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);
125
126static DEFINE_IDA(sd_index_ida);
127
128/* This semaphore is used to mediate the 0->1 reference get in the
129 * face of object destruction (i.e. we can't allow a get on an
130 * object after last put) */
131static DEFINE_MUTEX(sd_ref_mutex);
132
133static struct kmem_cache *sd_cdb_cache;
134static mempool_t *sd_cdb_pool;
135static mempool_t *sd_page_pool;
136
137static const char *sd_cache_types[] = {
138 "write through", "none", "write back",
139 "write back, no read (daft)"
140};
141
142static void sd_set_flush_flag(struct scsi_disk *sdkp)
143{
144 bool wc = false, fua = false;
145
146 if (sdkp->WCE) {
147 wc = true;
148 if (sdkp->DPOFUA)
149 fua = true;
150 }
151
152 blk_queue_write_cache(sdkp->disk->queue, wc, fua);
153}
154
155static ssize_t
156cache_type_store(struct device *dev, struct device_attribute *attr,
157 const char *buf, size_t count)
158{
159 int ct, rcd, wce, sp;
160 struct scsi_disk *sdkp = to_scsi_disk(dev);
161 struct scsi_device *sdp = sdkp->device;
162 char buffer[64];
163 char *buffer_data;
164 struct scsi_mode_data data;
165 struct scsi_sense_hdr sshdr;
166 static const char temp[] = "temporary ";
167 int len;
168
169 if (sdp->type != TYPE_DISK && sdp->type != TYPE_ZBC)
170 /* no cache control on RBC devices; theoretically they
171 * can do it, but there's probably so many exceptions
172 * it's not worth the risk */
173 return -EINVAL;
174
175 if (strncmp(buf, temp, sizeof(temp) - 1) == 0) {
176 buf += sizeof(temp) - 1;
177 sdkp->cache_override = 1;
178 } else {
179 sdkp->cache_override = 0;
180 }
181
182 ct = sysfs_match_string(sd_cache_types, buf);
183 if (ct < 0)
184 return -EINVAL;
185
186 rcd = ct & 0x01 ? 1 : 0;
187 wce = (ct & 0x02) && !sdkp->write_prot ? 1 : 0;
188
189 if (sdkp->cache_override) {
190 sdkp->WCE = wce;
191 sdkp->RCD = rcd;
192 sd_set_flush_flag(sdkp);
193 return count;
194 }
195
196 if (scsi_mode_sense(sdp, 0x08, 8, buffer, sizeof(buffer), SD_TIMEOUT,
197 SD_MAX_RETRIES, &data, NULL))
198 return -EINVAL;
199 len = min_t(size_t, sizeof(buffer), data.length - data.header_length -
200 data.block_descriptor_length);
201 buffer_data = buffer + data.header_length +
202 data.block_descriptor_length;
203 buffer_data[2] &= ~0x05;
204 buffer_data[2] |= wce << 2 | rcd;
205 sp = buffer_data[0] & 0x80 ? 1 : 0;
206 buffer_data[0] &= ~0x80;
207
208 /*
209 * Ensure WP, DPOFUA, and RESERVED fields are cleared in
210 * received mode parameter buffer before doing MODE SELECT.
211 */
212 data.device_specific = 0;
213
214 if (scsi_mode_select(sdp, 1, sp, 8, buffer_data, len, SD_TIMEOUT,
215 SD_MAX_RETRIES, &data, &sshdr)) {
216 if (scsi_sense_valid(&sshdr))
217 sd_print_sense_hdr(sdkp, &sshdr);
218 return -EINVAL;
219 }
220 revalidate_disk(sdkp->disk);
221 return count;
222}
223
224static ssize_t
225manage_start_stop_show(struct device *dev, struct device_attribute *attr,
226 char *buf)
227{
228 struct scsi_disk *sdkp = to_scsi_disk(dev);
229 struct scsi_device *sdp = sdkp->device;
230
231 return sprintf(buf, "%u\n", sdp->manage_start_stop);
232}
233
234static ssize_t
235manage_start_stop_store(struct device *dev, struct device_attribute *attr,
236 const char *buf, size_t count)
237{
238 struct scsi_disk *sdkp = to_scsi_disk(dev);
239 struct scsi_device *sdp = sdkp->device;
240 bool v;
241
242 if (!capable(CAP_SYS_ADMIN))
243 return -EACCES;
244
245 if (kstrtobool(buf, &v))
246 return -EINVAL;
247
248 sdp->manage_start_stop = v;
249
250 return count;
251}
252static DEVICE_ATTR_RW(manage_start_stop);
253
254static ssize_t
255allow_restart_show(struct device *dev, struct device_attribute *attr, char *buf)
256{
257 struct scsi_disk *sdkp = to_scsi_disk(dev);
258
259 return sprintf(buf, "%u\n", sdkp->device->allow_restart);
260}
261
262static ssize_t
263allow_restart_store(struct device *dev, struct device_attribute *attr,
264 const char *buf, size_t count)
265{
266 bool v;
267 struct scsi_disk *sdkp = to_scsi_disk(dev);
268 struct scsi_device *sdp = sdkp->device;
269
270 if (!capable(CAP_SYS_ADMIN))
271 return -EACCES;
272
273 if (sdp->type != TYPE_DISK && sdp->type != TYPE_ZBC)
274 return -EINVAL;
275
276 if (kstrtobool(buf, &v))
277 return -EINVAL;
278
279 sdp->allow_restart = v;
280
281 return count;
282}
283static DEVICE_ATTR_RW(allow_restart);
284
285static ssize_t
286cache_type_show(struct device *dev, struct device_attribute *attr, char *buf)
287{
288 struct scsi_disk *sdkp = to_scsi_disk(dev);
289 int ct = sdkp->RCD + 2*sdkp->WCE;
290
291 return sprintf(buf, "%s\n", sd_cache_types[ct]);
292}
293static DEVICE_ATTR_RW(cache_type);
294
295static ssize_t
296FUA_show(struct device *dev, struct device_attribute *attr, char *buf)
297{
298 struct scsi_disk *sdkp = to_scsi_disk(dev);
299
300 return sprintf(buf, "%u\n", sdkp->DPOFUA);
301}
302static DEVICE_ATTR_RO(FUA);
303
304static ssize_t
305protection_type_show(struct device *dev, struct device_attribute *attr,
306 char *buf)
307{
308 struct scsi_disk *sdkp = to_scsi_disk(dev);
309
310 return sprintf(buf, "%u\n", sdkp->protection_type);
311}
312
313static ssize_t
314protection_type_store(struct device *dev, struct device_attribute *attr,
315 const char *buf, size_t count)
316{
317 struct scsi_disk *sdkp = to_scsi_disk(dev);
318 unsigned int val;
319 int err;
320
321 if (!capable(CAP_SYS_ADMIN))
322 return -EACCES;
323
324 err = kstrtouint(buf, 10, &val);
325
326 if (err)
327 return err;
328
329 if (val <= T10_PI_TYPE3_PROTECTION)
330 sdkp->protection_type = val;
331
332 return count;
333}
334static DEVICE_ATTR_RW(protection_type);
335
336static ssize_t
337protection_mode_show(struct device *dev, struct device_attribute *attr,
338 char *buf)
339{
340 struct scsi_disk *sdkp = to_scsi_disk(dev);
341 struct scsi_device *sdp = sdkp->device;
342 unsigned int dif, dix;
343
344 dif = scsi_host_dif_capable(sdp->host, sdkp->protection_type);
345 dix = scsi_host_dix_capable(sdp->host, sdkp->protection_type);
346
347 if (!dix && scsi_host_dix_capable(sdp->host, T10_PI_TYPE0_PROTECTION)) {
348 dif = 0;
349 dix = 1;
350 }
351
352 if (!dif && !dix)
353 return sprintf(buf, "none\n");
354
355 return sprintf(buf, "%s%u\n", dix ? "dix" : "dif", dif);
356}
357static DEVICE_ATTR_RO(protection_mode);
358
359static ssize_t
360app_tag_own_show(struct device *dev, struct device_attribute *attr, char *buf)
361{
362 struct scsi_disk *sdkp = to_scsi_disk(dev);
363
364 return sprintf(buf, "%u\n", sdkp->ATO);
365}
366static DEVICE_ATTR_RO(app_tag_own);
367
368static ssize_t
369thin_provisioning_show(struct device *dev, struct device_attribute *attr,
370 char *buf)
371{
372 struct scsi_disk *sdkp = to_scsi_disk(dev);
373
374 return sprintf(buf, "%u\n", sdkp->lbpme);
375}
376static DEVICE_ATTR_RO(thin_provisioning);
377
378/* sysfs_match_string() requires dense arrays */
379static const char *lbp_mode[] = {
380 [SD_LBP_FULL] = "full",
381 [SD_LBP_UNMAP] = "unmap",
382 [SD_LBP_WS16] = "writesame_16",
383 [SD_LBP_WS10] = "writesame_10",
384 [SD_LBP_ZERO] = "writesame_zero",
385 [SD_LBP_DISABLE] = "disabled",
386};
387
388static ssize_t
389provisioning_mode_show(struct device *dev, struct device_attribute *attr,
390 char *buf)
391{
392 struct scsi_disk *sdkp = to_scsi_disk(dev);
393
394 return sprintf(buf, "%s\n", lbp_mode[sdkp->provisioning_mode]);
395}
396
397static ssize_t
398provisioning_mode_store(struct device *dev, struct device_attribute *attr,
399 const char *buf, size_t count)
400{
401 struct scsi_disk *sdkp = to_scsi_disk(dev);
402 struct scsi_device *sdp = sdkp->device;
403 int mode;
404
405 if (!capable(CAP_SYS_ADMIN))
406 return -EACCES;
407
408 if (sd_is_zoned(sdkp)) {
409 sd_config_discard(sdkp, SD_LBP_DISABLE);
410 return count;
411 }
412
413 if (sdp->type != TYPE_DISK)
414 return -EINVAL;
415
416 mode = sysfs_match_string(lbp_mode, buf);
417 if (mode < 0)
418 return -EINVAL;
419
420 sd_config_discard(sdkp, mode);
421
422 return count;
423}
424static DEVICE_ATTR_RW(provisioning_mode);
425
426/* sysfs_match_string() requires dense arrays */
427static const char *zeroing_mode[] = {
428 [SD_ZERO_WRITE] = "write",
429 [SD_ZERO_WS] = "writesame",
430 [SD_ZERO_WS16_UNMAP] = "writesame_16_unmap",
431 [SD_ZERO_WS10_UNMAP] = "writesame_10_unmap",
432};
433
434static ssize_t
435zeroing_mode_show(struct device *dev, struct device_attribute *attr,
436 char *buf)
437{
438 struct scsi_disk *sdkp = to_scsi_disk(dev);
439
440 return sprintf(buf, "%s\n", zeroing_mode[sdkp->zeroing_mode]);
441}
442
443static ssize_t
444zeroing_mode_store(struct device *dev, struct device_attribute *attr,
445 const char *buf, size_t count)
446{
447 struct scsi_disk *sdkp = to_scsi_disk(dev);
448 int mode;
449
450 if (!capable(CAP_SYS_ADMIN))
451 return -EACCES;
452
453 mode = sysfs_match_string(zeroing_mode, buf);
454 if (mode < 0)
455 return -EINVAL;
456
457 sdkp->zeroing_mode = mode;
458
459 return count;
460}
461static DEVICE_ATTR_RW(zeroing_mode);
462
463static ssize_t
464max_medium_access_timeouts_show(struct device *dev,
465 struct device_attribute *attr, char *buf)
466{
467 struct scsi_disk *sdkp = to_scsi_disk(dev);
468
469 return sprintf(buf, "%u\n", sdkp->max_medium_access_timeouts);
470}
471
472static ssize_t
473max_medium_access_timeouts_store(struct device *dev,
474 struct device_attribute *attr, const char *buf,
475 size_t count)
476{
477 struct scsi_disk *sdkp = to_scsi_disk(dev);
478 int err;
479
480 if (!capable(CAP_SYS_ADMIN))
481 return -EACCES;
482
483 err = kstrtouint(buf, 10, &sdkp->max_medium_access_timeouts);
484
485 return err ? err : count;
486}
487static DEVICE_ATTR_RW(max_medium_access_timeouts);
488
489static ssize_t
490max_write_same_blocks_show(struct device *dev, struct device_attribute *attr,
491 char *buf)
492{
493 struct scsi_disk *sdkp = to_scsi_disk(dev);
494
495 return sprintf(buf, "%u\n", sdkp->max_ws_blocks);
496}
497
498static ssize_t
499max_write_same_blocks_store(struct device *dev, struct device_attribute *attr,
500 const char *buf, size_t count)
501{
502 struct scsi_disk *sdkp = to_scsi_disk(dev);
503 struct scsi_device *sdp = sdkp->device;
504 unsigned long max;
505 int err;
506
507 if (!capable(CAP_SYS_ADMIN))
508 return -EACCES;
509
510 if (sdp->type != TYPE_DISK && sdp->type != TYPE_ZBC)
511 return -EINVAL;
512
513 err = kstrtoul(buf, 10, &max);
514
515 if (err)
516 return err;
517
518 if (max == 0)
519 sdp->no_write_same = 1;
520 else if (max <= SD_MAX_WS16_BLOCKS) {
521 sdp->no_write_same = 0;
522 sdkp->max_ws_blocks = max;
523 }
524
525 sd_config_write_same(sdkp);
526
527 return count;
528}
529static DEVICE_ATTR_RW(max_write_same_blocks);
530
531static ssize_t
532zoned_cap_show(struct device *dev, struct device_attribute *attr, char *buf)
533{
534 struct scsi_disk *sdkp = to_scsi_disk(dev);
535
536 if (sdkp->device->type == TYPE_ZBC)
537 return sprintf(buf, "host-managed\n");
538 if (sdkp->zoned == 1)
539 return sprintf(buf, "host-aware\n");
540 if (sdkp->zoned == 2)
541 return sprintf(buf, "drive-managed\n");
542 return sprintf(buf, "none\n");
543}
544static DEVICE_ATTR_RO(zoned_cap);
545
546static struct attribute *sd_disk_attrs[] = {
547 &dev_attr_cache_type.attr,
548 &dev_attr_FUA.attr,
549 &dev_attr_allow_restart.attr,
550 &dev_attr_manage_start_stop.attr,
551 &dev_attr_protection_type.attr,
552 &dev_attr_protection_mode.attr,
553 &dev_attr_app_tag_own.attr,
554 &dev_attr_thin_provisioning.attr,
555 &dev_attr_provisioning_mode.attr,
556 &dev_attr_zeroing_mode.attr,
557 &dev_attr_max_write_same_blocks.attr,
558 &dev_attr_max_medium_access_timeouts.attr,
559 &dev_attr_zoned_cap.attr,
560 NULL,
561};
562ATTRIBUTE_GROUPS(sd_disk);
563
564static struct class sd_disk_class = {
565 .name = "scsi_disk",
566 .owner = THIS_MODULE,
567 .dev_release = scsi_disk_release,
568 .dev_groups = sd_disk_groups,
569};
570
571static const struct dev_pm_ops sd_pm_ops = {
572 .suspend = sd_suspend_system,
573 .resume = sd_resume,
574 .poweroff = sd_suspend_system,
575 .restore = sd_resume,
576 .runtime_suspend = sd_suspend_runtime,
577 .runtime_resume = sd_resume,
578};
579
580static struct scsi_driver sd_template = {
581 .gendrv = {
582 .name = "sd",
583 .owner = THIS_MODULE,
584 .probe = sd_probe,
585 .probe_type = PROBE_PREFER_ASYNCHRONOUS,
586 .remove = sd_remove,
587 .shutdown = sd_shutdown,
588 .pm = &sd_pm_ops,
589 },
590 .rescan = sd_rescan,
591 .init_command = sd_init_command,
592 .uninit_command = sd_uninit_command,
593 .done = sd_done,
594 .eh_action = sd_eh_action,
595 .eh_reset = sd_eh_reset,
596};
597
598/*
599 * Dummy kobj_map->probe function.
600 * The default ->probe function will call modprobe, which is
601 * pointless as this module is already loaded.
602 */
603static struct kobject *sd_default_probe(dev_t devt, int *partno, void *data)
604{
605 return NULL;
606}
607
608/*
609 * Device no to disk mapping:
610 *
611 * major disc2 disc p1
612 * |............|.............|....|....| <- dev_t
613 * 31 20 19 8 7 4 3 0
614 *
615 * Inside a major, we have 16k disks, however mapped non-
616 * contiguously. The first 16 disks are for major0, the next
617 * ones with major1, ... Disk 256 is for major0 again, disk 272
618 * for major1, ...
619 * As we stay compatible with our numbering scheme, we can reuse
620 * the well-know SCSI majors 8, 65--71, 136--143.
621 */
622static int sd_major(int major_idx)
623{
624 switch (major_idx) {
625 case 0:
626 return SCSI_DISK0_MAJOR;
627 case 1 ... 7:
628 return SCSI_DISK1_MAJOR + major_idx - 1;
629 case 8 ... 15:
630 return SCSI_DISK8_MAJOR + major_idx - 8;
631 default:
632 BUG();
633 return 0; /* shut up gcc */
634 }
635}
636
637static struct scsi_disk *scsi_disk_get(struct gendisk *disk)
638{
639 struct scsi_disk *sdkp = NULL;
640
641 mutex_lock(&sd_ref_mutex);
642
643 if (disk->private_data) {
644 sdkp = scsi_disk(disk);
645 if (scsi_device_get(sdkp->device) == 0)
646 get_device(&sdkp->dev);
647 else
648 sdkp = NULL;
649 }
650 mutex_unlock(&sd_ref_mutex);
651 return sdkp;
652}
653
654static void scsi_disk_put(struct scsi_disk *sdkp)
655{
656 struct scsi_device *sdev = sdkp->device;
657
658 mutex_lock(&sd_ref_mutex);
659 put_device(&sdkp->dev);
660 scsi_device_put(sdev);
661 mutex_unlock(&sd_ref_mutex);
662}
663
664#ifdef CONFIG_BLK_SED_OPAL
665static int sd_sec_submit(void *data, u16 spsp, u8 secp, void *buffer,
666 size_t len, bool send)
667{
668 struct scsi_device *sdev = data;
669 u8 cdb[12] = { 0, };
670 int ret;
671
672 cdb[0] = send ? SECURITY_PROTOCOL_OUT : SECURITY_PROTOCOL_IN;
673 cdb[1] = secp;
674 put_unaligned_be16(spsp, &cdb[2]);
675 put_unaligned_be32(len, &cdb[6]);
676
677 ret = scsi_execute_req(sdev, cdb,
678 send ? DMA_TO_DEVICE : DMA_FROM_DEVICE,
679 buffer, len, NULL, SD_TIMEOUT, SD_MAX_RETRIES, NULL);
680 return ret <= 0 ? ret : -EIO;
681}
682#endif /* CONFIG_BLK_SED_OPAL */
683
684/*
685 * Look up the DIX operation based on whether the command is read or
686 * write and whether dix and dif are enabled.
687 */
688static unsigned int sd_prot_op(bool write, bool dix, bool dif)
689{
690 /* Lookup table: bit 2 (write), bit 1 (dix), bit 0 (dif) */
691 static const unsigned int ops[] = { /* wrt dix dif */
692 SCSI_PROT_NORMAL, /* 0 0 0 */
693 SCSI_PROT_READ_STRIP, /* 0 0 1 */
694 SCSI_PROT_READ_INSERT, /* 0 1 0 */
695 SCSI_PROT_READ_PASS, /* 0 1 1 */
696 SCSI_PROT_NORMAL, /* 1 0 0 */
697 SCSI_PROT_WRITE_INSERT, /* 1 0 1 */
698 SCSI_PROT_WRITE_STRIP, /* 1 1 0 */
699 SCSI_PROT_WRITE_PASS, /* 1 1 1 */
700 };
701
702 return ops[write << 2 | dix << 1 | dif];
703}
704
705/*
706 * Returns a mask of the protection flags that are valid for a given DIX
707 * operation.
708 */
709static unsigned int sd_prot_flag_mask(unsigned int prot_op)
710{
711 static const unsigned int flag_mask[] = {
712 [SCSI_PROT_NORMAL] = 0,
713
714 [SCSI_PROT_READ_STRIP] = SCSI_PROT_TRANSFER_PI |
715 SCSI_PROT_GUARD_CHECK |
716 SCSI_PROT_REF_CHECK |
717 SCSI_PROT_REF_INCREMENT,
718
719 [SCSI_PROT_READ_INSERT] = SCSI_PROT_REF_INCREMENT |
720 SCSI_PROT_IP_CHECKSUM,
721
722 [SCSI_PROT_READ_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 [SCSI_PROT_WRITE_INSERT] = SCSI_PROT_TRANSFER_PI |
729 SCSI_PROT_REF_INCREMENT,
730
731 [SCSI_PROT_WRITE_STRIP] = SCSI_PROT_GUARD_CHECK |
732 SCSI_PROT_REF_CHECK |
733 SCSI_PROT_REF_INCREMENT |
734 SCSI_PROT_IP_CHECKSUM,
735
736 [SCSI_PROT_WRITE_PASS] = SCSI_PROT_TRANSFER_PI |
737 SCSI_PROT_GUARD_CHECK |
738 SCSI_PROT_REF_CHECK |
739 SCSI_PROT_REF_INCREMENT |
740 SCSI_PROT_IP_CHECKSUM,
741 };
742
743 return flag_mask[prot_op];
744}
745
746static unsigned char sd_setup_protect_cmnd(struct scsi_cmnd *scmd,
747 unsigned int dix, unsigned int dif)
748{
749 struct bio *bio = scmd->request->bio;
750 unsigned int prot_op = sd_prot_op(rq_data_dir(scmd->request), dix, dif);
751 unsigned int protect = 0;
752
753 if (dix) { /* DIX Type 0, 1, 2, 3 */
754 if (bio_integrity_flagged(bio, BIP_IP_CHECKSUM))
755 scmd->prot_flags |= SCSI_PROT_IP_CHECKSUM;
756
757 if (bio_integrity_flagged(bio, BIP_CTRL_NOCHECK) == false)
758 scmd->prot_flags |= SCSI_PROT_GUARD_CHECK;
759 }
760
761 if (dif != T10_PI_TYPE3_PROTECTION) { /* DIX/DIF Type 0, 1, 2 */
762 scmd->prot_flags |= SCSI_PROT_REF_INCREMENT;
763
764 if (bio_integrity_flagged(bio, BIP_CTRL_NOCHECK) == false)
765 scmd->prot_flags |= SCSI_PROT_REF_CHECK;
766 }
767
768 if (dif) { /* DIX/DIF Type 1, 2, 3 */
769 scmd->prot_flags |= SCSI_PROT_TRANSFER_PI;
770
771 if (bio_integrity_flagged(bio, BIP_DISK_NOCHECK))
772 protect = 3 << 5; /* Disable target PI checking */
773 else
774 protect = 1 << 5; /* Enable target PI checking */
775 }
776
777 scsi_set_prot_op(scmd, prot_op);
778 scsi_set_prot_type(scmd, dif);
779 scmd->prot_flags &= sd_prot_flag_mask(prot_op);
780
781 return protect;
782}
783
784static void sd_config_discard(struct scsi_disk *sdkp, unsigned int mode)
785{
786 struct request_queue *q = sdkp->disk->queue;
787 unsigned int logical_block_size = sdkp->device->sector_size;
788 unsigned int max_blocks = 0;
789
790 q->limits.discard_alignment =
791 sdkp->unmap_alignment * logical_block_size;
792 q->limits.discard_granularity =
793 max(sdkp->physical_block_size,
794 sdkp->unmap_granularity * logical_block_size);
795 sdkp->provisioning_mode = mode;
796
797 switch (mode) {
798
799 case SD_LBP_FULL:
800 case SD_LBP_DISABLE:
801 blk_queue_max_discard_sectors(q, 0);
802 blk_queue_flag_clear(QUEUE_FLAG_DISCARD, q);
803 return;
804
805 case SD_LBP_UNMAP:
806 max_blocks = min_not_zero(sdkp->max_unmap_blocks,
807 (u32)SD_MAX_WS16_BLOCKS);
808 break;
809
810 case SD_LBP_WS16:
811 if (sdkp->device->unmap_limit_for_ws)
812 max_blocks = sdkp->max_unmap_blocks;
813 else
814 max_blocks = sdkp->max_ws_blocks;
815
816 max_blocks = min_not_zero(max_blocks, (u32)SD_MAX_WS16_BLOCKS);
817 break;
818
819 case SD_LBP_WS10:
820 if (sdkp->device->unmap_limit_for_ws)
821 max_blocks = sdkp->max_unmap_blocks;
822 else
823 max_blocks = sdkp->max_ws_blocks;
824
825 max_blocks = min_not_zero(max_blocks, (u32)SD_MAX_WS10_BLOCKS);
826 break;
827
828 case SD_LBP_ZERO:
829 max_blocks = min_not_zero(sdkp->max_ws_blocks,
830 (u32)SD_MAX_WS10_BLOCKS);
831 break;
832 }
833
834 blk_queue_max_discard_sectors(q, max_blocks * (logical_block_size >> 9));
835 blk_queue_flag_set(QUEUE_FLAG_DISCARD, q);
836}
837
838static blk_status_t sd_setup_unmap_cmnd(struct scsi_cmnd *cmd)
839{
840 struct scsi_device *sdp = cmd->device;
841 struct request *rq = cmd->request;
842 u64 lba = sectors_to_logical(sdp, blk_rq_pos(rq));
843 u32 nr_blocks = sectors_to_logical(sdp, blk_rq_sectors(rq));
844 unsigned int data_len = 24;
845 char *buf;
846
847 rq->special_vec.bv_page = mempool_alloc(sd_page_pool, GFP_ATOMIC);
848 if (!rq->special_vec.bv_page)
849 return BLK_STS_RESOURCE;
850 clear_highpage(rq->special_vec.bv_page);
851 rq->special_vec.bv_offset = 0;
852 rq->special_vec.bv_len = data_len;
853 rq->rq_flags |= RQF_SPECIAL_PAYLOAD;
854
855 cmd->cmd_len = 10;
856 cmd->cmnd[0] = UNMAP;
857 cmd->cmnd[8] = 24;
858
859 buf = page_address(rq->special_vec.bv_page);
860 put_unaligned_be16(6 + 16, &buf[0]);
861 put_unaligned_be16(16, &buf[2]);
862 put_unaligned_be64(lba, &buf[8]);
863 put_unaligned_be32(nr_blocks, &buf[16]);
864
865 cmd->allowed = SD_MAX_RETRIES;
866 cmd->transfersize = data_len;
867 rq->timeout = SD_TIMEOUT;
868
869 return scsi_init_io(cmd);
870}
871
872static blk_status_t sd_setup_write_same16_cmnd(struct scsi_cmnd *cmd,
873 bool unmap)
874{
875 struct scsi_device *sdp = cmd->device;
876 struct request *rq = cmd->request;
877 u64 lba = sectors_to_logical(sdp, blk_rq_pos(rq));
878 u32 nr_blocks = sectors_to_logical(sdp, blk_rq_sectors(rq));
879 u32 data_len = sdp->sector_size;
880
881 rq->special_vec.bv_page = mempool_alloc(sd_page_pool, GFP_ATOMIC);
882 if (!rq->special_vec.bv_page)
883 return BLK_STS_RESOURCE;
884 clear_highpage(rq->special_vec.bv_page);
885 rq->special_vec.bv_offset = 0;
886 rq->special_vec.bv_len = data_len;
887 rq->rq_flags |= RQF_SPECIAL_PAYLOAD;
888
889 cmd->cmd_len = 16;
890 cmd->cmnd[0] = WRITE_SAME_16;
891 if (unmap)
892 cmd->cmnd[1] = 0x8; /* UNMAP */
893 put_unaligned_be64(lba, &cmd->cmnd[2]);
894 put_unaligned_be32(nr_blocks, &cmd->cmnd[10]);
895
896 cmd->allowed = SD_MAX_RETRIES;
897 cmd->transfersize = data_len;
898 rq->timeout = unmap ? SD_TIMEOUT : SD_WRITE_SAME_TIMEOUT;
899
900 return scsi_init_io(cmd);
901}
902
903static blk_status_t sd_setup_write_same10_cmnd(struct scsi_cmnd *cmd,
904 bool unmap)
905{
906 struct scsi_device *sdp = cmd->device;
907 struct request *rq = cmd->request;
908 u64 lba = sectors_to_logical(sdp, blk_rq_pos(rq));
909 u32 nr_blocks = sectors_to_logical(sdp, blk_rq_sectors(rq));
910 u32 data_len = sdp->sector_size;
911
912 rq->special_vec.bv_page = mempool_alloc(sd_page_pool, GFP_ATOMIC);
913 if (!rq->special_vec.bv_page)
914 return BLK_STS_RESOURCE;
915 clear_highpage(rq->special_vec.bv_page);
916 rq->special_vec.bv_offset = 0;
917 rq->special_vec.bv_len = data_len;
918 rq->rq_flags |= RQF_SPECIAL_PAYLOAD;
919
920 cmd->cmd_len = 10;
921 cmd->cmnd[0] = WRITE_SAME;
922 if (unmap)
923 cmd->cmnd[1] = 0x8; /* UNMAP */
924 put_unaligned_be32(lba, &cmd->cmnd[2]);
925 put_unaligned_be16(nr_blocks, &cmd->cmnd[7]);
926
927 cmd->allowed = SD_MAX_RETRIES;
928 cmd->transfersize = data_len;
929 rq->timeout = unmap ? SD_TIMEOUT : SD_WRITE_SAME_TIMEOUT;
930
931 return scsi_init_io(cmd);
932}
933
934static blk_status_t sd_setup_write_zeroes_cmnd(struct scsi_cmnd *cmd)
935{
936 struct request *rq = cmd->request;
937 struct scsi_device *sdp = cmd->device;
938 struct scsi_disk *sdkp = scsi_disk(rq->rq_disk);
939 u64 lba = sectors_to_logical(sdp, blk_rq_pos(rq));
940 u32 nr_blocks = sectors_to_logical(sdp, blk_rq_sectors(rq));
941
942 if (!(rq->cmd_flags & REQ_NOUNMAP)) {
943 switch (sdkp->zeroing_mode) {
944 case SD_ZERO_WS16_UNMAP:
945 return sd_setup_write_same16_cmnd(cmd, true);
946 case SD_ZERO_WS10_UNMAP:
947 return sd_setup_write_same10_cmnd(cmd, true);
948 }
949 }
950
951 if (sdp->no_write_same)
952 return BLK_STS_TARGET;
953
954 if (sdkp->ws16 || lba > 0xffffffff || nr_blocks > 0xffff)
955 return sd_setup_write_same16_cmnd(cmd, false);
956
957 return sd_setup_write_same10_cmnd(cmd, false);
958}
959
960static void sd_config_write_same(struct scsi_disk *sdkp)
961{
962 struct request_queue *q = sdkp->disk->queue;
963 unsigned int logical_block_size = sdkp->device->sector_size;
964
965 if (sdkp->device->no_write_same) {
966 sdkp->max_ws_blocks = 0;
967 goto out;
968 }
969
970 /* Some devices can not handle block counts above 0xffff despite
971 * supporting WRITE SAME(16). Consequently we default to 64k
972 * blocks per I/O unless the device explicitly advertises a
973 * bigger limit.
974 */
975 if (sdkp->max_ws_blocks > SD_MAX_WS10_BLOCKS)
976 sdkp->max_ws_blocks = min_not_zero(sdkp->max_ws_blocks,
977 (u32)SD_MAX_WS16_BLOCKS);
978 else if (sdkp->ws16 || sdkp->ws10 || sdkp->device->no_report_opcodes)
979 sdkp->max_ws_blocks = min_not_zero(sdkp->max_ws_blocks,
980 (u32)SD_MAX_WS10_BLOCKS);
981 else {
982 sdkp->device->no_write_same = 1;
983 sdkp->max_ws_blocks = 0;
984 }
985
986 if (sdkp->lbprz && sdkp->lbpws)
987 sdkp->zeroing_mode = SD_ZERO_WS16_UNMAP;
988 else if (sdkp->lbprz && sdkp->lbpws10)
989 sdkp->zeroing_mode = SD_ZERO_WS10_UNMAP;
990 else if (sdkp->max_ws_blocks)
991 sdkp->zeroing_mode = SD_ZERO_WS;
992 else
993 sdkp->zeroing_mode = SD_ZERO_WRITE;
994
995 if (sdkp->max_ws_blocks &&
996 sdkp->physical_block_size > logical_block_size) {
997 /*
998 * Reporting a maximum number of blocks that is not aligned
999 * on the device physical size would cause a large write same
1000 * request to be split into physically unaligned chunks by
1001 * __blkdev_issue_write_zeroes() and __blkdev_issue_write_same()
1002 * even if the caller of these functions took care to align the
1003 * large request. So make sure the maximum reported is aligned
1004 * to the device physical block size. This is only an optional
1005 * optimization for regular disks, but this is mandatory to
1006 * avoid failure of large write same requests directed at
1007 * sequential write required zones of host-managed ZBC disks.
1008 */
1009 sdkp->max_ws_blocks =
1010 round_down(sdkp->max_ws_blocks,
1011 bytes_to_logical(sdkp->device,
1012 sdkp->physical_block_size));
1013 }
1014
1015out:
1016 blk_queue_max_write_same_sectors(q, sdkp->max_ws_blocks *
1017 (logical_block_size >> 9));
1018 blk_queue_max_write_zeroes_sectors(q, sdkp->max_ws_blocks *
1019 (logical_block_size >> 9));
1020}
1021
1022/**
1023 * sd_setup_write_same_cmnd - write the same data to multiple blocks
1024 * @cmd: command to prepare
1025 *
1026 * Will set up either WRITE SAME(10) or WRITE SAME(16) depending on
1027 * the preference indicated by the target device.
1028 **/
1029static blk_status_t sd_setup_write_same_cmnd(struct scsi_cmnd *cmd)
1030{
1031 struct request *rq = cmd->request;
1032 struct scsi_device *sdp = cmd->device;
1033 struct scsi_disk *sdkp = scsi_disk(rq->rq_disk);
1034 struct bio *bio = rq->bio;
1035 u64 lba = sectors_to_logical(sdp, blk_rq_pos(rq));
1036 u32 nr_blocks = sectors_to_logical(sdp, blk_rq_sectors(rq));
1037 blk_status_t ret;
1038
1039 if (sdkp->device->no_write_same)
1040 return BLK_STS_TARGET;
1041
1042 BUG_ON(bio_offset(bio) || bio_iovec(bio).bv_len != sdp->sector_size);
1043
1044 rq->timeout = SD_WRITE_SAME_TIMEOUT;
1045
1046 if (sdkp->ws16 || lba > 0xffffffff || nr_blocks > 0xffff) {
1047 cmd->cmd_len = 16;
1048 cmd->cmnd[0] = WRITE_SAME_16;
1049 put_unaligned_be64(lba, &cmd->cmnd[2]);
1050 put_unaligned_be32(nr_blocks, &cmd->cmnd[10]);
1051 } else {
1052 cmd->cmd_len = 10;
1053 cmd->cmnd[0] = WRITE_SAME;
1054 put_unaligned_be32(lba, &cmd->cmnd[2]);
1055 put_unaligned_be16(nr_blocks, &cmd->cmnd[7]);
1056 }
1057
1058 cmd->transfersize = sdp->sector_size;
1059 cmd->allowed = SD_MAX_RETRIES;
1060
1061 /*
1062 * For WRITE SAME the data transferred via the DATA OUT buffer is
1063 * different from the amount of data actually written to the target.
1064 *
1065 * We set up __data_len to the amount of data transferred via the
1066 * DATA OUT buffer so that blk_rq_map_sg sets up the proper S/G list
1067 * to transfer a single sector of data first, but then reset it to
1068 * the amount of data to be written right after so that the I/O path
1069 * knows how much to actually write.
1070 */
1071 rq->__data_len = sdp->sector_size;
1072 ret = scsi_init_io(cmd);
1073 rq->__data_len = blk_rq_bytes(rq);
1074
1075 return ret;
1076}
1077
1078static blk_status_t sd_setup_flush_cmnd(struct scsi_cmnd *cmd)
1079{
1080 struct request *rq = cmd->request;
1081
1082 /* flush requests don't perform I/O, zero the S/G table */
1083 memset(&cmd->sdb, 0, sizeof(cmd->sdb));
1084
1085 cmd->cmnd[0] = SYNCHRONIZE_CACHE;
1086 cmd->cmd_len = 10;
1087 cmd->transfersize = 0;
1088 cmd->allowed = SD_MAX_RETRIES;
1089
1090 rq->timeout = rq->q->rq_timeout * SD_FLUSH_TIMEOUT_MULTIPLIER;
1091 return BLK_STS_OK;
1092}
1093
1094static blk_status_t sd_setup_rw32_cmnd(struct scsi_cmnd *cmd, bool write,
1095 sector_t lba, unsigned int nr_blocks,
1096 unsigned char flags)
1097{
1098 cmd->cmnd = mempool_alloc(sd_cdb_pool, GFP_ATOMIC);
1099 if (unlikely(cmd->cmnd == NULL))
1100 return BLK_STS_RESOURCE;
1101
1102 cmd->cmd_len = SD_EXT_CDB_SIZE;
1103 memset(cmd->cmnd, 0, cmd->cmd_len);
1104
1105 cmd->cmnd[0] = VARIABLE_LENGTH_CMD;
1106 cmd->cmnd[7] = 0x18; /* Additional CDB len */
1107 cmd->cmnd[9] = write ? WRITE_32 : READ_32;
1108 cmd->cmnd[10] = flags;
1109 put_unaligned_be64(lba, &cmd->cmnd[12]);
1110 put_unaligned_be32(lba, &cmd->cmnd[20]); /* Expected Indirect LBA */
1111 put_unaligned_be32(nr_blocks, &cmd->cmnd[28]);
1112
1113 return BLK_STS_OK;
1114}
1115
1116static blk_status_t sd_setup_rw16_cmnd(struct scsi_cmnd *cmd, bool write,
1117 sector_t lba, unsigned int nr_blocks,
1118 unsigned char flags)
1119{
1120 cmd->cmd_len = 16;
1121 cmd->cmnd[0] = write ? WRITE_16 : READ_16;
1122 cmd->cmnd[1] = flags;
1123 cmd->cmnd[14] = 0;
1124 cmd->cmnd[15] = 0;
1125 put_unaligned_be64(lba, &cmd->cmnd[2]);
1126 put_unaligned_be32(nr_blocks, &cmd->cmnd[10]);
1127
1128 return BLK_STS_OK;
1129}
1130
1131static blk_status_t sd_setup_rw10_cmnd(struct scsi_cmnd *cmd, bool write,
1132 sector_t lba, unsigned int nr_blocks,
1133 unsigned char flags)
1134{
1135 cmd->cmd_len = 10;
1136 cmd->cmnd[0] = write ? WRITE_10 : READ_10;
1137 cmd->cmnd[1] = flags;
1138 cmd->cmnd[6] = 0;
1139 cmd->cmnd[9] = 0;
1140 put_unaligned_be32(lba, &cmd->cmnd[2]);
1141 put_unaligned_be16(nr_blocks, &cmd->cmnd[7]);
1142
1143 return BLK_STS_OK;
1144}
1145
1146static blk_status_t sd_setup_rw6_cmnd(struct scsi_cmnd *cmd, bool write,
1147 sector_t lba, unsigned int nr_blocks,
1148 unsigned char flags)
1149{
1150 /* Avoid that 0 blocks gets translated into 256 blocks. */
1151 if (WARN_ON_ONCE(nr_blocks == 0))
1152 return BLK_STS_IOERR;
1153
1154 if (unlikely(flags & 0x8)) {
1155 /*
1156 * This happens only if this drive failed 10byte rw
1157 * command with ILLEGAL_REQUEST during operation and
1158 * thus turned off use_10_for_rw.
1159 */
1160 scmd_printk(KERN_ERR, cmd, "FUA write on READ/WRITE(6) drive\n");
1161 return BLK_STS_IOERR;
1162 }
1163
1164 cmd->cmd_len = 6;
1165 cmd->cmnd[0] = write ? WRITE_6 : READ_6;
1166 cmd->cmnd[1] = (lba >> 16) & 0x1f;
1167 cmd->cmnd[2] = (lba >> 8) & 0xff;
1168 cmd->cmnd[3] = lba & 0xff;
1169 cmd->cmnd[4] = nr_blocks;
1170 cmd->cmnd[5] = 0;
1171
1172 return BLK_STS_OK;
1173}
1174
1175static blk_status_t sd_setup_read_write_cmnd(struct scsi_cmnd *cmd)
1176{
1177 struct request *rq = cmd->request;
1178 struct scsi_device *sdp = cmd->device;
1179 struct scsi_disk *sdkp = scsi_disk(rq->rq_disk);
1180 sector_t lba = sectors_to_logical(sdp, blk_rq_pos(rq));
1181 sector_t threshold;
1182 unsigned int nr_blocks = sectors_to_logical(sdp, blk_rq_sectors(rq));
1183 unsigned int mask = logical_to_sectors(sdp, 1) - 1;
1184 bool write = rq_data_dir(rq) == WRITE;
1185 unsigned char protect, fua;
1186 blk_status_t ret;
1187 unsigned int dif;
1188 bool dix;
1189
1190 ret = scsi_init_io(cmd);
1191 if (ret != BLK_STS_OK)
1192 return ret;
1193
1194 if (!scsi_device_online(sdp) || sdp->changed) {
1195 scmd_printk(KERN_ERR, cmd, "device offline or changed\n");
1196 return BLK_STS_IOERR;
1197 }
1198
1199 if (blk_rq_pos(rq) + blk_rq_sectors(rq) > get_capacity(rq->rq_disk)) {
1200 scmd_printk(KERN_ERR, cmd, "access beyond end of device\n");
1201 return BLK_STS_IOERR;
1202 }
1203
1204 if ((blk_rq_pos(rq) & mask) || (blk_rq_sectors(rq) & mask)) {
1205 scmd_printk(KERN_ERR, cmd, "request not aligned to the logical block size\n");
1206 return BLK_STS_IOERR;
1207 }
1208
1209 /*
1210 * Some SD card readers can't handle accesses which touch the
1211 * last one or two logical blocks. Split accesses as needed.
1212 */
1213 threshold = sdkp->capacity - SD_LAST_BUGGY_SECTORS;
1214
1215 if (unlikely(sdp->last_sector_bug && lba + nr_blocks > threshold)) {
1216 if (lba < threshold) {
1217 /* Access up to the threshold but not beyond */
1218 nr_blocks = threshold - lba;
1219 } else {
1220 /* Access only a single logical block */
1221 nr_blocks = 1;
1222 }
1223 }
1224
1225 if (req_op(rq) == REQ_OP_ZONE_APPEND) {
1226 ret = sd_zbc_prepare_zone_append(cmd, &lba, nr_blocks);
1227 if (ret)
1228 return ret;
1229 }
1230
1231 fua = rq->cmd_flags & REQ_FUA ? 0x8 : 0;
1232 dix = scsi_prot_sg_count(cmd);
1233 dif = scsi_host_dif_capable(cmd->device->host, sdkp->protection_type);
1234
1235 if (dif || dix)
1236 protect = sd_setup_protect_cmnd(cmd, dix, dif);
1237 else
1238 protect = 0;
1239
1240 if (protect && sdkp->protection_type == T10_PI_TYPE2_PROTECTION) {
1241 ret = sd_setup_rw32_cmnd(cmd, write, lba, nr_blocks,
1242 protect | fua);
1243 } else if (sdp->use_16_for_rw || (nr_blocks > 0xffff)) {
1244 ret = sd_setup_rw16_cmnd(cmd, write, lba, nr_blocks,
1245 protect | fua);
1246 } else if ((nr_blocks > 0xff) || (lba > 0x1fffff) ||
1247 sdp->use_10_for_rw || protect) {
1248 ret = sd_setup_rw10_cmnd(cmd, write, lba, nr_blocks,
1249 protect | fua);
1250 } else {
1251 ret = sd_setup_rw6_cmnd(cmd, write, lba, nr_blocks,
1252 protect | fua);
1253 }
1254
1255 if (unlikely(ret != BLK_STS_OK))
1256 return ret;
1257
1258 /*
1259 * We shouldn't disconnect in the middle of a sector, so with a dumb
1260 * host adapter, it's safe to assume that we can at least transfer
1261 * this many bytes between each connect / disconnect.
1262 */
1263 cmd->transfersize = sdp->sector_size;
1264 cmd->underflow = nr_blocks << 9;
1265 cmd->allowed = SD_MAX_RETRIES;
1266 cmd->sdb.length = nr_blocks * sdp->sector_size;
1267
1268 SCSI_LOG_HLQUEUE(1,
1269 scmd_printk(KERN_INFO, cmd,
1270 "%s: block=%llu, count=%d\n", __func__,
1271 (unsigned long long)blk_rq_pos(rq),
1272 blk_rq_sectors(rq)));
1273 SCSI_LOG_HLQUEUE(2,
1274 scmd_printk(KERN_INFO, cmd,
1275 "%s %d/%u 512 byte blocks.\n",
1276 write ? "writing" : "reading", nr_blocks,
1277 blk_rq_sectors(rq)));
1278
1279 /*
1280 * This indicates that the command is ready from our end to be
1281 * queued.
1282 */
1283 return BLK_STS_OK;
1284}
1285
1286static blk_status_t sd_init_command(struct scsi_cmnd *cmd)
1287{
1288 struct request *rq = cmd->request;
1289
1290 switch (req_op(rq)) {
1291 case REQ_OP_DISCARD:
1292 switch (scsi_disk(rq->rq_disk)->provisioning_mode) {
1293 case SD_LBP_UNMAP:
1294 return sd_setup_unmap_cmnd(cmd);
1295 case SD_LBP_WS16:
1296 return sd_setup_write_same16_cmnd(cmd, true);
1297 case SD_LBP_WS10:
1298 return sd_setup_write_same10_cmnd(cmd, true);
1299 case SD_LBP_ZERO:
1300 return sd_setup_write_same10_cmnd(cmd, false);
1301 default:
1302 return BLK_STS_TARGET;
1303 }
1304 case REQ_OP_WRITE_ZEROES:
1305 return sd_setup_write_zeroes_cmnd(cmd);
1306 case REQ_OP_WRITE_SAME:
1307 return sd_setup_write_same_cmnd(cmd);
1308 case REQ_OP_FLUSH:
1309 return sd_setup_flush_cmnd(cmd);
1310 case REQ_OP_READ:
1311 case REQ_OP_WRITE:
1312 case REQ_OP_ZONE_APPEND:
1313 return sd_setup_read_write_cmnd(cmd);
1314 case REQ_OP_ZONE_RESET:
1315 return sd_zbc_setup_zone_mgmt_cmnd(cmd, ZO_RESET_WRITE_POINTER,
1316 false);
1317 case REQ_OP_ZONE_RESET_ALL:
1318 return sd_zbc_setup_zone_mgmt_cmnd(cmd, ZO_RESET_WRITE_POINTER,
1319 true);
1320 case REQ_OP_ZONE_OPEN:
1321 return sd_zbc_setup_zone_mgmt_cmnd(cmd, ZO_OPEN_ZONE, false);
1322 case REQ_OP_ZONE_CLOSE:
1323 return sd_zbc_setup_zone_mgmt_cmnd(cmd, ZO_CLOSE_ZONE, false);
1324 case REQ_OP_ZONE_FINISH:
1325 return sd_zbc_setup_zone_mgmt_cmnd(cmd, ZO_FINISH_ZONE, false);
1326 default:
1327 WARN_ON_ONCE(1);
1328 return BLK_STS_NOTSUPP;
1329 }
1330}
1331
1332static void sd_uninit_command(struct scsi_cmnd *SCpnt)
1333{
1334 struct request *rq = SCpnt->request;
1335 u8 *cmnd;
1336
1337 if (rq->rq_flags & RQF_SPECIAL_PAYLOAD)
1338 mempool_free(rq->special_vec.bv_page, sd_page_pool);
1339
1340 if (SCpnt->cmnd != scsi_req(rq)->cmd) {
1341 cmnd = SCpnt->cmnd;
1342 SCpnt->cmnd = NULL;
1343 SCpnt->cmd_len = 0;
1344 mempool_free(cmnd, sd_cdb_pool);
1345 }
1346}
1347
1348/**
1349 * sd_open - open a scsi disk device
1350 * @bdev: Block device of the scsi disk to open
1351 * @mode: FMODE_* mask
1352 *
1353 * Returns 0 if successful. Returns a negated errno value in case
1354 * of error.
1355 *
1356 * Note: This can be called from a user context (e.g. fsck(1) )
1357 * or from within the kernel (e.g. as a result of a mount(1) ).
1358 * In the latter case @inode and @filp carry an abridged amount
1359 * of information as noted above.
1360 *
1361 * Locking: called with bdev->bd_mutex held.
1362 **/
1363static int sd_open(struct block_device *bdev, fmode_t mode)
1364{
1365 struct scsi_disk *sdkp = scsi_disk_get(bdev->bd_disk);
1366 struct scsi_device *sdev;
1367 int retval;
1368
1369 if (!sdkp)
1370 return -ENXIO;
1371
1372 SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp, "sd_open\n"));
1373
1374 sdev = sdkp->device;
1375
1376 /*
1377 * If the device is in error recovery, wait until it is done.
1378 * If the device is offline, then disallow any access to it.
1379 */
1380 retval = -ENXIO;
1381 if (!scsi_block_when_processing_errors(sdev))
1382 goto error_out;
1383
1384 if (sdev->removable || sdkp->write_prot)
1385 check_disk_change(bdev);
1386
1387 /*
1388 * If the drive is empty, just let the open fail.
1389 */
1390 retval = -ENOMEDIUM;
1391 if (sdev->removable && !sdkp->media_present && !(mode & FMODE_NDELAY))
1392 goto error_out;
1393
1394 /*
1395 * If the device has the write protect tab set, have the open fail
1396 * if the user expects to be able to write to the thing.
1397 */
1398 retval = -EROFS;
1399 if (sdkp->write_prot && (mode & FMODE_WRITE))
1400 goto error_out;
1401
1402 /*
1403 * It is possible that the disk changing stuff resulted in
1404 * the device being taken offline. If this is the case,
1405 * report this to the user, and don't pretend that the
1406 * open actually succeeded.
1407 */
1408 retval = -ENXIO;
1409 if (!scsi_device_online(sdev))
1410 goto error_out;
1411
1412 if ((atomic_inc_return(&sdkp->openers) == 1) && sdev->removable) {
1413 if (scsi_block_when_processing_errors(sdev))
1414 scsi_set_medium_removal(sdev, SCSI_REMOVAL_PREVENT);
1415 }
1416
1417 return 0;
1418
1419error_out:
1420 scsi_disk_put(sdkp);
1421 return retval;
1422}
1423
1424/**
1425 * sd_release - invoked when the (last) close(2) is called on this
1426 * scsi disk.
1427 * @disk: disk to release
1428 * @mode: FMODE_* mask
1429 *
1430 * Returns 0.
1431 *
1432 * Note: may block (uninterruptible) if error recovery is underway
1433 * on this disk.
1434 *
1435 * Locking: called with bdev->bd_mutex held.
1436 **/
1437static void sd_release(struct gendisk *disk, fmode_t mode)
1438{
1439 struct scsi_disk *sdkp = scsi_disk(disk);
1440 struct scsi_device *sdev = sdkp->device;
1441
1442 SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp, "sd_release\n"));
1443
1444 if (atomic_dec_return(&sdkp->openers) == 0 && sdev->removable) {
1445 if (scsi_block_when_processing_errors(sdev))
1446 scsi_set_medium_removal(sdev, SCSI_REMOVAL_ALLOW);
1447 }
1448
1449 scsi_disk_put(sdkp);
1450}
1451
1452static int sd_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1453{
1454 struct scsi_disk *sdkp = scsi_disk(bdev->bd_disk);
1455 struct scsi_device *sdp = sdkp->device;
1456 struct Scsi_Host *host = sdp->host;
1457 sector_t capacity = logical_to_sectors(sdp, sdkp->capacity);
1458 int diskinfo[4];
1459
1460 /* default to most commonly used values */
1461 diskinfo[0] = 0x40; /* 1 << 6 */
1462 diskinfo[1] = 0x20; /* 1 << 5 */
1463 diskinfo[2] = capacity >> 11;
1464
1465 /* override with calculated, extended default, or driver values */
1466 if (host->hostt->bios_param)
1467 host->hostt->bios_param(sdp, bdev, capacity, diskinfo);
1468 else
1469 scsicam_bios_param(bdev, capacity, diskinfo);
1470
1471 geo->heads = diskinfo[0];
1472 geo->sectors = diskinfo[1];
1473 geo->cylinders = diskinfo[2];
1474 return 0;
1475}
1476
1477/**
1478 * sd_ioctl - process an ioctl
1479 * @bdev: target block device
1480 * @mode: FMODE_* mask
1481 * @cmd: ioctl command number
1482 * @p: this is third argument given to ioctl(2) system call.
1483 * Often contains a pointer.
1484 *
1485 * Returns 0 if successful (some ioctls return positive numbers on
1486 * success as well). Returns a negated errno value in case of error.
1487 *
1488 * Note: most ioctls are forward onto the block subsystem or further
1489 * down in the scsi subsystem.
1490 **/
1491static int sd_ioctl_common(struct block_device *bdev, fmode_t mode,
1492 unsigned int cmd, void __user *p)
1493{
1494 struct gendisk *disk = bdev->bd_disk;
1495 struct scsi_disk *sdkp = scsi_disk(disk);
1496 struct scsi_device *sdp = sdkp->device;
1497 int error;
1498
1499 SCSI_LOG_IOCTL(1, sd_printk(KERN_INFO, sdkp, "sd_ioctl: disk=%s, "
1500 "cmd=0x%x\n", disk->disk_name, cmd));
1501
1502 error = scsi_verify_blk_ioctl(bdev, cmd);
1503 if (error < 0)
1504 return error;
1505
1506 /*
1507 * If we are in the middle of error recovery, don't let anyone
1508 * else try and use this device. Also, if error recovery fails, it
1509 * may try and take the device offline, in which case all further
1510 * access to the device is prohibited.
1511 */
1512 error = scsi_ioctl_block_when_processing_errors(sdp, cmd,
1513 (mode & FMODE_NDELAY) != 0);
1514 if (error)
1515 goto out;
1516
1517 if (is_sed_ioctl(cmd))
1518 return sed_ioctl(sdkp->opal_dev, cmd, p);
1519
1520 /*
1521 * Send SCSI addressing ioctls directly to mid level, send other
1522 * ioctls to block level and then onto mid level if they can't be
1523 * resolved.
1524 */
1525 switch (cmd) {
1526 case SCSI_IOCTL_GET_IDLUN:
1527 case SCSI_IOCTL_GET_BUS_NUMBER:
1528 error = scsi_ioctl(sdp, cmd, p);
1529 break;
1530 default:
1531 error = scsi_cmd_blk_ioctl(bdev, mode, cmd, p);
1532 break;
1533 }
1534out:
1535 return error;
1536}
1537
1538static void set_media_not_present(struct scsi_disk *sdkp)
1539{
1540 if (sdkp->media_present)
1541 sdkp->device->changed = 1;
1542
1543 if (sdkp->device->removable) {
1544 sdkp->media_present = 0;
1545 sdkp->capacity = 0;
1546 }
1547}
1548
1549static int media_not_present(struct scsi_disk *sdkp,
1550 struct scsi_sense_hdr *sshdr)
1551{
1552 if (!scsi_sense_valid(sshdr))
1553 return 0;
1554
1555 /* not invoked for commands that could return deferred errors */
1556 switch (sshdr->sense_key) {
1557 case UNIT_ATTENTION:
1558 case NOT_READY:
1559 /* medium not present */
1560 if (sshdr->asc == 0x3A) {
1561 set_media_not_present(sdkp);
1562 return 1;
1563 }
1564 }
1565 return 0;
1566}
1567
1568/**
1569 * sd_check_events - check media events
1570 * @disk: kernel device descriptor
1571 * @clearing: disk events currently being cleared
1572 *
1573 * Returns mask of DISK_EVENT_*.
1574 *
1575 * Note: this function is invoked from the block subsystem.
1576 **/
1577static unsigned int sd_check_events(struct gendisk *disk, unsigned int clearing)
1578{
1579 struct scsi_disk *sdkp = scsi_disk_get(disk);
1580 struct scsi_device *sdp;
1581 int retval;
1582
1583 if (!sdkp)
1584 return 0;
1585
1586 sdp = sdkp->device;
1587 SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp, "sd_check_events\n"));
1588
1589 /*
1590 * If the device is offline, don't send any commands - just pretend as
1591 * if the command failed. If the device ever comes back online, we
1592 * can deal with it then. It is only because of unrecoverable errors
1593 * that we would ever take a device offline in the first place.
1594 */
1595 if (!scsi_device_online(sdp)) {
1596 set_media_not_present(sdkp);
1597 goto out;
1598 }
1599
1600 /*
1601 * Using TEST_UNIT_READY enables differentiation between drive with
1602 * no cartridge loaded - NOT READY, drive with changed cartridge -
1603 * UNIT ATTENTION, or with same cartridge - GOOD STATUS.
1604 *
1605 * Drives that auto spin down. eg iomega jaz 1G, will be started
1606 * by sd_spinup_disk() from sd_revalidate_disk(), which happens whenever
1607 * sd_revalidate() is called.
1608 */
1609 if (scsi_block_when_processing_errors(sdp)) {
1610 struct scsi_sense_hdr sshdr = { 0, };
1611
1612 retval = scsi_test_unit_ready(sdp, SD_TIMEOUT, SD_MAX_RETRIES,
1613 &sshdr);
1614
1615 /* failed to execute TUR, assume media not present */
1616 if (host_byte(retval)) {
1617 set_media_not_present(sdkp);
1618 goto out;
1619 }
1620
1621 if (media_not_present(sdkp, &sshdr))
1622 goto out;
1623 }
1624
1625 /*
1626 * For removable scsi disk we have to recognise the presence
1627 * of a disk in the drive.
1628 */
1629 if (!sdkp->media_present)
1630 sdp->changed = 1;
1631 sdkp->media_present = 1;
1632out:
1633 /*
1634 * sdp->changed is set under the following conditions:
1635 *
1636 * Medium present state has changed in either direction.
1637 * Device has indicated UNIT_ATTENTION.
1638 */
1639 retval = sdp->changed ? DISK_EVENT_MEDIA_CHANGE : 0;
1640 sdp->changed = 0;
1641 scsi_disk_put(sdkp);
1642 return retval;
1643}
1644
1645static int sd_sync_cache(struct scsi_disk *sdkp, struct scsi_sense_hdr *sshdr)
1646{
1647 int retries, res;
1648 struct scsi_device *sdp = sdkp->device;
1649 const int timeout = sdp->request_queue->rq_timeout
1650 * SD_FLUSH_TIMEOUT_MULTIPLIER;
1651 struct scsi_sense_hdr my_sshdr;
1652
1653 if (!scsi_device_online(sdp))
1654 return -ENODEV;
1655
1656 /* caller might not be interested in sense, but we need it */
1657 if (!sshdr)
1658 sshdr = &my_sshdr;
1659
1660 for (retries = 3; retries > 0; --retries) {
1661 unsigned char cmd[10] = { 0 };
1662
1663 cmd[0] = SYNCHRONIZE_CACHE;
1664 /*
1665 * Leave the rest of the command zero to indicate
1666 * flush everything.
1667 */
1668 res = scsi_execute(sdp, cmd, DMA_NONE, NULL, 0, NULL, sshdr,
1669 timeout, SD_MAX_RETRIES, 0, RQF_PM, NULL);
1670 if (res == 0)
1671 break;
1672 }
1673
1674 if (res) {
1675 sd_print_result(sdkp, "Synchronize Cache(10) failed", res);
1676
1677 if (driver_byte(res) == DRIVER_SENSE)
1678 sd_print_sense_hdr(sdkp, sshdr);
1679
1680 /* we need to evaluate the error return */
1681 if (scsi_sense_valid(sshdr) &&
1682 (sshdr->asc == 0x3a || /* medium not present */
1683 sshdr->asc == 0x20 || /* invalid command */
1684 (sshdr->asc == 0x74 && sshdr->ascq == 0x71))) /* drive is password locked */
1685 /* this is no error here */
1686 return 0;
1687
1688 switch (host_byte(res)) {
1689 /* ignore errors due to racing a disconnection */
1690 case DID_BAD_TARGET:
1691 case DID_NO_CONNECT:
1692 return 0;
1693 /* signal the upper layer it might try again */
1694 case DID_BUS_BUSY:
1695 case DID_IMM_RETRY:
1696 case DID_REQUEUE:
1697 case DID_SOFT_ERROR:
1698 return -EBUSY;
1699 default:
1700 return -EIO;
1701 }
1702 }
1703 return 0;
1704}
1705
1706static void sd_rescan(struct device *dev)
1707{
1708 struct scsi_disk *sdkp = dev_get_drvdata(dev);
1709
1710 revalidate_disk(sdkp->disk);
1711}
1712
1713static int sd_ioctl(struct block_device *bdev, fmode_t mode,
1714 unsigned int cmd, unsigned long arg)
1715{
1716 void __user *p = (void __user *)arg;
1717 int ret;
1718
1719 ret = sd_ioctl_common(bdev, mode, cmd, p);
1720 if (ret != -ENOTTY)
1721 return ret;
1722
1723 return scsi_ioctl(scsi_disk(bdev->bd_disk)->device, cmd, p);
1724}
1725
1726#ifdef CONFIG_COMPAT
1727static int sd_compat_ioctl(struct block_device *bdev, fmode_t mode,
1728 unsigned int cmd, unsigned long arg)
1729{
1730 void __user *p = compat_ptr(arg);
1731 int ret;
1732
1733 ret = sd_ioctl_common(bdev, mode, cmd, p);
1734 if (ret != -ENOTTY)
1735 return ret;
1736
1737 return scsi_compat_ioctl(scsi_disk(bdev->bd_disk)->device, cmd, p);
1738}
1739#endif
1740
1741static char sd_pr_type(enum pr_type type)
1742{
1743 switch (type) {
1744 case PR_WRITE_EXCLUSIVE:
1745 return 0x01;
1746 case PR_EXCLUSIVE_ACCESS:
1747 return 0x03;
1748 case PR_WRITE_EXCLUSIVE_REG_ONLY:
1749 return 0x05;
1750 case PR_EXCLUSIVE_ACCESS_REG_ONLY:
1751 return 0x06;
1752 case PR_WRITE_EXCLUSIVE_ALL_REGS:
1753 return 0x07;
1754 case PR_EXCLUSIVE_ACCESS_ALL_REGS:
1755 return 0x08;
1756 default:
1757 return 0;
1758 }
1759};
1760
1761static int sd_pr_command(struct block_device *bdev, u8 sa,
1762 u64 key, u64 sa_key, u8 type, u8 flags)
1763{
1764 struct scsi_device *sdev = scsi_disk(bdev->bd_disk)->device;
1765 struct scsi_sense_hdr sshdr;
1766 int result;
1767 u8 cmd[16] = { 0, };
1768 u8 data[24] = { 0, };
1769
1770 cmd[0] = PERSISTENT_RESERVE_OUT;
1771 cmd[1] = sa;
1772 cmd[2] = type;
1773 put_unaligned_be32(sizeof(data), &cmd[5]);
1774
1775 put_unaligned_be64(key, &data[0]);
1776 put_unaligned_be64(sa_key, &data[8]);
1777 data[20] = flags;
1778
1779 result = scsi_execute_req(sdev, cmd, DMA_TO_DEVICE, &data, sizeof(data),
1780 &sshdr, SD_TIMEOUT, SD_MAX_RETRIES, NULL);
1781
1782 if (driver_byte(result) == DRIVER_SENSE &&
1783 scsi_sense_valid(&sshdr)) {
1784 sdev_printk(KERN_INFO, sdev, "PR command failed: %d\n", result);
1785 scsi_print_sense_hdr(sdev, NULL, &sshdr);
1786 }
1787
1788 return result;
1789}
1790
1791static int sd_pr_register(struct block_device *bdev, u64 old_key, u64 new_key,
1792 u32 flags)
1793{
1794 if (flags & ~PR_FL_IGNORE_KEY)
1795 return -EOPNOTSUPP;
1796 return sd_pr_command(bdev, (flags & PR_FL_IGNORE_KEY) ? 0x06 : 0x00,
1797 old_key, new_key, 0,
1798 (1 << 0) /* APTPL */);
1799}
1800
1801static int sd_pr_reserve(struct block_device *bdev, u64 key, enum pr_type type,
1802 u32 flags)
1803{
1804 if (flags)
1805 return -EOPNOTSUPP;
1806 return sd_pr_command(bdev, 0x01, key, 0, sd_pr_type(type), 0);
1807}
1808
1809static int sd_pr_release(struct block_device *bdev, u64 key, enum pr_type type)
1810{
1811 return sd_pr_command(bdev, 0x02, key, 0, sd_pr_type(type), 0);
1812}
1813
1814static int sd_pr_preempt(struct block_device *bdev, u64 old_key, u64 new_key,
1815 enum pr_type type, bool abort)
1816{
1817 return sd_pr_command(bdev, abort ? 0x05 : 0x04, old_key, new_key,
1818 sd_pr_type(type), 0);
1819}
1820
1821static int sd_pr_clear(struct block_device *bdev, u64 key)
1822{
1823 return sd_pr_command(bdev, 0x03, key, 0, 0, 0);
1824}
1825
1826static const struct pr_ops sd_pr_ops = {
1827 .pr_register = sd_pr_register,
1828 .pr_reserve = sd_pr_reserve,
1829 .pr_release = sd_pr_release,
1830 .pr_preempt = sd_pr_preempt,
1831 .pr_clear = sd_pr_clear,
1832};
1833
1834static const struct block_device_operations sd_fops = {
1835 .owner = THIS_MODULE,
1836 .open = sd_open,
1837 .release = sd_release,
1838 .ioctl = sd_ioctl,
1839 .getgeo = sd_getgeo,
1840#ifdef CONFIG_COMPAT
1841 .compat_ioctl = sd_compat_ioctl,
1842#endif
1843 .check_events = sd_check_events,
1844 .revalidate_disk = sd_revalidate_disk,
1845 .unlock_native_capacity = sd_unlock_native_capacity,
1846 .report_zones = sd_zbc_report_zones,
1847 .pr_ops = &sd_pr_ops,
1848};
1849
1850/**
1851 * sd_eh_reset - reset error handling callback
1852 * @scmd: sd-issued command that has failed
1853 *
1854 * This function is called by the SCSI midlayer before starting
1855 * SCSI EH. When counting medium access failures we have to be
1856 * careful to register it only only once per device and SCSI EH run;
1857 * there might be several timed out commands which will cause the
1858 * 'max_medium_access_timeouts' counter to trigger after the first
1859 * SCSI EH run already and set the device to offline.
1860 * So this function resets the internal counter before starting SCSI EH.
1861 **/
1862static void sd_eh_reset(struct scsi_cmnd *scmd)
1863{
1864 struct scsi_disk *sdkp = scsi_disk(scmd->request->rq_disk);
1865
1866 /* New SCSI EH run, reset gate variable */
1867 sdkp->ignore_medium_access_errors = false;
1868}
1869
1870/**
1871 * sd_eh_action - error handling callback
1872 * @scmd: sd-issued command that has failed
1873 * @eh_disp: The recovery disposition suggested by the midlayer
1874 *
1875 * This function is called by the SCSI midlayer upon completion of an
1876 * error test command (currently TEST UNIT READY). The result of sending
1877 * the eh command is passed in eh_disp. We're looking for devices that
1878 * fail medium access commands but are OK with non access commands like
1879 * test unit ready (so wrongly see the device as having a successful
1880 * recovery)
1881 **/
1882static int sd_eh_action(struct scsi_cmnd *scmd, int eh_disp)
1883{
1884 struct scsi_disk *sdkp = scsi_disk(scmd->request->rq_disk);
1885 struct scsi_device *sdev = scmd->device;
1886
1887 if (!scsi_device_online(sdev) ||
1888 !scsi_medium_access_command(scmd) ||
1889 host_byte(scmd->result) != DID_TIME_OUT ||
1890 eh_disp != SUCCESS)
1891 return eh_disp;
1892
1893 /*
1894 * The device has timed out executing a medium access command.
1895 * However, the TEST UNIT READY command sent during error
1896 * handling completed successfully. Either the device is in the
1897 * process of recovering or has it suffered an internal failure
1898 * that prevents access to the storage medium.
1899 */
1900 if (!sdkp->ignore_medium_access_errors) {
1901 sdkp->medium_access_timed_out++;
1902 sdkp->ignore_medium_access_errors = true;
1903 }
1904
1905 /*
1906 * If the device keeps failing read/write commands but TEST UNIT
1907 * READY always completes successfully we assume that medium
1908 * access is no longer possible and take the device offline.
1909 */
1910 if (sdkp->medium_access_timed_out >= sdkp->max_medium_access_timeouts) {
1911 scmd_printk(KERN_ERR, scmd,
1912 "Medium access timeout failure. Offlining disk!\n");
1913 mutex_lock(&sdev->state_mutex);
1914 scsi_device_set_state(sdev, SDEV_OFFLINE);
1915 mutex_unlock(&sdev->state_mutex);
1916
1917 return SUCCESS;
1918 }
1919
1920 return eh_disp;
1921}
1922
1923static unsigned int sd_completed_bytes(struct scsi_cmnd *scmd)
1924{
1925 struct request *req = scmd->request;
1926 struct scsi_device *sdev = scmd->device;
1927 unsigned int transferred, good_bytes;
1928 u64 start_lba, end_lba, bad_lba;
1929
1930 /*
1931 * Some commands have a payload smaller than the device logical
1932 * block size (e.g. INQUIRY on a 4K disk).
1933 */
1934 if (scsi_bufflen(scmd) <= sdev->sector_size)
1935 return 0;
1936
1937 /* Check if we have a 'bad_lba' information */
1938 if (!scsi_get_sense_info_fld(scmd->sense_buffer,
1939 SCSI_SENSE_BUFFERSIZE,
1940 &bad_lba))
1941 return 0;
1942
1943 /*
1944 * If the bad lba was reported incorrectly, we have no idea where
1945 * the error is.
1946 */
1947 start_lba = sectors_to_logical(sdev, blk_rq_pos(req));
1948 end_lba = start_lba + bytes_to_logical(sdev, scsi_bufflen(scmd));
1949 if (bad_lba < start_lba || bad_lba >= end_lba)
1950 return 0;
1951
1952 /*
1953 * resid is optional but mostly filled in. When it's unused,
1954 * its value is zero, so we assume the whole buffer transferred
1955 */
1956 transferred = scsi_bufflen(scmd) - scsi_get_resid(scmd);
1957
1958 /* This computation should always be done in terms of the
1959 * resolution of the device's medium.
1960 */
1961 good_bytes = logical_to_bytes(sdev, bad_lba - start_lba);
1962
1963 return min(good_bytes, transferred);
1964}
1965
1966/**
1967 * sd_done - bottom half handler: called when the lower level
1968 * driver has completed (successfully or otherwise) a scsi command.
1969 * @SCpnt: mid-level's per command structure.
1970 *
1971 * Note: potentially run from within an ISR. Must not block.
1972 **/
1973static int sd_done(struct scsi_cmnd *SCpnt)
1974{
1975 int result = SCpnt->result;
1976 unsigned int good_bytes = result ? 0 : scsi_bufflen(SCpnt);
1977 unsigned int sector_size = SCpnt->device->sector_size;
1978 unsigned int resid;
1979 struct scsi_sense_hdr sshdr;
1980 struct scsi_disk *sdkp = scsi_disk(SCpnt->request->rq_disk);
1981 struct request *req = SCpnt->request;
1982 int sense_valid = 0;
1983 int sense_deferred = 0;
1984
1985 switch (req_op(req)) {
1986 case REQ_OP_DISCARD:
1987 case REQ_OP_WRITE_ZEROES:
1988 case REQ_OP_WRITE_SAME:
1989 case REQ_OP_ZONE_RESET:
1990 case REQ_OP_ZONE_RESET_ALL:
1991 case REQ_OP_ZONE_OPEN:
1992 case REQ_OP_ZONE_CLOSE:
1993 case REQ_OP_ZONE_FINISH:
1994 if (!result) {
1995 good_bytes = blk_rq_bytes(req);
1996 scsi_set_resid(SCpnt, 0);
1997 } else {
1998 good_bytes = 0;
1999 scsi_set_resid(SCpnt, blk_rq_bytes(req));
2000 }
2001 break;
2002 default:
2003 /*
2004 * In case of bogus fw or device, we could end up having
2005 * an unaligned partial completion. Check this here and force
2006 * alignment.
2007 */
2008 resid = scsi_get_resid(SCpnt);
2009 if (resid & (sector_size - 1)) {
2010 sd_printk(KERN_INFO, sdkp,
2011 "Unaligned partial completion (resid=%u, sector_sz=%u)\n",
2012 resid, sector_size);
2013 scsi_print_command(SCpnt);
2014 resid = min(scsi_bufflen(SCpnt),
2015 round_up(resid, sector_size));
2016 scsi_set_resid(SCpnt, resid);
2017 }
2018 }
2019
2020 if (result) {
2021 sense_valid = scsi_command_normalize_sense(SCpnt, &sshdr);
2022 if (sense_valid)
2023 sense_deferred = scsi_sense_is_deferred(&sshdr);
2024 }
2025 sdkp->medium_access_timed_out = 0;
2026
2027 if (driver_byte(result) != DRIVER_SENSE &&
2028 (!sense_valid || sense_deferred))
2029 goto out;
2030
2031 switch (sshdr.sense_key) {
2032 case HARDWARE_ERROR:
2033 case MEDIUM_ERROR:
2034 good_bytes = sd_completed_bytes(SCpnt);
2035 break;
2036 case RECOVERED_ERROR:
2037 good_bytes = scsi_bufflen(SCpnt);
2038 break;
2039 case NO_SENSE:
2040 /* This indicates a false check condition, so ignore it. An
2041 * unknown amount of data was transferred so treat it as an
2042 * error.
2043 */
2044 SCpnt->result = 0;
2045 memset(SCpnt->sense_buffer, 0, SCSI_SENSE_BUFFERSIZE);
2046 break;
2047 case ABORTED_COMMAND:
2048 if (sshdr.asc == 0x10) /* DIF: Target detected corruption */
2049 good_bytes = sd_completed_bytes(SCpnt);
2050 break;
2051 case ILLEGAL_REQUEST:
2052 switch (sshdr.asc) {
2053 case 0x10: /* DIX: Host detected corruption */
2054 good_bytes = sd_completed_bytes(SCpnt);
2055 break;
2056 case 0x20: /* INVALID COMMAND OPCODE */
2057 case 0x24: /* INVALID FIELD IN CDB */
2058 switch (SCpnt->cmnd[0]) {
2059 case UNMAP:
2060 sd_config_discard(sdkp, SD_LBP_DISABLE);
2061 break;
2062 case WRITE_SAME_16:
2063 case WRITE_SAME:
2064 if (SCpnt->cmnd[1] & 8) { /* UNMAP */
2065 sd_config_discard(sdkp, SD_LBP_DISABLE);
2066 } else {
2067 sdkp->device->no_write_same = 1;
2068 sd_config_write_same(sdkp);
2069 req->rq_flags |= RQF_QUIET;
2070 }
2071 break;
2072 }
2073 }
2074 break;
2075 default:
2076 break;
2077 }
2078
2079 out:
2080 if (sd_is_zoned(sdkp))
2081 good_bytes = sd_zbc_complete(SCpnt, good_bytes, &sshdr);
2082
2083 SCSI_LOG_HLCOMPLETE(1, scmd_printk(KERN_INFO, SCpnt,
2084 "sd_done: completed %d of %d bytes\n",
2085 good_bytes, scsi_bufflen(SCpnt)));
2086
2087 return good_bytes;
2088}
2089
2090/*
2091 * spinup disk - called only in sd_revalidate_disk()
2092 */
2093static void
2094sd_spinup_disk(struct scsi_disk *sdkp)
2095{
2096 unsigned char cmd[10];
2097 unsigned long spintime_expire = 0;
2098 int retries, spintime;
2099 unsigned int the_result;
2100 struct scsi_sense_hdr sshdr;
2101 int sense_valid = 0;
2102
2103 spintime = 0;
2104
2105 /* Spin up drives, as required. Only do this at boot time */
2106 /* Spinup needs to be done for module loads too. */
2107 do {
2108 retries = 0;
2109
2110 do {
2111 cmd[0] = TEST_UNIT_READY;
2112 memset((void *) &cmd[1], 0, 9);
2113
2114 the_result = scsi_execute_req(sdkp->device, cmd,
2115 DMA_NONE, NULL, 0,
2116 &sshdr, SD_TIMEOUT,
2117 SD_MAX_RETRIES, NULL);
2118
2119 /*
2120 * If the drive has indicated to us that it
2121 * doesn't have any media in it, don't bother
2122 * with any more polling.
2123 */
2124 if (media_not_present(sdkp, &sshdr))
2125 return;
2126
2127 if (the_result)
2128 sense_valid = scsi_sense_valid(&sshdr);
2129 retries++;
2130 } while (retries < 3 &&
2131 (!scsi_status_is_good(the_result) ||
2132 ((driver_byte(the_result) == DRIVER_SENSE) &&
2133 sense_valid && sshdr.sense_key == UNIT_ATTENTION)));
2134
2135 if (driver_byte(the_result) != DRIVER_SENSE) {
2136 /* no sense, TUR either succeeded or failed
2137 * with a status error */
2138 if(!spintime && !scsi_status_is_good(the_result)) {
2139 sd_print_result(sdkp, "Test Unit Ready failed",
2140 the_result);
2141 }
2142 break;
2143 }
2144
2145 /*
2146 * The device does not want the automatic start to be issued.
2147 */
2148 if (sdkp->device->no_start_on_add)
2149 break;
2150
2151 if (sense_valid && sshdr.sense_key == NOT_READY) {
2152 if (sshdr.asc == 4 && sshdr.ascq == 3)
2153 break; /* manual intervention required */
2154 if (sshdr.asc == 4 && sshdr.ascq == 0xb)
2155 break; /* standby */
2156 if (sshdr.asc == 4 && sshdr.ascq == 0xc)
2157 break; /* unavailable */
2158 if (sshdr.asc == 4 && sshdr.ascq == 0x1b)
2159 break; /* sanitize in progress */
2160 /*
2161 * Issue command to spin up drive when not ready
2162 */
2163 if (!spintime) {
2164 sd_printk(KERN_NOTICE, sdkp, "Spinning up disk...");
2165 cmd[0] = START_STOP;
2166 cmd[1] = 1; /* Return immediately */
2167 memset((void *) &cmd[2], 0, 8);
2168 cmd[4] = 1; /* Start spin cycle */
2169 if (sdkp->device->start_stop_pwr_cond)
2170 cmd[4] |= 1 << 4;
2171 scsi_execute_req(sdkp->device, cmd, DMA_NONE,
2172 NULL, 0, &sshdr,
2173 SD_TIMEOUT, SD_MAX_RETRIES,
2174 NULL);
2175 spintime_expire = jiffies + 100 * HZ;
2176 spintime = 1;
2177 }
2178 /* Wait 1 second for next try */
2179 msleep(1000);
2180 printk(KERN_CONT ".");
2181
2182 /*
2183 * Wait for USB flash devices with slow firmware.
2184 * Yes, this sense key/ASC combination shouldn't
2185 * occur here. It's characteristic of these devices.
2186 */
2187 } else if (sense_valid &&
2188 sshdr.sense_key == UNIT_ATTENTION &&
2189 sshdr.asc == 0x28) {
2190 if (!spintime) {
2191 spintime_expire = jiffies + 5 * HZ;
2192 spintime = 1;
2193 }
2194 /* Wait 1 second for next try */
2195 msleep(1000);
2196 } else {
2197 /* we don't understand the sense code, so it's
2198 * probably pointless to loop */
2199 if(!spintime) {
2200 sd_printk(KERN_NOTICE, sdkp, "Unit Not Ready\n");
2201 sd_print_sense_hdr(sdkp, &sshdr);
2202 }
2203 break;
2204 }
2205
2206 } while (spintime && time_before_eq(jiffies, spintime_expire));
2207
2208 if (spintime) {
2209 if (scsi_status_is_good(the_result))
2210 printk(KERN_CONT "ready\n");
2211 else
2212 printk(KERN_CONT "not responding...\n");
2213 }
2214}
2215
2216/*
2217 * Determine whether disk supports Data Integrity Field.
2218 */
2219static int sd_read_protection_type(struct scsi_disk *sdkp, unsigned char *buffer)
2220{
2221 struct scsi_device *sdp = sdkp->device;
2222 u8 type;
2223 int ret = 0;
2224
2225 if (scsi_device_protection(sdp) == 0 || (buffer[12] & 1) == 0) {
2226 sdkp->protection_type = 0;
2227 return ret;
2228 }
2229
2230 type = ((buffer[12] >> 1) & 7) + 1; /* P_TYPE 0 = Type 1 */
2231
2232 if (type > T10_PI_TYPE3_PROTECTION)
2233 ret = -ENODEV;
2234 else if (scsi_host_dif_capable(sdp->host, type))
2235 ret = 1;
2236
2237 if (sdkp->first_scan || type != sdkp->protection_type)
2238 switch (ret) {
2239 case -ENODEV:
2240 sd_printk(KERN_ERR, sdkp, "formatted with unsupported" \
2241 " protection type %u. Disabling disk!\n",
2242 type);
2243 break;
2244 case 1:
2245 sd_printk(KERN_NOTICE, sdkp,
2246 "Enabling DIF Type %u protection\n", type);
2247 break;
2248 case 0:
2249 sd_printk(KERN_NOTICE, sdkp,
2250 "Disabling DIF Type %u protection\n", type);
2251 break;
2252 }
2253
2254 sdkp->protection_type = type;
2255
2256 return ret;
2257}
2258
2259static void read_capacity_error(struct scsi_disk *sdkp, struct scsi_device *sdp,
2260 struct scsi_sense_hdr *sshdr, int sense_valid,
2261 int the_result)
2262{
2263 if (driver_byte(the_result) == DRIVER_SENSE)
2264 sd_print_sense_hdr(sdkp, sshdr);
2265 else
2266 sd_printk(KERN_NOTICE, sdkp, "Sense not available.\n");
2267
2268 /*
2269 * Set dirty bit for removable devices if not ready -
2270 * sometimes drives will not report this properly.
2271 */
2272 if (sdp->removable &&
2273 sense_valid && sshdr->sense_key == NOT_READY)
2274 set_media_not_present(sdkp);
2275
2276 /*
2277 * We used to set media_present to 0 here to indicate no media
2278 * in the drive, but some drives fail read capacity even with
2279 * media present, so we can't do that.
2280 */
2281 sdkp->capacity = 0; /* unknown mapped to zero - as usual */
2282}
2283
2284#define RC16_LEN 32
2285#if RC16_LEN > SD_BUF_SIZE
2286#error RC16_LEN must not be more than SD_BUF_SIZE
2287#endif
2288
2289#define READ_CAPACITY_RETRIES_ON_RESET 10
2290
2291static int read_capacity_16(struct scsi_disk *sdkp, struct scsi_device *sdp,
2292 unsigned char *buffer)
2293{
2294 unsigned char cmd[16];
2295 struct scsi_sense_hdr sshdr;
2296 int sense_valid = 0;
2297 int the_result;
2298 int retries = 3, reset_retries = READ_CAPACITY_RETRIES_ON_RESET;
2299 unsigned int alignment;
2300 unsigned long long lba;
2301 unsigned sector_size;
2302
2303 if (sdp->no_read_capacity_16)
2304 return -EINVAL;
2305
2306 do {
2307 memset(cmd, 0, 16);
2308 cmd[0] = SERVICE_ACTION_IN_16;
2309 cmd[1] = SAI_READ_CAPACITY_16;
2310 cmd[13] = RC16_LEN;
2311 memset(buffer, 0, RC16_LEN);
2312
2313 the_result = scsi_execute_req(sdp, cmd, DMA_FROM_DEVICE,
2314 buffer, RC16_LEN, &sshdr,
2315 SD_TIMEOUT, SD_MAX_RETRIES, NULL);
2316
2317 if (media_not_present(sdkp, &sshdr))
2318 return -ENODEV;
2319
2320 if (the_result) {
2321 sense_valid = scsi_sense_valid(&sshdr);
2322 if (sense_valid &&
2323 sshdr.sense_key == ILLEGAL_REQUEST &&
2324 (sshdr.asc == 0x20 || sshdr.asc == 0x24) &&
2325 sshdr.ascq == 0x00)
2326 /* Invalid Command Operation Code or
2327 * Invalid Field in CDB, just retry
2328 * silently with RC10 */
2329 return -EINVAL;
2330 if (sense_valid &&
2331 sshdr.sense_key == UNIT_ATTENTION &&
2332 sshdr.asc == 0x29 && sshdr.ascq == 0x00)
2333 /* Device reset might occur several times,
2334 * give it one more chance */
2335 if (--reset_retries > 0)
2336 continue;
2337 }
2338 retries--;
2339
2340 } while (the_result && retries);
2341
2342 if (the_result) {
2343 sd_print_result(sdkp, "Read Capacity(16) failed", the_result);
2344 read_capacity_error(sdkp, sdp, &sshdr, sense_valid, the_result);
2345 return -EINVAL;
2346 }
2347
2348 sector_size = get_unaligned_be32(&buffer[8]);
2349 lba = get_unaligned_be64(&buffer[0]);
2350
2351 if (sd_read_protection_type(sdkp, buffer) < 0) {
2352 sdkp->capacity = 0;
2353 return -ENODEV;
2354 }
2355
2356 /* Logical blocks per physical block exponent */
2357 sdkp->physical_block_size = (1 << (buffer[13] & 0xf)) * sector_size;
2358
2359 /* RC basis */
2360 sdkp->rc_basis = (buffer[12] >> 4) & 0x3;
2361
2362 /* Lowest aligned logical block */
2363 alignment = ((buffer[14] & 0x3f) << 8 | buffer[15]) * sector_size;
2364 blk_queue_alignment_offset(sdp->request_queue, alignment);
2365 if (alignment && sdkp->first_scan)
2366 sd_printk(KERN_NOTICE, sdkp,
2367 "physical block alignment offset: %u\n", alignment);
2368
2369 if (buffer[14] & 0x80) { /* LBPME */
2370 sdkp->lbpme = 1;
2371
2372 if (buffer[14] & 0x40) /* LBPRZ */
2373 sdkp->lbprz = 1;
2374
2375 sd_config_discard(sdkp, SD_LBP_WS16);
2376 }
2377
2378 sdkp->capacity = lba + 1;
2379 return sector_size;
2380}
2381
2382static int read_capacity_10(struct scsi_disk *sdkp, struct scsi_device *sdp,
2383 unsigned char *buffer)
2384{
2385 unsigned char cmd[16];
2386 struct scsi_sense_hdr sshdr;
2387 int sense_valid = 0;
2388 int the_result;
2389 int retries = 3, reset_retries = READ_CAPACITY_RETRIES_ON_RESET;
2390 sector_t lba;
2391 unsigned sector_size;
2392
2393 do {
2394 cmd[0] = READ_CAPACITY;
2395 memset(&cmd[1], 0, 9);
2396 memset(buffer, 0, 8);
2397
2398 the_result = scsi_execute_req(sdp, cmd, DMA_FROM_DEVICE,
2399 buffer, 8, &sshdr,
2400 SD_TIMEOUT, SD_MAX_RETRIES, NULL);
2401
2402 if (media_not_present(sdkp, &sshdr))
2403 return -ENODEV;
2404
2405 if (the_result) {
2406 sense_valid = scsi_sense_valid(&sshdr);
2407 if (sense_valid &&
2408 sshdr.sense_key == UNIT_ATTENTION &&
2409 sshdr.asc == 0x29 && sshdr.ascq == 0x00)
2410 /* Device reset might occur several times,
2411 * give it one more chance */
2412 if (--reset_retries > 0)
2413 continue;
2414 }
2415 retries--;
2416
2417 } while (the_result && retries);
2418
2419 if (the_result) {
2420 sd_print_result(sdkp, "Read Capacity(10) failed", the_result);
2421 read_capacity_error(sdkp, sdp, &sshdr, sense_valid, the_result);
2422 return -EINVAL;
2423 }
2424
2425 sector_size = get_unaligned_be32(&buffer[4]);
2426 lba = get_unaligned_be32(&buffer[0]);
2427
2428 if (sdp->no_read_capacity_16 && (lba == 0xffffffff)) {
2429 /* Some buggy (usb cardreader) devices return an lba of
2430 0xffffffff when the want to report a size of 0 (with
2431 which they really mean no media is present) */
2432 sdkp->capacity = 0;
2433 sdkp->physical_block_size = sector_size;
2434 return sector_size;
2435 }
2436
2437 sdkp->capacity = lba + 1;
2438 sdkp->physical_block_size = sector_size;
2439 return sector_size;
2440}
2441
2442static int sd_try_rc16_first(struct scsi_device *sdp)
2443{
2444 if (sdp->host->max_cmd_len < 16)
2445 return 0;
2446 if (sdp->try_rc_10_first)
2447 return 0;
2448 if (sdp->scsi_level > SCSI_SPC_2)
2449 return 1;
2450 if (scsi_device_protection(sdp))
2451 return 1;
2452 return 0;
2453}
2454
2455/*
2456 * read disk capacity
2457 */
2458static void
2459sd_read_capacity(struct scsi_disk *sdkp, unsigned char *buffer)
2460{
2461 int sector_size;
2462 struct scsi_device *sdp = sdkp->device;
2463
2464 if (sd_try_rc16_first(sdp)) {
2465 sector_size = read_capacity_16(sdkp, sdp, buffer);
2466 if (sector_size == -EOVERFLOW)
2467 goto got_data;
2468 if (sector_size == -ENODEV)
2469 return;
2470 if (sector_size < 0)
2471 sector_size = read_capacity_10(sdkp, sdp, buffer);
2472 if (sector_size < 0)
2473 return;
2474 } else {
2475 sector_size = read_capacity_10(sdkp, sdp, buffer);
2476 if (sector_size == -EOVERFLOW)
2477 goto got_data;
2478 if (sector_size < 0)
2479 return;
2480 if ((sizeof(sdkp->capacity) > 4) &&
2481 (sdkp->capacity > 0xffffffffULL)) {
2482 int old_sector_size = sector_size;
2483 sd_printk(KERN_NOTICE, sdkp, "Very big device. "
2484 "Trying to use READ CAPACITY(16).\n");
2485 sector_size = read_capacity_16(sdkp, sdp, buffer);
2486 if (sector_size < 0) {
2487 sd_printk(KERN_NOTICE, sdkp,
2488 "Using 0xffffffff as device size\n");
2489 sdkp->capacity = 1 + (sector_t) 0xffffffff;
2490 sector_size = old_sector_size;
2491 goto got_data;
2492 }
2493 /* Remember that READ CAPACITY(16) succeeded */
2494 sdp->try_rc_10_first = 0;
2495 }
2496 }
2497
2498 /* Some devices are known to return the total number of blocks,
2499 * not the highest block number. Some devices have versions
2500 * which do this and others which do not. Some devices we might
2501 * suspect of doing this but we don't know for certain.
2502 *
2503 * If we know the reported capacity is wrong, decrement it. If
2504 * we can only guess, then assume the number of blocks is even
2505 * (usually true but not always) and err on the side of lowering
2506 * the capacity.
2507 */
2508 if (sdp->fix_capacity ||
2509 (sdp->guess_capacity && (sdkp->capacity & 0x01))) {
2510 sd_printk(KERN_INFO, sdkp, "Adjusting the sector count "
2511 "from its reported value: %llu\n",
2512 (unsigned long long) sdkp->capacity);
2513 --sdkp->capacity;
2514 }
2515
2516got_data:
2517 if (sector_size == 0) {
2518 sector_size = 512;
2519 sd_printk(KERN_NOTICE, sdkp, "Sector size 0 reported, "
2520 "assuming 512.\n");
2521 }
2522
2523 if (sector_size != 512 &&
2524 sector_size != 1024 &&
2525 sector_size != 2048 &&
2526 sector_size != 4096) {
2527 sd_printk(KERN_NOTICE, sdkp, "Unsupported sector size %d.\n",
2528 sector_size);
2529 /*
2530 * The user might want to re-format the drive with
2531 * a supported sectorsize. Once this happens, it
2532 * would be relatively trivial to set the thing up.
2533 * For this reason, we leave the thing in the table.
2534 */
2535 sdkp->capacity = 0;
2536 /*
2537 * set a bogus sector size so the normal read/write
2538 * logic in the block layer will eventually refuse any
2539 * request on this device without tripping over power
2540 * of two sector size assumptions
2541 */
2542 sector_size = 512;
2543 }
2544 blk_queue_logical_block_size(sdp->request_queue, sector_size);
2545 blk_queue_physical_block_size(sdp->request_queue,
2546 sdkp->physical_block_size);
2547 sdkp->device->sector_size = sector_size;
2548
2549 if (sdkp->capacity > 0xffffffff)
2550 sdp->use_16_for_rw = 1;
2551
2552}
2553
2554/*
2555 * Print disk capacity
2556 */
2557static void
2558sd_print_capacity(struct scsi_disk *sdkp,
2559 sector_t old_capacity)
2560{
2561 int sector_size = sdkp->device->sector_size;
2562 char cap_str_2[10], cap_str_10[10];
2563
2564 if (!sdkp->first_scan && old_capacity == sdkp->capacity)
2565 return;
2566
2567 string_get_size(sdkp->capacity, sector_size,
2568 STRING_UNITS_2, cap_str_2, sizeof(cap_str_2));
2569 string_get_size(sdkp->capacity, sector_size,
2570 STRING_UNITS_10, cap_str_10, sizeof(cap_str_10));
2571
2572 sd_printk(KERN_NOTICE, sdkp,
2573 "%llu %d-byte logical blocks: (%s/%s)\n",
2574 (unsigned long long)sdkp->capacity,
2575 sector_size, cap_str_10, cap_str_2);
2576
2577 if (sdkp->physical_block_size != sector_size)
2578 sd_printk(KERN_NOTICE, sdkp,
2579 "%u-byte physical blocks\n",
2580 sdkp->physical_block_size);
2581}
2582
2583/* called with buffer of length 512 */
2584static inline int
2585sd_do_mode_sense(struct scsi_device *sdp, int dbd, int modepage,
2586 unsigned char *buffer, int len, struct scsi_mode_data *data,
2587 struct scsi_sense_hdr *sshdr)
2588{
2589 return scsi_mode_sense(sdp, dbd, modepage, buffer, len,
2590 SD_TIMEOUT, SD_MAX_RETRIES, data,
2591 sshdr);
2592}
2593
2594/*
2595 * read write protect setting, if possible - called only in sd_revalidate_disk()
2596 * called with buffer of length SD_BUF_SIZE
2597 */
2598static void
2599sd_read_write_protect_flag(struct scsi_disk *sdkp, unsigned char *buffer)
2600{
2601 int res;
2602 struct scsi_device *sdp = sdkp->device;
2603 struct scsi_mode_data data;
2604 int old_wp = sdkp->write_prot;
2605
2606 set_disk_ro(sdkp->disk, 0);
2607 if (sdp->skip_ms_page_3f) {
2608 sd_first_printk(KERN_NOTICE, sdkp, "Assuming Write Enabled\n");
2609 return;
2610 }
2611
2612 if (sdp->use_192_bytes_for_3f) {
2613 res = sd_do_mode_sense(sdp, 0, 0x3F, buffer, 192, &data, NULL);
2614 } else {
2615 /*
2616 * First attempt: ask for all pages (0x3F), but only 4 bytes.
2617 * We have to start carefully: some devices hang if we ask
2618 * for more than is available.
2619 */
2620 res = sd_do_mode_sense(sdp, 0, 0x3F, buffer, 4, &data, NULL);
2621
2622 /*
2623 * Second attempt: ask for page 0 When only page 0 is
2624 * implemented, a request for page 3F may return Sense Key
2625 * 5: Illegal Request, Sense Code 24: Invalid field in
2626 * CDB.
2627 */
2628 if (!scsi_status_is_good(res))
2629 res = sd_do_mode_sense(sdp, 0, 0, buffer, 4, &data, NULL);
2630
2631 /*
2632 * Third attempt: ask 255 bytes, as we did earlier.
2633 */
2634 if (!scsi_status_is_good(res))
2635 res = sd_do_mode_sense(sdp, 0, 0x3F, buffer, 255,
2636 &data, NULL);
2637 }
2638
2639 if (!scsi_status_is_good(res)) {
2640 sd_first_printk(KERN_WARNING, sdkp,
2641 "Test WP failed, assume Write Enabled\n");
2642 } else {
2643 sdkp->write_prot = ((data.device_specific & 0x80) != 0);
2644 set_disk_ro(sdkp->disk, sdkp->write_prot);
2645 if (sdkp->first_scan || old_wp != sdkp->write_prot) {
2646 sd_printk(KERN_NOTICE, sdkp, "Write Protect is %s\n",
2647 sdkp->write_prot ? "on" : "off");
2648 sd_printk(KERN_DEBUG, sdkp, "Mode Sense: %4ph\n", buffer);
2649 }
2650 }
2651}
2652
2653/*
2654 * sd_read_cache_type - called only from sd_revalidate_disk()
2655 * called with buffer of length SD_BUF_SIZE
2656 */
2657static void
2658sd_read_cache_type(struct scsi_disk *sdkp, unsigned char *buffer)
2659{
2660 int len = 0, res;
2661 struct scsi_device *sdp = sdkp->device;
2662
2663 int dbd;
2664 int modepage;
2665 int first_len;
2666 struct scsi_mode_data data;
2667 struct scsi_sense_hdr sshdr;
2668 int old_wce = sdkp->WCE;
2669 int old_rcd = sdkp->RCD;
2670 int old_dpofua = sdkp->DPOFUA;
2671
2672
2673 if (sdkp->cache_override)
2674 return;
2675
2676 first_len = 4;
2677 if (sdp->skip_ms_page_8) {
2678 if (sdp->type == TYPE_RBC)
2679 goto defaults;
2680 else {
2681 if (sdp->skip_ms_page_3f)
2682 goto defaults;
2683 modepage = 0x3F;
2684 if (sdp->use_192_bytes_for_3f)
2685 first_len = 192;
2686 dbd = 0;
2687 }
2688 } else if (sdp->type == TYPE_RBC) {
2689 modepage = 6;
2690 dbd = 8;
2691 } else {
2692 modepage = 8;
2693 dbd = 0;
2694 }
2695
2696 /* cautiously ask */
2697 res = sd_do_mode_sense(sdp, dbd, modepage, buffer, first_len,
2698 &data, &sshdr);
2699
2700 if (!scsi_status_is_good(res))
2701 goto bad_sense;
2702
2703 if (!data.header_length) {
2704 modepage = 6;
2705 first_len = 0;
2706 sd_first_printk(KERN_ERR, sdkp,
2707 "Missing header in MODE_SENSE response\n");
2708 }
2709
2710 /* that went OK, now ask for the proper length */
2711 len = data.length;
2712
2713 /*
2714 * We're only interested in the first three bytes, actually.
2715 * But the data cache page is defined for the first 20.
2716 */
2717 if (len < 3)
2718 goto bad_sense;
2719 else if (len > SD_BUF_SIZE) {
2720 sd_first_printk(KERN_NOTICE, sdkp, "Truncating mode parameter "
2721 "data from %d to %d bytes\n", len, SD_BUF_SIZE);
2722 len = SD_BUF_SIZE;
2723 }
2724 if (modepage == 0x3F && sdp->use_192_bytes_for_3f)
2725 len = 192;
2726
2727 /* Get the data */
2728 if (len > first_len)
2729 res = sd_do_mode_sense(sdp, dbd, modepage, buffer, len,
2730 &data, &sshdr);
2731
2732 if (scsi_status_is_good(res)) {
2733 int offset = data.header_length + data.block_descriptor_length;
2734
2735 while (offset < len) {
2736 u8 page_code = buffer[offset] & 0x3F;
2737 u8 spf = buffer[offset] & 0x40;
2738
2739 if (page_code == 8 || page_code == 6) {
2740 /* We're interested only in the first 3 bytes.
2741 */
2742 if (len - offset <= 2) {
2743 sd_first_printk(KERN_ERR, sdkp,
2744 "Incomplete mode parameter "
2745 "data\n");
2746 goto defaults;
2747 } else {
2748 modepage = page_code;
2749 goto Page_found;
2750 }
2751 } else {
2752 /* Go to the next page */
2753 if (spf && len - offset > 3)
2754 offset += 4 + (buffer[offset+2] << 8) +
2755 buffer[offset+3];
2756 else if (!spf && len - offset > 1)
2757 offset += 2 + buffer[offset+1];
2758 else {
2759 sd_first_printk(KERN_ERR, sdkp,
2760 "Incomplete mode "
2761 "parameter data\n");
2762 goto defaults;
2763 }
2764 }
2765 }
2766
2767 sd_first_printk(KERN_ERR, sdkp, "No Caching mode page found\n");
2768 goto defaults;
2769
2770 Page_found:
2771 if (modepage == 8) {
2772 sdkp->WCE = ((buffer[offset + 2] & 0x04) != 0);
2773 sdkp->RCD = ((buffer[offset + 2] & 0x01) != 0);
2774 } else {
2775 sdkp->WCE = ((buffer[offset + 2] & 0x01) == 0);
2776 sdkp->RCD = 0;
2777 }
2778
2779 sdkp->DPOFUA = (data.device_specific & 0x10) != 0;
2780 if (sdp->broken_fua) {
2781 sd_first_printk(KERN_NOTICE, sdkp, "Disabling FUA\n");
2782 sdkp->DPOFUA = 0;
2783 } else if (sdkp->DPOFUA && !sdkp->device->use_10_for_rw &&
2784 !sdkp->device->use_16_for_rw) {
2785 sd_first_printk(KERN_NOTICE, sdkp,
2786 "Uses READ/WRITE(6), disabling FUA\n");
2787 sdkp->DPOFUA = 0;
2788 }
2789
2790 /* No cache flush allowed for write protected devices */
2791 if (sdkp->WCE && sdkp->write_prot)
2792 sdkp->WCE = 0;
2793
2794 if (sdkp->first_scan || old_wce != sdkp->WCE ||
2795 old_rcd != sdkp->RCD || old_dpofua != sdkp->DPOFUA)
2796 sd_printk(KERN_NOTICE, sdkp,
2797 "Write cache: %s, read cache: %s, %s\n",
2798 sdkp->WCE ? "enabled" : "disabled",
2799 sdkp->RCD ? "disabled" : "enabled",
2800 sdkp->DPOFUA ? "supports DPO and FUA"
2801 : "doesn't support DPO or FUA");
2802
2803 return;
2804 }
2805
2806bad_sense:
2807 if (scsi_sense_valid(&sshdr) &&
2808 sshdr.sense_key == ILLEGAL_REQUEST &&
2809 sshdr.asc == 0x24 && sshdr.ascq == 0x0)
2810 /* Invalid field in CDB */
2811 sd_first_printk(KERN_NOTICE, sdkp, "Cache data unavailable\n");
2812 else
2813 sd_first_printk(KERN_ERR, sdkp,
2814 "Asking for cache data failed\n");
2815
2816defaults:
2817 if (sdp->wce_default_on) {
2818 sd_first_printk(KERN_NOTICE, sdkp,
2819 "Assuming drive cache: write back\n");
2820 sdkp->WCE = 1;
2821 } else {
2822 sd_first_printk(KERN_ERR, sdkp,
2823 "Assuming drive cache: write through\n");
2824 sdkp->WCE = 0;
2825 }
2826 sdkp->RCD = 0;
2827 sdkp->DPOFUA = 0;
2828}
2829
2830/*
2831 * The ATO bit indicates whether the DIF application tag is available
2832 * for use by the operating system.
2833 */
2834static void sd_read_app_tag_own(struct scsi_disk *sdkp, unsigned char *buffer)
2835{
2836 int res, offset;
2837 struct scsi_device *sdp = sdkp->device;
2838 struct scsi_mode_data data;
2839 struct scsi_sense_hdr sshdr;
2840
2841 if (sdp->type != TYPE_DISK && sdp->type != TYPE_ZBC)
2842 return;
2843
2844 if (sdkp->protection_type == 0)
2845 return;
2846
2847 res = scsi_mode_sense(sdp, 1, 0x0a, buffer, 36, SD_TIMEOUT,
2848 SD_MAX_RETRIES, &data, &sshdr);
2849
2850 if (!scsi_status_is_good(res) || !data.header_length ||
2851 data.length < 6) {
2852 sd_first_printk(KERN_WARNING, sdkp,
2853 "getting Control mode page failed, assume no ATO\n");
2854
2855 if (scsi_sense_valid(&sshdr))
2856 sd_print_sense_hdr(sdkp, &sshdr);
2857
2858 return;
2859 }
2860
2861 offset = data.header_length + data.block_descriptor_length;
2862
2863 if ((buffer[offset] & 0x3f) != 0x0a) {
2864 sd_first_printk(KERN_ERR, sdkp, "ATO Got wrong page\n");
2865 return;
2866 }
2867
2868 if ((buffer[offset + 5] & 0x80) == 0)
2869 return;
2870
2871 sdkp->ATO = 1;
2872
2873 return;
2874}
2875
2876/**
2877 * sd_read_block_limits - Query disk device for preferred I/O sizes.
2878 * @sdkp: disk to query
2879 */
2880static void sd_read_block_limits(struct scsi_disk *sdkp)
2881{
2882 unsigned int sector_sz = sdkp->device->sector_size;
2883 const int vpd_len = 64;
2884 unsigned char *buffer = kmalloc(vpd_len, GFP_KERNEL);
2885
2886 if (!buffer ||
2887 /* Block Limits VPD */
2888 scsi_get_vpd_page(sdkp->device, 0xb0, buffer, vpd_len))
2889 goto out;
2890
2891 blk_queue_io_min(sdkp->disk->queue,
2892 get_unaligned_be16(&buffer[6]) * sector_sz);
2893
2894 sdkp->max_xfer_blocks = get_unaligned_be32(&buffer[8]);
2895 sdkp->opt_xfer_blocks = get_unaligned_be32(&buffer[12]);
2896
2897 if (buffer[3] == 0x3c) {
2898 unsigned int lba_count, desc_count;
2899
2900 sdkp->max_ws_blocks = (u32)get_unaligned_be64(&buffer[36]);
2901
2902 if (!sdkp->lbpme)
2903 goto out;
2904
2905 lba_count = get_unaligned_be32(&buffer[20]);
2906 desc_count = get_unaligned_be32(&buffer[24]);
2907
2908 if (lba_count && desc_count)
2909 sdkp->max_unmap_blocks = lba_count;
2910
2911 sdkp->unmap_granularity = get_unaligned_be32(&buffer[28]);
2912
2913 if (buffer[32] & 0x80)
2914 sdkp->unmap_alignment =
2915 get_unaligned_be32(&buffer[32]) & ~(1 << 31);
2916
2917 if (!sdkp->lbpvpd) { /* LBP VPD page not provided */
2918
2919 if (sdkp->max_unmap_blocks)
2920 sd_config_discard(sdkp, SD_LBP_UNMAP);
2921 else
2922 sd_config_discard(sdkp, SD_LBP_WS16);
2923
2924 } else { /* LBP VPD page tells us what to use */
2925 if (sdkp->lbpu && sdkp->max_unmap_blocks)
2926 sd_config_discard(sdkp, SD_LBP_UNMAP);
2927 else if (sdkp->lbpws)
2928 sd_config_discard(sdkp, SD_LBP_WS16);
2929 else if (sdkp->lbpws10)
2930 sd_config_discard(sdkp, SD_LBP_WS10);
2931 else
2932 sd_config_discard(sdkp, SD_LBP_DISABLE);
2933 }
2934 }
2935
2936 out:
2937 kfree(buffer);
2938}
2939
2940/**
2941 * sd_read_block_characteristics - Query block dev. characteristics
2942 * @sdkp: disk to query
2943 */
2944static void sd_read_block_characteristics(struct scsi_disk *sdkp)
2945{
2946 struct request_queue *q = sdkp->disk->queue;
2947 unsigned char *buffer;
2948 u16 rot;
2949 const int vpd_len = 64;
2950
2951 buffer = kmalloc(vpd_len, GFP_KERNEL);
2952
2953 if (!buffer ||
2954 /* Block Device Characteristics VPD */
2955 scsi_get_vpd_page(sdkp->device, 0xb1, buffer, vpd_len))
2956 goto out;
2957
2958 rot = get_unaligned_be16(&buffer[4]);
2959
2960 if (rot == 1) {
2961 blk_queue_flag_set(QUEUE_FLAG_NONROT, q);
2962 blk_queue_flag_clear(QUEUE_FLAG_ADD_RANDOM, q);
2963 }
2964
2965 if (sdkp->device->type == TYPE_ZBC) {
2966 /* Host-managed */
2967 blk_queue_set_zoned(sdkp->disk, BLK_ZONED_HM);
2968 } else {
2969 sdkp->zoned = (buffer[8] >> 4) & 3;
2970 if (sdkp->zoned == 1) {
2971 /* Host-aware */
2972 blk_queue_set_zoned(sdkp->disk, BLK_ZONED_HA);
2973 } else {
2974 /* Regular disk or drive managed disk */
2975 blk_queue_set_zoned(sdkp->disk, BLK_ZONED_NONE);
2976 }
2977 }
2978
2979 if (!sdkp->first_scan)
2980 goto out;
2981
2982 if (blk_queue_is_zoned(q)) {
2983 sd_printk(KERN_NOTICE, sdkp, "Host-%s zoned block device\n",
2984 q->limits.zoned == BLK_ZONED_HM ? "managed" : "aware");
2985 } else {
2986 if (sdkp->zoned == 1)
2987 sd_printk(KERN_NOTICE, sdkp,
2988 "Host-aware SMR disk used as regular disk\n");
2989 else if (sdkp->zoned == 2)
2990 sd_printk(KERN_NOTICE, sdkp,
2991 "Drive-managed SMR disk\n");
2992 }
2993
2994 out:
2995 kfree(buffer);
2996}
2997
2998/**
2999 * sd_read_block_provisioning - Query provisioning VPD page
3000 * @sdkp: disk to query
3001 */
3002static void sd_read_block_provisioning(struct scsi_disk *sdkp)
3003{
3004 unsigned char *buffer;
3005 const int vpd_len = 8;
3006
3007 if (sdkp->lbpme == 0)
3008 return;
3009
3010 buffer = kmalloc(vpd_len, GFP_KERNEL);
3011
3012 if (!buffer || scsi_get_vpd_page(sdkp->device, 0xb2, buffer, vpd_len))
3013 goto out;
3014
3015 sdkp->lbpvpd = 1;
3016 sdkp->lbpu = (buffer[5] >> 7) & 1; /* UNMAP */
3017 sdkp->lbpws = (buffer[5] >> 6) & 1; /* WRITE SAME(16) with UNMAP */
3018 sdkp->lbpws10 = (buffer[5] >> 5) & 1; /* WRITE SAME(10) with UNMAP */
3019
3020 out:
3021 kfree(buffer);
3022}
3023
3024static void sd_read_write_same(struct scsi_disk *sdkp, unsigned char *buffer)
3025{
3026 struct scsi_device *sdev = sdkp->device;
3027
3028 if (sdev->host->no_write_same) {
3029 sdev->no_write_same = 1;
3030
3031 return;
3032 }
3033
3034 if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE, INQUIRY) < 0) {
3035 /* too large values might cause issues with arcmsr */
3036 int vpd_buf_len = 64;
3037
3038 sdev->no_report_opcodes = 1;
3039
3040 /* Disable WRITE SAME if REPORT SUPPORTED OPERATION
3041 * CODES is unsupported and the device has an ATA
3042 * Information VPD page (SAT).
3043 */
3044 if (!scsi_get_vpd_page(sdev, 0x89, buffer, vpd_buf_len))
3045 sdev->no_write_same = 1;
3046 }
3047
3048 if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE, WRITE_SAME_16) == 1)
3049 sdkp->ws16 = 1;
3050
3051 if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE, WRITE_SAME) == 1)
3052 sdkp->ws10 = 1;
3053}
3054
3055static void sd_read_security(struct scsi_disk *sdkp, unsigned char *buffer)
3056{
3057 struct scsi_device *sdev = sdkp->device;
3058
3059 if (!sdev->security_supported)
3060 return;
3061
3062 if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE,
3063 SECURITY_PROTOCOL_IN) == 1 &&
3064 scsi_report_opcode(sdev, buffer, SD_BUF_SIZE,
3065 SECURITY_PROTOCOL_OUT) == 1)
3066 sdkp->security = 1;
3067}
3068
3069/*
3070 * Determine the device's preferred I/O size for reads and writes
3071 * unless the reported value is unreasonably small, large, not a
3072 * multiple of the physical block size, or simply garbage.
3073 */
3074static bool sd_validate_opt_xfer_size(struct scsi_disk *sdkp,
3075 unsigned int dev_max)
3076{
3077 struct scsi_device *sdp = sdkp->device;
3078 unsigned int opt_xfer_bytes =
3079 logical_to_bytes(sdp, sdkp->opt_xfer_blocks);
3080
3081 if (sdkp->opt_xfer_blocks == 0)
3082 return false;
3083
3084 if (sdkp->opt_xfer_blocks > dev_max) {
3085 sd_first_printk(KERN_WARNING, sdkp,
3086 "Optimal transfer size %u logical blocks " \
3087 "> dev_max (%u logical blocks)\n",
3088 sdkp->opt_xfer_blocks, dev_max);
3089 return false;
3090 }
3091
3092 if (sdkp->opt_xfer_blocks > SD_DEF_XFER_BLOCKS) {
3093 sd_first_printk(KERN_WARNING, sdkp,
3094 "Optimal transfer size %u logical blocks " \
3095 "> sd driver limit (%u logical blocks)\n",
3096 sdkp->opt_xfer_blocks, SD_DEF_XFER_BLOCKS);
3097 return false;
3098 }
3099
3100 if (opt_xfer_bytes < PAGE_SIZE) {
3101 sd_first_printk(KERN_WARNING, sdkp,
3102 "Optimal transfer size %u bytes < " \
3103 "PAGE_SIZE (%u bytes)\n",
3104 opt_xfer_bytes, (unsigned int)PAGE_SIZE);
3105 return false;
3106 }
3107
3108 if (opt_xfer_bytes & (sdkp->physical_block_size - 1)) {
3109 sd_first_printk(KERN_WARNING, sdkp,
3110 "Optimal transfer size %u bytes not a " \
3111 "multiple of physical block size (%u bytes)\n",
3112 opt_xfer_bytes, sdkp->physical_block_size);
3113 return false;
3114 }
3115
3116 sd_first_printk(KERN_INFO, sdkp, "Optimal transfer size %u bytes\n",
3117 opt_xfer_bytes);
3118 return true;
3119}
3120
3121/**
3122 * sd_revalidate_disk - called the first time a new disk is seen,
3123 * performs disk spin up, read_capacity, etc.
3124 * @disk: struct gendisk we care about
3125 **/
3126static int sd_revalidate_disk(struct gendisk *disk)
3127{
3128 struct scsi_disk *sdkp = scsi_disk(disk);
3129 struct scsi_device *sdp = sdkp->device;
3130 struct request_queue *q = sdkp->disk->queue;
3131 sector_t old_capacity = sdkp->capacity;
3132 unsigned char *buffer;
3133 unsigned int dev_max, rw_max;
3134
3135 SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp,
3136 "sd_revalidate_disk\n"));
3137
3138 /*
3139 * If the device is offline, don't try and read capacity or any
3140 * of the other niceties.
3141 */
3142 if (!scsi_device_online(sdp))
3143 goto out;
3144
3145 buffer = kmalloc(SD_BUF_SIZE, GFP_KERNEL);
3146 if (!buffer) {
3147 sd_printk(KERN_WARNING, sdkp, "sd_revalidate_disk: Memory "
3148 "allocation failure.\n");
3149 goto out;
3150 }
3151
3152 sd_spinup_disk(sdkp);
3153
3154 /*
3155 * Without media there is no reason to ask; moreover, some devices
3156 * react badly if we do.
3157 */
3158 if (sdkp->media_present) {
3159 sd_read_capacity(sdkp, buffer);
3160
3161 /*
3162 * set the default to rotational. All non-rotational devices
3163 * support the block characteristics VPD page, which will
3164 * cause this to be updated correctly and any device which
3165 * doesn't support it should be treated as rotational.
3166 */
3167 blk_queue_flag_clear(QUEUE_FLAG_NONROT, q);
3168 blk_queue_flag_set(QUEUE_FLAG_ADD_RANDOM, q);
3169
3170 if (scsi_device_supports_vpd(sdp)) {
3171 sd_read_block_provisioning(sdkp);
3172 sd_read_block_limits(sdkp);
3173 sd_read_block_characteristics(sdkp);
3174 sd_zbc_read_zones(sdkp, buffer);
3175 }
3176
3177 sd_print_capacity(sdkp, old_capacity);
3178
3179 sd_read_write_protect_flag(sdkp, buffer);
3180 sd_read_cache_type(sdkp, buffer);
3181 sd_read_app_tag_own(sdkp, buffer);
3182 sd_read_write_same(sdkp, buffer);
3183 sd_read_security(sdkp, buffer);
3184 }
3185
3186 /*
3187 * We now have all cache related info, determine how we deal
3188 * with flush requests.
3189 */
3190 sd_set_flush_flag(sdkp);
3191
3192 /* Initial block count limit based on CDB TRANSFER LENGTH field size. */
3193 dev_max = sdp->use_16_for_rw ? SD_MAX_XFER_BLOCKS : SD_DEF_XFER_BLOCKS;
3194
3195 /* Some devices report a maximum block count for READ/WRITE requests. */
3196 dev_max = min_not_zero(dev_max, sdkp->max_xfer_blocks);
3197 q->limits.max_dev_sectors = logical_to_sectors(sdp, dev_max);
3198
3199 if (sd_validate_opt_xfer_size(sdkp, dev_max)) {
3200 q->limits.io_opt = logical_to_bytes(sdp, sdkp->opt_xfer_blocks);
3201 rw_max = logical_to_sectors(sdp, sdkp->opt_xfer_blocks);
3202 } else {
3203 q->limits.io_opt = 0;
3204 rw_max = min_not_zero(logical_to_sectors(sdp, dev_max),
3205 (sector_t)BLK_DEF_MAX_SECTORS);
3206 }
3207
3208 /* Do not exceed controller limit */
3209 rw_max = min(rw_max, queue_max_hw_sectors(q));
3210
3211 /*
3212 * Only update max_sectors if previously unset or if the current value
3213 * exceeds the capabilities of the hardware.
3214 */
3215 if (sdkp->first_scan ||
3216 q->limits.max_sectors > q->limits.max_dev_sectors ||
3217 q->limits.max_sectors > q->limits.max_hw_sectors)
3218 q->limits.max_sectors = rw_max;
3219
3220 sdkp->first_scan = 0;
3221
3222 set_capacity_revalidate_and_notify(disk,
3223 logical_to_sectors(sdp, sdkp->capacity), false);
3224 sd_config_write_same(sdkp);
3225 kfree(buffer);
3226
3227 /*
3228 * For a zoned drive, revalidating the zones can be done only once
3229 * the gendisk capacity is set. So if this fails, set back the gendisk
3230 * capacity to 0.
3231 */
3232 if (sd_zbc_revalidate_zones(sdkp))
3233 set_capacity_revalidate_and_notify(disk, 0, false);
3234
3235 out:
3236 return 0;
3237}
3238
3239/**
3240 * sd_unlock_native_capacity - unlock native capacity
3241 * @disk: struct gendisk to set capacity for
3242 *
3243 * Block layer calls this function if it detects that partitions
3244 * on @disk reach beyond the end of the device. If the SCSI host
3245 * implements ->unlock_native_capacity() method, it's invoked to
3246 * give it a chance to adjust the device capacity.
3247 *
3248 * CONTEXT:
3249 * Defined by block layer. Might sleep.
3250 */
3251static void sd_unlock_native_capacity(struct gendisk *disk)
3252{
3253 struct scsi_device *sdev = scsi_disk(disk)->device;
3254
3255 if (sdev->host->hostt->unlock_native_capacity)
3256 sdev->host->hostt->unlock_native_capacity(sdev);
3257}
3258
3259/**
3260 * sd_format_disk_name - format disk name
3261 * @prefix: name prefix - ie. "sd" for SCSI disks
3262 * @index: index of the disk to format name for
3263 * @buf: output buffer
3264 * @buflen: length of the output buffer
3265 *
3266 * SCSI disk names starts at sda. The 26th device is sdz and the
3267 * 27th is sdaa. The last one for two lettered suffix is sdzz
3268 * which is followed by sdaaa.
3269 *
3270 * This is basically 26 base counting with one extra 'nil' entry
3271 * at the beginning from the second digit on and can be
3272 * determined using similar method as 26 base conversion with the
3273 * index shifted -1 after each digit is computed.
3274 *
3275 * CONTEXT:
3276 * Don't care.
3277 *
3278 * RETURNS:
3279 * 0 on success, -errno on failure.
3280 */
3281static int sd_format_disk_name(char *prefix, int index, char *buf, int buflen)
3282{
3283 const int base = 'z' - 'a' + 1;
3284 char *begin = buf + strlen(prefix);
3285 char *end = buf + buflen;
3286 char *p;
3287 int unit;
3288
3289 p = end - 1;
3290 *p = '\0';
3291 unit = base;
3292 do {
3293 if (p == begin)
3294 return -EINVAL;
3295 *--p = 'a' + (index % unit);
3296 index = (index / unit) - 1;
3297 } while (index >= 0);
3298
3299 memmove(begin, p, end - p);
3300 memcpy(buf, prefix, strlen(prefix));
3301
3302 return 0;
3303}
3304
3305/**
3306 * sd_probe - called during driver initialization and whenever a
3307 * new scsi device is attached to the system. It is called once
3308 * for each scsi device (not just disks) present.
3309 * @dev: pointer to device object
3310 *
3311 * Returns 0 if successful (or not interested in this scsi device
3312 * (e.g. scanner)); 1 when there is an error.
3313 *
3314 * Note: this function is invoked from the scsi mid-level.
3315 * This function sets up the mapping between a given
3316 * <host,channel,id,lun> (found in sdp) and new device name
3317 * (e.g. /dev/sda). More precisely it is the block device major
3318 * and minor number that is chosen here.
3319 *
3320 * Assume sd_probe is not re-entrant (for time being)
3321 * Also think about sd_probe() and sd_remove() running coincidentally.
3322 **/
3323static int sd_probe(struct device *dev)
3324{
3325 struct scsi_device *sdp = to_scsi_device(dev);
3326 struct scsi_disk *sdkp;
3327 struct gendisk *gd;
3328 int index;
3329 int error;
3330
3331 scsi_autopm_get_device(sdp);
3332 error = -ENODEV;
3333 if (sdp->type != TYPE_DISK &&
3334 sdp->type != TYPE_ZBC &&
3335 sdp->type != TYPE_MOD &&
3336 sdp->type != TYPE_RBC)
3337 goto out;
3338
3339#ifndef CONFIG_BLK_DEV_ZONED
3340 if (sdp->type == TYPE_ZBC)
3341 goto out;
3342#endif
3343 SCSI_LOG_HLQUEUE(3, sdev_printk(KERN_INFO, sdp,
3344 "sd_probe\n"));
3345
3346 error = -ENOMEM;
3347 sdkp = kzalloc(sizeof(*sdkp), GFP_KERNEL);
3348 if (!sdkp)
3349 goto out;
3350
3351 gd = alloc_disk(SD_MINORS);
3352 if (!gd)
3353 goto out_free;
3354
3355 index = ida_alloc(&sd_index_ida, GFP_KERNEL);
3356 if (index < 0) {
3357 sdev_printk(KERN_WARNING, sdp, "sd_probe: memory exhausted.\n");
3358 goto out_put;
3359 }
3360
3361 error = sd_format_disk_name("sd", index, gd->disk_name, DISK_NAME_LEN);
3362 if (error) {
3363 sdev_printk(KERN_WARNING, sdp, "SCSI disk (sd) name length exceeded.\n");
3364 goto out_free_index;
3365 }
3366
3367 sdkp->device = sdp;
3368 sdkp->driver = &sd_template;
3369 sdkp->disk = gd;
3370 sdkp->index = index;
3371 atomic_set(&sdkp->openers, 0);
3372 atomic_set(&sdkp->device->ioerr_cnt, 0);
3373
3374 if (!sdp->request_queue->rq_timeout) {
3375 if (sdp->type != TYPE_MOD)
3376 blk_queue_rq_timeout(sdp->request_queue, SD_TIMEOUT);
3377 else
3378 blk_queue_rq_timeout(sdp->request_queue,
3379 SD_MOD_TIMEOUT);
3380 }
3381
3382 device_initialize(&sdkp->dev);
3383 sdkp->dev.parent = dev;
3384 sdkp->dev.class = &sd_disk_class;
3385 dev_set_name(&sdkp->dev, "%s", dev_name(dev));
3386
3387 error = device_add(&sdkp->dev);
3388 if (error)
3389 goto out_free_index;
3390
3391 get_device(dev);
3392 dev_set_drvdata(dev, sdkp);
3393
3394 gd->major = sd_major((index & 0xf0) >> 4);
3395 gd->first_minor = ((index & 0xf) << 4) | (index & 0xfff00);
3396
3397 gd->fops = &sd_fops;
3398 gd->private_data = &sdkp->driver;
3399 gd->queue = sdkp->device->request_queue;
3400
3401 /* defaults, until the device tells us otherwise */
3402 sdp->sector_size = 512;
3403 sdkp->capacity = 0;
3404 sdkp->media_present = 1;
3405 sdkp->write_prot = 0;
3406 sdkp->cache_override = 0;
3407 sdkp->WCE = 0;
3408 sdkp->RCD = 0;
3409 sdkp->ATO = 0;
3410 sdkp->first_scan = 1;
3411 sdkp->max_medium_access_timeouts = SD_MAX_MEDIUM_TIMEOUTS;
3412
3413 sd_revalidate_disk(gd);
3414
3415 gd->flags = GENHD_FL_EXT_DEVT;
3416 if (sdp->removable) {
3417 gd->flags |= GENHD_FL_REMOVABLE;
3418 gd->events |= DISK_EVENT_MEDIA_CHANGE;
3419 gd->event_flags = DISK_EVENT_FLAG_POLL | DISK_EVENT_FLAG_UEVENT;
3420 }
3421
3422 blk_pm_runtime_init(sdp->request_queue, dev);
3423 if (sdp->rpm_autosuspend) {
3424 pm_runtime_set_autosuspend_delay(dev,
3425 sdp->host->hostt->rpm_autosuspend_delay);
3426 }
3427 device_add_disk(dev, gd, NULL);
3428 if (sdkp->capacity)
3429 sd_dif_config_host(sdkp);
3430
3431 sd_revalidate_disk(gd);
3432
3433 if (sdkp->security) {
3434 sdkp->opal_dev = init_opal_dev(sdp, &sd_sec_submit);
3435 if (sdkp->opal_dev)
3436 sd_printk(KERN_NOTICE, sdkp, "supports TCG Opal\n");
3437 }
3438
3439 sd_printk(KERN_NOTICE, sdkp, "Attached SCSI %sdisk\n",
3440 sdp->removable ? "removable " : "");
3441 scsi_autopm_put_device(sdp);
3442
3443 return 0;
3444
3445 out_free_index:
3446 ida_free(&sd_index_ida, index);
3447 out_put:
3448 put_disk(gd);
3449 out_free:
3450 sd_zbc_release_disk(sdkp);
3451 kfree(sdkp);
3452 out:
3453 scsi_autopm_put_device(sdp);
3454 return error;
3455}
3456
3457/**
3458 * sd_remove - called whenever a scsi disk (previously recognized by
3459 * sd_probe) is detached from the system. It is called (potentially
3460 * multiple times) during sd module unload.
3461 * @dev: pointer to device object
3462 *
3463 * Note: this function is invoked from the scsi mid-level.
3464 * This function potentially frees up a device name (e.g. /dev/sdc)
3465 * that could be re-used by a subsequent sd_probe().
3466 * This function is not called when the built-in sd driver is "exit-ed".
3467 **/
3468static int sd_remove(struct device *dev)
3469{
3470 struct scsi_disk *sdkp;
3471 dev_t devt;
3472
3473 sdkp = dev_get_drvdata(dev);
3474 devt = disk_devt(sdkp->disk);
3475 scsi_autopm_get_device(sdkp->device);
3476
3477 async_synchronize_full_domain(&scsi_sd_pm_domain);
3478 device_del(&sdkp->dev);
3479 del_gendisk(sdkp->disk);
3480 sd_shutdown(dev);
3481
3482 free_opal_dev(sdkp->opal_dev);
3483
3484 blk_register_region(devt, SD_MINORS, NULL,
3485 sd_default_probe, NULL, NULL);
3486
3487 mutex_lock(&sd_ref_mutex);
3488 dev_set_drvdata(dev, NULL);
3489 put_device(&sdkp->dev);
3490 mutex_unlock(&sd_ref_mutex);
3491
3492 return 0;
3493}
3494
3495/**
3496 * scsi_disk_release - Called to free the scsi_disk structure
3497 * @dev: pointer to embedded class device
3498 *
3499 * sd_ref_mutex must be held entering this routine. Because it is
3500 * called on last put, you should always use the scsi_disk_get()
3501 * scsi_disk_put() helpers which manipulate the semaphore directly
3502 * and never do a direct put_device.
3503 **/
3504static void scsi_disk_release(struct device *dev)
3505{
3506 struct scsi_disk *sdkp = to_scsi_disk(dev);
3507 struct gendisk *disk = sdkp->disk;
3508 struct request_queue *q = disk->queue;
3509
3510 ida_free(&sd_index_ida, sdkp->index);
3511
3512 /*
3513 * Wait until all requests that are in progress have completed.
3514 * This is necessary to avoid that e.g. scsi_end_request() crashes
3515 * due to clearing the disk->private_data pointer. Wait from inside
3516 * scsi_disk_release() instead of from sd_release() to avoid that
3517 * freezing and unfreezing the request queue affects user space I/O
3518 * in case multiple processes open a /dev/sd... node concurrently.
3519 */
3520 blk_mq_freeze_queue(q);
3521 blk_mq_unfreeze_queue(q);
3522
3523 disk->private_data = NULL;
3524 put_disk(disk);
3525 put_device(&sdkp->device->sdev_gendev);
3526
3527 sd_zbc_release_disk(sdkp);
3528
3529 kfree(sdkp);
3530}
3531
3532static int sd_start_stop_device(struct scsi_disk *sdkp, int start)
3533{
3534 unsigned char cmd[6] = { START_STOP }; /* START_VALID */
3535 struct scsi_sense_hdr sshdr;
3536 struct scsi_device *sdp = sdkp->device;
3537 int res;
3538
3539 if (start)
3540 cmd[4] |= 1; /* START */
3541
3542 if (sdp->start_stop_pwr_cond)
3543 cmd[4] |= start ? 1 << 4 : 3 << 4; /* Active or Standby */
3544
3545 if (!scsi_device_online(sdp))
3546 return -ENODEV;
3547
3548 res = scsi_execute(sdp, cmd, DMA_NONE, NULL, 0, NULL, &sshdr,
3549 SD_TIMEOUT, SD_MAX_RETRIES, 0, RQF_PM, NULL);
3550 if (res) {
3551 sd_print_result(sdkp, "Start/Stop Unit failed", res);
3552 if (driver_byte(res) == DRIVER_SENSE)
3553 sd_print_sense_hdr(sdkp, &sshdr);
3554 if (scsi_sense_valid(&sshdr) &&
3555 /* 0x3a is medium not present */
3556 sshdr.asc == 0x3a)
3557 res = 0;
3558 }
3559
3560 /* SCSI error codes must not go to the generic layer */
3561 if (res)
3562 return -EIO;
3563
3564 return 0;
3565}
3566
3567/*
3568 * Send a SYNCHRONIZE CACHE instruction down to the device through
3569 * the normal SCSI command structure. Wait for the command to
3570 * complete.
3571 */
3572static void sd_shutdown(struct device *dev)
3573{
3574 struct scsi_disk *sdkp = dev_get_drvdata(dev);
3575
3576 if (!sdkp)
3577 return; /* this can happen */
3578
3579 if (pm_runtime_suspended(dev))
3580 return;
3581
3582 if (sdkp->WCE && sdkp->media_present) {
3583 sd_printk(KERN_NOTICE, sdkp, "Synchronizing SCSI cache\n");
3584 sd_sync_cache(sdkp, NULL);
3585 }
3586
3587 if (system_state != SYSTEM_RESTART && sdkp->device->manage_start_stop) {
3588 sd_printk(KERN_NOTICE, sdkp, "Stopping disk\n");
3589 sd_start_stop_device(sdkp, 0);
3590 }
3591}
3592
3593static int sd_suspend_common(struct device *dev, bool ignore_stop_errors)
3594{
3595 struct scsi_disk *sdkp = dev_get_drvdata(dev);
3596 struct scsi_sense_hdr sshdr;
3597 int ret = 0;
3598
3599 if (!sdkp) /* E.g.: runtime suspend following sd_remove() */
3600 return 0;
3601
3602 if (sdkp->WCE && sdkp->media_present) {
3603 sd_printk(KERN_NOTICE, sdkp, "Synchronizing SCSI cache\n");
3604 ret = sd_sync_cache(sdkp, &sshdr);
3605
3606 if (ret) {
3607 /* ignore OFFLINE device */
3608 if (ret == -ENODEV)
3609 return 0;
3610
3611 if (!scsi_sense_valid(&sshdr) ||
3612 sshdr.sense_key != ILLEGAL_REQUEST)
3613 return ret;
3614
3615 /*
3616 * sshdr.sense_key == ILLEGAL_REQUEST means this drive
3617 * doesn't support sync. There's not much to do and
3618 * suspend shouldn't fail.
3619 */
3620 ret = 0;
3621 }
3622 }
3623
3624 if (sdkp->device->manage_start_stop) {
3625 sd_printk(KERN_NOTICE, sdkp, "Stopping disk\n");
3626 /* an error is not worth aborting a system sleep */
3627 ret = sd_start_stop_device(sdkp, 0);
3628 if (ignore_stop_errors)
3629 ret = 0;
3630 }
3631
3632 return ret;
3633}
3634
3635static int sd_suspend_system(struct device *dev)
3636{
3637 return sd_suspend_common(dev, true);
3638}
3639
3640static int sd_suspend_runtime(struct device *dev)
3641{
3642 return sd_suspend_common(dev, false);
3643}
3644
3645static int sd_resume(struct device *dev)
3646{
3647 struct scsi_disk *sdkp = dev_get_drvdata(dev);
3648 int ret;
3649
3650 if (!sdkp) /* E.g.: runtime resume at the start of sd_probe() */
3651 return 0;
3652
3653 if (!sdkp->device->manage_start_stop)
3654 return 0;
3655
3656 sd_printk(KERN_NOTICE, sdkp, "Starting disk\n");
3657 ret = sd_start_stop_device(sdkp, 1);
3658 if (!ret)
3659 opal_unlock_from_suspend(sdkp->opal_dev);
3660 return ret;
3661}
3662
3663/**
3664 * init_sd - entry point for this driver (both when built in or when
3665 * a module).
3666 *
3667 * Note: this function registers this driver with the scsi mid-level.
3668 **/
3669static int __init init_sd(void)
3670{
3671 int majors = 0, i, err;
3672
3673 SCSI_LOG_HLQUEUE(3, printk("init_sd: sd driver entry point\n"));
3674
3675 for (i = 0; i < SD_MAJORS; i++) {
3676 if (register_blkdev(sd_major(i), "sd") != 0)
3677 continue;
3678 majors++;
3679 blk_register_region(sd_major(i), SD_MINORS, NULL,
3680 sd_default_probe, NULL, NULL);
3681 }
3682
3683 if (!majors)
3684 return -ENODEV;
3685
3686 err = class_register(&sd_disk_class);
3687 if (err)
3688 goto err_out;
3689
3690 sd_cdb_cache = kmem_cache_create("sd_ext_cdb", SD_EXT_CDB_SIZE,
3691 0, 0, NULL);
3692 if (!sd_cdb_cache) {
3693 printk(KERN_ERR "sd: can't init extended cdb cache\n");
3694 err = -ENOMEM;
3695 goto err_out_class;
3696 }
3697
3698 sd_cdb_pool = mempool_create_slab_pool(SD_MEMPOOL_SIZE, sd_cdb_cache);
3699 if (!sd_cdb_pool) {
3700 printk(KERN_ERR "sd: can't init extended cdb pool\n");
3701 err = -ENOMEM;
3702 goto err_out_cache;
3703 }
3704
3705 sd_page_pool = mempool_create_page_pool(SD_MEMPOOL_SIZE, 0);
3706 if (!sd_page_pool) {
3707 printk(KERN_ERR "sd: can't init discard page pool\n");
3708 err = -ENOMEM;
3709 goto err_out_ppool;
3710 }
3711
3712 err = scsi_register_driver(&sd_template.gendrv);
3713 if (err)
3714 goto err_out_driver;
3715
3716 return 0;
3717
3718err_out_driver:
3719 mempool_destroy(sd_page_pool);
3720
3721err_out_ppool:
3722 mempool_destroy(sd_cdb_pool);
3723
3724err_out_cache:
3725 kmem_cache_destroy(sd_cdb_cache);
3726
3727err_out_class:
3728 class_unregister(&sd_disk_class);
3729err_out:
3730 for (i = 0; i < SD_MAJORS; i++)
3731 unregister_blkdev(sd_major(i), "sd");
3732 return err;
3733}
3734
3735/**
3736 * exit_sd - exit point for this driver (when it is a module).
3737 *
3738 * Note: this function unregisters this driver from the scsi mid-level.
3739 **/
3740static void __exit exit_sd(void)
3741{
3742 int i;
3743
3744 SCSI_LOG_HLQUEUE(3, printk("exit_sd: exiting sd driver\n"));
3745
3746 scsi_unregister_driver(&sd_template.gendrv);
3747 mempool_destroy(sd_cdb_pool);
3748 mempool_destroy(sd_page_pool);
3749 kmem_cache_destroy(sd_cdb_cache);
3750
3751 class_unregister(&sd_disk_class);
3752
3753 for (i = 0; i < SD_MAJORS; i++) {
3754 blk_unregister_region(sd_major(i), SD_MINORS);
3755 unregister_blkdev(sd_major(i), "sd");
3756 }
3757}
3758
3759module_init(init_sd);
3760module_exit(exit_sd);
3761
3762void sd_print_sense_hdr(struct scsi_disk *sdkp, struct scsi_sense_hdr *sshdr)
3763{
3764 scsi_print_sense_hdr(sdkp->device,
3765 sdkp->disk ? sdkp->disk->disk_name : NULL, sshdr);
3766}
3767
3768void sd_print_result(const struct scsi_disk *sdkp, const char *msg, int result)
3769{
3770 const char *hb_string = scsi_hostbyte_string(result);
3771 const char *db_string = scsi_driverbyte_string(result);
3772
3773 if (hb_string || db_string)
3774 sd_printk(KERN_INFO, sdkp,
3775 "%s: Result: hostbyte=%s driverbyte=%s\n", msg,
3776 hb_string ? hb_string : "invalid",
3777 db_string ? db_string : "invalid");
3778 else
3779 sd_printk(KERN_INFO, sdkp,
3780 "%s: Result: hostbyte=0x%02x driverbyte=0x%02x\n",
3781 msg, host_byte(result), driver_byte(result));
3782}