp, blk_rq_sectors(rq));
865 u32 data_len = sdp->sector_size;
866
867 rq->special_vec.bv_page = mempool_alloc(sd_page_pool, GFP_ATOMIC);
868 if (!rq->special_vec.bv_page)
869 return BLK_STS_RESOURCE;
870 clear_highpage(rq->special_vec.bv_page);
871 rq->special_vec.bv_offset = 0;
872 rq->special_vec.bv_len = data_len;
873 rq->rq_flags |= RQF_SPECIAL_PAYLOAD;
874
875 cmd->cmd_len = 16;
876 cmd->cmnd[0] = WRITE_SAME_16;
877 if (unmap)
878 cmd->cmnd[1] = 0x8; /* UNMAP */
879 put_unaligned_be64(lba, &cmd->cmnd[2]);
880 put_unaligned_be32(nr_blocks, &cmd->cmnd[10]);
881
882 cmd->allowed = SD_MAX_RETRIES;
883 cmd->transfersize = data_len;
884 rq->timeout = unmap ? SD_TIMEOUT : SD_WRITE_SAME_TIMEOUT;
885
886 return scsi_init_io(cmd);
887}
888
889static blk_status_t sd_setup_write_same10_cmnd(struct scsi_cmnd *cmd,
890 bool unmap)
891{
892 struct scsi_device *sdp = cmd->device;
893 struct request *rq = cmd->request;
894 u64 lba = sectors_to_logical(sdp, blk_rq_pos(rq));
895 u32 nr_blocks = sectors_to_logical(sdp, blk_rq_sectors(rq));
896 u32 data_len = sdp->sector_size;
897
898 rq->special_vec.bv_page = mempool_alloc(sd_page_pool, GFP_ATOMIC);
899 if (!rq->special_vec.bv_page)
900 return BLK_STS_RESOURCE;
901 clear_highpage(rq->special_vec.bv_page);
902 rq->special_vec.bv_offset = 0;
903 rq->special_vec.bv_len = data_len;
904 rq->rq_flags |= RQF_SPECIAL_PAYLOAD;
905
906 cmd->cmd_len = 10;
907 cmd->cmnd[0] = WRITE_SAME;
908 if (unmap)
909 cmd->cmnd[1] = 0x8; /* UNMAP */
910 put_unaligned_be32(lba, &cmd->cmnd[2]);
911 put_unaligned_be16(nr_blocks, &cmd->cmnd[7]);
912
913 cmd->allowed = SD_MAX_RETRIES;
914 cmd->transfersize = data_len;
915 rq->timeout = unmap ? SD_TIMEOUT : SD_WRITE_SAME_TIMEOUT;
916
917 return scsi_init_io(cmd);
918}
919
920static blk_status_t sd_setup_write_zeroes_cmnd(struct scsi_cmnd *cmd)
921{
922 struct request *rq = cmd->request;
923 struct scsi_device *sdp = cmd->device;
924 struct scsi_disk *sdkp = scsi_disk(rq->rq_disk);
925 u64 lba = sectors_to_logical(sdp, blk_rq_pos(rq));
926 u32 nr_blocks = sectors_to_logical(sdp, blk_rq_sectors(rq));
927
928 if (!(rq->cmd_flags & REQ_NOUNMAP)) {
929 switch (sdkp->zeroing_mode) {
930 case SD_ZERO_WS16_UNMAP:
931 return sd_setup_write_same16_cmnd(cmd, true);
932 case SD_ZERO_WS10_UNMAP:
933 return sd_setup_write_same10_cmnd(cmd, true);
934 }
935 }
936
937 if (sdp->no_write_same)
938 return BLK_STS_TARGET;
939
940 if (sdkp->ws16 || lba > 0xffffffff || nr_blocks > 0xffff)
941 return sd_setup_write_same16_cmnd(cmd, false);
942
943 return sd_setup_write_same10_cmnd(cmd, false);
944}
945
946static void sd_config_write_same(struct scsi_disk *sdkp)
947{
948 struct request_queue *q = sdkp->disk->queue;
949 unsigned int logical_block_size = sdkp->device->sector_size;
950
951 if (sdkp->device->no_write_same) {
952 sdkp->max_ws_blocks = 0;
953 goto out;
954 }
955
956 /* Some devices can not handle block counts above 0xffff despite
957 * supporting WRITE SAME(16). Consequently we default to 64k
958 * blocks per I/O unless the device explicitly advertises a
959 * bigger limit.
960 */
961 if (sdkp->max_ws_blocks > SD_MAX_WS10_BLOCKS)
962 sdkp->max_ws_blocks = min_not_zero(sdkp->max_ws_blocks,
963 (u32)SD_MAX_WS16_BLOCKS);
964 else if (sdkp->ws16 || sdkp->ws10 || sdkp->device->no_report_opcodes)
965 sdkp->max_ws_blocks = min_not_zero(sdkp->max_ws_blocks,
966 (u32)SD_MAX_WS10_BLOCKS);
967 else {
968 sdkp->device->no_write_same = 1;
969 sdkp->max_ws_blocks = 0;
970 }
971
972 if (sdkp->lbprz && sdkp->lbpws)
973 sdkp->zeroing_mode = SD_ZERO_WS16_UNMAP;
974 else if (sdkp->lbprz && sdkp->lbpws10)
975 sdkp->zeroing_mode = SD_ZERO_WS10_UNMAP;
976 else if (sdkp->max_ws_blocks)
977 sdkp->zeroing_mode = SD_ZERO_WS;
978 else
979 sdkp->zeroing_mode = SD_ZERO_WRITE;
980
981 if (sdkp->max_ws_blocks &&
982 sdkp->physical_block_size > logical_block_size) {
983 /*
984 * Reporting a maximum number of blocks that is not aligned
985 * on the device physical size would cause a large write same
986 * request to be split into physically unaligned chunks by
987 * __blkdev_issue_write_zeroes() and __blkdev_issue_write_same()
988 * even if the caller of these functions took care to align the
989 * large request. So make sure the maximum reported is aligned
990 * to the device physical block size. This is only an optional
991 * optimization for regular disks, but this is mandatory to
992 * avoid failure of large write same requests directed at
993 * sequential write required zones of host-managed ZBC disks.
994 */
995 sdkp->max_ws_blocks =
996 round_down(sdkp->max_ws_blocks,
997 bytes_to_logical(sdkp->device,
998 sdkp->physical_block_size));
999 }
1000
1001out:
1002 blk_queue_max_write_same_sectors(q, sdkp->max_ws_blocks *
1003 (logical_block_size >> 9));
1004 blk_queue_max_write_zeroes_sectors(q, sdkp->max_ws_blocks *
1005 (logical_block_size >> 9));
1006}
1007
1008/**
1009 * sd_setup_write_same_cmnd - write the same data to multiple blocks
1010 * @cmd: command to prepare
1011 *
1012 * Will set up either WRITE SAME(10) or WRITE SAME(16) depending on
1013 * the preference indicated by the target device.
1014 **/
1015static blk_status_t sd_setup_write_same_cmnd(struct scsi_cmnd *cmd)
1016{
1017 struct request *rq = cmd->request;
1018 struct scsi_device *sdp = cmd->device;
1019 struct scsi_disk *sdkp = scsi_disk(rq->rq_disk);
1020 struct bio *bio = rq->bio;
1021 u64 lba = sectors_to_logical(sdp, blk_rq_pos(rq));
1022 u32 nr_blocks = sectors_to_logical(sdp, blk_rq_sectors(rq));
1023 blk_status_t ret;
1024
1025 if (sdkp->device->no_write_same)
1026 return BLK_STS_TARGET;
1027
1028 BUG_ON(bio_offset(bio) || bio_iovec(bio).bv_len != sdp->sector_size);
1029
1030 rq->timeout = SD_WRITE_SAME_TIMEOUT;
1031
1032 if (sdkp->ws16 || lba > 0xffffffff || nr_blocks > 0xffff) {
1033 cmd->cmd_len = 16;
1034 cmd->cmnd[0] = WRITE_SAME_16;
1035 put_unaligned_be64(lba, &cmd->cmnd[2]);
1036 put_unaligned_be32(nr_blocks, &cmd->cmnd[10]);
1037 } else {
1038 cmd->cmd_len = 10;
1039 cmd->cmnd[0] = WRITE_SAME;
1040 put_unaligned_be32(lba, &cmd->cmnd[2]);
1041 put_unaligned_be16(nr_blocks, &cmd->cmnd[7]);
1042 }
1043
1044 cmd->transfersize = sdp->sector_size;
1045 cmd->allowed = SD_MAX_RETRIES;
1046
1047 /*
1048 * For WRITE SAME the data transferred via the DATA OUT buffer is
1049 * different from the amount of data actually written to the target.
1050 *
1051 * We set up __data_len to the amount of data transferred via the
1052 * DATA OUT buffer so that blk_rq_map_sg sets up the proper S/G list
1053 * to transfer a single sector of data first, but then reset it to
1054 * the amount of data to be written right after so that the I/O path
1055 * knows how much to actually write.
1056 */
1057 rq->__data_len = sdp->sector_size;
1058 ret = scsi_init_io(cmd);
1059 rq->__data_len = blk_rq_bytes(rq);
1060
1061 return ret;
1062}
1063
1064static blk_status_t sd_setup_flush_cmnd(struct scsi_cmnd *cmd)
1065{
1066 struct request *rq = cmd->request;
1067
1068 /* flush requests don't perform I/O, zero the S/G table */
1069 memset(&cmd->sdb, 0, sizeof(cmd->sdb));
1070
1071 cmd->cmnd[0] = SYNCHRONIZE_CACHE;
1072 cmd->cmd_len = 10;
1073 cmd->transfersize = 0;
1074 cmd->allowed = SD_MAX_RETRIES;
1075
1076 rq->timeout = rq->q->rq_timeout * SD_FLUSH_TIMEOUT_MULTIPLIER;
1077 return BLK_STS_OK;
1078}
1079
1080static blk_status_t sd_setup_rw32_cmnd(struct scsi_cmnd *cmd, bool write,
1081 sector_t lba, unsigned int nr_blocks,
1082 unsigned char flags)
1083{
1084 cmd->cmnd = mempool_alloc(sd_cdb_pool, GFP_ATOMIC);
1085 if (unlikely(cmd->cmnd == NULL))
1086 return BLK_STS_RESOURCE;
1087
1088 cmd->cmd_len = SD_EXT_CDB_SIZE;
1089 memset(cmd->cmnd, 0, cmd->cmd_len);
1090
1091 cmd->cmnd[0] = VARIABLE_LENGTH_CMD;
1092 cmd->cmnd[7] = 0x18; /* Additional CDB len */
1093 cmd->cmnd[9] = write ? WRITE_32 : READ_32;
1094 cmd->cmnd[10] = flags;
1095 put_unaligned_be64(lba, &cmd->cmnd[12]);
1096 put_unaligned_be32(lba, &cmd->cmnd[20]); /* Expected Indirect LBA */
1097 put_unaligned_be32(nr_blocks, &cmd->cmnd[28]);
1098
1099 return BLK_STS_OK;
1100}
1101
1102static blk_status_t sd_setup_rw16_cmnd(struct scsi_cmnd *cmd, bool write,
1103 sector_t lba, unsigned int nr_blocks,
1104 unsigned char flags)
1105{
1106 cmd->cmd_len = 16;
1107 cmd->cmnd[0] = write ? WRITE_16 : READ_16;
1108 cmd->cmnd[1] = flags;
1109 cmd->cmnd[14] = 0;
1110 cmd->cmnd[15] = 0;
1111 put_unaligned_be64(lba, &cmd->cmnd[2]);
1112 put_unaligned_be32(nr_blocks, &cmd->cmnd[10]);
1113
1114 return BLK_STS_OK;
1115}
1116
1117static blk_status_t sd_setup_rw10_cmnd(struct scsi_cmnd *cmd, bool write,
1118 sector_t lba, unsigned int nr_blocks,
1119 unsigned char flags)
1120{
1121 cmd->cmd_len = 10;
1122 cmd->cmnd[0] = write ? WRITE_10 : READ_10;
1123 cmd->cmnd[1] = flags;
1124 cmd->cmnd[6] = 0;
1125 cmd->cmnd[9] = 0;
1126 put_unaligned_be32(lba, &cmd->cmnd[2]);
1127 put_unaligned_be16(nr_blocks, &cmd->cmnd[7]);
1128
1129 return BLK_STS_OK;
1130}
1131
1132static blk_status_t sd_setup_rw6_cmnd(struct scsi_cmnd *cmd, bool write,
1133 sector_t lba, unsigned int nr_blocks,
1134 unsigned char flags)
1135{
1136 /* Avoid that 0 blocks gets translated into 256 blocks. */
1137 if (WARN_ON_ONCE(nr_blocks == 0))
1138 return BLK_STS_IOERR;
1139
1140 if (unlikely(flags & 0x8)) {
1141 /*
1142 * This happens only if this drive failed 10byte rw
1143 * command with ILLEGAL_REQUEST during operation and
1144 * thus turned off use_10_for_rw.
1145 */
1146 scmd_printk(KERN_ERR, cmd, "FUA write on READ/WRITE(6) drive\n");
1147 return BLK_STS_IOERR;
1148 }
1149
1150 cmd->cmd_len = 6;
1151 cmd->cmnd[0] = write ? WRITE_6 : READ_6;
1152 cmd->cmnd[1] = (lba >> 16) & 0x1f;
1153 cmd->cmnd[2] = (lba >> 8) & 0xff;
1154 cmd->cmnd[3] = lba & 0xff;
1155 cmd->cmnd[4] = nr_blocks;
1156 cmd->cmnd[5] = 0;
1157
1158 return BLK_STS_OK;
1159}
1160
1161static blk_status_t sd_setup_read_write_cmnd(struct scsi_cmnd *cmd)
1162{
1163 struct request *rq = cmd->request;
1164 struct scsi_device *sdp = cmd->device;
1165 struct scsi_disk *sdkp = scsi_disk(rq->rq_disk);
1166 sector_t lba = sectors_to_logical(sdp, blk_rq_pos(rq));
1167 sector_t threshold;
1168 unsigned int nr_blocks = sectors_to_logical(sdp, blk_rq_sectors(rq));
1169 unsigned int mask = logical_to_sectors(sdp, 1) - 1;
1170 bool write = rq_data_dir(rq) == WRITE;
1171 unsigned char protect, fua;
1172 blk_status_t ret;
1173 unsigned int dif;
1174 bool dix;
1175
1176 ret = scsi_init_io(cmd);
1177 if (ret != BLK_STS_OK)
1178 return ret;
1179
1180 if (!scsi_device_online(sdp) || sdp->changed) {
1181 scmd_printk(KERN_ERR, cmd, "device offline or changed\n");
1182 return BLK_STS_IOERR;
1183 }
1184
1185 if (blk_rq_pos(rq) + blk_rq_sectors(rq) > get_capacity(rq->rq_disk)) {
1186 scmd_printk(KERN_ERR, cmd, "access beyond end of device\n");
1187 return BLK_STS_IOERR;
1188 }
1189
1190 if ((blk_rq_pos(rq) & mask) || (blk_rq_sectors(rq) & mask)) {
1191 scmd_printk(KERN_ERR, cmd, "request not aligned to the logical block size\n");
1192 return BLK_STS_IOERR;
1193 }
1194
1195 /*
1196 * Some SD card readers can't handle accesses which touch the
1197 * last one or two logical blocks. Split accesses as needed.
1198 */
1199 threshold = sdkp->capacity - SD_LAST_BUGGY_SECTORS;
1200
1201 if (unlikely(sdp->last_sector_bug && lba + nr_blocks > threshold)) {
1202 if (lba < threshold) {
1203 /* Access up to the threshold but not beyond */
1204 nr_blocks = threshold - lba;
1205 } else {
1206 /* Access only a single logical block */
1207 nr_blocks = 1;
1208 }
1209 }
1210
1211 fua = rq->cmd_flags & REQ_FUA ? 0x8 : 0;
1212 dix = scsi_prot_sg_count(cmd);
1213 dif = scsi_host_dif_capable(cmd->device->host, sdkp->protection_type);
1214
1215 if (dif || dix)
1216 protect = sd_setup_protect_cmnd(cmd, dix, dif);
1217 else
1218 protect = 0;
1219
1220 if (protect && sdkp->protection_type == T10_PI_TYPE2_PROTECTION) {
1221 ret = sd_setup_rw32_cmnd(cmd, write, lba, nr_blocks,
1222 protect | fua);
1223 } else if (sdp->use_16_for_rw || (nr_blocks > 0xffff)) {
1224 ret = sd_setup_rw16_cmnd(cmd, write, lba, nr_blocks,
1225 protect | fua);
1226 } else if ((nr_blocks > 0xff) || (lba > 0x1fffff) ||
1227 sdp->use_10_for_rw || protect) {
1228 ret = sd_setup_rw10_cmnd(cmd, write, lba, nr_blocks,
1229 protect | fua);
1230 } else {
1231 ret = sd_setup_rw6_cmnd(cmd, write, lba, nr_blocks,
1232 protect | fua);
1233 }
1234
1235 if (unlikely(ret != BLK_STS_OK))
1236 return ret;
1237
1238 /*
1239 * We shouldn't disconnect in the middle of a sector, so with a dumb
1240 * host adapter, it's safe to assume that we can at least transfer
1241 * this many bytes between each connect / disconnect.
1242 */
1243 cmd->transfersize = sdp->sector_size;
1244 cmd->underflow = nr_blocks << 9;
1245 cmd->allowed = SD_MAX_RETRIES;
1246 cmd->sdb.length = nr_blocks * sdp->sector_size;
1247
1248 SCSI_LOG_HLQUEUE(1,
1249 scmd_printk(KERN_INFO, cmd,
1250 "%s: block=%llu, count=%d\n", __func__,
1251 (unsigned long long)blk_rq_pos(rq),
1252 blk_rq_sectors(rq)));
1253 SCSI_LOG_HLQUEUE(2,
1254 scmd_printk(KERN_INFO, cmd,
1255 "%s %d/%u 512 byte blocks.\n",
1256 write ? "writing" : "reading", nr_blocks,
1257 blk_rq_sectors(rq)));
1258
1259 /*
1260 * This indicates that the command is ready from our end to be
1261 * queued.
1262 */
1263 return BLK_STS_OK;
1264}
1265
1266static blk_status_t sd_init_command(struct scsi_cmnd *cmd)
1267{
1268 struct request *rq = cmd->request;
1269
1270 switch (req_op(rq)) {
1271 case REQ_OP_DISCARD:
1272 switch (scsi_disk(rq->rq_disk)->provisioning_mode) {
1273 case SD_LBP_UNMAP:
1274 return sd_setup_unmap_cmnd(cmd);
1275 case SD_LBP_WS16:
1276 return sd_setup_write_same16_cmnd(cmd, true);
1277 case SD_LBP_WS10:
1278 return sd_setup_write_same10_cmnd(cmd, true);
1279 case SD_LBP_ZERO:
1280 return sd_setup_write_same10_cmnd(cmd, false);
1281 default:
1282 return BLK_STS_TARGET;
1283 }
1284 case REQ_OP_WRITE_ZEROES:
1285 return sd_setup_write_zeroes_cmnd(cmd);
1286 case REQ_OP_WRITE_SAME:
1287 return sd_setup_write_same_cmnd(cmd);
1288 case REQ_OP_FLUSH:
1289 return sd_setup_flush_cmnd(cmd);
1290 case REQ_OP_READ:
1291 case REQ_OP_WRITE:
1292 return sd_setup_read_write_cmnd(cmd);
1293 case REQ_OP_ZONE_RESET:
1294 return sd_zbc_setup_reset_cmnd(cmd, false);
1295 case REQ_OP_ZONE_RESET_ALL:
1296 return sd_zbc_setup_reset_cmnd(cmd, true);
1297 default:
1298 WARN_ON_ONCE(1);
1299 return BLK_STS_NOTSUPP;
1300 }
1301}
1302
1303static void sd_uninit_command(struct scsi_cmnd *SCpnt)
1304{
1305 struct request *rq = SCpnt->request;
1306 u8 *cmnd;
1307
1308 if (rq->rq_flags & RQF_SPECIAL_PAYLOAD)
1309 mempool_free(rq->special_vec.bv_page, sd_page_pool);
1310
1311 if (SCpnt->cmnd != scsi_req(rq)->cmd) {
1312 cmnd = SCpnt->cmnd;
1313 SCpnt->cmnd = NULL;
1314 SCpnt->cmd_len = 0;
1315 mempool_free(cmnd, sd_cdb_pool);
1316 }
1317}
1318
1319/**
1320 * sd_open - open a scsi disk device
1321 * @bdev: Block device of the scsi disk to open
1322 * @mode: FMODE_* mask
1323 *
1324 * Returns 0 if successful. Returns a negated errno value in case
1325 * of error.
1326 *
1327 * Note: This can be called from a user context (e.g. fsck(1) )
1328 * or from within the kernel (e.g. as a result of a mount(1) ).
1329 * In the latter case @inode and @filp carry an abridged amount
1330 * of information as noted above.
1331 *
1332 * Locking: called with bdev->bd_mutex held.
1333 **/
1334static int sd_open(struct block_device *bdev, fmode_t mode)
1335{
1336 struct scsi_disk *sdkp = scsi_disk_get(bdev->bd_disk);
1337 struct scsi_device *sdev;
1338 int retval;
1339
1340 if (!sdkp)
1341 return -ENXIO;
1342
1343 SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp, "sd_open\n"));
1344
1345 sdev = sdkp->device;
1346
1347 /*
1348 * If the device is in error recovery, wait until it is done.
1349 * If the device is offline, then disallow any access to it.
1350 */
1351 retval = -ENXIO;
1352 if (!scsi_block_when_processing_errors(sdev))
1353 goto error_out;
1354
1355 if (sdev->removable || sdkp->write_prot)
1356 check_disk_change(bdev);
1357
1358 /*
1359 * If the drive is empty, just let the open fail.
1360 */
1361 retval = -ENOMEDIUM;
1362 if (sdev->removable && !sdkp->media_present && !(mode & FMODE_NDELAY))
1363 goto error_out;
1364
1365 /*
1366 * If the device has the write protect tab set, have the open fail
1367 * if the user expects to be able to write to the thing.
1368 */
1369 retval = -EROFS;
1370 if (sdkp->write_prot && (mode & FMODE_WRITE))
1371 goto error_out;
1372
1373 /*
1374 * It is possible that the disk changing stuff resulted in
1375 * the device being taken offline. If this is the case,
1376 * report this to the user, and don't pretend that the
1377 * open actually succeeded.
1378 */
1379 retval = -ENXIO;
1380 if (!scsi_device_online(sdev))
1381 goto error_out;
1382
1383 if ((atomic_inc_return(&sdkp->openers) == 1) && sdev->removable) {
1384 if (scsi_block_when_processing_errors(sdev))
1385 scsi_set_medium_removal(sdev, SCSI_REMOVAL_PREVENT);
1386 }
1387
1388 return 0;
1389
1390error_out:
1391 scsi_disk_put(sdkp);
1392 return retval;
1393}
1394
1395/**
1396 * sd_release - invoked when the (last) close(2) is called on this
1397 * scsi disk.
1398 * @disk: disk to release
1399 * @mode: FMODE_* mask
1400 *
1401 * Returns 0.
1402 *
1403 * Note: may block (uninterruptible) if error recovery is underway
1404 * on this disk.
1405 *
1406 * Locking: called with bdev->bd_mutex held.
1407 **/
1408static void sd_release(struct gendisk *disk, fmode_t mode)
1409{
1410 struct scsi_disk *sdkp = scsi_disk(disk);
1411 struct scsi_device *sdev = sdkp->device;
1412
1413 SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp, "sd_release\n"));
1414
1415 if (atomic_dec_return(&sdkp->openers) == 0 && sdev->removable) {
1416 if (scsi_block_when_processing_errors(sdev))
1417 scsi_set_medium_removal(sdev, SCSI_REMOVAL_ALLOW);
1418 }
1419
1420 scsi_disk_put(sdkp);
1421}
1422
1423static int sd_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1424{
1425 struct scsi_disk *sdkp = scsi_disk(bdev->bd_disk);
1426 struct scsi_device *sdp = sdkp->device;
1427 struct Scsi_Host *host = sdp->host;
1428 sector_t capacity = logical_to_sectors(sdp, sdkp->capacity);
1429 int diskinfo[4];
1430
1431 /* default to most commonly used values */
1432 diskinfo[0] = 0x40; /* 1 << 6 */
1433 diskinfo[1] = 0x20; /* 1 << 5 */
1434 diskinfo[2] = capacity >> 11;
1435
1436 /* override with calculated, extended default, or driver values */
1437 if (host->hostt->bios_param)
1438 host->hostt->bios_param(sdp, bdev, capacity, diskinfo);
1439 else
1440 scsicam_bios_param(bdev, capacity, diskinfo);
1441
1442 geo->heads = diskinfo[0];
1443 geo->sectors = diskinfo[1];
1444 geo->cylinders = diskinfo[2];
1445 return 0;
1446}
1447
1448/**
1449 * sd_ioctl - process an ioctl
1450 * @bdev: target block device
1451 * @mode: FMODE_* mask
1452 * @cmd: ioctl command number
1453 * @arg: this is third argument given to ioctl(2) system call.
1454 * Often contains a pointer.
1455 *
1456 * Returns 0 if successful (some ioctls return positive numbers on
1457 * success as well). Returns a negated errno value in case of error.
1458 *
1459 * Note: most ioctls are forward onto the block subsystem or further
1460 * down in the scsi subsystem.
1461 **/
1462static int sd_ioctl(struct block_device *bdev, fmode_t mode,
1463 unsigned int cmd, unsigned long arg)
1464{
1465 struct gendisk *disk = bdev->bd_disk;
1466 struct scsi_disk *sdkp = scsi_disk(disk);
1467 struct scsi_device *sdp = sdkp->device;
1468 void __user *p = (void __user *)arg;
1469 int error;
1470
1471 SCSI_LOG_IOCTL(1, sd_printk(KERN_INFO, sdkp, "sd_ioctl: disk=%s, "
1472 "cmd=0x%x\n", disk->disk_name, cmd));
1473
1474 error = scsi_verify_blk_ioctl(bdev, cmd);
1475 if (error < 0)
1476 return error;
1477
1478 /*
1479 * If we are in the middle of error recovery, don't let anyone
1480 * else try and use this device. Also, if error recovery fails, it
1481 * may try and take the device offline, in which case all further
1482 * access to the device is prohibited.
1483 */
1484 error = scsi_ioctl_block_when_processing_errors(sdp, cmd,
1485 (mode & FMODE_NDELAY) != 0);
1486 if (error)
1487 goto out;
1488
1489 if (is_sed_ioctl(cmd))
1490 return sed_ioctl(sdkp->opal_dev, cmd, p);
1491
1492 /*
1493 * Send SCSI addressing ioctls directly to mid level, send other
1494 * ioctls to block level and then onto mid level if they can't be
1495 * resolved.
1496 */
1497 switch (cmd) {
1498 case SCSI_IOCTL_GET_IDLUN:
1499 case SCSI_IOCTL_GET_BUS_NUMBER:
1500 error = scsi_ioctl(sdp, cmd, p);
1501 break;
1502 default:
1503 error = scsi_cmd_blk_ioctl(bdev, mode, cmd, p);
1504 if (error != -ENOTTY)
1505 break;
1506 error = scsi_ioctl(sdp, cmd, p);
1507 break;
1508 }
1509out:
1510 return error;
1511}
1512
1513static void set_media_not_present(struct scsi_disk *sdkp)
1514{
1515 if (sdkp->media_present)
1516 sdkp->device->changed = 1;
1517
1518 if (sdkp->device->removable) {
1519 sdkp->media_present = 0;
1520 sdkp->capacity = 0;
1521 }
1522}
1523
1524static int media_not_present(struct scsi_disk *sdkp,
1525 struct scsi_sense_hdr *sshdr)
1526{
1527 if (!scsi_sense_valid(sshdr))
1528 return 0;
1529
1530 /* not invoked for commands that could return deferred errors */
1531 switch (sshdr->sense_key) {
1532 case UNIT_ATTENTION:
1533 case NOT_READY:
1534 /* medium not present */
1535 if (sshdr->asc == 0x3A) {
1536 set_media_not_present(sdkp);
1537 return 1;
1538 }
1539 }
1540 return 0;
1541}
1542
1543/**
1544 * sd_check_events - check media events
1545 * @disk: kernel device descriptor
1546 * @clearing: disk events currently being cleared
1547 *
1548 * Returns mask of DISK_EVENT_*.
1549 *
1550 * Note: this function is invoked from the block subsystem.
1551 **/
1552static unsigned int sd_check_events(struct gendisk *disk, unsigned int clearing)
1553{
1554 struct scsi_disk *sdkp = scsi_disk_get(disk);
1555 struct scsi_device *sdp;
1556 int retval;
1557
1558 if (!sdkp)
1559 return 0;
1560
1561 sdp = sdkp->device;
1562 SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp, "sd_check_events\n"));
1563
1564 /*
1565 * If the device is offline, don't send any commands - just pretend as
1566 * if the command failed. If the device ever comes back online, we
1567 * can deal with it then. It is only because of unrecoverable errors
1568 * that we would ever take a device offline in the first place.
1569 */
1570 if (!scsi_device_online(sdp)) {
1571 set_media_not_present(sdkp);
1572 goto out;
1573 }
1574
1575 /*
1576 * Using TEST_UNIT_READY enables differentiation between drive with
1577 * no cartridge loaded - NOT READY, drive with changed cartridge -
1578 * UNIT ATTENTION, or with same cartridge - GOOD STATUS.
1579 *
1580 * Drives that auto spin down. eg iomega jaz 1G, will be started
1581 * by sd_spinup_disk() from sd_revalidate_disk(), which happens whenever
1582 * sd_revalidate() is called.
1583 */
1584 if (scsi_block_when_processing_errors(sdp)) {
1585 struct scsi_sense_hdr sshdr = { 0, };
1586
1587 retval = scsi_test_unit_ready(sdp, SD_TIMEOUT, SD_MAX_RETRIES,
1588 &sshdr);
1589
1590 /* failed to execute TUR, assume media not present */
1591 if (host_byte(retval)) {
1592 set_media_not_present(sdkp);
1593 goto out;
1594 }
1595
1596 if (media_not_present(sdkp, &sshdr))
1597 goto out;
1598 }
1599
1600 /*
1601 * For removable scsi disk we have to recognise the presence
1602 * of a disk in the drive.
1603 */
1604 if (!sdkp->media_present)
1605 sdp->changed = 1;
1606 sdkp->media_present = 1;
1607out:
1608 /*
1609 * sdp->changed is set under the following conditions:
1610 *
1611 * Medium present state has changed in either direction.
1612 * Device has indicated UNIT_ATTENTION.
1613 */
1614 retval = sdp->changed ? DISK_EVENT_MEDIA_CHANGE : 0;
1615 sdp->changed = 0;
1616 scsi_disk_put(sdkp);
1617 return retval;
1618}
1619
1620static int sd_sync_cache(struct scsi_disk *sdkp, struct scsi_sense_hdr *sshdr)
1621{
1622 int retries, res;
1623 struct scsi_device *sdp = sdkp->device;
1624 const int timeout = sdp->request_queue->rq_timeout
1625 * SD_FLUSH_TIMEOUT_MULTIPLIER;
1626 struct scsi_sense_hdr my_sshdr;
1627
1628 if (!scsi_device_online(sdp))
1629 return -ENODEV;
1630
1631 /* caller might not be interested in sense, but we need it */
1632 if (!sshdr)
1633 sshdr = &my_sshdr;
1634
1635 for (retries = 3; retries > 0; --retries) {
1636 unsigned char cmd[10] = { 0 };
1637
1638 cmd[0] = SYNCHRONIZE_CACHE;
1639 /*
1640 * Leave the rest of the command zero to indicate
1641 * flush everything.
1642 */
1643 res = scsi_execute(sdp, cmd, DMA_NONE, NULL, 0, NULL, sshdr,
1644 timeout, SD_MAX_RETRIES, 0, RQF_PM, NULL);
1645 if (res == 0)
1646 break;
1647 }
1648
1649 if (res) {
1650 sd_print_result(sdkp, "Synchronize Cache(10) failed", res);
1651
1652 if (driver_byte(res) == DRIVER_SENSE)
1653 sd_print_sense_hdr(sdkp, sshdr);
1654
1655 /* we need to evaluate the error return */
1656 if (scsi_sense_valid(sshdr) &&
1657 (sshdr->asc == 0x3a || /* medium not present */
1658 sshdr->asc == 0x20 || /* invalid command */
1659 (sshdr->asc == 0x74 && sshdr->ascq == 0x71))) /* drive is password locked */
1660 /* this is no error here */
1661 return 0;
1662
1663 switch (host_byte(res)) {
1664 /* ignore errors due to racing a disconnection */
1665 case DID_BAD_TARGET:
1666 case DID_NO_CONNECT:
1667 return 0;
1668 /* signal the upper layer it might try again */
1669 case DID_BUS_BUSY:
1670 case DID_IMM_RETRY:
1671 case DID_REQUEUE:
1672 case DID_SOFT_ERROR:
1673 return -EBUSY;
1674 default:
1675 return -EIO;
1676 }
1677 }
1678 return 0;
1679}
1680
1681static void sd_rescan(struct device *dev)
1682{
1683 struct scsi_disk *sdkp = dev_get_drvdata(dev);
1684
1685 revalidate_disk(sdkp->disk);
1686}
1687
1688
1689#ifdef CONFIG_COMPAT
1690/*
1691 * This gets directly called from VFS. When the ioctl
1692 * is not recognized we go back to the other translation paths.
1693 */
1694static int sd_compat_ioctl(struct block_device *bdev, fmode_t mode,
1695 unsigned int cmd, unsigned long arg)
1696{
1697 struct scsi_device *sdev = scsi_disk(bdev->bd_disk)->device;
1698 int error;
1699
1700 error = scsi_ioctl_block_when_processing_errors(sdev, cmd,
1701 (mode & FMODE_NDELAY) != 0);
1702 if (error)
1703 return error;
1704
1705 /*
1706 * Let the static ioctl translation table take care of it.
1707 */
1708 if (!sdev->host->hostt->compat_ioctl)
1709 return -ENOIOCTLCMD;
1710 return sdev->host->hostt->compat_ioctl(sdev, cmd, (void __user *)arg);
1711}
1712#endif
1713
1714static char sd_pr_type(enum pr_type type)
1715{
1716 switch (type) {
1717 case PR_WRITE_EXCLUSIVE:
1718 return 0x01;
1719 case PR_EXCLUSIVE_ACCESS:
1720 return 0x03;
1721 case PR_WRITE_EXCLUSIVE_REG_ONLY:
1722 return 0x05;
1723 case PR_EXCLUSIVE_ACCESS_REG_ONLY:
1724 return 0x06;
1725 case PR_WRITE_EXCLUSIVE_ALL_REGS:
1726 return 0x07;
1727 case PR_EXCLUSIVE_ACCESS_ALL_REGS:
1728 return 0x08;
1729 default:
1730 return 0;
1731 }
1732};
1733
1734static int sd_pr_command(struct block_device *bdev, u8 sa,
1735 u64 key, u64 sa_key, u8 type, u8 flags)
1736{
1737 struct scsi_device *sdev = scsi_disk(bdev->bd_disk)->device;
1738 struct scsi_sense_hdr sshdr;
1739 int result;
1740 u8 cmd[16] = { 0, };
1741 u8 data[24] = { 0, };
1742
1743 cmd[0] = PERSISTENT_RESERVE_OUT;
1744 cmd[1] = sa;
1745 cmd[2] = type;
1746 put_unaligned_be32(sizeof(data), &cmd[5]);
1747
1748 put_unaligned_be64(key, &data[0]);
1749 put_unaligned_be64(sa_key, &data[8]);
1750 data[20] = flags;
1751
1752 result = scsi_execute_req(sdev, cmd, DMA_TO_DEVICE, &data, sizeof(data),
1753 &sshdr, SD_TIMEOUT, SD_MAX_RETRIES, NULL);
1754
1755 if (driver_byte(result) == DRIVER_SENSE &&
1756 scsi_sense_valid(&sshdr)) {
1757 sdev_printk(KERN_INFO, sdev, "PR command failed: %d\n", result);
1758 scsi_print_sense_hdr(sdev, NULL, &sshdr);
1759 }
1760
1761 return result;
1762}
1763
1764static int sd_pr_register(struct block_device *bdev, u64 old_key, u64 new_key,
1765 u32 flags)
1766{
1767 if (flags & ~PR_FL_IGNORE_KEY)
1768 return -EOPNOTSUPP;
1769 return sd_pr_command(bdev, (flags & PR_FL_IGNORE_KEY) ? 0x06 : 0x00,
1770 old_key, new_key, 0,
1771 (1 << 0) /* APTPL */);
1772}
1773
1774static int sd_pr_reserve(struct block_device *bdev, u64 key, enum pr_type type,
1775 u32 flags)
1776{
1777 if (flags)
1778 return -EOPNOTSUPP;
1779 return sd_pr_command(bdev, 0x01, key, 0, sd_pr_type(type), 0);
1780}
1781
1782static int sd_pr_release(struct block_device *bdev, u64 key, enum pr_type type)
1783{
1784 return sd_pr_command(bdev, 0x02, key, 0, sd_pr_type(type), 0);
1785}
1786
1787static int sd_pr_preempt(struct block_device *bdev, u64 old_key, u64 new_key,
1788 enum pr_type type, bool abort)
1789{
1790 return sd_pr_command(bdev, abort ? 0x05 : 0x04, old_key, new_key,
1791 sd_pr_type(type), 0);
1792}
1793
1794static int sd_pr_clear(struct block_device *bdev, u64 key)
1795{
1796 return sd_pr_command(bdev, 0x03, key, 0, 0, 0);
1797}
1798
1799static const struct pr_ops sd_pr_ops = {
1800 .pr_register = sd_pr_register,
1801 .pr_reserve = sd_pr_reserve,
1802 .pr_release = sd_pr_release,
1803 .pr_preempt = sd_pr_preempt,
1804 .pr_clear = sd_pr_clear,
1805};
1806
1807static const struct block_device_operations sd_fops = {
1808 .owner = THIS_MODULE,
1809 .open = sd_open,
1810 .release = sd_release,
1811 .ioctl = sd_ioctl,
1812 .getgeo = sd_getgeo,
1813#ifdef CONFIG_COMPAT
1814 .compat_ioctl = sd_compat_ioctl,
1815#endif
1816 .check_events = sd_check_events,
1817 .revalidate_disk = sd_revalidate_disk,
1818 .unlock_native_capacity = sd_unlock_native_capacity,
1819 .report_zones = sd_zbc_report_zones,
1820 .pr_ops = &sd_pr_ops,
1821};
1822
1823/**
1824 * sd_eh_reset - reset error handling callback
1825 * @scmd: sd-issued command that has failed
1826 *
1827 * This function is called by the SCSI midlayer before starting
1828 * SCSI EH. When counting medium access failures we have to be
1829 * careful to register it only only once per device and SCSI EH run;
1830 * there might be several timed out commands which will cause the
1831 * 'max_medium_access_timeouts' counter to trigger after the first
1832 * SCSI EH run already and set the device to offline.
1833 * So this function resets the internal counter before starting SCSI EH.
1834 **/
1835static void sd_eh_reset(struct scsi_cmnd *scmd)
1836{
1837 struct scsi_disk *sdkp = scsi_disk(scmd->request->rq_disk);
1838
1839 /* New SCSI EH run, reset gate variable */
1840 sdkp->ignore_medium_access_errors = false;
1841}
1842
1843/**
1844 * sd_eh_action - error handling callback
1845 * @scmd: sd-issued command that has failed
1846 * @eh_disp: The recovery disposition suggested by the midlayer
1847 *
1848 * This function is called by the SCSI midlayer upon completion of an
1849 * error test command (currently TEST UNIT READY). The result of sending
1850 * the eh command is passed in eh_disp. We're looking for devices that
1851 * fail medium access commands but are OK with non access commands like
1852 * test unit ready (so wrongly see the device as having a successful
1853 * recovery)
1854 **/
1855static int sd_eh_action(struct scsi_cmnd *scmd, int eh_disp)
1856{
1857 struct scsi_disk *sdkp = scsi_disk(scmd->request->rq_disk);
1858 struct scsi_device *sdev = scmd->device;
1859
1860 if (!scsi_device_online(sdev) ||
1861 !scsi_medium_access_command(scmd) ||
1862 host_byte(scmd->result) != DID_TIME_OUT ||
1863 eh_disp != SUCCESS)
1864 return eh_disp;
1865
1866 /*
1867 * The device has timed out executing a medium access command.
1868 * However, the TEST UNIT READY command sent during error
1869 * handling completed successfully. Either the device is in the
1870 * process of recovering or has it suffered an internal failure
1871 * that prevents access to the storage medium.
1872 */
1873 if (!sdkp->ignore_medium_access_errors) {
1874 sdkp->medium_access_timed_out++;
1875 sdkp->ignore_medium_access_errors = true;
1876 }
1877
1878 /*
1879 * If the device keeps failing read/write commands but TEST UNIT
1880 * READY always completes successfully we assume that medium
1881 * access is no longer possible and take the device offline.
1882 */
1883 if (sdkp->medium_access_timed_out >= sdkp->max_medium_access_timeouts) {
1884 scmd_printk(KERN_ERR, scmd,
1885 "Medium access timeout failure. Offlining disk!\n");
1886 mutex_lock(&sdev->state_mutex);
1887 scsi_device_set_state(sdev, SDEV_OFFLINE);
1888 mutex_unlock(&sdev->state_mutex);
1889
1890 return SUCCESS;
1891 }
1892
1893 return eh_disp;
1894}
1895
1896static unsigned int sd_completed_bytes(struct scsi_cmnd *scmd)
1897{
1898 struct request *req = scmd->request;
1899 struct scsi_device *sdev = scmd->device;
1900 unsigned int transferred, good_bytes;
1901 u64 start_lba, end_lba, bad_lba;
1902
1903 /*
1904 * Some commands have a payload smaller than the device logical
1905 * block size (e.g. INQUIRY on a 4K disk).
1906 */
1907 if (scsi_bufflen(scmd) <= sdev->sector_size)
1908 return 0;
1909
1910 /* Check if we have a 'bad_lba' information */
1911 if (!scsi_get_sense_info_fld(scmd->sense_buffer,
1912 SCSI_SENSE_BUFFERSIZE,
1913 &bad_lba))
1914 return 0;
1915
1916 /*
1917 * If the bad lba was reported incorrectly, we have no idea where
1918 * the error is.
1919 */
1920 start_lba = sectors_to_logical(sdev, blk_rq_pos(req));
1921 end_lba = start_lba + bytes_to_logical(sdev, scsi_bufflen(scmd));
1922 if (bad_lba < start_lba || bad_lba >= end_lba)
1923 return 0;
1924
1925 /*
1926 * resid is optional but mostly filled in. When it's unused,
1927 * its value is zero, so we assume the whole buffer transferred
1928 */
1929 transferred = scsi_bufflen(scmd) - scsi_get_resid(scmd);
1930
1931 /* This computation should always be done in terms of the
1932 * resolution of the device's medium.
1933 */
1934 good_bytes = logical_to_bytes(sdev, bad_lba - start_lba);
1935
1936 return min(good_bytes, transferred);
1937}
1938
1939/**
1940 * sd_done - bottom half handler: called when the lower level
1941 * driver has completed (successfully or otherwise) a scsi command.
1942 * @SCpnt: mid-level's per command structure.
1943 *
1944 * Note: potentially run from within an ISR. Must not block.
1945 **/
1946static int sd_done(struct scsi_cmnd *SCpnt)
1947{
1948 int result = SCpnt->result;
1949 unsigned int good_bytes = result ? 0 : scsi_bufflen(SCpnt);
1950 unsigned int sector_size = SCpnt->device->sector_size;
1951 unsigned int resid;
1952 struct scsi_sense_hdr sshdr;
1953 struct scsi_disk *sdkp = scsi_disk(SCpnt->request->rq_disk);
1954 struct request *req = SCpnt->request;
1955 int sense_valid = 0;
1956 int sense_deferred = 0;
1957
1958 switch (req_op(req)) {
1959 case REQ_OP_DISCARD:
1960 case REQ_OP_WRITE_ZEROES:
1961 case REQ_OP_WRITE_SAME:
1962 case REQ_OP_ZONE_RESET:
1963 case REQ_OP_ZONE_RESET_ALL:
1964 if (!result) {
1965 good_bytes = blk_rq_bytes(req);
1966 scsi_set_resid(SCpnt, 0);
1967 } else {
1968 good_bytes = 0;
1969 scsi_set_resid(SCpnt, blk_rq_bytes(req));
1970 }
1971 break;
1972 default:
1973 /*
1974 * In case of bogus fw or device, we could end up having
1975 * an unaligned partial completion. Check this here and force
1976 * alignment.
1977 */
1978 resid = scsi_get_resid(SCpnt);
1979 if (resid & (sector_size - 1)) {
1980 sd_printk(KERN_INFO, sdkp,
1981 "Unaligned partial completion (resid=%u, sector_sz=%u)\n",
1982 resid, sector_size);
1983 scsi_print_command(SCpnt);
1984 resid = min(scsi_bufflen(SCpnt),
1985 round_up(resid, sector_size));
1986 scsi_set_resid(SCpnt, resid);
1987 }
1988 }
1989
1990 if (result) {
1991 sense_valid = scsi_command_normalize_sense(SCpnt, &sshdr);
1992 if (sense_valid)
1993 sense_deferred = scsi_sense_is_deferred(&sshdr);
1994 }
1995 sdkp->medium_access_timed_out = 0;
1996
1997 if (driver_byte(result) != DRIVER_SENSE &&
1998 (!sense_valid || sense_deferred))
1999 goto out;
2000
2001 switch (sshdr.sense_key) {
2002 case HARDWARE_ERROR:
2003 case MEDIUM_ERROR:
2004 good_bytes = sd_completed_bytes(SCpnt);
2005 break;
2006 case RECOVERED_ERROR:
2007 good_bytes = scsi_bufflen(SCpnt);
2008 break;
2009 case NO_SENSE:
2010 /* This indicates a false check condition, so ignore it. An
2011 * unknown amount of data was transferred so treat it as an
2012 * error.
2013 */
2014 SCpnt->result = 0;
2015 memset(SCpnt->sense_buffer, 0, SCSI_SENSE_BUFFERSIZE);
2016 break;
2017 case ABORTED_COMMAND:
2018 if (sshdr.asc == 0x10) /* DIF: Target detected corruption */
2019 good_bytes = sd_completed_bytes(SCpnt);
2020 break;
2021 case ILLEGAL_REQUEST:
2022 switch (sshdr.asc) {
2023 case 0x10: /* DIX: Host detected corruption */
2024 good_bytes = sd_completed_bytes(SCpnt);
2025 break;
2026 case 0x20: /* INVALID COMMAND OPCODE */
2027 case 0x24: /* INVALID FIELD IN CDB */
2028 switch (SCpnt->cmnd[0]) {
2029 case UNMAP:
2030 sd_config_discard(sdkp, SD_LBP_DISABLE);
2031 break;
2032 case WRITE_SAME_16:
2033 case WRITE_SAME:
2034 if (SCpnt->cmnd[1] & 8) { /* UNMAP */
2035 sd_config_discard(sdkp, SD_LBP_DISABLE);
2036 } else {
2037 sdkp->device->no_write_same = 1;
2038 sd_config_write_same(sdkp);
2039 req->rq_flags |= RQF_QUIET;
2040 }
2041 break;
2042 }
2043 }
2044 break;
2045 default:
2046 break;
2047 }
2048
2049 out:
2050 if (sd_is_zoned(sdkp))
2051 sd_zbc_complete(SCpnt, good_bytes, &sshdr);
2052
2053 SCSI_LOG_HLCOMPLETE(1, scmd_printk(KERN_INFO, SCpnt,
2054 "sd_done: completed %d of %d bytes\n",
2055 good_bytes, scsi_bufflen(SCpnt)));
2056
2057 return good_bytes;
2058}
2059
2060/*
2061 * spinup disk - called only in sd_revalidate_disk()
2062 */
2063static void
2064sd_spinup_disk(struct scsi_disk *sdkp)
2065{
2066 unsigned char cmd[10];
2067 unsigned long spintime_expire = 0;
2068 int retries, spintime;
2069 unsigned int the_result;
2070 struct scsi_sense_hdr sshdr;
2071 int sense_valid = 0;
2072
2073 spintime = 0;
2074
2075 /* Spin up drives, as required. Only do this at boot time */
2076 /* Spinup needs to be done for module loads too. */
2077 do {
2078 retries = 0;
2079
2080 do {
2081 cmd[0] = TEST_UNIT_READY;
2082 memset((void *) &cmd[1], 0, 9);
2083
2084 the_result = scsi_execute_req(sdkp->device, cmd,
2085 DMA_NONE, NULL, 0,
2086 &sshdr, SD_TIMEOUT,
2087 SD_MAX_RETRIES, NULL);
2088
2089 /*
2090 * If the drive has indicated to us that it
2091 * doesn't have any media in it, don't bother
2092 * with any more polling.
2093 */
2094 if (media_not_present(sdkp, &sshdr))
2095 return;
2096
2097 if (the_result)
2098 sense_valid = scsi_sense_valid(&sshdr);
2099 retries++;
2100 } while (retries < 3 &&
2101 (!scsi_status_is_good(the_result) ||
2102 ((driver_byte(the_result) == DRIVER_SENSE) &&
2103 sense_valid && sshdr.sense_key == UNIT_ATTENTION)));
2104
2105 if (driver_byte(the_result) != DRIVER_SENSE) {
2106 /* no sense, TUR either succeeded or failed
2107 * with a status error */
2108 if(!spintime && !scsi_status_is_good(the_result)) {
2109 sd_print_result(sdkp, "Test Unit Ready failed",
2110 the_result);
2111 }
2112 break;
2113 }
2114
2115 /*
2116 * The device does not want the automatic start to be issued.
2117 */
2118 if (sdkp->device->no_start_on_add)
2119 break;
2120
2121 if (sense_valid && sshdr.sense_key == NOT_READY) {
2122 if (sshdr.asc == 4 && sshdr.ascq == 3)
2123 break; /* manual intervention required */
2124 if (sshdr.asc == 4 && sshdr.ascq == 0xb)
2125 break; /* standby */
2126 if (sshdr.asc == 4 && sshdr.ascq == 0xc)
2127 break; /* unavailable */
2128 if (sshdr.asc == 4 && sshdr.ascq == 0x1b)
2129 break; /* sanitize in progress */
2130 /*
2131 * Issue command to spin up drive when not ready
2132 */
2133 if (!spintime) {
2134 sd_printk(KERN_NOTICE, sdkp, "Spinning up disk...");
2135 cmd[0] = START_STOP;
2136 cmd[1] = 1; /* Return immediately */
2137 memset((void *) &cmd[2], 0, 8);
2138 cmd[4] = 1; /* Start spin cycle */
2139 if (sdkp->device->start_stop_pwr_cond)
2140 cmd[4] |= 1 << 4;
2141 scsi_execute_req(sdkp->device, cmd, DMA_NONE,
2142 NULL, 0, &sshdr,
2143 SD_TIMEOUT, SD_MAX_RETRIES,
2144 NULL);
2145 spintime_expire = jiffies + 100 * HZ;
2146 spintime = 1;
2147 }
2148 /* Wait 1 second for next try */
2149 msleep(1000);
2150 printk(KERN_CONT ".");
2151
2152 /*
2153 * Wait for USB flash devices with slow firmware.
2154 * Yes, this sense key/ASC combination shouldn't
2155 * occur here. It's characteristic of these devices.
2156 */
2157 } else if (sense_valid &&
2158 sshdr.sense_key == UNIT_ATTENTION &&
2159 sshdr.asc == 0x28) {
2160 if (!spintime) {
2161 spintime_expire = jiffies + 5 * HZ;
2162 spintime = 1;
2163 }
2164 /* Wait 1 second for next try */
2165 msleep(1000);
2166 } else {
2167 /* we don't understand the sense code, so it's
2168 * probably pointless to loop */
2169 if(!spintime) {
2170 sd_printk(KERN_NOTICE, sdkp, "Unit Not Ready\n");
2171 sd_print_sense_hdr(sdkp, &sshdr);
2172 }
2173 break;
2174 }
2175
2176 } while (spintime && time_before_eq(jiffies, spintime_expire));
2177
2178 if (spintime) {
2179 if (scsi_status_is_good(the_result))
2180 printk(KERN_CONT "ready\n");
2181 else
2182 printk(KERN_CONT "not responding...\n");
2183 }
2184}
2185
2186/*
2187 * Determine whether disk supports Data Integrity Field.
2188 */
2189static int sd_read_protection_type(struct scsi_disk *sdkp, unsigned char *buffer)
2190{
2191 struct scsi_device *sdp = sdkp->device;
2192 u8 type;
2193 int ret = 0;
2194
2195 if (scsi_device_protection(sdp) == 0 || (buffer[12] & 1) == 0)
2196 return ret;
2197
2198 type = ((buffer[12] >> 1) & 7) + 1; /* P_TYPE 0 = Type 1 */
2199
2200 if (type > T10_PI_TYPE3_PROTECTION)
2201 ret = -ENODEV;
2202 else if (scsi_host_dif_capable(sdp->host, type))
2203 ret = 1;
2204
2205 if (sdkp->first_scan || type != sdkp->protection_type)
2206 switch (ret) {
2207 case -ENODEV:
2208 sd_printk(KERN_ERR, sdkp, "formatted with unsupported" \
2209 " protection type %u. Disabling disk!\n",
2210 type);
2211 break;
2212 case 1:
2213 sd_printk(KERN_NOTICE, sdkp,
2214 "Enabling DIF Type %u protection\n", type);
2215 break;
2216 case 0:
2217 sd_printk(KERN_NOTICE, sdkp,
2218 "Disabling DIF Type %u protection\n", type);
2219 break;
2220 }
2221
2222 sdkp->protection_type = type;
2223
2224 return ret;
2225}
2226
2227static void read_capacity_error(struct scsi_disk *sdkp, struct scsi_device *sdp,
2228 struct scsi_sense_hdr *sshdr, int sense_valid,
2229 int the_result)
2230{
2231 if (driver_byte(the_result) == DRIVER_SENSE)
2232 sd_print_sense_hdr(sdkp, sshdr);
2233 else
2234 sd_printk(KERN_NOTICE, sdkp, "Sense not available.\n");
2235
2236 /*
2237 * Set dirty bit for removable devices if not ready -
2238 * sometimes drives will not report this properly.
2239 */
2240 if (sdp->removable &&
2241 sense_valid && sshdr->sense_key == NOT_READY)
2242 set_media_not_present(sdkp);
2243
2244 /*
2245 * We used to set media_present to 0 here to indicate no media
2246 * in the drive, but some drives fail read capacity even with
2247 * media present, so we can't do that.
2248 */
2249 sdkp->capacity = 0; /* unknown mapped to zero - as usual */
2250}
2251
2252#define RC16_LEN 32
2253#if RC16_LEN > SD_BUF_SIZE
2254#error RC16_LEN must not be more than SD_BUF_SIZE
2255#endif
2256
2257#define READ_CAPACITY_RETRIES_ON_RESET 10
2258
2259static int read_capacity_16(struct scsi_disk *sdkp, struct scsi_device *sdp,
2260 unsigned char *buffer)
2261{
2262 unsigned char cmd[16];
2263 struct scsi_sense_hdr sshdr;
2264 int sense_valid = 0;
2265 int the_result;
2266 int retries = 3, reset_retries = READ_CAPACITY_RETRIES_ON_RESET;
2267 unsigned int alignment;
2268 unsigned long long lba;
2269 unsigned sector_size;
2270
2271 if (sdp->no_read_capacity_16)
2272 return -EINVAL;
2273
2274 do {
2275 memset(cmd, 0, 16);
2276 cmd[0] = SERVICE_ACTION_IN_16;
2277 cmd[1] = SAI_READ_CAPACITY_16;
2278 cmd[13] = RC16_LEN;
2279 memset(buffer, 0, RC16_LEN);
2280
2281 the_result = scsi_execute_req(sdp, cmd, DMA_FROM_DEVICE,
2282 buffer, RC16_LEN, &sshdr,
2283 SD_TIMEOUT, SD_MAX_RETRIES, NULL);
2284
2285 if (media_not_present(sdkp, &sshdr))
2286 return -ENODEV;
2287
2288 if (the_result) {
2289 sense_valid = scsi_sense_valid(&sshdr);
2290 if (sense_valid &&
2291 sshdr.sense_key == ILLEGAL_REQUEST &&
2292 (sshdr.asc == 0x20 || sshdr.asc == 0x24) &&
2293 sshdr.ascq == 0x00)
2294 /* Invalid Command Operation Code or
2295 * Invalid Field in CDB, just retry
2296 * silently with RC10 */
2297 return -EINVAL;
2298 if (sense_valid &&
2299 sshdr.sense_key == UNIT_ATTENTION &&
2300 sshdr.asc == 0x29 && sshdr.ascq == 0x00)
2301 /* Device reset might occur several times,
2302 * give it one more chance */
2303 if (--reset_retries > 0)
2304 continue;
2305 }
2306 retries--;
2307
2308 } while (the_result && retries);
2309
2310 if (the_result) {
2311 sd_print_result(sdkp, "Read Capacity(16) failed", the_result);
2312 read_capacity_error(sdkp, sdp, &sshdr, sense_valid, the_result);
2313 return -EINVAL;
2314 }
2315
2316 sector_size = get_unaligned_be32(&buffer[8]);
2317 lba = get_unaligned_be64(&buffer[0]);
2318
2319 if (sd_read_protection_type(sdkp, buffer) < 0) {
2320 sdkp->capacity = 0;
2321 return -ENODEV;
2322 }
2323
2324 /* Logical blocks per physical block exponent */
2325 sdkp->physical_block_size = (1 << (buffer[13] & 0xf)) * sector_size;
2326
2327 /* RC basis */
2328 sdkp->rc_basis = (buffer[12] >> 4) & 0x3;
2329
2330 /* Lowest aligned logical block */
2331 alignment = ((buffer[14] & 0x3f) << 8 | buffer[15]) * sector_size;
2332 blk_queue_alignment_offset(sdp->request_queue, alignment);
2333 if (alignment && sdkp->first_scan)
2334 sd_printk(KERN_NOTICE, sdkp,
2335 "physical block alignment offset: %u\n", alignment);
2336
2337 if (buffer[14] & 0x80) { /* LBPME */
2338 sdkp->lbpme = 1;
2339
2340 if (buffer[14] & 0x40) /* LBPRZ */
2341 sdkp->lbprz = 1;
2342
2343 sd_config_discard(sdkp, SD_LBP_WS16);
2344 }
2345
2346 sdkp->capacity = lba + 1;
2347 return sector_size;
2348}
2349
2350static int read_capacity_10(struct scsi_disk *sdkp, struct scsi_device *sdp,
2351 unsigned char *buffer)
2352{
2353 unsigned char cmd[16];
2354 struct scsi_sense_hdr sshdr;
2355 int sense_valid = 0;
2356 int the_result;
2357 int retries = 3, reset_retries = READ_CAPACITY_RETRIES_ON_RESET;
2358 sector_t lba;
2359 unsigned sector_size;
2360
2361 do {
2362 cmd[0] = READ_CAPACITY;
2363 memset(&cmd[1], 0, 9);
2364 memset(buffer, 0, 8);
2365
2366 the_result = scsi_execute_req(sdp, cmd, DMA_FROM_DEVICE,
2367 buffer, 8, &sshdr,
2368 SD_TIMEOUT, SD_MAX_RETRIES, NULL);
2369
2370 if (media_not_present(sdkp, &sshdr))
2371 return -ENODEV;
2372
2373 if (the_result) {
2374 sense_valid = scsi_sense_valid(&sshdr);
2375 if (sense_valid &&
2376 sshdr.sense_key == UNIT_ATTENTION &&
2377 sshdr.asc == 0x29 && sshdr.ascq == 0x00)
2378 /* Device reset might occur several times,
2379 * give it one more chance */
2380 if (--reset_retries > 0)
2381 continue;
2382 }
2383 retries--;
2384
2385 } while (the_result && retries);
2386
2387 if (the_result) {
2388 sd_print_result(sdkp, "Read Capacity(10) failed", the_result);
2389 read_capacity_error(sdkp, sdp, &sshdr, sense_valid, the_result);
2390 return -EINVAL;
2391 }
2392
2393 sector_size = get_unaligned_be32(&buffer[4]);
2394 lba = get_unaligned_be32(&buffer[0]);
2395
2396 if (sdp->no_read_capacity_16 && (lba == 0xffffffff)) {
2397 /* Some buggy (usb cardreader) devices return an lba of
2398 0xffffffff when the want to report a size of 0 (with
2399 which they really mean no media is present) */
2400 sdkp->capacity = 0;
2401 sdkp->physical_block_size = sector_size;
2402 return sector_size;
2403 }
2404
2405 sdkp->capacity = lba + 1;
2406 sdkp->physical_block_size = sector_size;
2407 return sector_size;
2408}
2409
2410static int sd_try_rc16_first(struct scsi_device *sdp)
2411{
2412 if (sdp->host->max_cmd_len < 16)
2413 return 0;
2414 if (sdp->try_rc_10_first)
2415 return 0;
2416 if (sdp->scsi_level > SCSI_SPC_2)
2417 return 1;
2418 if (scsi_device_protection(sdp))
2419 return 1;
2420 return 0;
2421}
2422
2423/*
2424 * read disk capacity
2425 */
2426static void
2427sd_read_capacity(struct scsi_disk *sdkp, unsigned char *buffer)
2428{
2429 int sector_size;
2430 struct scsi_device *sdp = sdkp->device;
2431
2432 if (sd_try_rc16_first(sdp)) {
2433 sector_size = read_capacity_16(sdkp, sdp, buffer);
2434 if (sector_size == -EOVERFLOW)
2435 goto got_data;
2436 if (sector_size == -ENODEV)
2437 return;
2438 if (sector_size < 0)
2439 sector_size = read_capacity_10(sdkp, sdp, buffer);
2440 if (sector_size < 0)
2441 return;
2442 } else {
2443 sector_size = read_capacity_10(sdkp, sdp, buffer);
2444 if (sector_size == -EOVERFLOW)
2445 goto got_data;
2446 if (sector_size < 0)
2447 return;
2448 if ((sizeof(sdkp->capacity) > 4) &&
2449 (sdkp->capacity > 0xffffffffULL)) {
2450 int old_sector_size = sector_size;
2451 sd_printk(KERN_NOTICE, sdkp, "Very big device. "
2452 "Trying to use READ CAPACITY(16).\n");
2453 sector_size = read_capacity_16(sdkp, sdp, buffer);
2454 if (sector_size < 0) {
2455 sd_printk(KERN_NOTICE, sdkp,
2456 "Using 0xffffffff as device size\n");
2457 sdkp->capacity = 1 + (sector_t) 0xffffffff;
2458 sector_size = old_sector_size;
2459 goto got_data;
2460 }
2461 /* Remember that READ CAPACITY(16) succeeded */
2462 sdp->try_rc_10_first = 0;
2463 }
2464 }
2465
2466 /* Some devices are known to return the total number of blocks,
2467 * not the highest block number. Some devices have versions
2468 * which do this and others which do not. Some devices we might
2469 * suspect of doing this but we don't know for certain.
2470 *
2471 * If we know the reported capacity is wrong, decrement it. If
2472 * we can only guess, then assume the number of blocks is even
2473 * (usually true but not always) and err on the side of lowering
2474 * the capacity.
2475 */
2476 if (sdp->fix_capacity ||
2477 (sdp->guess_capacity && (sdkp->capacity & 0x01))) {
2478 sd_printk(KERN_INFO, sdkp, "Adjusting the sector count "
2479 "from its reported value: %llu\n",
2480 (unsigned long long) sdkp->capacity);
2481 --sdkp->capacity;
2482 }
2483
2484got_data:
2485 if (sector_size == 0) {
2486 sector_size = 512;
2487 sd_printk(KERN_NOTICE, sdkp, "Sector size 0 reported, "
2488 "assuming 512.\n");
2489 }
2490
2491 if (sector_size != 512 &&
2492 sector_size != 1024 &&
2493 sector_size != 2048 &&
2494 sector_size != 4096) {
2495 sd_printk(KERN_NOTICE, sdkp, "Unsupported sector size %d.\n",
2496 sector_size);
2497 /*
2498 * The user might want to re-format the drive with
2499 * a supported sectorsize. Once this happens, it
2500 * would be relatively trivial to set the thing up.
2501 * For this reason, we leave the thing in the table.
2502 */
2503 sdkp->capacity = 0;
2504 /*
2505 * set a bogus sector size so the normal read/write
2506 * logic in the block layer will eventually refuse any
2507 * request on this device without tripping over power
2508 * of two sector size assumptions
2509 */
2510 sector_size = 512;
2511 }
2512 blk_queue_logical_block_size(sdp->request_queue, sector_size);
2513 blk_queue_physical_block_size(sdp->request_queue,
2514 sdkp->physical_block_size);
2515 sdkp->device->sector_size = sector_size;
2516
2517 if (sdkp->capacity > 0xffffffff)
2518 sdp->use_16_for_rw = 1;
2519
2520}
2521
2522/*
2523 * Print disk capacity
2524 */
2525static void
2526sd_print_capacity(struct scsi_disk *sdkp,
2527 sector_t old_capacity)
2528{
2529 int sector_size = sdkp->device->sector_size;
2530 char cap_str_2[10], cap_str_10[10];
2531
2532 if (!sdkp->first_scan && old_capacity == sdkp->capacity)
2533 return;
2534
2535 string_get_size(sdkp->capacity, sector_size,
2536 STRING_UNITS_2, cap_str_2, sizeof(cap_str_2));
2537 string_get_size(sdkp->capacity, sector_size,
2538 STRING_UNITS_10, cap_str_10, sizeof(cap_str_10));
2539
2540 sd_printk(KERN_NOTICE, sdkp,
2541 "%llu %d-byte logical blocks: (%s/%s)\n",
2542 (unsigned long long)sdkp->capacity,
2543 sector_size, cap_str_10, cap_str_2);
2544
2545 if (sdkp->physical_block_size != sector_size)
2546 sd_printk(KERN_NOTICE, sdkp,
2547 "%u-byte physical blocks\n",
2548 sdkp->physical_block_size);
2549
2550 sd_zbc_print_zones(sdkp);
2551}
2552
2553/* called with buffer of length 512 */
2554static inline int
2555sd_do_mode_sense(struct scsi_device *sdp, int dbd, int modepage,
2556 unsigned char *buffer, int len, struct scsi_mode_data *data,
2557 struct scsi_sense_hdr *sshdr)
2558{
2559 return scsi_mode_sense(sdp, dbd, modepage, buffer, len,
2560 SD_TIMEOUT, SD_MAX_RETRIES, data,
2561 sshdr);
2562}
2563
2564/*
2565 * read write protect setting, if possible - called only in sd_revalidate_disk()
2566 * called with buffer of length SD_BUF_SIZE
2567 */
2568static void
2569sd_read_write_protect_flag(struct scsi_disk *sdkp, unsigned char *buffer)
2570{
2571 int res;
2572 struct scsi_device *sdp = sdkp->device;
2573 struct scsi_mode_data data;
2574 int old_wp = sdkp->write_prot;
2575
2576 set_disk_ro(sdkp->disk, 0);
2577 if (sdp->skip_ms_page_3f) {
2578 sd_first_printk(KERN_NOTICE, sdkp, "Assuming Write Enabled\n");
2579 return;
2580 }
2581
2582 if (sdp->use_192_bytes_for_3f) {
2583 res = sd_do_mode_sense(sdp, 0, 0x3F, buffer, 192, &data, NULL);
2584 } else {
2585 /*
2586 * First attempt: ask for all pages (0x3F), but only 4 bytes.
2587 * We have to start carefully: some devices hang if we ask
2588 * for more than is available.
2589 */
2590 res = sd_do_mode_sense(sdp, 0, 0x3F, buffer, 4, &data, NULL);
2591
2592 /*
2593 * Second attempt: ask for page 0 When only page 0 is
2594 * implemented, a request for page 3F may return Sense Key
2595 * 5: Illegal Request, Sense Code 24: Invalid field in
2596 * CDB.
2597 */
2598 if (!scsi_status_is_good(res))
2599 res = sd_do_mode_sense(sdp, 0, 0, buffer, 4, &data, NULL);
2600
2601 /*
2602 * Third attempt: ask 255 bytes, as we did earlier.
2603 */
2604 if (!scsi_status_is_good(res))
2605 res = sd_do_mode_sense(sdp, 0, 0x3F, buffer, 255,
2606 &data, NULL);
2607 }
2608
2609 if (!scsi_status_is_good(res)) {
2610 sd_first_printk(KERN_WARNING, sdkp,
2611 "Test WP failed, assume Write Enabled\n");
2612 } else {
2613 sdkp->write_prot = ((data.device_specific & 0x80) != 0);
2614 set_disk_ro(sdkp->disk, sdkp->write_prot);
2615 if (sdkp->first_scan || old_wp != sdkp->write_prot) {
2616 sd_printk(KERN_NOTICE, sdkp, "Write Protect is %s\n",
2617 sdkp->write_prot ? "on" : "off");
2618 sd_printk(KERN_DEBUG, sdkp, "Mode Sense: %4ph\n", buffer);
2619 }
2620 }
2621}
2622
2623/*
2624 * sd_read_cache_type - called only from sd_revalidate_disk()
2625 * called with buffer of length SD_BUF_SIZE
2626 */
2627static void
2628sd_read_cache_type(struct scsi_disk *sdkp, unsigned char *buffer)
2629{
2630 int len = 0, res;
2631 struct scsi_device *sdp = sdkp->device;
2632
2633 int dbd;
2634 int modepage;
2635 int first_len;
2636 struct scsi_mode_data data;
2637 struct scsi_sense_hdr sshdr;
2638 int old_wce = sdkp->WCE;
2639 int old_rcd = sdkp->RCD;
2640 int old_dpofua = sdkp->DPOFUA;
2641
2642
2643 if (sdkp->cache_override)
2644 return;
2645
2646 first_len = 4;
2647 if (sdp->skip_ms_page_8) {
2648 if (sdp->type == TYPE_RBC)
2649 goto defaults;
2650 else {
2651 if (sdp->skip_ms_page_3f)
2652 goto defaults;
2653 modepage = 0x3F;
2654 if (sdp->use_192_bytes_for_3f)
2655 first_len = 192;
2656 dbd = 0;
2657 }
2658 } else if (sdp->type == TYPE_RBC) {
2659 modepage = 6;
2660 dbd = 8;
2661 } else {
2662 modepage = 8;
2663 dbd = 0;
2664 }
2665
2666 /* cautiously ask */
2667 res = sd_do_mode_sense(sdp, dbd, modepage, buffer, first_len,
2668 &data, &sshdr);
2669
2670 if (!scsi_status_is_good(res))
2671 goto bad_sense;
2672
2673 if (!data.header_length) {
2674 modepage = 6;
2675 first_len = 0;
2676 sd_first_printk(KERN_ERR, sdkp,
2677 "Missing header in MODE_SENSE response\n");
2678 }
2679
2680 /* that went OK, now ask for the proper length */
2681 len = data.length;
2682
2683 /*
2684 * We're only interested in the first three bytes, actually.
2685 * But the data cache page is defined for the first 20.
2686 */
2687 if (len < 3)
2688 goto bad_sense;
2689 else if (len > SD_BUF_SIZE) {
2690 sd_first_printk(KERN_NOTICE, sdkp, "Truncating mode parameter "
2691 "data from %d to %d bytes\n", len, SD_BUF_SIZE);
2692 len = SD_BUF_SIZE;
2693 }
2694 if (modepage == 0x3F && sdp->use_192_bytes_for_3f)
2695 len = 192;
2696
2697 /* Get the data */
2698 if (len > first_len)
2699 res = sd_do_mode_sense(sdp, dbd, modepage, buffer, len,
2700 &data, &sshdr);
2701
2702 if (scsi_status_is_good(res)) {
2703 int offset = data.header_length + data.block_descriptor_length;
2704
2705 while (offset < len) {
2706 u8 page_code = buffer[offset] & 0x3F;
2707 u8 spf = buffer[offset] & 0x40;
2708
2709 if (page_code == 8 || page_code == 6) {
2710 /* We're interested only in the first 3 bytes.
2711 */
2712 if (len - offset <= 2) {
2713 sd_first_printk(KERN_ERR, sdkp,
2714 "Incomplete mode parameter "
2715 "data\n");
2716 goto defaults;
2717 } else {
2718 modepage = page_code;
2719 goto Page_found;
2720 }
2721 } else {
2722 /* Go to the next page */
2723 if (spf && len - offset > 3)
2724 offset += 4 + (buffer[offset+2] << 8) +
2725 buffer[offset+3];
2726 else if (!spf && len - offset > 1)
2727 offset += 2 + buffer[offset+1];
2728 else {
2729 sd_first_printk(KERN_ERR, sdkp,
2730 "Incomplete mode "
2731 "parameter data\n");
2732 goto defaults;
2733 }
2734 }
2735 }
2736
2737 sd_first_printk(KERN_ERR, sdkp, "No Caching mode page found\n");
2738 goto defaults;
2739
2740 Page_found:
2741 if (modepage == 8) {
2742 sdkp->WCE = ((buffer[offset + 2] & 0x04) != 0);
2743 sdkp->RCD = ((buffer[offset + 2] & 0x01) != 0);
2744 } else {
2745 sdkp->WCE = ((buffer[offset + 2] & 0x01) == 0);
2746 sdkp->RCD = 0;
2747 }
2748
2749 sdkp->DPOFUA = (data.device_specific & 0x10) != 0;
2750 if (sdp->broken_fua) {
2751 sd_first_printk(KERN_NOTICE, sdkp, "Disabling FUA\n");
2752 sdkp->DPOFUA = 0;
2753 } else if (sdkp->DPOFUA && !sdkp->device->use_10_for_rw &&
2754 !sdkp->device->use_16_for_rw) {
2755 sd_first_printk(KERN_NOTICE, sdkp,
2756 "Uses READ/WRITE(6), disabling FUA\n");
2757 sdkp->DPOFUA = 0;
2758 }
2759
2760 /* No cache flush allowed for write protected devices */
2761 if (sdkp->WCE && sdkp->write_prot)
2762 sdkp->WCE = 0;
2763
2764 if (sdkp->first_scan || old_wce != sdkp->WCE ||
2765 old_rcd != sdkp->RCD || old_dpofua != sdkp->DPOFUA)
2766 sd_printk(KERN_NOTICE, sdkp,
2767 "Write cache: %s, read cache: %s, %s\n",
2768 sdkp->WCE ? "enabled" : "disabled",
2769 sdkp->RCD ? "disabled" : "enabled",
2770 sdkp->DPOFUA ? "supports DPO and FUA"
2771 : "doesn't support DPO or FUA");
2772
2773 return;
2774 }
2775
2776bad_sense:
2777 if (scsi_sense_valid(&sshdr) &&
2778 sshdr.sense_key == ILLEGAL_REQUEST &&
2779 sshdr.asc == 0x24 && sshdr.ascq == 0x0)
2780 /* Invalid field in CDB */
2781 sd_first_printk(KERN_NOTICE, sdkp, "Cache data unavailable\n");
2782 else
2783 sd_first_printk(KERN_ERR, sdkp,
2784 "Asking for cache data failed\n");
2785
2786defaults:
2787 if (sdp->wce_default_on) {
2788 sd_first_printk(KERN_NOTICE, sdkp,
2789 "Assuming drive cache: write back\n");
2790 sdkp->WCE = 1;
2791 } else {
2792 sd_first_printk(KERN_ERR, sdkp,
2793 "Assuming drive cache: write through\n");
2794 sdkp->WCE = 0;
2795 }
2796 sdkp->RCD = 0;
2797 sdkp->DPOFUA = 0;
2798}
2799
2800/*
2801 * The ATO bit indicates whether the DIF application tag is available
2802 * for use by the operating system.
2803 */
2804static void sd_read_app_tag_own(struct scsi_disk *sdkp, unsigned char *buffer)
2805{
2806 int res, offset;
2807 struct scsi_device *sdp = sdkp->device;
2808 struct scsi_mode_data data;
2809 struct scsi_sense_hdr sshdr;
2810
2811 if (sdp->type != TYPE_DISK && sdp->type != TYPE_ZBC)
2812 return;
2813
2814 if (sdkp->protection_type == 0)
2815 return;
2816
2817 res = scsi_mode_sense(sdp, 1, 0x0a, buffer, 36, SD_TIMEOUT,
2818 SD_MAX_RETRIES, &data, &sshdr);
2819
2820 if (!scsi_status_is_good(res) || !data.header_length ||
2821 data.length < 6) {
2822 sd_first_printk(KERN_WARNING, sdkp,
2823 "getting Control mode page failed, assume no ATO\n");
2824
2825 if (scsi_sense_valid(&sshdr))
2826 sd_print_sense_hdr(sdkp, &sshdr);
2827
2828 return;
2829 }
2830
2831 offset = data.header_length + data.block_descriptor_length;
2832
2833 if ((buffer[offset] & 0x3f) != 0x0a) {
2834 sd_first_printk(KERN_ERR, sdkp, "ATO Got wrong page\n");
2835 return;
2836 }
2837
2838 if ((buffer[offset + 5] & 0x80) == 0)
2839 return;
2840
2841 sdkp->ATO = 1;
2842
2843 return;
2844}
2845
2846/**
2847 * sd_read_block_limits - Query disk device for preferred I/O sizes.
2848 * @sdkp: disk to query
2849 */
2850static void sd_read_block_limits(struct scsi_disk *sdkp)
2851{
2852 unsigned int sector_sz = sdkp->device->sector_size;
2853 const int vpd_len = 64;
2854 unsigned char *buffer = kmalloc(vpd_len, GFP_KERNEL);
2855
2856 if (!buffer ||
2857 /* Block Limits VPD */
2858 scsi_get_vpd_page(sdkp->device, 0xb0, buffer, vpd_len))
2859 goto out;
2860
2861 blk_queue_io_min(sdkp->disk->queue,
2862 get_unaligned_be16(&buffer[6]) * sector_sz);
2863
2864 sdkp->max_xfer_blocks = get_unaligned_be32(&buffer[8]);
2865 sdkp->opt_xfer_blocks = get_unaligned_be32(&buffer[12]);
2866
2867 if (buffer[3] == 0x3c) {
2868 unsigned int lba_count, desc_count;
2869
2870 sdkp->max_ws_blocks = (u32)get_unaligned_be64(&buffer[36]);
2871
2872 if (!sdkp->lbpme)
2873 goto out;
2874
2875 lba_count = get_unaligned_be32(&buffer[20]);
2876 desc_count = get_unaligned_be32(&buffer[24]);
2877
2878 if (lba_count && desc_count)
2879 sdkp->max_unmap_blocks = lba_count;
2880
2881 sdkp->unmap_granularity = get_unaligned_be32(&buffer[28]);
2882
2883 if (buffer[32] & 0x80)
2884 sdkp->unmap_alignment =
2885 get_unaligned_be32(&buffer[32]) & ~(1 << 31);
2886
2887 if (!sdkp->lbpvpd) { /* LBP VPD page not provided */
2888
2889 if (sdkp->max_unmap_blocks)
2890 sd_config_discard(sdkp, SD_LBP_UNMAP);
2891 else
2892 sd_config_discard(sdkp, SD_LBP_WS16);
2893
2894 } else { /* LBP VPD page tells us what to use */
2895 if (sdkp->lbpu && sdkp->max_unmap_blocks)
2896 sd_config_discard(sdkp, SD_LBP_UNMAP);
2897 else if (sdkp->lbpws)
2898 sd_config_discard(sdkp, SD_LBP_WS16);
2899 else if (sdkp->lbpws10)
2900 sd_config_discard(sdkp, SD_LBP_WS10);
2901 else
2902 sd_config_discard(sdkp, SD_LBP_DISABLE);
2903 }
2904 }
2905
2906 out:
2907 kfree(buffer);
2908}
2909
2910/**
2911 * sd_read_block_characteristics - Query block dev. characteristics
2912 * @sdkp: disk to query
2913 */
2914static void sd_read_block_characteristics(struct scsi_disk *sdkp)
2915{
2916 struct request_queue *q = sdkp->disk->queue;
2917 unsigned char *buffer;
2918 u16 rot;
2919 const int vpd_len = 64;
2920
2921 buffer = kmalloc(vpd_len, GFP_KERNEL);
2922
2923 if (!buffer ||
2924 /* Block Device Characteristics VPD */
2925 scsi_get_vpd_page(sdkp->device, 0xb1, buffer, vpd_len))
2926 goto out;
2927
2928 rot = get_unaligned_be16(&buffer[4]);
2929
2930 if (rot == 1) {
2931 blk_queue_flag_set(QUEUE_FLAG_NONROT, q);
2932 blk_queue_flag_clear(QUEUE_FLAG_ADD_RANDOM, q);
2933 }
2934
2935 if (sdkp->device->type == TYPE_ZBC) {
2936 /* Host-managed */
2937 q->limits.zoned = BLK_ZONED_HM;
2938 } else {
2939 sdkp->zoned = (buffer[8] >> 4) & 3;
2940 if (sdkp->zoned == 1)
2941 /* Host-aware */
2942 q->limits.zoned = BLK_ZONED_HA;
2943 else
2944 /*
2945 * Treat drive-managed devices as
2946 * regular block devices.
2947 */
2948 q->limits.zoned = BLK_ZONED_NONE;
2949 }
2950 if (blk_queue_is_zoned(q) && sdkp->first_scan)
2951 sd_printk(KERN_NOTICE, sdkp, "Host-%s zoned block device\n",
2952 q->limits.zoned == BLK_ZONED_HM ? "managed" : "aware");
2953
2954 out:
2955 kfree(buffer);
2956}
2957
2958/**
2959 * sd_read_block_provisioning - Query provisioning VPD page
2960 * @sdkp: disk to query
2961 */
2962static void sd_read_block_provisioning(struct scsi_disk *sdkp)
2963{
2964 unsigned char *buffer;
2965 const int vpd_len = 8;
2966
2967 if (sdkp->lbpme == 0)
2968 return;
2969
2970 buffer = kmalloc(vpd_len, GFP_KERNEL);
2971
2972 if (!buffer || scsi_get_vpd_page(sdkp->device, 0xb2, buffer, vpd_len))
2973 goto out;
2974
2975 sdkp->lbpvpd = 1;
2976 sdkp->lbpu = (buffer[5] >> 7) & 1; /* UNMAP */
2977 sdkp->lbpws = (buffer[5] >> 6) & 1; /* WRITE SAME(16) with UNMAP */
2978 sdkp->lbpws10 = (buffer[5] >> 5) & 1; /* WRITE SAME(10) with UNMAP */
2979
2980 out:
2981 kfree(buffer);
2982}
2983
2984static void sd_read_write_same(struct scsi_disk *sdkp, unsigned char *buffer)
2985{
2986 struct scsi_device *sdev = sdkp->device;
2987
2988 if (sdev->host->no_write_same) {
2989 sdev->no_write_same = 1;
2990
2991 return;
2992 }
2993
2994 if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE, INQUIRY) < 0) {
2995 /* too large values might cause issues with arcmsr */
2996 int vpd_buf_len = 64;
2997
2998 sdev->no_report_opcodes = 1;
2999
3000 /* Disable WRITE SAME if REPORT SUPPORTED OPERATION
3001 * CODES is unsupported and the device has an ATA
3002 * Information VPD page (SAT).
3003 */
3004 if (!scsi_get_vpd_page(sdev, 0x89, buffer, vpd_buf_len))
3005 sdev->no_write_same = 1;
3006 }
3007
3008 if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE, WRITE_SAME_16) == 1)
3009 sdkp->ws16 = 1;
3010
3011 if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE, WRITE_SAME) == 1)
3012 sdkp->ws10 = 1;
3013}
3014
3015static void sd_read_security(struct scsi_disk *sdkp, unsigned char *buffer)
3016{
3017 struct scsi_device *sdev = sdkp->device;
3018
3019 if (!sdev->security_supported)
3020 return;
3021
3022 if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE,
3023 SECURITY_PROTOCOL_IN) == 1 &&
3024 scsi_report_opcode(sdev, buffer, SD_BUF_SIZE,
3025 SECURITY_PROTOCOL_OUT) == 1)
3026 sdkp->security = 1;
3027}
3028
3029/*
3030 * Determine the device's preferred I/O size for reads and writes
3031 * unless the reported value is unreasonably small, large, not a
3032 * multiple of the physical block size, or simply garbage.
3033 */
3034static bool sd_validate_opt_xfer_size(struct scsi_disk *sdkp,
3035 unsigned int dev_max)
3036{
3037 struct scsi_device *sdp = sdkp->device;
3038 unsigned int opt_xfer_bytes =
3039 logical_to_bytes(sdp, sdkp->opt_xfer_blocks);
3040
3041 if (sdkp->opt_xfer_blocks == 0)
3042 return false;
3043
3044 if (sdkp->opt_xfer_blocks > dev_max) {
3045 sd_first_printk(KERN_WARNING, sdkp,
3046 "Optimal transfer size %u logical blocks " \
3047 "> dev_max (%u logical blocks)\n",
3048 sdkp->opt_xfer_blocks, dev_max);
3049 return false;
3050 }
3051
3052 if (sdkp->opt_xfer_blocks > SD_DEF_XFER_BLOCKS) {
3053 sd_first_printk(KERN_WARNING, sdkp,
3054 "Optimal transfer size %u logical blocks " \
3055 "> sd driver limit (%u logical blocks)\n",
3056 sdkp->opt_xfer_blocks, SD_DEF_XFER_BLOCKS);
3057 return false;
3058 }
3059
3060 if (opt_xfer_bytes < PAGE_SIZE) {
3061 sd_first_printk(KERN_WARNING, sdkp,
3062 "Optimal transfer size %u bytes < " \
3063 "PAGE_SIZE (%u bytes)\n",
3064 opt_xfer_bytes, (unsigned int)PAGE_SIZE);
3065 return false;
3066 }
3067
3068 if (opt_xfer_bytes & (sdkp->physical_block_size - 1)) {
3069 sd_first_printk(KERN_WARNING, sdkp,
3070 "Optimal transfer size %u bytes not a " \
3071 "multiple of physical block size (%u bytes)\n",
3072 opt_xfer_bytes, sdkp->physical_block_size);
3073 return false;
3074 }
3075
3076 sd_first_printk(KERN_INFO, sdkp, "Optimal transfer size %u bytes\n",
3077 opt_xfer_bytes);
3078 return true;
3079}
3080
3081/**
3082 * sd_revalidate_disk - called the first time a new disk is seen,
3083 * performs disk spin up, read_capacity, etc.
3084 * @disk: struct gendisk we care about
3085 **/
3086static int sd_revalidate_disk(struct gendisk *disk)
3087{
3088 struct scsi_disk *sdkp = scsi_disk(disk);
3089 struct scsi_device *sdp = sdkp->device;
3090 struct request_queue *q = sdkp->disk->queue;
3091 sector_t old_capacity = sdkp->capacity;
3092 unsigned char *buffer;
3093 unsigned int dev_max, rw_max;
3094
3095 SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp,
3096 "sd_revalidate_disk\n"));
3097
3098 /*
3099 * If the device is offline, don't try and read capacity or any
3100 * of the other niceties.
3101 */
3102 if (!scsi_device_online(sdp))
3103 goto out;
3104
3105 buffer = kmalloc(SD_BUF_SIZE, GFP_KERNEL);
3106 if (!buffer) {
3107 sd_printk(KERN_WARNING, sdkp, "sd_revalidate_disk: Memory "
3108 "allocation failure.\n");
3109 goto out;
3110 }
3111
3112 sd_spinup_disk(sdkp);
3113
3114 /*
3115 * Without media there is no reason to ask; moreover, some devices
3116 * react badly if we do.
3117 */
3118 if (sdkp->media_present) {
3119 sd_read_capacity(sdkp, buffer);
3120
3121 /*
3122 * set the default to rotational. All non-rotational devices
3123 * support the block characteristics VPD page, which will
3124 * cause this to be updated correctly and any device which
3125 * doesn't support it should be treated as rotational.
3126 */
3127 blk_queue_flag_clear(QUEUE_FLAG_NONROT, q);
3128 blk_queue_flag_set(QUEUE_FLAG_ADD_RANDOM, q);
3129
3130 if (scsi_device_supports_vpd(sdp)) {
3131 sd_read_block_provisioning(sdkp);
3132 sd_read_block_limits(sdkp);
3133 sd_read_block_characteristics(sdkp);
3134 sd_zbc_read_zones(sdkp, buffer);
3135 }
3136
3137 sd_print_capacity(sdkp, old_capacity);
3138
3139 sd_read_write_protect_flag(sdkp, buffer);
3140 sd_read_cache_type(sdkp, buffer);
3141 sd_read_app_tag_own(sdkp, buffer);
3142 sd_read_write_same(sdkp, buffer);
3143 sd_read_security(sdkp, buffer);
3144 }
3145
3146 /*
3147 * We now have all cache related info, determine how we deal
3148 * with flush requests.
3149 */
3150 sd_set_flush_flag(sdkp);
3151
3152 /* Initial block count limit based on CDB TRANSFER LENGTH field size. */
3153 dev_max = sdp->use_16_for_rw ? SD_MAX_XFER_BLOCKS : SD_DEF_XFER_BLOCKS;
3154
3155 /* Some devices report a maximum block count for READ/WRITE requests. */
3156 dev_max = min_not_zero(dev_max, sdkp->max_xfer_blocks);
3157 q->limits.max_dev_sectors = logical_to_sectors(sdp, dev_max);
3158
3159 if (sd_validate_opt_xfer_size(sdkp, dev_max)) {
3160 q->limits.io_opt = logical_to_bytes(sdp, sdkp->opt_xfer_blocks);
3161 rw_max = logical_to_sectors(sdp, sdkp->opt_xfer_blocks);
3162 } else
3163 rw_max = min_not_zero(logical_to_sectors(sdp, dev_max),
3164 (sector_t)BLK_DEF_MAX_SECTORS);
3165
3166 /* Do not exceed controller limit */
3167 rw_max = min(rw_max, queue_max_hw_sectors(q));
3168
3169 /*
3170 * Only update max_sectors if previously unset or if the current value
3171 * exceeds the capabilities of the hardware.
3172 */
3173 if (sdkp->first_scan ||
3174 q->limits.max_sectors > q->limits.max_dev_sectors ||
3175 q->limits.max_sectors > q->limits.max_hw_sectors)
3176 q->limits.max_sectors = rw_max;
3177
3178 sdkp->first_scan = 0;
3179
3180 set_capacity(disk, logical_to_sectors(sdp, sdkp->capacity));
3181 sd_config_write_same(sdkp);
3182 kfree(buffer);
3183
3184 out:
3185 return 0;
3186}
3187
3188/**
3189 * sd_unlock_native_capacity - unlock native capacity
3190 * @disk: struct gendisk to set capacity for
3191 *
3192 * Block layer calls this function if it detects that partitions
3193 * on @disk reach beyond the end of the device. If the SCSI host
3194 * implements ->unlock_native_capacity() method, it's invoked to
3195 * give it a chance to adjust the device capacity.
3196 *
3197 * CONTEXT:
3198 * Defined by block layer. Might sleep.
3199 */
3200static void sd_unlock_native_capacity(struct gendisk *disk)
3201{
3202 struct scsi_device *sdev = scsi_disk(disk)->device;
3203
3204 if (sdev->host->hostt->unlock_native_capacity)
3205 sdev->host->hostt->unlock_native_capacity(sdev);
3206}
3207
3208/**
3209 * sd_format_disk_name - format disk name
3210 * @prefix: name prefix - ie. "sd" for SCSI disks
3211 * @index: index of the disk to format name for
3212 * @buf: output buffer
3213 * @buflen: length of the output buffer
3214 *
3215 * SCSI disk names starts at sda. The 26th device is sdz and the
3216 * 27th is sdaa. The last one for two lettered suffix is sdzz
3217 * which is followed by sdaaa.
3218 *
3219 * This is basically 26 base counting with one extra 'nil' entry
3220 * at the beginning from the second digit on and can be
3221 * determined using similar method as 26 base conversion with the
3222 * index shifted -1 after each digit is computed.
3223 *
3224 * CONTEXT:
3225 * Don't care.
3226 *
3227 * RETURNS:
3228 * 0 on success, -errno on failure.
3229 */
3230static int sd_format_disk_name(char *prefix, int index, char *buf, int buflen)
3231{
3232 const int base = 'z' - 'a' + 1;
3233 char *begin = buf + strlen(prefix);
3234 char *end = buf + buflen;
3235 char *p;
3236 int unit;
3237
3238 p = end - 1;
3239 *p = '\0';
3240 unit = base;
3241 do {
3242 if (p == begin)
3243 return -EINVAL;
3244 *--p = 'a' + (index % unit);
3245 index = (index / unit) - 1;
3246 } while (index >= 0);
3247
3248 memmove(begin, p, end - p);
3249 memcpy(buf, prefix, strlen(prefix));
3250
3251 return 0;
3252}
3253
3254/**
3255 * sd_probe - called during driver initialization and whenever a
3256 * new scsi device is attached to the system. It is called once
3257 * for each scsi device (not just disks) present.
3258 * @dev: pointer to device object
3259 *
3260 * Returns 0 if successful (or not interested in this scsi device
3261 * (e.g. scanner)); 1 when there is an error.
3262 *
3263 * Note: this function is invoked from the scsi mid-level.
3264 * This function sets up the mapping between a given
3265 * <host,channel,id,lun> (found in sdp) and new device name
3266 * (e.g. /dev/sda). More precisely it is the block device major
3267 * and minor number that is chosen here.
3268 *
3269 * Assume sd_probe is not re-entrant (for time being)
3270 * Also think about sd_probe() and sd_remove() running coincidentally.
3271 **/
3272static int sd_probe(struct device *dev)
3273{
3274 struct scsi_device *sdp = to_scsi_device(dev);
3275 struct scsi_disk *sdkp;
3276 struct gendisk *gd;
3277 int index;
3278 int error;
3279
3280 scsi_autopm_get_device(sdp);
3281 error = -ENODEV;
3282 if (sdp->type != TYPE_DISK &&
3283 sdp->type != TYPE_ZBC &&
3284 sdp->type != TYPE_MOD &&
3285 sdp->type != TYPE_RBC)
3286 goto out;
3287
3288#ifndef CONFIG_BLK_DEV_ZONED
3289 if (sdp->type == TYPE_ZBC)
3290 goto out;
3291#endif
3292 SCSI_LOG_HLQUEUE(3, sdev_printk(KERN_INFO, sdp,
3293 "sd_probe\n"));
3294
3295 error = -ENOMEM;
3296 sdkp = kzalloc(sizeof(*sdkp), GFP_KERNEL);
3297 if (!sdkp)
3298 goto out;
3299
3300 gd = alloc_disk(SD_MINORS);
3301 if (!gd)
3302 goto out_free;
3303
3304 index = ida_alloc(&sd_index_ida, GFP_KERNEL);
3305 if (index < 0) {
3306 sdev_printk(KERN_WARNING, sdp, "sd_probe: memory exhausted.\n");
3307 goto out_put;
3308 }
3309
3310 error = sd_format_disk_name("sd", index, gd->disk_name, DISK_NAME_LEN);
3311 if (error) {
3312 sdev_printk(KERN_WARNING, sdp, "SCSI disk (sd) name length exceeded.\n");
3313 goto out_free_index;
3314 }
3315
3316 sdkp->device = sdp;
3317 sdkp->driver = &sd_template;
3318 sdkp->disk = gd;
3319 sdkp->index = index;
3320 atomic_set(&sdkp->openers, 0);
3321 atomic_set(&sdkp->device->ioerr_cnt, 0);
3322
3323 if (!sdp->request_queue->rq_timeout) {
3324 if (sdp->type != TYPE_MOD)
3325 blk_queue_rq_timeout(sdp->request_queue, SD_TIMEOUT);
3326 else
3327 blk_queue_rq_timeout(sdp->request_queue,
3328 SD_MOD_TIMEOUT);
3329 }
3330
3331 device_initialize(&sdkp->dev);
3332 sdkp->dev.parent = dev;
3333 sdkp->dev.class = &sd_disk_class;
3334 dev_set_name(&sdkp->dev, "%s", dev_name(dev));
3335
3336 error = device_add(&sdkp->dev);
3337 if (error)
3338 goto out_free_index;
3339
3340 get_device(dev);
3341 dev_set_drvdata(dev, sdkp);
3342
3343 gd->major = sd_major((index & 0xf0) >> 4);
3344 gd->first_minor = ((index & 0xf) << 4) | (index & 0xfff00);
3345
3346 gd->fops = &sd_fops;
3347 gd->private_data = &sdkp->driver;
3348 gd->queue = sdkp->device->request_queue;
3349
3350 /* defaults, until the device tells us otherwise */
3351 sdp->sector_size = 512;
3352 sdkp->capacity = 0;
3353 sdkp->media_present = 1;
3354 sdkp->write_prot = 0;
3355 sdkp->cache_override = 0;
3356 sdkp->WCE = 0;
3357 sdkp->RCD = 0;
3358 sdkp->ATO = 0;
3359 sdkp->first_scan = 1;
3360 sdkp->max_medium_access_timeouts = SD_MAX_MEDIUM_TIMEOUTS;
3361
3362 sd_revalidate_disk(gd);
3363
3364 gd->flags = GENHD_FL_EXT_DEVT;
3365 if (sdp->removable) {
3366 gd->flags |= GENHD_FL_REMOVABLE;
3367 gd->events |= DISK_EVENT_MEDIA_CHANGE;
3368 gd->event_flags = DISK_EVENT_FLAG_POLL | DISK_EVENT_FLAG_UEVENT;
3369 }
3370
3371 blk_pm_runtime_init(sdp->request_queue, dev);
3372 device_add_disk(dev, gd, NULL);
3373 if (sdkp->capacity)
3374 sd_dif_config_host(sdkp);
3375
3376 sd_revalidate_disk(gd);
3377
3378 if (sdkp->security) {
3379 sdkp->opal_dev = init_opal_dev(sdp, &sd_sec_submit);
3380 if (sdkp->opal_dev)
3381 sd_printk(KERN_NOTICE, sdkp, "supports TCG Opal\n");
3382 }
3383
3384 sd_printk(KERN_NOTICE, sdkp, "Attached SCSI %sdisk\n",
3385 sdp->removable ? "removable " : "");
3386 scsi_autopm_put_device(sdp);
3387
3388 return 0;
3389
3390 out_free_index:
3391 ida_free(&sd_index_ida, index);
3392 out_put:
3393 put_disk(gd);
3394 out_free:
3395 kfree(sdkp);
3396 out:
3397 scsi_autopm_put_device(sdp);
3398 return error;
3399}
3400
3401/**
3402 * sd_remove - called whenever a scsi disk (previously recognized by
3403 * sd_probe) is detached from the system. It is called (potentially
3404 * multiple times) during sd module unload.
3405 * @dev: pointer to device object
3406 *
3407 * Note: this function is invoked from the scsi mid-level.
3408 * This function potentially frees up a device name (e.g. /dev/sdc)
3409 * that could be re-used by a subsequent sd_probe().
3410 * This function is not called when the built-in sd driver is "exit-ed".
3411 **/
3412static int sd_remove(struct device *dev)
3413{
3414 struct scsi_disk *sdkp;
3415 dev_t devt;
3416
3417 sdkp = dev_get_drvdata(dev);
3418 devt = disk_devt(sdkp->disk);
3419 scsi_autopm_get_device(sdkp->device);
3420
3421 async_synchronize_full_domain(&scsi_sd_pm_domain);
3422 device_del(&sdkp->dev);
3423 del_gendisk(sdkp->disk);
3424 sd_shutdown(dev);
3425
3426 free_opal_dev(sdkp->opal_dev);
3427
3428 blk_register_region(devt, SD_MINORS, NULL,
3429 sd_default_probe, NULL, NULL);
3430
3431 mutex_lock(&sd_ref_mutex);
3432 dev_set_drvdata(dev, NULL);
3433 put_device(&sdkp->dev);
3434 mutex_unlock(&sd_ref_mutex);
3435
3436 return 0;
3437}
3438
3439/**
3440 * scsi_disk_release - Called to free the scsi_disk structure
3441 * @dev: pointer to embedded class device
3442 *
3443 * sd_ref_mutex must be held entering this routine. Because it is
3444 * called on last put, you should always use the scsi_disk_get()
3445 * scsi_disk_put() helpers which manipulate the semaphore directly
3446 * and never do a direct put_device.
3447 **/
3448static void scsi_disk_release(struct device *dev)
3449{
3450 struct scsi_disk *sdkp = to_scsi_disk(dev);
3451 struct gendisk *disk = sdkp->disk;
3452 struct request_queue *q = disk->queue;
3453
3454 ida_free(&sd_index_ida, sdkp->index);
3455
3456 /*
3457 * Wait until all requests that are in progress have completed.
3458 * This is necessary to avoid that e.g. scsi_end_request() crashes
3459 * due to clearing the disk->private_data pointer. Wait from inside
3460 * scsi_disk_release() instead of from sd_release() to avoid that
3461 * freezing and unfreezing the request queue affects user space I/O
3462 * in case multiple processes open a /dev/sd... node concurrently.
3463 */
3464 blk_mq_freeze_queue(q);
3465 blk_mq_unfreeze_queue(q);
3466
3467 disk->private_data = NULL;
3468 put_disk(disk);
3469 put_device(&sdkp->device->sdev_gendev);
3470
3471 kfree(sdkp);
3472}
3473
3474static int sd_start_stop_device(struct scsi_disk *sdkp, int start)
3475{
3476 unsigned char cmd[6] = { START_STOP }; /* START_VALID */
3477 struct scsi_sense_hdr sshdr;
3478 struct scsi_device *sdp = sdkp->device;
3479 int res;
3480
3481 if (start)
3482 cmd[4] |= 1; /* START */
3483
3484 if (sdp->start_stop_pwr_cond)
3485 cmd[4] |= start ? 1 << 4 : 3 << 4; /* Active or Standby */
3486
3487 if (!scsi_device_online(sdp))
3488 return -ENODEV;
3489
3490 res = scsi_execute(sdp, cmd, DMA_NONE, NULL, 0, NULL, &sshdr,
3491 SD_TIMEOUT, SD_MAX_RETRIES, 0, RQF_PM, NULL);
3492 if (res) {
3493 sd_print_result(sdkp, "Start/Stop Unit failed", res);
3494 if (driver_byte(res) == DRIVER_SENSE)
3495 sd_print_sense_hdr(sdkp, &sshdr);
3496 if (scsi_sense_valid(&sshdr) &&
3497 /* 0x3a is medium not present */
3498 sshdr.asc == 0x3a)
3499 res = 0;
3500 }
3501
3502 /* SCSI error codes must not go to the generic layer */
3503 if (res)
3504 return -EIO;
3505
3506 return 0;
3507}
3508
3509/*
3510 * Send a SYNCHRONIZE CACHE instruction down to the device through
3511 * the normal SCSI command structure. Wait for the command to
3512 * complete.
3513 */
3514static void sd_shutdown(struct device *dev)
3515{
3516 struct scsi_disk *sdkp = dev_get_drvdata(dev);
3517
3518 if (!sdkp)
3519 return; /* this can happen */
3520
3521 if (pm_runtime_suspended(dev))
3522 return;
3523
3524 if (sdkp->WCE && sdkp->media_present) {
3525 sd_printk(KERN_NOTICE, sdkp, "Synchronizing SCSI cache\n");
3526 sd_sync_cache(sdkp, NULL);
3527 }
3528
3529 if (system_state != SYSTEM_RESTART && sdkp->device->manage_start_stop) {
3530 sd_printk(KERN_NOTICE, sdkp, "Stopping disk\n");
3531 sd_start_stop_device(sdkp, 0);
3532 }
3533}
3534
3535static int sd_suspend_common(struct device *dev, bool ignore_stop_errors)
3536{
3537 struct scsi_disk *sdkp = dev_get_drvdata(dev);
3538 struct scsi_sense_hdr sshdr;
3539 int ret = 0;
3540
3541 if (!sdkp) /* E.g.: runtime suspend following sd_remove() */
3542 return 0;
3543
3544 if (sdkp->WCE && sdkp->media_present) {
3545 sd_printk(KERN_NOTICE, sdkp, "Synchronizing SCSI cache\n");
3546 ret = sd_sync_cache(sdkp, &sshdr);
3547
3548 if (ret) {
3549 /* ignore OFFLINE device */
3550 if (ret == -ENODEV)
3551 return 0;
3552
3553 if (!scsi_sense_valid(&sshdr) ||
3554 sshdr.sense_key != ILLEGAL_REQUEST)
3555 return ret;
3556
3557 /*
3558 * sshdr.sense_key == ILLEGAL_REQUEST means this drive
3559 * doesn't support sync. There's not much to do and
3560 * suspend shouldn't fail.
3561 */
3562 ret = 0;
3563 }
3564 }
3565
3566 if (sdkp->device->manage_start_stop) {
3567 sd_printk(KERN_NOTICE, sdkp, "Stopping disk\n");
3568 /* an error is not worth aborting a system sleep */
3569 ret = sd_start_stop_device(sdkp, 0);
3570 if (ignore_stop_errors)
3571 ret = 0;
3572 }
3573
3574 return ret;
3575}
3576
3577static int sd_suspend_system(struct device *dev)
3578{
3579 return sd_suspend_common(dev, true);
3580}
3581
3582static int sd_suspend_runtime(struct device *dev)
3583{
3584 return sd_suspend_common(dev, false);
3585}
3586
3587static int sd_resume(struct device *dev)
3588{
3589 struct scsi_disk *sdkp = dev_get_drvdata(dev);
3590 int ret;
3591
3592 if (!sdkp) /* E.g.: runtime resume at the start of sd_probe() */
3593 return 0;
3594
3595 if (!sdkp->device->manage_start_stop)
3596 return 0;
3597
3598 sd_printk(KERN_NOTICE, sdkp, "Starting disk\n");
3599 ret = sd_start_stop_device(sdkp, 1);
3600 if (!ret)
3601 opal_unlock_from_suspend(sdkp->opal_dev);
3602 return ret;
3603}
3604
3605/**
3606 * init_sd - entry point for this driver (both when built in or when
3607 * a module).
3608 *
3609 * Note: this function registers this driver with the scsi mid-level.
3610 **/
3611static int __init init_sd(void)
3612{
3613 int majors = 0, i, err;
3614
3615 SCSI_LOG_HLQUEUE(3, printk("init_sd: sd driver entry point\n"));
3616
3617 for (i = 0; i < SD_MAJORS; i++) {
3618 if (register_blkdev(sd_major(i), "sd") != 0)
3619 continue;
3620 majors++;
3621 blk_register_region(sd_major(i), SD_MINORS, NULL,
3622 sd_default_probe, NULL, NULL);
3623 }
3624
3625 if (!majors)
3626 return -ENODEV;
3627
3628 err = class_register(&sd_disk_class);
3629 if (err)
3630 goto err_out;
3631
3632 sd_cdb_cache = kmem_cache_create("sd_ext_cdb", SD_EXT_CDB_SIZE,
3633 0, 0, NULL);
3634 if (!sd_cdb_cache) {
3635 printk(KERN_ERR "sd: can't init extended cdb cache\n");
3636 err = -ENOMEM;
3637 goto err_out_class;
3638 }
3639
3640 sd_cdb_pool = mempool_create_slab_pool(SD_MEMPOOL_SIZE, sd_cdb_cache);
3641 if (!sd_cdb_pool) {
3642 printk(KERN_ERR "sd: can't init extended cdb pool\n");
3643 err = -ENOMEM;
3644 goto err_out_cache;
3645 }
3646
3647 sd_page_pool = mempool_create_page_pool(SD_MEMPOOL_SIZE, 0);
3648 if (!sd_page_pool) {
3649 printk(KERN_ERR "sd: can't init discard page pool\n");
3650 err = -ENOMEM;
3651 goto err_out_ppool;
3652 }
3653
3654 err = scsi_register_driver(&sd_template.gendrv);
3655 if (err)
3656 goto err_out_driver;
3657
3658 return 0;
3659
3660err_out_driver:
3661 mempool_destroy(sd_page_pool);
3662
3663err_out_ppool:
3664 mempool_destroy(sd_cdb_pool);
3665
3666err_out_cache:
3667 kmem_cache_destroy(sd_cdb_cache);
3668
3669err_out_class:
3670 class_unregister(&sd_disk_class);
3671err_out:
3672 for (i = 0; i < SD_MAJORS; i++)
3673 unregister_blkdev(sd_major(i), "sd");
3674 return err;
3675}
3676
3677/**
3678 * exit_sd - exit point for this driver (when it is a module).
3679 *
3680 * Note: this function unregisters this driver from the scsi mid-level.
3681 **/
3682static void __exit exit_sd(void)
3683{
3684 int i;
3685
3686 SCSI_LOG_HLQUEUE(3, printk("exit_sd: exiting sd driver\n"));
3687
3688 scsi_unregister_driver(&sd_template.gendrv);
3689 mempool_destroy(sd_cdb_pool);
3690 mempool_destroy(sd_page_pool);
3691 kmem_cache_destroy(sd_cdb_cache);
3692
3693 class_unregister(&sd_disk_class);
3694
3695 for (i = 0; i < SD_MAJORS; i++) {
3696 blk_unregister_region(sd_major(i), SD_MINORS);
3697 unregister_blkdev(sd_major(i), "sd");
3698 }
3699}
3700
3701module_init(init_sd);
3702module_exit(exit_sd);
3703
3704static void sd_print_sense_hdr(struct scsi_disk *sdkp,
3705 struct scsi_sense_hdr *sshdr)
3706{
3707 scsi_print_sense_hdr(sdkp->device,
3708 sdkp->disk ? sdkp->disk->disk_name : NULL, sshdr);
3709}
3710
3711static void sd_print_result(const struct scsi_disk *sdkp, const char *msg,
3712 int result)
3713{
3714 const char *hb_string = scsi_hostbyte_string(result);
3715 const char *db_string = scsi_driverbyte_string(result);
3716
3717 if (hb_string || db_string)
3718 sd_printk(KERN_INFO, sdkp,
3719 "%s: Result: hostbyte=%s driverbyte=%s\n", msg,
3720 hb_string ? hb_string : "invalid",
3721 db_string ? db_string : "invalid");
3722 else
3723 sd_printk(KERN_INFO, sdkp,
3724 "%s: Result: hostbyte=0x%02x driverbyte=0x%02x\n",
3725 msg, host_byte(result), driver_byte(result));
3726}
3727