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/sed-opal.h>
54#include <linux/pm_runtime.h>
55#include <linux/pr.h>
56#include <linux/t10-pi.h>
57#include <linux/uaccess.h>
58#include <asm/unaligned.h>
59
60#include <scsi/scsi.h>
61#include <scsi/scsi_cmnd.h>
62#include <scsi/scsi_dbg.h>
63#include <scsi/scsi_device.h>
64#include <scsi/scsi_driver.h>
65#include <scsi/scsi_eh.h>
66#include <scsi/scsi_host.h>
67#include <scsi/scsi_ioctl.h>
68#include <scsi/scsicam.h>
69
70#include "sd.h"
71#include "scsi_priv.h"
72#include "scsi_logging.h"
73
74MODULE_AUTHOR("Eric Youngdale");
75MODULE_DESCRIPTION("SCSI disk (sd) driver");
76MODULE_LICENSE("GPL");
77
78MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK0_MAJOR);
79MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK1_MAJOR);
80MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK2_MAJOR);
81MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK3_MAJOR);
82MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK4_MAJOR);
83MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK5_MAJOR);
84MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK6_MAJOR);
85MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK7_MAJOR);
86MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK8_MAJOR);
87MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK9_MAJOR);
88MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK10_MAJOR);
89MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK11_MAJOR);
90MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK12_MAJOR);
91MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK13_MAJOR);
92MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK14_MAJOR);
93MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK15_MAJOR);
94MODULE_ALIAS_SCSI_DEVICE(TYPE_DISK);
95MODULE_ALIAS_SCSI_DEVICE(TYPE_MOD);
96MODULE_ALIAS_SCSI_DEVICE(TYPE_RBC);
97MODULE_ALIAS_SCSI_DEVICE(TYPE_ZBC);
98
99#if !defined(CONFIG_DEBUG_BLOCK_EXT_DEVT)
100#define SD_MINORS 16
101#else
102#define SD_MINORS 0
103#endif
104
105static void sd_config_discard(struct scsi_disk *, unsigned int);
106static void sd_config_write_same(struct scsi_disk *);
107static int sd_revalidate_disk(struct gendisk *);
108static void sd_unlock_native_capacity(struct gendisk *disk);
109static int sd_probe(struct device *);
110static int sd_remove(struct device *);
111static void sd_shutdown(struct device *);
112static int sd_suspend_system(struct device *);
113static int sd_suspend_runtime(struct device *);
114static int sd_resume(struct device *);
115static void sd_rescan(struct device *);
116static int sd_init_command(struct scsi_cmnd *SCpnt);
117static void sd_uninit_command(struct scsi_cmnd *SCpnt);
118static int sd_done(struct scsi_cmnd *);
119static void sd_eh_reset(struct scsi_cmnd *);
120static int sd_eh_action(struct scsi_cmnd *, int);
121static void sd_read_capacity(struct scsi_disk *sdkp, unsigned char *buffer);
122static void scsi_disk_release(struct device *cdev);
123static void sd_print_sense_hdr(struct scsi_disk *, struct scsi_sense_hdr *);
124static void sd_print_result(const struct scsi_disk *, const char *, int);
125
126static DEFINE_SPINLOCK(sd_index_lock);
127static DEFINE_IDA(sd_index_ida);
128
129/* This semaphore is used to mediate the 0->1 reference get in the
130 * face of object destruction (i.e. we can't allow a get on an
131 * object after last put) */
132static DEFINE_MUTEX(sd_ref_mutex);
133
134static struct kmem_cache *sd_cdb_cache;
135static mempool_t *sd_cdb_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 if (scsi_mode_select(sdp, 1, sp, 8, buffer_data, len, SD_TIMEOUT,
209 SD_MAX_RETRIES, &data, &sshdr)) {
210 if (scsi_sense_valid(&sshdr))
211 sd_print_sense_hdr(sdkp, &sshdr);
212 return -EINVAL;
213 }
214 revalidate_disk(sdkp->disk);
215 return count;
216}
217
218static ssize_t
219manage_start_stop_show(struct device *dev, struct device_attribute *attr,
220 char *buf)
221{
222 struct scsi_disk *sdkp = to_scsi_disk(dev);
223 struct scsi_device *sdp = sdkp->device;
224
225 return sprintf(buf, "%u\n", sdp->manage_start_stop);
226}
227
228static ssize_t
229manage_start_stop_store(struct device *dev, struct device_attribute *attr,
230 const char *buf, size_t count)
231{
232 struct scsi_disk *sdkp = to_scsi_disk(dev);
233 struct scsi_device *sdp = sdkp->device;
234 bool v;
235
236 if (!capable(CAP_SYS_ADMIN))
237 return -EACCES;
238
239 if (kstrtobool(buf, &v))
240 return -EINVAL;
241
242 sdp->manage_start_stop = v;
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 sprintf(buf, "%u\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 bool v;
261 struct scsi_disk *sdkp = to_scsi_disk(dev);
262 struct scsi_device *sdp = sdkp->device;
263
264 if (!capable(CAP_SYS_ADMIN))
265 return -EACCES;
266
267 if (sdp->type != TYPE_DISK && sdp->type != TYPE_ZBC)
268 return -EINVAL;
269
270 if (kstrtobool(buf, &v))
271 return -EINVAL;
272
273 sdp->allow_restart = v;
274
275 return count;
276}
277static DEVICE_ATTR_RW(allow_restart);
278
279static ssize_t
280cache_type_show(struct device *dev, struct device_attribute *attr, char *buf)
281{
282 struct scsi_disk *sdkp = to_scsi_disk(dev);
283 int ct = sdkp->RCD + 2*sdkp->WCE;
284
285 return sprintf(buf, "%s\n", sd_cache_types[ct]);
286}
287static DEVICE_ATTR_RW(cache_type);
288
289static ssize_t
290FUA_show(struct device *dev, struct device_attribute *attr, char *buf)
291{
292 struct scsi_disk *sdkp = to_scsi_disk(dev);
293
294 return sprintf(buf, "%u\n", sdkp->DPOFUA);
295}
296static DEVICE_ATTR_RO(FUA);
297
298static ssize_t
299protection_type_show(struct device *dev, struct device_attribute *attr,
300 char *buf)
301{
302 struct scsi_disk *sdkp = to_scsi_disk(dev);
303
304 return sprintf(buf, "%u\n", sdkp->protection_type);
305}
306
307static ssize_t
308protection_type_store(struct device *dev, struct device_attribute *attr,
309 const char *buf, size_t count)
310{
311 struct scsi_disk *sdkp = to_scsi_disk(dev);
312 unsigned int val;
313 int err;
314
315 if (!capable(CAP_SYS_ADMIN))
316 return -EACCES;
317
318 err = kstrtouint(buf, 10, &val);
319
320 if (err)
321 return err;
322
323 if (val <= T10_PI_TYPE3_PROTECTION)
324 sdkp->protection_type = val;
325
326 return count;
327}
328static DEVICE_ATTR_RW(protection_type);
329
330static ssize_t
331protection_mode_show(struct device *dev, struct device_attribute *attr,
332 char *buf)
333{
334 struct scsi_disk *sdkp = to_scsi_disk(dev);
335 struct scsi_device *sdp = sdkp->device;
336 unsigned int dif, dix;
337
338 dif = scsi_host_dif_capable(sdp->host, sdkp->protection_type);
339 dix = scsi_host_dix_capable(sdp->host, sdkp->protection_type);
340
341 if (!dix && scsi_host_dix_capable(sdp->host, T10_PI_TYPE0_PROTECTION)) {
342 dif = 0;
343 dix = 1;
344 }
345
346 if (!dif && !dix)
347 return sprintf(buf, "none\n");
348
349 return sprintf(buf, "%s%u\n", dix ? "dix" : "dif", dif);
350}
351static DEVICE_ATTR_RO(protection_mode);
352
353static ssize_t
354app_tag_own_show(struct device *dev, struct device_attribute *attr, char *buf)
355{
356 struct scsi_disk *sdkp = to_scsi_disk(dev);
357
358 return sprintf(buf, "%u\n", sdkp->ATO);
359}
360static DEVICE_ATTR_RO(app_tag_own);
361
362static ssize_t
363thin_provisioning_show(struct device *dev, struct device_attribute *attr,
364 char *buf)
365{
366 struct scsi_disk *sdkp = to_scsi_disk(dev);
367
368 return sprintf(buf, "%u\n", sdkp->lbpme);
369}
370static DEVICE_ATTR_RO(thin_provisioning);
371
372/* sysfs_match_string() requires dense arrays */
373static const char *lbp_mode[] = {
374 [SD_LBP_FULL] = "full",
375 [SD_LBP_UNMAP] = "unmap",
376 [SD_LBP_WS16] = "writesame_16",
377 [SD_LBP_WS10] = "writesame_10",
378 [SD_LBP_ZERO] = "writesame_zero",
379 [SD_LBP_DISABLE] = "disabled",
380};
381
382static ssize_t
383provisioning_mode_show(struct device *dev, struct device_attribute *attr,
384 char *buf)
385{
386 struct scsi_disk *sdkp = to_scsi_disk(dev);
387
388 return sprintf(buf, "%s\n", lbp_mode[sdkp->provisioning_mode]);
389}
390
391static ssize_t
392provisioning_mode_store(struct device *dev, struct device_attribute *attr,
393 const char *buf, size_t count)
394{
395 struct scsi_disk *sdkp = to_scsi_disk(dev);
396 struct scsi_device *sdp = sdkp->device;
397 int mode;
398
399 if (!capable(CAP_SYS_ADMIN))
400 return -EACCES;
401
402 if (sd_is_zoned(sdkp)) {
403 sd_config_discard(sdkp, SD_LBP_DISABLE);
404 return count;
405 }
406
407 if (sdp->type != TYPE_DISK)
408 return -EINVAL;
409
410 mode = sysfs_match_string(lbp_mode, buf);
411 if (mode < 0)
412 return -EINVAL;
413
414 sd_config_discard(sdkp, mode);
415
416 return count;
417}
418static DEVICE_ATTR_RW(provisioning_mode);
419
420/* sysfs_match_string() requires dense arrays */
421static const char *zeroing_mode[] = {
422 [SD_ZERO_WRITE] = "write",
423 [SD_ZERO_WS] = "writesame",
424 [SD_ZERO_WS16_UNMAP] = "writesame_16_unmap",
425 [SD_ZERO_WS10_UNMAP] = "writesame_10_unmap",
426};
427
428static ssize_t
429zeroing_mode_show(struct device *dev, struct device_attribute *attr,
430 char *buf)
431{
432 struct scsi_disk *sdkp = to_scsi_disk(dev);
433
434 return sprintf(buf, "%s\n", zeroing_mode[sdkp->zeroing_mode]);
435}
436
437static ssize_t
438zeroing_mode_store(struct device *dev, struct device_attribute *attr,
439 const char *buf, size_t count)
440{
441 struct scsi_disk *sdkp = to_scsi_disk(dev);
442 int mode;
443
444 if (!capable(CAP_SYS_ADMIN))
445 return -EACCES;
446
447 mode = sysfs_match_string(zeroing_mode, buf);
448 if (mode < 0)
449 return -EINVAL;
450
451 sdkp->zeroing_mode = mode;
452
453 return count;
454}
455static DEVICE_ATTR_RW(zeroing_mode);
456
457static ssize_t
458max_medium_access_timeouts_show(struct device *dev,
459 struct device_attribute *attr, char *buf)
460{
461 struct scsi_disk *sdkp = to_scsi_disk(dev);
462
463 return sprintf(buf, "%u\n", sdkp->max_medium_access_timeouts);
464}
465
466static ssize_t
467max_medium_access_timeouts_store(struct device *dev,
468 struct device_attribute *attr, const char *buf,
469 size_t count)
470{
471 struct scsi_disk *sdkp = to_scsi_disk(dev);
472 int err;
473
474 if (!capable(CAP_SYS_ADMIN))
475 return -EACCES;
476
477 err = kstrtouint(buf, 10, &sdkp->max_medium_access_timeouts);
478
479 return err ? err : count;
480}
481static DEVICE_ATTR_RW(max_medium_access_timeouts);
482
483static ssize_t
484max_write_same_blocks_show(struct device *dev, struct device_attribute *attr,
485 char *buf)
486{
487 struct scsi_disk *sdkp = to_scsi_disk(dev);
488
489 return sprintf(buf, "%u\n", sdkp->max_ws_blocks);
490}
491
492static ssize_t
493max_write_same_blocks_store(struct device *dev, struct device_attribute *attr,
494 const char *buf, size_t count)
495{
496 struct scsi_disk *sdkp = to_scsi_disk(dev);
497 struct scsi_device *sdp = sdkp->device;
498 unsigned long max;
499 int err;
500
501 if (!capable(CAP_SYS_ADMIN))
502 return -EACCES;
503
504 if (sdp->type != TYPE_DISK && sdp->type != TYPE_ZBC)
505 return -EINVAL;
506
507 err = kstrtoul(buf, 10, &max);
508
509 if (err)
510 return err;
511
512 if (max == 0)
513 sdp->no_write_same = 1;
514 else if (max <= SD_MAX_WS16_BLOCKS) {
515 sdp->no_write_same = 0;
516 sdkp->max_ws_blocks = max;
517 }
518
519 sd_config_write_same(sdkp);
520
521 return count;
522}
523static DEVICE_ATTR_RW(max_write_same_blocks);
524
525static struct attribute *sd_disk_attrs[] = {
526 &dev_attr_cache_type.attr,
527 &dev_attr_FUA.attr,
528 &dev_attr_allow_restart.attr,
529 &dev_attr_manage_start_stop.attr,
530 &dev_attr_protection_type.attr,
531 &dev_attr_protection_mode.attr,
532 &dev_attr_app_tag_own.attr,
533 &dev_attr_thin_provisioning.attr,
534 &dev_attr_provisioning_mode.attr,
535 &dev_attr_zeroing_mode.attr,
536 &dev_attr_max_write_same_blocks.attr,
537 &dev_attr_max_medium_access_timeouts.attr,
538 NULL,
539};
540ATTRIBUTE_GROUPS(sd_disk);
541
542static struct class sd_disk_class = {
543 .name = "scsi_disk",
544 .owner = THIS_MODULE,
545 .dev_release = scsi_disk_release,
546 .dev_groups = sd_disk_groups,
547};
548
549static const struct dev_pm_ops sd_pm_ops = {
550 .suspend = sd_suspend_system,
551 .resume = sd_resume,
552 .poweroff = sd_suspend_system,
553 .restore = sd_resume,
554 .runtime_suspend = sd_suspend_runtime,
555 .runtime_resume = sd_resume,
556};
557
558static struct scsi_driver sd_template = {
559 .gendrv = {
560 .name = "sd",
561 .owner = THIS_MODULE,
562 .probe = sd_probe,
563 .remove = sd_remove,
564 .shutdown = sd_shutdown,
565 .pm = &sd_pm_ops,
566 },
567 .rescan = sd_rescan,
568 .init_command = sd_init_command,
569 .uninit_command = sd_uninit_command,
570 .done = sd_done,
571 .eh_action = sd_eh_action,
572 .eh_reset = sd_eh_reset,
573};
574
575/*
576 * Dummy kobj_map->probe function.
577 * The default ->probe function will call modprobe, which is
578 * pointless as this module is already loaded.
579 */
580static struct kobject *sd_default_probe(dev_t devt, int *partno, void *data)
581{
582 return NULL;
583}
584
585/*
586 * Device no to disk mapping:
587 *
588 * major disc2 disc p1
589 * |............|.............|....|....| <- dev_t
590 * 31 20 19 8 7 4 3 0
591 *
592 * Inside a major, we have 16k disks, however mapped non-
593 * contiguously. The first 16 disks are for major0, the next
594 * ones with major1, ... Disk 256 is for major0 again, disk 272
595 * for major1, ...
596 * As we stay compatible with our numbering scheme, we can reuse
597 * the well-know SCSI majors 8, 65--71, 136--143.
598 */
599static int sd_major(int major_idx)
600{
601 switch (major_idx) {
602 case 0:
603 return SCSI_DISK0_MAJOR;
604 case 1 ... 7:
605 return SCSI_DISK1_MAJOR + major_idx - 1;
606 case 8 ... 15:
607 return SCSI_DISK8_MAJOR + major_idx - 8;
608 default:
609 BUG();
610 return 0; /* shut up gcc */
611 }
612}
613
614static struct scsi_disk *scsi_disk_get(struct gendisk *disk)
615{
616 struct scsi_disk *sdkp = NULL;
617
618 mutex_lock(&sd_ref_mutex);
619
620 if (disk->private_data) {
621 sdkp = scsi_disk(disk);
622 if (scsi_device_get(sdkp->device) == 0)
623 get_device(&sdkp->dev);
624 else
625 sdkp = NULL;
626 }
627 mutex_unlock(&sd_ref_mutex);
628 return sdkp;
629}
630
631static void scsi_disk_put(struct scsi_disk *sdkp)
632{
633 struct scsi_device *sdev = sdkp->device;
634
635 mutex_lock(&sd_ref_mutex);
636 put_device(&sdkp->dev);
637 scsi_device_put(sdev);
638 mutex_unlock(&sd_ref_mutex);
639}
640
641#ifdef CONFIG_BLK_SED_OPAL
642static int sd_sec_submit(void *data, u16 spsp, u8 secp, void *buffer,
643 size_t len, bool send)
644{
645 struct scsi_device *sdev = data;
646 u8 cdb[12] = { 0, };
647 int ret;
648
649 cdb[0] = send ? SECURITY_PROTOCOL_OUT : SECURITY_PROTOCOL_IN;
650 cdb[1] = secp;
651 put_unaligned_be16(spsp, &cdb[2]);
652 put_unaligned_be32(len, &cdb[6]);
653
654 ret = scsi_execute_req(sdev, cdb,
655 send ? DMA_TO_DEVICE : DMA_FROM_DEVICE,
656 buffer, len, NULL, SD_TIMEOUT, SD_MAX_RETRIES, NULL);
657 return ret <= 0 ? ret : -EIO;
658}
659#endif /* CONFIG_BLK_SED_OPAL */
660
661static unsigned char sd_setup_protect_cmnd(struct scsi_cmnd *scmd,
662 unsigned int dix, unsigned int dif)
663{
664 struct bio *bio = scmd->request->bio;
665 unsigned int prot_op = sd_prot_op(rq_data_dir(scmd->request), dix, dif);
666 unsigned int protect = 0;
667
668 if (dix) { /* DIX Type 0, 1, 2, 3 */
669 if (bio_integrity_flagged(bio, BIP_IP_CHECKSUM))
670 scmd->prot_flags |= SCSI_PROT_IP_CHECKSUM;
671
672 if (bio_integrity_flagged(bio, BIP_CTRL_NOCHECK) == false)
673 scmd->prot_flags |= SCSI_PROT_GUARD_CHECK;
674 }
675
676 if (dif != T10_PI_TYPE3_PROTECTION) { /* DIX/DIF Type 0, 1, 2 */
677 scmd->prot_flags |= SCSI_PROT_REF_INCREMENT;
678
679 if (bio_integrity_flagged(bio, BIP_CTRL_NOCHECK) == false)
680 scmd->prot_flags |= SCSI_PROT_REF_CHECK;
681 }
682
683 if (dif) { /* DIX/DIF Type 1, 2, 3 */
684 scmd->prot_flags |= SCSI_PROT_TRANSFER_PI;
685
686 if (bio_integrity_flagged(bio, BIP_DISK_NOCHECK))
687 protect = 3 << 5; /* Disable target PI checking */
688 else
689 protect = 1 << 5; /* Enable target PI checking */
690 }
691
692 scsi_set_prot_op(scmd, prot_op);
693 scsi_set_prot_type(scmd, dif);
694 scmd->prot_flags &= sd_prot_flag_mask(prot_op);
695
696 return protect;
697}
698
699static void sd_config_discard(struct scsi_disk *sdkp, unsigned int mode)
700{
701 struct request_queue *q = sdkp->disk->queue;
702 unsigned int logical_block_size = sdkp->device->sector_size;
703 unsigned int max_blocks = 0;
704
705 q->limits.discard_alignment =
706 sdkp->unmap_alignment * logical_block_size;
707 q->limits.discard_granularity =
708 max(sdkp->physical_block_size,
709 sdkp->unmap_granularity * logical_block_size);
710 sdkp->provisioning_mode = mode;
711
712 switch (mode) {
713
714 case SD_LBP_FULL:
715 case SD_LBP_DISABLE:
716 blk_queue_max_discard_sectors(q, 0);
717 blk_queue_flag_clear(QUEUE_FLAG_DISCARD, q);
718 return;
719
720 case SD_LBP_UNMAP:
721 max_blocks = min_not_zero(sdkp->max_unmap_blocks,
722 (u32)SD_MAX_WS16_BLOCKS);
723 break;
724
725 case SD_LBP_WS16:
726 if (sdkp->device->unmap_limit_for_ws)
727 max_blocks = sdkp->max_unmap_blocks;
728 else
729 max_blocks = sdkp->max_ws_blocks;
730
731 max_blocks = min_not_zero(max_blocks, (u32)SD_MAX_WS16_BLOCKS);
732 break;
733
734 case SD_LBP_WS10:
735 if (sdkp->device->unmap_limit_for_ws)
736 max_blocks = sdkp->max_unmap_blocks;
737 else
738 max_blocks = sdkp->max_ws_blocks;
739
740 max_blocks = min_not_zero(max_blocks, (u32)SD_MAX_WS10_BLOCKS);
741 break;
742
743 case SD_LBP_ZERO:
744 max_blocks = min_not_zero(sdkp->max_ws_blocks,
745 (u32)SD_MAX_WS10_BLOCKS);
746 break;
747 }
748
749 blk_queue_max_discard_sectors(q, max_blocks * (logical_block_size >> 9));
750 blk_queue_flag_set(QUEUE_FLAG_DISCARD, q);
751}
752
753static int sd_setup_unmap_cmnd(struct scsi_cmnd *cmd)
754{
755 struct scsi_device *sdp = cmd->device;
756 struct request *rq = cmd->request;
757 u64 sector = blk_rq_pos(rq) >> (ilog2(sdp->sector_size) - 9);
758 u32 nr_sectors = blk_rq_sectors(rq) >> (ilog2(sdp->sector_size) - 9);
759 unsigned int data_len = 24;
760 char *buf;
761
762 rq->special_vec.bv_page = alloc_page(GFP_ATOMIC | __GFP_ZERO);
763 if (!rq->special_vec.bv_page)
764 return BLKPREP_DEFER;
765 rq->special_vec.bv_offset = 0;
766 rq->special_vec.bv_len = data_len;
767 rq->rq_flags |= RQF_SPECIAL_PAYLOAD;
768
769 cmd->cmd_len = 10;
770 cmd->cmnd[0] = UNMAP;
771 cmd->cmnd[8] = 24;
772
773 buf = page_address(rq->special_vec.bv_page);
774 put_unaligned_be16(6 + 16, &buf[0]);
775 put_unaligned_be16(16, &buf[2]);
776 put_unaligned_be64(sector, &buf[8]);
777 put_unaligned_be32(nr_sectors, &buf[16]);
778
779 cmd->allowed = SD_MAX_RETRIES;
780 cmd->transfersize = data_len;
781 rq->timeout = SD_TIMEOUT;
782 scsi_req(rq)->resid_len = data_len;
783
784 return scsi_init_io(cmd);
785}
786
787static int sd_setup_write_same16_cmnd(struct scsi_cmnd *cmd, bool unmap)
788{
789 struct scsi_device *sdp = cmd->device;
790 struct request *rq = cmd->request;
791 u64 sector = blk_rq_pos(rq) >> (ilog2(sdp->sector_size) - 9);
792 u32 nr_sectors = blk_rq_sectors(rq) >> (ilog2(sdp->sector_size) - 9);
793 u32 data_len = sdp->sector_size;
794
795 rq->special_vec.bv_page = alloc_page(GFP_ATOMIC | __GFP_ZERO);
796 if (!rq->special_vec.bv_page)
797 return BLKPREP_DEFER;
798 rq->special_vec.bv_offset = 0;
799 rq->special_vec.bv_len = data_len;
800 rq->rq_flags |= RQF_SPECIAL_PAYLOAD;
801
802 cmd->cmd_len = 16;
803 cmd->cmnd[0] = WRITE_SAME_16;
804 if (unmap)
805 cmd->cmnd[1] = 0x8; /* UNMAP */
806 put_unaligned_be64(sector, &cmd->cmnd[2]);
807 put_unaligned_be32(nr_sectors, &cmd->cmnd[10]);
808
809 cmd->allowed = SD_MAX_RETRIES;
810 cmd->transfersize = data_len;
811 rq->timeout = unmap ? SD_TIMEOUT : SD_WRITE_SAME_TIMEOUT;
812 scsi_req(rq)->resid_len = data_len;
813
814 return scsi_init_io(cmd);
815}
816
817static int sd_setup_write_same10_cmnd(struct scsi_cmnd *cmd, bool unmap)
818{
819 struct scsi_device *sdp = cmd->device;
820 struct request *rq = cmd->request;
821 u64 sector = blk_rq_pos(rq) >> (ilog2(sdp->sector_size) - 9);
822 u32 nr_sectors = blk_rq_sectors(rq) >> (ilog2(sdp->sector_size) - 9);
823 u32 data_len = sdp->sector_size;
824
825 rq->special_vec.bv_page = alloc_page(GFP_ATOMIC | __GFP_ZERO);
826 if (!rq->special_vec.bv_page)
827 return BLKPREP_DEFER;
828 rq->special_vec.bv_offset = 0;
829 rq->special_vec.bv_len = data_len;
830 rq->rq_flags |= RQF_SPECIAL_PAYLOAD;
831
832 cmd->cmd_len = 10;
833 cmd->cmnd[0] = WRITE_SAME;
834 if (unmap)
835 cmd->cmnd[1] = 0x8; /* UNMAP */
836 put_unaligned_be32(sector, &cmd->cmnd[2]);
837 put_unaligned_be16(nr_sectors, &cmd->cmnd[7]);
838
839 cmd->allowed = SD_MAX_RETRIES;
840 cmd->transfersize = data_len;
841 rq->timeout = unmap ? SD_TIMEOUT : SD_WRITE_SAME_TIMEOUT;
842 scsi_req(rq)->resid_len = data_len;
843
844 return scsi_init_io(cmd);
845}
846
847static int sd_setup_write_zeroes_cmnd(struct scsi_cmnd *cmd)
848{
849 struct request *rq = cmd->request;
850 struct scsi_device *sdp = cmd->device;
851 struct scsi_disk *sdkp = scsi_disk(rq->rq_disk);
852 u64 sector = blk_rq_pos(rq) >> (ilog2(sdp->sector_size) - 9);
853 u32 nr_sectors = blk_rq_sectors(rq) >> (ilog2(sdp->sector_size) - 9);
854
855 if (!(rq->cmd_flags & REQ_NOUNMAP)) {
856 switch (sdkp->zeroing_mode) {
857 case SD_ZERO_WS16_UNMAP:
858 return sd_setup_write_same16_cmnd(cmd, true);
859 case SD_ZERO_WS10_UNMAP:
860 return sd_setup_write_same10_cmnd(cmd, true);
861 }
862 }
863
864 if (sdp->no_write_same)
865 return BLKPREP_INVALID;
866
867 if (sdkp->ws16 || sector > 0xffffffff || nr_sectors > 0xffff)
868 return sd_setup_write_same16_cmnd(cmd, false);
869
870 return sd_setup_write_same10_cmnd(cmd, false);
871}
872
873static void sd_config_write_same(struct scsi_disk *sdkp)
874{
875 struct request_queue *q = sdkp->disk->queue;
876 unsigned int logical_block_size = sdkp->device->sector_size;
877
878 if (sdkp->device->no_write_same) {
879 sdkp->max_ws_blocks = 0;
880 goto out;
881 }
882
883 /* Some devices can not handle block counts above 0xffff despite
884 * supporting WRITE SAME(16). Consequently we default to 64k
885 * blocks per I/O unless the device explicitly advertises a
886 * bigger limit.
887 */
888 if (sdkp->max_ws_blocks > SD_MAX_WS10_BLOCKS)
889 sdkp->max_ws_blocks = min_not_zero(sdkp->max_ws_blocks,
890 (u32)SD_MAX_WS16_BLOCKS);
891 else if (sdkp->ws16 || sdkp->ws10 || sdkp->device->no_report_opcodes)
892 sdkp->max_ws_blocks = min_not_zero(sdkp->max_ws_blocks,
893 (u32)SD_MAX_WS10_BLOCKS);
894 else {
895 sdkp->device->no_write_same = 1;
896 sdkp->max_ws_blocks = 0;
897 }
898
899 if (sdkp->lbprz && sdkp->lbpws)
900 sdkp->zeroing_mode = SD_ZERO_WS16_UNMAP;
901 else if (sdkp->lbprz && sdkp->lbpws10)
902 sdkp->zeroing_mode = SD_ZERO_WS10_UNMAP;
903 else if (sdkp->max_ws_blocks)
904 sdkp->zeroing_mode = SD_ZERO_WS;
905 else
906 sdkp->zeroing_mode = SD_ZERO_WRITE;
907
908 if (sdkp->max_ws_blocks &&
909 sdkp->physical_block_size > logical_block_size) {
910 /*
911 * Reporting a maximum number of blocks that is not aligned
912 * on the device physical size would cause a large write same
913 * request to be split into physically unaligned chunks by
914 * __blkdev_issue_write_zeroes() and __blkdev_issue_write_same()
915 * even if the caller of these functions took care to align the
916 * large request. So make sure the maximum reported is aligned
917 * to the device physical block size. This is only an optional
918 * optimization for regular disks, but this is mandatory to
919 * avoid failure of large write same requests directed at
920 * sequential write required zones of host-managed ZBC disks.
921 */
922 sdkp->max_ws_blocks =
923 round_down(sdkp->max_ws_blocks,
924 bytes_to_logical(sdkp->device,
925 sdkp->physical_block_size));
926 }
927
928out:
929 blk_queue_max_write_same_sectors(q, sdkp->max_ws_blocks *
930 (logical_block_size >> 9));
931 blk_queue_max_write_zeroes_sectors(q, sdkp->max_ws_blocks *
932 (logical_block_size >> 9));
933}
934
935/**
936 * sd_setup_write_same_cmnd - write the same data to multiple blocks
937 * @cmd: command to prepare
938 *
939 * Will set up either WRITE SAME(10) or WRITE SAME(16) depending on
940 * the preference indicated by the target device.
941 **/
942static int sd_setup_write_same_cmnd(struct scsi_cmnd *cmd)
943{
944 struct request *rq = cmd->request;
945 struct scsi_device *sdp = cmd->device;
946 struct scsi_disk *sdkp = scsi_disk(rq->rq_disk);
947 struct bio *bio = rq->bio;
948 sector_t sector = blk_rq_pos(rq);
949 unsigned int nr_sectors = blk_rq_sectors(rq);
950 unsigned int nr_bytes = blk_rq_bytes(rq);
951 int ret;
952
953 if (sdkp->device->no_write_same)
954 return BLKPREP_INVALID;
955
956 BUG_ON(bio_offset(bio) || bio_iovec(bio).bv_len != sdp->sector_size);
957
958 sector >>= ilog2(sdp->sector_size) - 9;
959 nr_sectors >>= ilog2(sdp->sector_size) - 9;
960
961 rq->timeout = SD_WRITE_SAME_TIMEOUT;
962
963 if (sdkp->ws16 || sector > 0xffffffff || nr_sectors > 0xffff) {
964 cmd->cmd_len = 16;
965 cmd->cmnd[0] = WRITE_SAME_16;
966 put_unaligned_be64(sector, &cmd->cmnd[2]);
967 put_unaligned_be32(nr_sectors, &cmd->cmnd[10]);
968 } else {
969 cmd->cmd_len = 10;
970 cmd->cmnd[0] = WRITE_SAME;
971 put_unaligned_be32(sector, &cmd->cmnd[2]);
972 put_unaligned_be16(nr_sectors, &cmd->cmnd[7]);
973 }
974
975 cmd->transfersize = sdp->sector_size;
976 cmd->allowed = SD_MAX_RETRIES;
977
978 /*
979 * For WRITE SAME the data transferred via the DATA OUT buffer is
980 * different from the amount of data actually written to the target.
981 *
982 * We set up __data_len to the amount of data transferred via the
983 * DATA OUT buffer so that blk_rq_map_sg sets up the proper S/G list
984 * to transfer a single sector of data first, but then reset it to
985 * the amount of data to be written right after so that the I/O path
986 * knows how much to actually write.
987 */
988 rq->__data_len = sdp->sector_size;
989 ret = scsi_init_io(cmd);
990 rq->__data_len = nr_bytes;
991
992 return ret;
993}
994
995static int sd_setup_flush_cmnd(struct scsi_cmnd *cmd)
996{
997 struct request *rq = cmd->request;
998
999 /* flush requests don't perform I/O, zero the S/G table */
1000 memset(&cmd->sdb, 0, sizeof(cmd->sdb));
1001
1002 cmd->cmnd[0] = SYNCHRONIZE_CACHE;
1003 cmd->cmd_len = 10;
1004 cmd->transfersize = 0;
1005 cmd->allowed = SD_MAX_RETRIES;
1006
1007 rq->timeout = rq->q->rq_timeout * SD_FLUSH_TIMEOUT_MULTIPLIER;
1008 return BLKPREP_OK;
1009}
1010
1011static int sd_setup_read_write_cmnd(struct scsi_cmnd *SCpnt)
1012{
1013 struct request *rq = SCpnt->request;
1014 struct scsi_device *sdp = SCpnt->device;
1015 struct gendisk *disk = rq->rq_disk;
1016 struct scsi_disk *sdkp = scsi_disk(disk);
1017 sector_t block = blk_rq_pos(rq);
1018 sector_t threshold;
1019 unsigned int this_count = blk_rq_sectors(rq);
1020 unsigned int dif, dix;
1021 int ret;
1022 unsigned char protect;
1023
1024 ret = scsi_init_io(SCpnt);
1025 if (ret != BLKPREP_OK)
1026 return ret;
1027 WARN_ON_ONCE(SCpnt != rq->special);
1028
1029 /* from here on until we're complete, any goto out
1030 * is used for a killable error condition */
1031 ret = BLKPREP_KILL;
1032
1033 SCSI_LOG_HLQUEUE(1,
1034 scmd_printk(KERN_INFO, SCpnt,
1035 "%s: block=%llu, count=%d\n",
1036 __func__, (unsigned long long)block, this_count));
1037
1038 if (!sdp || !scsi_device_online(sdp) ||
1039 block + blk_rq_sectors(rq) > get_capacity(disk)) {
1040 SCSI_LOG_HLQUEUE(2, scmd_printk(KERN_INFO, SCpnt,
1041 "Finishing %u sectors\n",
1042 blk_rq_sectors(rq)));
1043 SCSI_LOG_HLQUEUE(2, scmd_printk(KERN_INFO, SCpnt,
1044 "Retry with 0x%p\n", SCpnt));
1045 goto out;
1046 }
1047
1048 if (sdp->changed) {
1049 /*
1050 * quietly refuse to do anything to a changed disc until
1051 * the changed bit has been reset
1052 */
1053 /* printk("SCSI disk has been changed or is not present. Prohibiting further I/O.\n"); */
1054 goto out;
1055 }
1056
1057 /*
1058 * Some SD card readers can't handle multi-sector accesses which touch
1059 * the last one or two hardware sectors. Split accesses as needed.
1060 */
1061 threshold = get_capacity(disk) - SD_LAST_BUGGY_SECTORS *
1062 (sdp->sector_size / 512);
1063
1064 if (unlikely(sdp->last_sector_bug && block + this_count > threshold)) {
1065 if (block < threshold) {
1066 /* Access up to the threshold but not beyond */
1067 this_count = threshold - block;
1068 } else {
1069 /* Access only a single hardware sector */
1070 this_count = sdp->sector_size / 512;
1071 }
1072 }
1073
1074 SCSI_LOG_HLQUEUE(2, scmd_printk(KERN_INFO, SCpnt, "block=%llu\n",
1075 (unsigned long long)block));
1076
1077 /*
1078 * If we have a 1K hardware sectorsize, prevent access to single
1079 * 512 byte sectors. In theory we could handle this - in fact
1080 * the scsi cdrom driver must be able to handle this because
1081 * we typically use 1K blocksizes, and cdroms typically have
1082 * 2K hardware sectorsizes. Of course, things are simpler
1083 * with the cdrom, since it is read-only. For performance
1084 * reasons, the filesystems should be able to handle this
1085 * and not force the scsi disk driver to use bounce buffers
1086 * for this.
1087 */
1088 if (sdp->sector_size == 1024) {
1089 if ((block & 1) || (blk_rq_sectors(rq) & 1)) {
1090 scmd_printk(KERN_ERR, SCpnt,
1091 "Bad block number requested\n");
1092 goto out;
1093 } else {
1094 block = block >> 1;
1095 this_count = this_count >> 1;
1096 }
1097 }
1098 if (sdp->sector_size == 2048) {
1099 if ((block & 3) || (blk_rq_sectors(rq) & 3)) {
1100 scmd_printk(KERN_ERR, SCpnt,
1101 "Bad block number requested\n");
1102 goto out;
1103 } else {
1104 block = block >> 2;
1105 this_count = this_count >> 2;
1106 }
1107 }
1108 if (sdp->sector_size == 4096) {
1109 if ((block & 7) || (blk_rq_sectors(rq) & 7)) {
1110 scmd_printk(KERN_ERR, SCpnt,
1111 "Bad block number requested\n");
1112 goto out;
1113 } else {
1114 block = block >> 3;
1115 this_count = this_count >> 3;
1116 }
1117 }
1118 if (rq_data_dir(rq) == WRITE) {
1119 SCpnt->cmnd[0] = WRITE_6;
1120
1121 if (blk_integrity_rq(rq))
1122 sd_dif_prepare(SCpnt);
1123
1124 } else if (rq_data_dir(rq) == READ) {
1125 SCpnt->cmnd[0] = READ_6;
1126 } else {
1127 scmd_printk(KERN_ERR, SCpnt, "Unknown command %d\n", req_op(rq));
1128 goto out;
1129 }
1130
1131 SCSI_LOG_HLQUEUE(2, scmd_printk(KERN_INFO, SCpnt,
1132 "%s %d/%u 512 byte blocks.\n",
1133 (rq_data_dir(rq) == WRITE) ?
1134 "writing" : "reading", this_count,
1135 blk_rq_sectors(rq)));
1136
1137 dix = scsi_prot_sg_count(SCpnt);
1138 dif = scsi_host_dif_capable(SCpnt->device->host, sdkp->protection_type);
1139
1140 if (dif || dix)
1141 protect = sd_setup_protect_cmnd(SCpnt, dix, dif);
1142 else
1143 protect = 0;
1144
1145 if (protect && sdkp->protection_type == T10_PI_TYPE2_PROTECTION) {
1146 SCpnt->cmnd = mempool_alloc(sd_cdb_pool, GFP_ATOMIC);
1147
1148 if (unlikely(SCpnt->cmnd == NULL)) {
1149 ret = BLKPREP_DEFER;
1150 goto out;
1151 }
1152
1153 SCpnt->cmd_len = SD_EXT_CDB_SIZE;
1154 memset(SCpnt->cmnd, 0, SCpnt->cmd_len);
1155 SCpnt->cmnd[0] = VARIABLE_LENGTH_CMD;
1156 SCpnt->cmnd[7] = 0x18;
1157 SCpnt->cmnd[9] = (rq_data_dir(rq) == READ) ? READ_32 : WRITE_32;
1158 SCpnt->cmnd[10] = protect | ((rq->cmd_flags & REQ_FUA) ? 0x8 : 0);
1159
1160 /* LBA */
1161 SCpnt->cmnd[12] = sizeof(block) > 4 ? (unsigned char) (block >> 56) & 0xff : 0;
1162 SCpnt->cmnd[13] = sizeof(block) > 4 ? (unsigned char) (block >> 48) & 0xff : 0;
1163 SCpnt->cmnd[14] = sizeof(block) > 4 ? (unsigned char) (block >> 40) & 0xff : 0;
1164 SCpnt->cmnd[15] = sizeof(block) > 4 ? (unsigned char) (block >> 32) & 0xff : 0;
1165 SCpnt->cmnd[16] = (unsigned char) (block >> 24) & 0xff;
1166 SCpnt->cmnd[17] = (unsigned char) (block >> 16) & 0xff;
1167 SCpnt->cmnd[18] = (unsigned char) (block >> 8) & 0xff;
1168 SCpnt->cmnd[19] = (unsigned char) block & 0xff;
1169
1170 /* Expected Indirect LBA */
1171 SCpnt->cmnd[20] = (unsigned char) (block >> 24) & 0xff;
1172 SCpnt->cmnd[21] = (unsigned char) (block >> 16) & 0xff;
1173 SCpnt->cmnd[22] = (unsigned char) (block >> 8) & 0xff;
1174 SCpnt->cmnd[23] = (unsigned char) block & 0xff;
1175
1176 /* Transfer length */
1177 SCpnt->cmnd[28] = (unsigned char) (this_count >> 24) & 0xff;
1178 SCpnt->cmnd[29] = (unsigned char) (this_count >> 16) & 0xff;
1179 SCpnt->cmnd[30] = (unsigned char) (this_count >> 8) & 0xff;
1180 SCpnt->cmnd[31] = (unsigned char) this_count & 0xff;
1181 } else if (sdp->use_16_for_rw || (this_count > 0xffff)) {
1182 SCpnt->cmnd[0] += READ_16 - READ_6;
1183 SCpnt->cmnd[1] = protect | ((rq->cmd_flags & REQ_FUA) ? 0x8 : 0);
1184 SCpnt->cmnd[2] = sizeof(block) > 4 ? (unsigned char) (block >> 56) & 0xff : 0;
1185 SCpnt->cmnd[3] = sizeof(block) > 4 ? (unsigned char) (block >> 48) & 0xff : 0;
1186 SCpnt->cmnd[4] = sizeof(block) > 4 ? (unsigned char) (block >> 40) & 0xff : 0;
1187 SCpnt->cmnd[5] = sizeof(block) > 4 ? (unsigned char) (block >> 32) & 0xff : 0;
1188 SCpnt->cmnd[6] = (unsigned char) (block >> 24) & 0xff;
1189 SCpnt->cmnd[7] = (unsigned char) (block >> 16) & 0xff;
1190 SCpnt->cmnd[8] = (unsigned char) (block >> 8) & 0xff;
1191 SCpnt->cmnd[9] = (unsigned char) block & 0xff;
1192 SCpnt->cmnd[10] = (unsigned char) (this_count >> 24) & 0xff;
1193 SCpnt->cmnd[11] = (unsigned char) (this_count >> 16) & 0xff;
1194 SCpnt->cmnd[12] = (unsigned char) (this_count >> 8) & 0xff;
1195 SCpnt->cmnd[13] = (unsigned char) this_count & 0xff;
1196 SCpnt->cmnd[14] = SCpnt->cmnd[15] = 0;
1197 } else if ((this_count > 0xff) || (block > 0x1fffff) ||
1198 scsi_device_protection(SCpnt->device) ||
1199 SCpnt->device->use_10_for_rw) {
1200 SCpnt->cmnd[0] += READ_10 - READ_6;
1201 SCpnt->cmnd[1] = protect | ((rq->cmd_flags & REQ_FUA) ? 0x8 : 0);
1202 SCpnt->cmnd[2] = (unsigned char) (block >> 24) & 0xff;
1203 SCpnt->cmnd[3] = (unsigned char) (block >> 16) & 0xff;
1204 SCpnt->cmnd[4] = (unsigned char) (block >> 8) & 0xff;
1205 SCpnt->cmnd[5] = (unsigned char) block & 0xff;
1206 SCpnt->cmnd[6] = SCpnt->cmnd[9] = 0;
1207 SCpnt->cmnd[7] = (unsigned char) (this_count >> 8) & 0xff;
1208 SCpnt->cmnd[8] = (unsigned char) this_count & 0xff;
1209 } else {
1210 if (unlikely(rq->cmd_flags & REQ_FUA)) {
1211 /*
1212 * This happens only if this drive failed
1213 * 10byte rw command with ILLEGAL_REQUEST
1214 * during operation and thus turned off
1215 * use_10_for_rw.
1216 */
1217 scmd_printk(KERN_ERR, SCpnt,
1218 "FUA write on READ/WRITE(6) drive\n");
1219 goto out;
1220 }
1221
1222 SCpnt->cmnd[1] |= (unsigned char) ((block >> 16) & 0x1f);
1223 SCpnt->cmnd[2] = (unsigned char) ((block >> 8) & 0xff);
1224 SCpnt->cmnd[3] = (unsigned char) block & 0xff;
1225 SCpnt->cmnd[4] = (unsigned char) this_count;
1226 SCpnt->cmnd[5] = 0;
1227 }
1228 SCpnt->sdb.length = this_count * sdp->sector_size;
1229
1230 /*
1231 * We shouldn't disconnect in the middle of a sector, so with a dumb
1232 * host adapter, it's safe to assume that we can at least transfer
1233 * this many bytes between each connect / disconnect.
1234 */
1235 SCpnt->transfersize = sdp->sector_size;
1236 SCpnt->underflow = this_count << 9;
1237 SCpnt->allowed = SD_MAX_RETRIES;
1238
1239 /*
1240 * This indicates that the command is ready from our end to be
1241 * queued.
1242 */
1243 ret = BLKPREP_OK;
1244 out:
1245 return ret;
1246}
1247
1248static int sd_init_command(struct scsi_cmnd *cmd)
1249{
1250 struct request *rq = cmd->request;
1251
1252 switch (req_op(rq)) {
1253 case REQ_OP_DISCARD:
1254 switch (scsi_disk(rq->rq_disk)->provisioning_mode) {
1255 case SD_LBP_UNMAP:
1256 return sd_setup_unmap_cmnd(cmd);
1257 case SD_LBP_WS16:
1258 return sd_setup_write_same16_cmnd(cmd, true);
1259 case SD_LBP_WS10:
1260 return sd_setup_write_same10_cmnd(cmd, true);
1261 case SD_LBP_ZERO:
1262 return sd_setup_write_same10_cmnd(cmd, false);
1263 default:
1264 return BLKPREP_INVALID;
1265 }
1266 case REQ_OP_WRITE_ZEROES:
1267 return sd_setup_write_zeroes_cmnd(cmd);
1268 case REQ_OP_WRITE_SAME:
1269 return sd_setup_write_same_cmnd(cmd);
1270 case REQ_OP_FLUSH:
1271 return sd_setup_flush_cmnd(cmd);
1272 case REQ_OP_READ:
1273 case REQ_OP_WRITE:
1274 return sd_setup_read_write_cmnd(cmd);
1275 case REQ_OP_ZONE_REPORT:
1276 return sd_zbc_setup_report_cmnd(cmd);
1277 case REQ_OP_ZONE_RESET:
1278 return sd_zbc_setup_reset_cmnd(cmd);
1279 default:
1280 BUG();
1281 }
1282}
1283
1284static void sd_uninit_command(struct scsi_cmnd *SCpnt)
1285{
1286 struct request *rq = SCpnt->request;
1287 u8 *cmnd;
1288
1289 if (rq->rq_flags & RQF_SPECIAL_PAYLOAD)
1290 __free_page(rq->special_vec.bv_page);
1291
1292 if (SCpnt->cmnd != scsi_req(rq)->cmd) {
1293 cmnd = SCpnt->cmnd;
1294 SCpnt->cmnd = NULL;
1295 SCpnt->cmd_len = 0;
1296 mempool_free(cmnd, sd_cdb_pool);
1297 }
1298}
1299
1300/**
1301 * sd_open - open a scsi disk device
1302 * @bdev: Block device of the scsi disk to open
1303 * @mode: FMODE_* mask
1304 *
1305 * Returns 0 if successful. Returns a negated errno value in case
1306 * of error.
1307 *
1308 * Note: This can be called from a user context (e.g. fsck(1) )
1309 * or from within the kernel (e.g. as a result of a mount(1) ).
1310 * In the latter case @inode and @filp carry an abridged amount
1311 * of information as noted above.
1312 *
1313 * Locking: called with bdev->bd_mutex held.
1314 **/
1315static int sd_open(struct block_device *bdev, fmode_t mode)
1316{
1317 struct scsi_disk *sdkp = scsi_disk_get(bdev->bd_disk);
1318 struct scsi_device *sdev;
1319 int retval;
1320
1321 if (!sdkp)
1322 return -ENXIO;
1323
1324 SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp, "sd_open\n"));
1325
1326 sdev = sdkp->device;
1327
1328 /*
1329 * If the device is in error recovery, wait until it is done.
1330 * If the device is offline, then disallow any access to it.
1331 */
1332 retval = -ENXIO;
1333 if (!scsi_block_when_processing_errors(sdev))
1334 goto error_out;
1335
1336 if (sdev->removable || sdkp->write_prot)
1337 check_disk_change(bdev);
1338
1339 /*
1340 * If the drive is empty, just let the open fail.
1341 */
1342 retval = -ENOMEDIUM;
1343 if (sdev->removable && !sdkp->media_present && !(mode & FMODE_NDELAY))
1344 goto error_out;
1345
1346 /*
1347 * If the device has the write protect tab set, have the open fail
1348 * if the user expects to be able to write to the thing.
1349 */
1350 retval = -EROFS;
1351 if (sdkp->write_prot && (mode & FMODE_WRITE))
1352 goto error_out;
1353
1354 /*
1355 * It is possible that the disk changing stuff resulted in
1356 * the device being taken offline. If this is the case,
1357 * report this to the user, and don't pretend that the
1358 * open actually succeeded.
1359 */
1360 retval = -ENXIO;
1361 if (!scsi_device_online(sdev))
1362 goto error_out;
1363
1364 if ((atomic_inc_return(&sdkp->openers) == 1) && sdev->removable) {
1365 if (scsi_block_when_processing_errors(sdev))
1366 scsi_set_medium_removal(sdev, SCSI_REMOVAL_PREVENT);
1367 }
1368
1369 return 0;
1370
1371error_out:
1372 scsi_disk_put(sdkp);
1373 return retval;
1374}
1375
1376/**
1377 * sd_release - invoked when the (last) close(2) is called on this
1378 * scsi disk.
1379 * @disk: disk to release
1380 * @mode: FMODE_* mask
1381 *
1382 * Returns 0.
1383 *
1384 * Note: may block (uninterruptible) if error recovery is underway
1385 * on this disk.
1386 *
1387 * Locking: called with bdev->bd_mutex held.
1388 **/
1389static void sd_release(struct gendisk *disk, fmode_t mode)
1390{
1391 struct scsi_disk *sdkp = scsi_disk(disk);
1392 struct scsi_device *sdev = sdkp->device;
1393
1394 SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp, "sd_release\n"));
1395
1396 if (atomic_dec_return(&sdkp->openers) == 0 && sdev->removable) {
1397 if (scsi_block_when_processing_errors(sdev))
1398 scsi_set_medium_removal(sdev, SCSI_REMOVAL_ALLOW);
1399 }
1400
1401 /*
1402 * XXX and what if there are packets in flight and this close()
1403 * XXX is followed by a "rmmod sd_mod"?
1404 */
1405
1406 scsi_disk_put(sdkp);
1407}
1408
1409static int sd_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1410{
1411 struct scsi_disk *sdkp = scsi_disk(bdev->bd_disk);
1412 struct scsi_device *sdp = sdkp->device;
1413 struct Scsi_Host *host = sdp->host;
1414 sector_t capacity = logical_to_sectors(sdp, sdkp->capacity);
1415 int diskinfo[4];
1416
1417 /* default to most commonly used values */
1418 diskinfo[0] = 0x40; /* 1 << 6 */
1419 diskinfo[1] = 0x20; /* 1 << 5 */
1420 diskinfo[2] = capacity >> 11;
1421
1422 /* override with calculated, extended default, or driver values */
1423 if (host->hostt->bios_param)
1424 host->hostt->bios_param(sdp, bdev, capacity, diskinfo);
1425 else
1426 scsicam_bios_param(bdev, capacity, diskinfo);
1427
1428 geo->heads = diskinfo[0];
1429 geo->sectors = diskinfo[1];
1430 geo->cylinders = diskinfo[2];
1431 return 0;
1432}
1433
1434/**
1435 * sd_ioctl - process an ioctl
1436 * @bdev: target block device
1437 * @mode: FMODE_* mask
1438 * @cmd: ioctl command number
1439 * @arg: this is third argument given to ioctl(2) system call.
1440 * Often contains a pointer.
1441 *
1442 * Returns 0 if successful (some ioctls return positive numbers on
1443 * success as well). Returns a negated errno value in case of error.
1444 *
1445 * Note: most ioctls are forward onto the block subsystem or further
1446 * down in the scsi subsystem.
1447 **/
1448static int sd_ioctl(struct block_device *bdev, fmode_t mode,
1449 unsigned int cmd, unsigned long arg)
1450{
1451 struct gendisk *disk = bdev->bd_disk;
1452 struct scsi_disk *sdkp = scsi_disk(disk);
1453 struct scsi_device *sdp = sdkp->device;
1454 void __user *p = (void __user *)arg;
1455 int error;
1456
1457 SCSI_LOG_IOCTL(1, sd_printk(KERN_INFO, sdkp, "sd_ioctl: disk=%s, "
1458 "cmd=0x%x\n", disk->disk_name, cmd));
1459
1460 error = scsi_verify_blk_ioctl(bdev, cmd);
1461 if (error < 0)
1462 return error;
1463
1464 /*
1465 * If we are in the middle of error recovery, don't let anyone
1466 * else try and use this device. Also, if error recovery fails, it
1467 * may try and take the device offline, in which case all further
1468 * access to the device is prohibited.
1469 */
1470 error = scsi_ioctl_block_when_processing_errors(sdp, cmd,
1471 (mode & FMODE_NDELAY) != 0);
1472 if (error)
1473 goto out;
1474
1475 if (is_sed_ioctl(cmd))
1476 return sed_ioctl(sdkp->opal_dev, cmd, p);
1477
1478 /*
1479 * Send SCSI addressing ioctls directly to mid level, send other
1480 * ioctls to block level and then onto mid level if they can't be
1481 * resolved.
1482 */
1483 switch (cmd) {
1484 case SCSI_IOCTL_GET_IDLUN:
1485 case SCSI_IOCTL_GET_BUS_NUMBER:
1486 error = scsi_ioctl(sdp, cmd, p);
1487 break;
1488 default:
1489 error = scsi_cmd_blk_ioctl(bdev, mode, cmd, p);
1490 if (error != -ENOTTY)
1491 break;
1492 error = scsi_ioctl(sdp, cmd, p);
1493 break;
1494 }
1495out:
1496 return error;
1497}
1498
1499static void set_media_not_present(struct scsi_disk *sdkp)
1500{
1501 if (sdkp->media_present)
1502 sdkp->device->changed = 1;
1503
1504 if (sdkp->device->removable) {
1505 sdkp->media_present = 0;
1506 sdkp->capacity = 0;
1507 }
1508}
1509
1510static int media_not_present(struct scsi_disk *sdkp,
1511 struct scsi_sense_hdr *sshdr)
1512{
1513 if (!scsi_sense_valid(sshdr))
1514 return 0;
1515
1516 /* not invoked for commands that could return deferred errors */
1517 switch (sshdr->sense_key) {
1518 case UNIT_ATTENTION:
1519 case NOT_READY:
1520 /* medium not present */
1521 if (sshdr->asc == 0x3A) {
1522 set_media_not_present(sdkp);
1523 return 1;
1524 }
1525 }
1526 return 0;
1527}
1528
1529/**
1530 * sd_check_events - check media events
1531 * @disk: kernel device descriptor
1532 * @clearing: disk events currently being cleared
1533 *
1534 * Returns mask of DISK_EVENT_*.
1535 *
1536 * Note: this function is invoked from the block subsystem.
1537 **/
1538static unsigned int sd_check_events(struct gendisk *disk, unsigned int clearing)
1539{
1540 struct scsi_disk *sdkp = scsi_disk_get(disk);
1541 struct scsi_device *sdp;
1542 int retval;
1543
1544 if (!sdkp)
1545 return 0;
1546
1547 sdp = sdkp->device;
1548 SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp, "sd_check_events\n"));
1549
1550 /*
1551 * If the device is offline, don't send any commands - just pretend as
1552 * if the command failed. If the device ever comes back online, we
1553 * can deal with it then. It is only because of unrecoverable errors
1554 * that we would ever take a device offline in the first place.
1555 */
1556 if (!scsi_device_online(sdp)) {
1557 set_media_not_present(sdkp);
1558 goto out;
1559 }
1560
1561 /*
1562 * Using TEST_UNIT_READY enables differentiation between drive with
1563 * no cartridge loaded - NOT READY, drive with changed cartridge -
1564 * UNIT ATTENTION, or with same cartridge - GOOD STATUS.
1565 *
1566 * Drives that auto spin down. eg iomega jaz 1G, will be started
1567 * by sd_spinup_disk() from sd_revalidate_disk(), which happens whenever
1568 * sd_revalidate() is called.
1569 */
1570 if (scsi_block_when_processing_errors(sdp)) {
1571 struct scsi_sense_hdr sshdr = { 0, };
1572
1573 retval = scsi_test_unit_ready(sdp, SD_TIMEOUT, SD_MAX_RETRIES,
1574 &sshdr);
1575
1576 /* failed to execute TUR, assume media not present */
1577 if (host_byte(retval)) {
1578 set_media_not_present(sdkp);
1579 goto out;
1580 }
1581
1582 if (media_not_present(sdkp, &sshdr))
1583 goto out;
1584 }
1585
1586 /*
1587 * For removable scsi disk we have to recognise the presence
1588 * of a disk in the drive.
1589 */
1590 if (!sdkp->media_present)
1591 sdp->changed = 1;
1592 sdkp->media_present = 1;
1593out:
1594 /*
1595 * sdp->changed is set under the following conditions:
1596 *
1597 * Medium present state has changed in either direction.
1598 * Device has indicated UNIT_ATTENTION.
1599 */
1600 retval = sdp->changed ? DISK_EVENT_MEDIA_CHANGE : 0;
1601 sdp->changed = 0;
1602 scsi_disk_put(sdkp);
1603 return retval;
1604}
1605
1606static int sd_sync_cache(struct scsi_disk *sdkp, struct scsi_sense_hdr *sshdr)
1607{
1608 int retries, res;
1609 struct scsi_device *sdp = sdkp->device;
1610 const int timeout = sdp->request_queue->rq_timeout
1611 * SD_FLUSH_TIMEOUT_MULTIPLIER;
1612 struct scsi_sense_hdr my_sshdr;
1613
1614 if (!scsi_device_online(sdp))
1615 return -ENODEV;
1616
1617 /* caller might not be interested in sense, but we need it */
1618 if (!sshdr)
1619 sshdr = &my_sshdr;
1620
1621 for (retries = 3; retries > 0; --retries) {
1622 unsigned char cmd[10] = { 0 };
1623
1624 cmd[0] = SYNCHRONIZE_CACHE;
1625 /*
1626 * Leave the rest of the command zero to indicate
1627 * flush everything.
1628 */
1629 res = scsi_execute(sdp, cmd, DMA_NONE, NULL, 0, NULL, sshdr,
1630 timeout, SD_MAX_RETRIES, 0, RQF_PM, NULL);
1631 if (res == 0)
1632 break;
1633 }
1634
1635 if (res) {
1636 sd_print_result(sdkp, "Synchronize Cache(10) failed", res);
1637
1638 if (driver_byte(res) & DRIVER_SENSE)
1639 sd_print_sense_hdr(sdkp, sshdr);
1640
1641 /* we need to evaluate the error return */
1642 if (scsi_sense_valid(sshdr) &&
1643 (sshdr->asc == 0x3a || /* medium not present */
1644 sshdr->asc == 0x20)) /* invalid command */
1645 /* this is no error here */
1646 return 0;
1647
1648 switch (host_byte(res)) {
1649 /* ignore errors due to racing a disconnection */
1650 case DID_BAD_TARGET:
1651 case DID_NO_CONNECT:
1652 return 0;
1653 /* signal the upper layer it might try again */
1654 case DID_BUS_BUSY:
1655 case DID_IMM_RETRY:
1656 case DID_REQUEUE:
1657 case DID_SOFT_ERROR:
1658 return -EBUSY;
1659 default:
1660 return -EIO;
1661 }
1662 }
1663 return 0;
1664}
1665
1666static void sd_rescan(struct device *dev)
1667{
1668 struct scsi_disk *sdkp = dev_get_drvdata(dev);
1669
1670 revalidate_disk(sdkp->disk);
1671}
1672
1673
1674#ifdef CONFIG_COMPAT
1675/*
1676 * This gets directly called from VFS. When the ioctl
1677 * is not recognized we go back to the other translation paths.
1678 */
1679static int sd_compat_ioctl(struct block_device *bdev, fmode_t mode,
1680 unsigned int cmd, unsigned long arg)
1681{
1682 struct scsi_device *sdev = scsi_disk(bdev->bd_disk)->device;
1683 int error;
1684
1685 error = scsi_ioctl_block_when_processing_errors(sdev, cmd,
1686 (mode & FMODE_NDELAY) != 0);
1687 if (error)
1688 return error;
1689
1690 /*
1691 * Let the static ioctl translation table take care of it.
1692 */
1693 if (!sdev->host->hostt->compat_ioctl)
1694 return -ENOIOCTLCMD;
1695 return sdev->host->hostt->compat_ioctl(sdev, cmd, (void __user *)arg);
1696}
1697#endif
1698
1699static char sd_pr_type(enum pr_type type)
1700{
1701 switch (type) {
1702 case PR_WRITE_EXCLUSIVE:
1703 return 0x01;
1704 case PR_EXCLUSIVE_ACCESS:
1705 return 0x03;
1706 case PR_WRITE_EXCLUSIVE_REG_ONLY:
1707 return 0x05;
1708 case PR_EXCLUSIVE_ACCESS_REG_ONLY:
1709 return 0x06;
1710 case PR_WRITE_EXCLUSIVE_ALL_REGS:
1711 return 0x07;
1712 case PR_EXCLUSIVE_ACCESS_ALL_REGS:
1713 return 0x08;
1714 default:
1715 return 0;
1716 }
1717};
1718
1719static int sd_pr_command(struct block_device *bdev, u8 sa,
1720 u64 key, u64 sa_key, u8 type, u8 flags)
1721{
1722 struct scsi_device *sdev = scsi_disk(bdev->bd_disk)->device;
1723 struct scsi_sense_hdr sshdr;
1724 int result;
1725 u8 cmd[16] = { 0, };
1726 u8 data[24] = { 0, };
1727
1728 cmd[0] = PERSISTENT_RESERVE_OUT;
1729 cmd[1] = sa;
1730 cmd[2] = type;
1731 put_unaligned_be32(sizeof(data), &cmd[5]);
1732
1733 put_unaligned_be64(key, &data[0]);
1734 put_unaligned_be64(sa_key, &data[8]);
1735 data[20] = flags;
1736
1737 result = scsi_execute_req(sdev, cmd, DMA_TO_DEVICE, &data, sizeof(data),
1738 &sshdr, SD_TIMEOUT, SD_MAX_RETRIES, NULL);
1739
1740 if ((driver_byte(result) & DRIVER_SENSE) &&
1741 (scsi_sense_valid(&sshdr))) {
1742 sdev_printk(KERN_INFO, sdev, "PR command failed: %d\n", result);
1743 scsi_print_sense_hdr(sdev, NULL, &sshdr);
1744 }
1745
1746 return result;
1747}
1748
1749static int sd_pr_register(struct block_device *bdev, u64 old_key, u64 new_key,
1750 u32 flags)
1751{
1752 if (flags & ~PR_FL_IGNORE_KEY)
1753 return -EOPNOTSUPP;
1754 return sd_pr_command(bdev, (flags & PR_FL_IGNORE_KEY) ? 0x06 : 0x00,
1755 old_key, new_key, 0,
1756 (1 << 0) /* APTPL */);
1757}
1758
1759static int sd_pr_reserve(struct block_device *bdev, u64 key, enum pr_type type,
1760 u32 flags)
1761{
1762 if (flags)
1763 return -EOPNOTSUPP;
1764 return sd_pr_command(bdev, 0x01, key, 0, sd_pr_type(type), 0);
1765}
1766
1767static int sd_pr_release(struct block_device *bdev, u64 key, enum pr_type type)
1768{
1769 return sd_pr_command(bdev, 0x02, key, 0, sd_pr_type(type), 0);
1770}
1771
1772static int sd_pr_preempt(struct block_device *bdev, u64 old_key, u64 new_key,
1773 enum pr_type type, bool abort)
1774{
1775 return sd_pr_command(bdev, abort ? 0x05 : 0x04, old_key, new_key,
1776 sd_pr_type(type), 0);
1777}
1778
1779static int sd_pr_clear(struct block_device *bdev, u64 key)
1780{
1781 return sd_pr_command(bdev, 0x03, key, 0, 0, 0);
1782}
1783
1784static const struct pr_ops sd_pr_ops = {
1785 .pr_register = sd_pr_register,
1786 .pr_reserve = sd_pr_reserve,
1787 .pr_release = sd_pr_release,
1788 .pr_preempt = sd_pr_preempt,
1789 .pr_clear = sd_pr_clear,
1790};
1791
1792static const struct block_device_operations sd_fops = {
1793 .owner = THIS_MODULE,
1794 .open = sd_open,
1795 .release = sd_release,
1796 .ioctl = sd_ioctl,
1797 .getgeo = sd_getgeo,
1798#ifdef CONFIG_COMPAT
1799 .compat_ioctl = sd_compat_ioctl,
1800#endif
1801 .check_events = sd_check_events,
1802 .revalidate_disk = sd_revalidate_disk,
1803 .unlock_native_capacity = sd_unlock_native_capacity,
1804 .pr_ops = &sd_pr_ops,
1805};
1806
1807/**
1808 * sd_eh_reset - reset error handling callback
1809 * @scmd: sd-issued command that has failed
1810 *
1811 * This function is called by the SCSI midlayer before starting
1812 * SCSI EH. When counting medium access failures we have to be
1813 * careful to register it only only once per device and SCSI EH run;
1814 * there might be several timed out commands which will cause the
1815 * 'max_medium_access_timeouts' counter to trigger after the first
1816 * SCSI EH run already and set the device to offline.
1817 * So this function resets the internal counter before starting SCSI EH.
1818 **/
1819static void sd_eh_reset(struct scsi_cmnd *scmd)
1820{
1821 struct scsi_disk *sdkp = scsi_disk(scmd->request->rq_disk);
1822
1823 /* New SCSI EH run, reset gate variable */
1824 sdkp->ignore_medium_access_errors = false;
1825}
1826
1827/**
1828 * sd_eh_action - error handling callback
1829 * @scmd: sd-issued command that has failed
1830 * @eh_disp: The recovery disposition suggested by the midlayer
1831 *
1832 * This function is called by the SCSI midlayer upon completion of an
1833 * error test command (currently TEST UNIT READY). The result of sending
1834 * the eh command is passed in eh_disp. We're looking for devices that
1835 * fail medium access commands but are OK with non access commands like
1836 * test unit ready (so wrongly see the device as having a successful
1837 * recovery)
1838 **/
1839static int sd_eh_action(struct scsi_cmnd *scmd, int eh_disp)
1840{
1841 struct scsi_disk *sdkp = scsi_disk(scmd->request->rq_disk);
1842 struct scsi_device *sdev = scmd->device;
1843
1844 if (!scsi_device_online(sdev) ||
1845 !scsi_medium_access_command(scmd) ||
1846 host_byte(scmd->result) != DID_TIME_OUT ||
1847 eh_disp != SUCCESS)
1848 return eh_disp;
1849
1850 /*
1851 * The device has timed out executing a medium access command.
1852 * However, the TEST UNIT READY command sent during error
1853 * handling completed successfully. Either the device is in the
1854 * process of recovering or has it suffered an internal failure
1855 * that prevents access to the storage medium.
1856 */
1857 if (!sdkp->ignore_medium_access_errors) {
1858 sdkp->medium_access_timed_out++;
1859 sdkp->ignore_medium_access_errors = true;
1860 }
1861
1862 /*
1863 * If the device keeps failing read/write commands but TEST UNIT
1864 * READY always completes successfully we assume that medium
1865 * access is no longer possible and take the device offline.
1866 */
1867 if (sdkp->medium_access_timed_out >= sdkp->max_medium_access_timeouts) {
1868 scmd_printk(KERN_ERR, scmd,
1869 "Medium access timeout failure. Offlining disk!\n");
1870 mutex_lock(&sdev->state_mutex);
1871 scsi_device_set_state(sdev, SDEV_OFFLINE);
1872 mutex_unlock(&sdev->state_mutex);
1873
1874 return SUCCESS;
1875 }
1876
1877 return eh_disp;
1878}
1879
1880static unsigned int sd_completed_bytes(struct scsi_cmnd *scmd)
1881{
1882 struct request *req = scmd->request;
1883 struct scsi_device *sdev = scmd->device;
1884 unsigned int transferred, good_bytes;
1885 u64 start_lba, end_lba, bad_lba;
1886
1887 /*
1888 * Some commands have a payload smaller than the device logical
1889 * block size (e.g. INQUIRY on a 4K disk).
1890 */
1891 if (scsi_bufflen(scmd) <= sdev->sector_size)
1892 return 0;
1893
1894 /* Check if we have a 'bad_lba' information */
1895 if (!scsi_get_sense_info_fld(scmd->sense_buffer,
1896 SCSI_SENSE_BUFFERSIZE,
1897 &bad_lba))
1898 return 0;
1899
1900 /*
1901 * If the bad lba was reported incorrectly, we have no idea where
1902 * the error is.
1903 */
1904 start_lba = sectors_to_logical(sdev, blk_rq_pos(req));
1905 end_lba = start_lba + bytes_to_logical(sdev, scsi_bufflen(scmd));
1906 if (bad_lba < start_lba || bad_lba >= end_lba)
1907 return 0;
1908
1909 /*
1910 * resid is optional but mostly filled in. When it's unused,
1911 * its value is zero, so we assume the whole buffer transferred
1912 */
1913 transferred = scsi_bufflen(scmd) - scsi_get_resid(scmd);
1914
1915 /* This computation should always be done in terms of the
1916 * resolution of the device's medium.
1917 */
1918 good_bytes = logical_to_bytes(sdev, bad_lba - start_lba);
1919
1920 return min(good_bytes, transferred);
1921}
1922
1923/**
1924 * sd_done - bottom half handler: called when the lower level
1925 * driver has completed (successfully or otherwise) a scsi command.
1926 * @SCpnt: mid-level's per command structure.
1927 *
1928 * Note: potentially run from within an ISR. Must not block.
1929 **/
1930static int sd_done(struct scsi_cmnd *SCpnt)
1931{
1932 int result = SCpnt->result;
1933 unsigned int good_bytes = result ? 0 : scsi_bufflen(SCpnt);
1934 unsigned int sector_size = SCpnt->device->sector_size;
1935 unsigned int resid;
1936 struct scsi_sense_hdr sshdr;
1937 struct scsi_disk *sdkp = scsi_disk(SCpnt->request->rq_disk);
1938 struct request *req = SCpnt->request;
1939 int sense_valid = 0;
1940 int sense_deferred = 0;
1941
1942 switch (req_op(req)) {
1943 case REQ_OP_DISCARD:
1944 case REQ_OP_WRITE_ZEROES:
1945 case REQ_OP_WRITE_SAME:
1946 case REQ_OP_ZONE_RESET:
1947 if (!result) {
1948 good_bytes = blk_rq_bytes(req);
1949 scsi_set_resid(SCpnt, 0);
1950 } else {
1951 good_bytes = 0;
1952 scsi_set_resid(SCpnt, blk_rq_bytes(req));
1953 }
1954 break;
1955 case REQ_OP_ZONE_REPORT:
1956 if (!result) {
1957 good_bytes = scsi_bufflen(SCpnt)
1958 - scsi_get_resid(SCpnt);
1959 scsi_set_resid(SCpnt, 0);
1960 } else {
1961 good_bytes = 0;
1962 scsi_set_resid(SCpnt, blk_rq_bytes(req));
1963 }
1964 break;
1965 default:
1966 /*
1967 * In case of bogus fw or device, we could end up having
1968 * an unaligned partial completion. Check this here and force
1969 * alignment.
1970 */
1971 resid = scsi_get_resid(SCpnt);
1972 if (resid & (sector_size - 1)) {
1973 sd_printk(KERN_INFO, sdkp,
1974 "Unaligned partial completion (resid=%u, sector_sz=%u)\n",
1975 resid, sector_size);
1976 resid = min(scsi_bufflen(SCpnt),
1977 round_up(resid, sector_size));
1978 scsi_set_resid(SCpnt, resid);
1979 }
1980 }
1981
1982 if (result) {
1983 sense_valid = scsi_command_normalize_sense(SCpnt, &sshdr);
1984 if (sense_valid)
1985 sense_deferred = scsi_sense_is_deferred(&sshdr);
1986 }
1987 sdkp->medium_access_timed_out = 0;
1988
1989 if (driver_byte(result) != DRIVER_SENSE &&
1990 (!sense_valid || sense_deferred))
1991 goto out;
1992
1993 switch (sshdr.sense_key) {
1994 case HARDWARE_ERROR:
1995 case MEDIUM_ERROR:
1996 good_bytes = sd_completed_bytes(SCpnt);
1997 break;
1998 case RECOVERED_ERROR:
1999 good_bytes = scsi_bufflen(SCpnt);
2000 break;
2001 case NO_SENSE:
2002 /* This indicates a false check condition, so ignore it. An
2003 * unknown amount of data was transferred so treat it as an
2004 * error.
2005 */
2006 SCpnt->result = 0;
2007 memset(SCpnt->sense_buffer, 0, SCSI_SENSE_BUFFERSIZE);
2008 break;
2009 case ABORTED_COMMAND:
2010 if (sshdr.asc == 0x10) /* DIF: Target detected corruption */
2011 good_bytes = sd_completed_bytes(SCpnt);
2012 break;
2013 case ILLEGAL_REQUEST:
2014 switch (sshdr.asc) {
2015 case 0x10: /* DIX: Host detected corruption */
2016 good_bytes = sd_completed_bytes(SCpnt);
2017 break;
2018 case 0x20: /* INVALID COMMAND OPCODE */
2019 case 0x24: /* INVALID FIELD IN CDB */
2020 switch (SCpnt->cmnd[0]) {
2021 case UNMAP:
2022 sd_config_discard(sdkp, SD_LBP_DISABLE);
2023 break;
2024 case WRITE_SAME_16:
2025 case WRITE_SAME:
2026 if (SCpnt->cmnd[1] & 8) { /* UNMAP */
2027 sd_config_discard(sdkp, SD_LBP_DISABLE);
2028 } else {
2029 sdkp->device->no_write_same = 1;
2030 sd_config_write_same(sdkp);
2031 req->__data_len = blk_rq_bytes(req);
2032 req->rq_flags |= RQF_QUIET;
2033 }
2034 break;
2035 }
2036 }
2037 break;
2038 default:
2039 break;
2040 }
2041
2042 out:
2043 if (sd_is_zoned(sdkp))
2044 sd_zbc_complete(SCpnt, good_bytes, &sshdr);
2045
2046 SCSI_LOG_HLCOMPLETE(1, scmd_printk(KERN_INFO, SCpnt,
2047 "sd_done: completed %d of %d bytes\n",
2048 good_bytes, scsi_bufflen(SCpnt)));
2049
2050 if (rq_data_dir(SCpnt->request) == READ && scsi_prot_sg_count(SCpnt))
2051 sd_dif_complete(SCpnt, good_bytes);
2052
2053 return good_bytes;
2054}
2055
2056/*
2057 * spinup disk - called only in sd_revalidate_disk()
2058 */
2059static void
2060sd_spinup_disk(struct scsi_disk *sdkp)
2061{
2062 unsigned char cmd[10];
2063 unsigned long spintime_expire = 0;
2064 int retries, spintime;
2065 unsigned int the_result;
2066 struct scsi_sense_hdr sshdr;
2067 int sense_valid = 0;
2068
2069 spintime = 0;
2070
2071 /* Spin up drives, as required. Only do this at boot time */
2072 /* Spinup needs to be done for module loads too. */
2073 do {
2074 retries = 0;
2075
2076 do {
2077 cmd[0] = TEST_UNIT_READY;
2078 memset((void *) &cmd[1], 0, 9);
2079
2080 the_result = scsi_execute_req(sdkp->device, cmd,
2081 DMA_NONE, NULL, 0,
2082 &sshdr, SD_TIMEOUT,
2083 SD_MAX_RETRIES, NULL);
2084
2085 /*
2086 * If the drive has indicated to us that it
2087 * doesn't have any media in it, don't bother
2088 * with any more polling.
2089 */
2090 if (media_not_present(sdkp, &sshdr))
2091 return;
2092
2093 if (the_result)
2094 sense_valid = scsi_sense_valid(&sshdr);
2095 retries++;
2096 } while (retries < 3 &&
2097 (!scsi_status_is_good(the_result) ||
2098 ((driver_byte(the_result) & DRIVER_SENSE) &&
2099 sense_valid && sshdr.sense_key == UNIT_ATTENTION)));
2100
2101 if ((driver_byte(the_result) & DRIVER_SENSE) == 0) {
2102 /* no sense, TUR either succeeded or failed
2103 * with a status error */
2104 if(!spintime && !scsi_status_is_good(the_result)) {
2105 sd_print_result(sdkp, "Test Unit Ready failed",
2106 the_result);
2107 }
2108 break;
2109 }
2110
2111 /*
2112 * The device does not want the automatic start to be issued.
2113 */
2114 if (sdkp->device->no_start_on_add)
2115 break;
2116
2117 if (sense_valid && sshdr.sense_key == NOT_READY) {
2118 if (sshdr.asc == 4 && sshdr.ascq == 3)
2119 break; /* manual intervention required */
2120 if (sshdr.asc == 4 && sshdr.ascq == 0xb)
2121 break; /* standby */
2122 if (sshdr.asc == 4 && sshdr.ascq == 0xc)
2123 break; /* unavailable */
2124 if (sshdr.asc == 4 && sshdr.ascq == 0x1b)
2125 break; /* sanitize in progress */
2126 /*
2127 * Issue command to spin up drive when not ready
2128 */
2129 if (!spintime) {
2130 sd_printk(KERN_NOTICE, sdkp, "Spinning up disk...");
2131 cmd[0] = START_STOP;
2132 cmd[1] = 1; /* Return immediately */
2133 memset((void *) &cmd[2], 0, 8);
2134 cmd[4] = 1; /* Start spin cycle */
2135 if (sdkp->device->start_stop_pwr_cond)
2136 cmd[4] |= 1 << 4;
2137 scsi_execute_req(sdkp->device, cmd, DMA_NONE,
2138 NULL, 0, &sshdr,
2139 SD_TIMEOUT, SD_MAX_RETRIES,
2140 NULL);
2141 spintime_expire = jiffies + 100 * HZ;
2142 spintime = 1;
2143 }
2144 /* Wait 1 second for next try */
2145 msleep(1000);
2146 printk(KERN_CONT ".");
2147
2148 /*
2149 * Wait for USB flash devices with slow firmware.
2150 * Yes, this sense key/ASC combination shouldn't
2151 * occur here. It's characteristic of these devices.
2152 */
2153 } else if (sense_valid &&
2154 sshdr.sense_key == UNIT_ATTENTION &&
2155 sshdr.asc == 0x28) {
2156 if (!spintime) {
2157 spintime_expire = jiffies + 5 * HZ;
2158 spintime = 1;
2159 }
2160 /* Wait 1 second for next try */
2161 msleep(1000);
2162 } else {
2163 /* we don't understand the sense code, so it's
2164 * probably pointless to loop */
2165 if(!spintime) {
2166 sd_printk(KERN_NOTICE, sdkp, "Unit Not Ready\n");
2167 sd_print_sense_hdr(sdkp, &sshdr);
2168 }
2169 break;
2170 }
2171
2172 } while (spintime && time_before_eq(jiffies, spintime_expire));
2173
2174 if (spintime) {
2175 if (scsi_status_is_good(the_result))
2176 printk(KERN_CONT "ready\n");
2177 else
2178 printk(KERN_CONT "not responding...\n");
2179 }
2180}
2181
2182/*
2183 * Determine whether disk supports Data Integrity Field.
2184 */
2185static int sd_read_protection_type(struct scsi_disk *sdkp, unsigned char *buffer)
2186{
2187 struct scsi_device *sdp = sdkp->device;
2188 u8 type;
2189 int ret = 0;
2190
2191 if (scsi_device_protection(sdp) == 0 || (buffer[12] & 1) == 0)
2192 return ret;
2193
2194 type = ((buffer[12] >> 1) & 7) + 1; /* P_TYPE 0 = Type 1 */
2195
2196 if (type > T10_PI_TYPE3_PROTECTION)
2197 ret = -ENODEV;
2198 else if (scsi_host_dif_capable(sdp->host, type))
2199 ret = 1;
2200
2201 if (sdkp->first_scan || type != sdkp->protection_type)
2202 switch (ret) {
2203 case -ENODEV:
2204 sd_printk(KERN_ERR, sdkp, "formatted with unsupported" \
2205 " protection type %u. Disabling disk!\n",
2206 type);
2207 break;
2208 case 1:
2209 sd_printk(KERN_NOTICE, sdkp,
2210 "Enabling DIF Type %u protection\n", type);
2211 break;
2212 case 0:
2213 sd_printk(KERN_NOTICE, sdkp,
2214 "Disabling DIF Type %u protection\n", type);
2215 break;
2216 }
2217
2218 sdkp->protection_type = type;
2219
2220 return ret;
2221}
2222
2223static void read_capacity_error(struct scsi_disk *sdkp, struct scsi_device *sdp,
2224 struct scsi_sense_hdr *sshdr, int sense_valid,
2225 int the_result)
2226{
2227 if (driver_byte(the_result) & DRIVER_SENSE)
2228 sd_print_sense_hdr(sdkp, sshdr);
2229 else
2230 sd_printk(KERN_NOTICE, sdkp, "Sense not available.\n");
2231
2232 /*
2233 * Set dirty bit for removable devices if not ready -
2234 * sometimes drives will not report this properly.
2235 */
2236 if (sdp->removable &&
2237 sense_valid && sshdr->sense_key == NOT_READY)
2238 set_media_not_present(sdkp);
2239
2240 /*
2241 * We used to set media_present to 0 here to indicate no media
2242 * in the drive, but some drives fail read capacity even with
2243 * media present, so we can't do that.
2244 */
2245 sdkp->capacity = 0; /* unknown mapped to zero - as usual */
2246}
2247
2248#define RC16_LEN 32
2249#if RC16_LEN > SD_BUF_SIZE
2250#error RC16_LEN must not be more than SD_BUF_SIZE
2251#endif
2252
2253#define READ_CAPACITY_RETRIES_ON_RESET 10
2254
2255/*
2256 * Ensure that we don't overflow sector_t when CONFIG_LBDAF is not set
2257 * and the reported logical block size is bigger than 512 bytes. Note
2258 * that last_sector is a u64 and therefore logical_to_sectors() is not
2259 * applicable.
2260 */
2261static bool sd_addressable_capacity(u64 lba, unsigned int sector_size)
2262{
2263 u64 last_sector = (lba + 1ULL) << (ilog2(sector_size) - 9);
2264
2265 if (sizeof(sector_t) == 4 && last_sector > U32_MAX)
2266 return false;
2267
2268 return true;
2269}
2270
2271static int read_capacity_16(struct scsi_disk *sdkp, struct scsi_device *sdp,
2272 unsigned char *buffer)
2273{
2274 unsigned char cmd[16];
2275 struct scsi_sense_hdr sshdr;
2276 int sense_valid = 0;
2277 int the_result;
2278 int retries = 3, reset_retries = READ_CAPACITY_RETRIES_ON_RESET;
2279 unsigned int alignment;
2280 unsigned long long lba;
2281 unsigned sector_size;
2282
2283 if (sdp->no_read_capacity_16)
2284 return -EINVAL;
2285
2286 do {
2287 memset(cmd, 0, 16);
2288 cmd[0] = SERVICE_ACTION_IN_16;
2289 cmd[1] = SAI_READ_CAPACITY_16;
2290 cmd[13] = RC16_LEN;
2291 memset(buffer, 0, RC16_LEN);
2292
2293 the_result = scsi_execute_req(sdp, cmd, DMA_FROM_DEVICE,
2294 buffer, RC16_LEN, &sshdr,
2295 SD_TIMEOUT, SD_MAX_RETRIES, NULL);
2296
2297 if (media_not_present(sdkp, &sshdr))
2298 return -ENODEV;
2299
2300 if (the_result) {
2301 sense_valid = scsi_sense_valid(&sshdr);
2302 if (sense_valid &&
2303 sshdr.sense_key == ILLEGAL_REQUEST &&
2304 (sshdr.asc == 0x20 || sshdr.asc == 0x24) &&
2305 sshdr.ascq == 0x00)
2306 /* Invalid Command Operation Code or
2307 * Invalid Field in CDB, just retry
2308 * silently with RC10 */
2309 return -EINVAL;
2310 if (sense_valid &&
2311 sshdr.sense_key == UNIT_ATTENTION &&
2312 sshdr.asc == 0x29 && sshdr.ascq == 0x00)
2313 /* Device reset might occur several times,
2314 * give it one more chance */
2315 if (--reset_retries > 0)
2316 continue;
2317 }
2318 retries--;
2319
2320 } while (the_result && retries);
2321
2322 if (the_result) {
2323 sd_print_result(sdkp, "Read Capacity(16) failed", the_result);
2324 read_capacity_error(sdkp, sdp, &sshdr, sense_valid, the_result);
2325 return -EINVAL;
2326 }
2327
2328 sector_size = get_unaligned_be32(&buffer[8]);
2329 lba = get_unaligned_be64(&buffer[0]);
2330
2331 if (sd_read_protection_type(sdkp, buffer) < 0) {
2332 sdkp->capacity = 0;
2333 return -ENODEV;
2334 }
2335
2336 if (!sd_addressable_capacity(lba, sector_size)) {
2337 sd_printk(KERN_ERR, sdkp, "Too big for this kernel. Use a "
2338 "kernel compiled with support for large block "
2339 "devices.\n");
2340 sdkp->capacity = 0;
2341 return -EOVERFLOW;
2342 }
2343
2344 /* Logical blocks per physical block exponent */
2345 sdkp->physical_block_size = (1 << (buffer[13] & 0xf)) * sector_size;
2346
2347 /* RC basis */
2348 sdkp->rc_basis = (buffer[12] >> 4) & 0x3;
2349
2350 /* Lowest aligned logical block */
2351 alignment = ((buffer[14] & 0x3f) << 8 | buffer[15]) * sector_size;
2352 blk_queue_alignment_offset(sdp->request_queue, alignment);
2353 if (alignment && sdkp->first_scan)
2354 sd_printk(KERN_NOTICE, sdkp,
2355 "physical block alignment offset: %u\n", alignment);
2356
2357 if (buffer[14] & 0x80) { /* LBPME */
2358 sdkp->lbpme = 1;
2359
2360 if (buffer[14] & 0x40) /* LBPRZ */
2361 sdkp->lbprz = 1;
2362
2363 sd_config_discard(sdkp, SD_LBP_WS16);
2364 }
2365
2366 sdkp->capacity = lba + 1;
2367 return sector_size;
2368}
2369
2370static int read_capacity_10(struct scsi_disk *sdkp, struct scsi_device *sdp,
2371 unsigned char *buffer)
2372{
2373 unsigned char cmd[16];
2374 struct scsi_sense_hdr sshdr;
2375 int sense_valid = 0;
2376 int the_result;
2377 int retries = 3, reset_retries = READ_CAPACITY_RETRIES_ON_RESET;
2378 sector_t lba;
2379 unsigned sector_size;
2380
2381 do {
2382 cmd[0] = READ_CAPACITY;
2383 memset(&cmd[1], 0, 9);
2384 memset(buffer, 0, 8);
2385
2386 the_result = scsi_execute_req(sdp, cmd, DMA_FROM_DEVICE,
2387 buffer, 8, &sshdr,
2388 SD_TIMEOUT, SD_MAX_RETRIES, NULL);
2389
2390 if (media_not_present(sdkp, &sshdr))
2391 return -ENODEV;
2392
2393 if (the_result) {
2394 sense_valid = scsi_sense_valid(&sshdr);
2395 if (sense_valid &&
2396 sshdr.sense_key == UNIT_ATTENTION &&
2397 sshdr.asc == 0x29 && sshdr.ascq == 0x00)
2398 /* Device reset might occur several times,
2399 * give it one more chance */
2400 if (--reset_retries > 0)
2401 continue;
2402 }
2403 retries--;
2404
2405 } while (the_result && retries);
2406
2407 if (the_result) {
2408 sd_print_result(sdkp, "Read Capacity(10) failed", the_result);
2409 read_capacity_error(sdkp, sdp, &sshdr, sense_valid, the_result);
2410 return -EINVAL;
2411 }
2412
2413 sector_size = get_unaligned_be32(&buffer[4]);
2414 lba = get_unaligned_be32(&buffer[0]);
2415
2416 if (sdp->no_read_capacity_16 && (lba == 0xffffffff)) {
2417 /* Some buggy (usb cardreader) devices return an lba of
2418 0xffffffff when the want to report a size of 0 (with
2419 which they really mean no media is present) */
2420 sdkp->capacity = 0;
2421 sdkp->physical_block_size = sector_size;
2422 return sector_size;
2423 }
2424
2425 if (!sd_addressable_capacity(lba, sector_size)) {
2426 sd_printk(KERN_ERR, sdkp, "Too big for this kernel. Use a "
2427 "kernel compiled with support for large block "
2428 "devices.\n");
2429 sdkp->capacity = 0;
2430 return -EOVERFLOW;
2431 }
2432
2433 sdkp->capacity = lba + 1;
2434 sdkp->physical_block_size = sector_size;
2435 return sector_size;
2436}
2437
2438static int sd_try_rc16_first(struct scsi_device *sdp)
2439{
2440 if (sdp->host->max_cmd_len < 16)
2441 return 0;
2442 if (sdp->try_rc_10_first)
2443 return 0;
2444 if (sdp->scsi_level > SCSI_SPC_2)
2445 return 1;
2446 if (scsi_device_protection(sdp))
2447 return 1;
2448 return 0;
2449}
2450
2451/*
2452 * read disk capacity
2453 */
2454static void
2455sd_read_capacity(struct scsi_disk *sdkp, unsigned char *buffer)
2456{
2457 int sector_size;
2458 struct scsi_device *sdp = sdkp->device;
2459
2460 if (sd_try_rc16_first(sdp)) {
2461 sector_size = read_capacity_16(sdkp, sdp, buffer);
2462 if (sector_size == -EOVERFLOW)
2463 goto got_data;
2464 if (sector_size == -ENODEV)
2465 return;
2466 if (sector_size < 0)
2467 sector_size = read_capacity_10(sdkp, sdp, buffer);
2468 if (sector_size < 0)
2469 return;
2470 } else {
2471 sector_size = read_capacity_10(sdkp, sdp, buffer);
2472 if (sector_size == -EOVERFLOW)
2473 goto got_data;
2474 if (sector_size < 0)
2475 return;
2476 if ((sizeof(sdkp->capacity) > 4) &&
2477 (sdkp->capacity > 0xffffffffULL)) {
2478 int old_sector_size = sector_size;
2479 sd_printk(KERN_NOTICE, sdkp, "Very big device. "
2480 "Trying to use READ CAPACITY(16).\n");
2481 sector_size = read_capacity_16(sdkp, sdp, buffer);
2482 if (sector_size < 0) {
2483 sd_printk(KERN_NOTICE, sdkp,
2484 "Using 0xffffffff as device size\n");
2485 sdkp->capacity = 1 + (sector_t) 0xffffffff;
2486 sector_size = old_sector_size;
2487 goto got_data;
2488 }
2489 /* Remember that READ CAPACITY(16) succeeded */
2490 sdp->try_rc_10_first = 0;
2491 }
2492 }
2493
2494 /* Some devices are known to return the total number of blocks,
2495 * not the highest block number. Some devices have versions
2496 * which do this and others which do not. Some devices we might
2497 * suspect of doing this but we don't know for certain.
2498 *
2499 * If we know the reported capacity is wrong, decrement it. If
2500 * we can only guess, then assume the number of blocks is even
2501 * (usually true but not always) and err on the side of lowering
2502 * the capacity.
2503 */
2504 if (sdp->fix_capacity ||
2505 (sdp->guess_capacity && (sdkp->capacity & 0x01))) {
2506 sd_printk(KERN_INFO, sdkp, "Adjusting the sector count "
2507 "from its reported value: %llu\n",
2508 (unsigned long long) sdkp->capacity);
2509 --sdkp->capacity;
2510 }
2511
2512got_data:
2513 if (sector_size == 0) {
2514 sector_size = 512;
2515 sd_printk(KERN_NOTICE, sdkp, "Sector size 0 reported, "
2516 "assuming 512.\n");
2517 }
2518
2519 if (sector_size != 512 &&
2520 sector_size != 1024 &&
2521 sector_size != 2048 &&
2522 sector_size != 4096) {
2523 sd_printk(KERN_NOTICE, sdkp, "Unsupported sector size %d.\n",
2524 sector_size);
2525 /*
2526 * The user might want to re-format the drive with
2527 * a supported sectorsize. Once this happens, it
2528 * would be relatively trivial to set the thing up.
2529 * For this reason, we leave the thing in the table.
2530 */
2531 sdkp->capacity = 0;
2532 /*
2533 * set a bogus sector size so the normal read/write
2534 * logic in the block layer will eventually refuse any
2535 * request on this device without tripping over power
2536 * of two sector size assumptions
2537 */
2538 sector_size = 512;
2539 }
2540 blk_queue_logical_block_size(sdp->request_queue, sector_size);
2541 blk_queue_physical_block_size(sdp->request_queue,
2542 sdkp->physical_block_size);
2543 sdkp->device->sector_size = sector_size;
2544
2545 if (sdkp->capacity > 0xffffffff)
2546 sdp->use_16_for_rw = 1;
2547
2548}
2549
2550/*
2551 * Print disk capacity
2552 */
2553static void
2554sd_print_capacity(struct scsi_disk *sdkp,
2555 sector_t old_capacity)
2556{
2557 int sector_size = sdkp->device->sector_size;
2558 char cap_str_2[10], cap_str_10[10];
2559
2560 string_get_size(sdkp->capacity, sector_size,
2561 STRING_UNITS_2, cap_str_2, sizeof(cap_str_2));
2562 string_get_size(sdkp->capacity, sector_size,
2563 STRING_UNITS_10, cap_str_10,
2564 sizeof(cap_str_10));
2565
2566 if (sdkp->first_scan || old_capacity != sdkp->capacity) {
2567 sd_printk(KERN_NOTICE, sdkp,
2568 "%llu %d-byte logical blocks: (%s/%s)\n",
2569 (unsigned long long)sdkp->capacity,
2570 sector_size, cap_str_10, cap_str_2);
2571
2572 if (sdkp->physical_block_size != sector_size)
2573 sd_printk(KERN_NOTICE, sdkp,
2574 "%u-byte physical blocks\n",
2575 sdkp->physical_block_size);
2576
2577 sd_zbc_print_zones(sdkp);
2578 }
2579}
2580
2581/* called with buffer of length 512 */
2582static inline int
2583sd_do_mode_sense(struct scsi_device *sdp, int dbd, int modepage,
2584 unsigned char *buffer, int len, struct scsi_mode_data *data,
2585 struct scsi_sense_hdr *sshdr)
2586{
2587 return scsi_mode_sense(sdp, dbd, modepage, buffer, len,
2588 SD_TIMEOUT, SD_MAX_RETRIES, data,
2589 sshdr);
2590}
2591
2592/*
2593 * read write protect setting, if possible - called only in sd_revalidate_disk()
2594 * called with buffer of length SD_BUF_SIZE
2595 */
2596static void
2597sd_read_write_protect_flag(struct scsi_disk *sdkp, unsigned char *buffer)
2598{
2599 int res;
2600 struct scsi_device *sdp = sdkp->device;
2601 struct scsi_mode_data data;
2602 int disk_ro = get_disk_ro(sdkp->disk);
2603 int old_wp = sdkp->write_prot;
2604
2605 set_disk_ro(sdkp->disk, 0);
2606 if (sdp->skip_ms_page_3f) {
2607 sd_first_printk(KERN_NOTICE, sdkp, "Assuming Write Enabled\n");
2608 return;
2609 }
2610
2611 if (sdp->use_192_bytes_for_3f) {
2612 res = sd_do_mode_sense(sdp, 0, 0x3F, buffer, 192, &data, NULL);
2613 } else {
2614 /*
2615 * First attempt: ask for all pages (0x3F), but only 4 bytes.
2616 * We have to start carefully: some devices hang if we ask
2617 * for more than is available.
2618 */
2619 res = sd_do_mode_sense(sdp, 0, 0x3F, buffer, 4, &data, NULL);
2620
2621 /*
2622 * Second attempt: ask for page 0 When only page 0 is
2623 * implemented, a request for page 3F may return Sense Key
2624 * 5: Illegal Request, Sense Code 24: Invalid field in
2625 * CDB.
2626 */
2627 if (!scsi_status_is_good(res))
2628 res = sd_do_mode_sense(sdp, 0, 0, buffer, 4, &data, NULL);
2629
2630 /*
2631 * Third attempt: ask 255 bytes, as we did earlier.
2632 */
2633 if (!scsi_status_is_good(res))
2634 res = sd_do_mode_sense(sdp, 0, 0x3F, buffer, 255,
2635 &data, NULL);
2636 }
2637
2638 if (!scsi_status_is_good(res)) {
2639 sd_first_printk(KERN_WARNING, sdkp,
2640 "Test WP failed, assume Write Enabled\n");
2641 } else {
2642 sdkp->write_prot = ((data.device_specific & 0x80) != 0);
2643 set_disk_ro(sdkp->disk, sdkp->write_prot || disk_ro);
2644 if (sdkp->first_scan || old_wp != sdkp->write_prot) {
2645 sd_printk(KERN_NOTICE, sdkp, "Write Protect is %s\n",
2646 sdkp->write_prot ? "on" : "off");
2647 sd_printk(KERN_DEBUG, sdkp, "Mode Sense: %4ph\n", buffer);
2648 }
2649 }
2650}
2651
2652/*
2653 * sd_read_cache_type - called only from sd_revalidate_disk()
2654 * called with buffer of length SD_BUF_SIZE
2655 */
2656static void
2657sd_read_cache_type(struct scsi_disk *sdkp, unsigned char *buffer)
2658{
2659 int len = 0, res;
2660 struct scsi_device *sdp = sdkp->device;
2661
2662 int dbd;
2663 int modepage;
2664 int first_len;
2665 struct scsi_mode_data data;
2666 struct scsi_sense_hdr sshdr;
2667 int old_wce = sdkp->WCE;
2668 int old_rcd = sdkp->RCD;
2669 int old_dpofua = sdkp->DPOFUA;
2670
2671
2672 if (sdkp->cache_override)
2673 return;
2674
2675 first_len = 4;
2676 if (sdp->skip_ms_page_8) {
2677 if (sdp->type == TYPE_RBC)
2678 goto defaults;
2679 else {
2680 if (sdp->skip_ms_page_3f)
2681 goto defaults;
2682 modepage = 0x3F;
2683 if (sdp->use_192_bytes_for_3f)
2684 first_len = 192;
2685 dbd = 0;
2686 }
2687 } else if (sdp->type == TYPE_RBC) {
2688 modepage = 6;
2689 dbd = 8;
2690 } else {
2691 modepage = 8;
2692 dbd = 0;
2693 }
2694
2695 /* cautiously ask */
2696 res = sd_do_mode_sense(sdp, dbd, modepage, buffer, first_len,
2697 &data, &sshdr);
2698
2699 if (!scsi_status_is_good(res))
2700 goto bad_sense;
2701
2702 if (!data.header_length) {
2703 modepage = 6;
2704 first_len = 0;
2705 sd_first_printk(KERN_ERR, sdkp,
2706 "Missing header in MODE_SENSE response\n");
2707 }
2708
2709 /* that went OK, now ask for the proper length */
2710 len = data.length;
2711
2712 /*
2713 * We're only interested in the first three bytes, actually.
2714 * But the data cache page is defined for the first 20.
2715 */
2716 if (len < 3)
2717 goto bad_sense;
2718 else if (len > SD_BUF_SIZE) {
2719 sd_first_printk(KERN_NOTICE, sdkp, "Truncating mode parameter "
2720 "data from %d to %d bytes\n", len, SD_BUF_SIZE);
2721 len = SD_BUF_SIZE;
2722 }
2723 if (modepage == 0x3F && sdp->use_192_bytes_for_3f)
2724 len = 192;
2725
2726 /* Get the data */
2727 if (len > first_len)
2728 res = sd_do_mode_sense(sdp, dbd, modepage, buffer, len,
2729 &data, &sshdr);
2730
2731 if (scsi_status_is_good(res)) {
2732 int offset = data.header_length + data.block_descriptor_length;
2733
2734 while (offset < len) {
2735 u8 page_code = buffer[offset] & 0x3F;
2736 u8 spf = buffer[offset] & 0x40;
2737
2738 if (page_code == 8 || page_code == 6) {
2739 /* We're interested only in the first 3 bytes.
2740 */
2741 if (len - offset <= 2) {
2742 sd_first_printk(KERN_ERR, sdkp,
2743 "Incomplete mode parameter "
2744 "data\n");
2745 goto defaults;
2746 } else {
2747 modepage = page_code;
2748 goto Page_found;
2749 }
2750 } else {
2751 /* Go to the next page */
2752 if (spf && len - offset > 3)
2753 offset += 4 + (buffer[offset+2] << 8) +
2754 buffer[offset+3];
2755 else if (!spf && len - offset > 1)
2756 offset += 2 + buffer[offset+1];
2757 else {
2758 sd_first_printk(KERN_ERR, sdkp,
2759 "Incomplete mode "
2760 "parameter data\n");
2761 goto defaults;
2762 }
2763 }
2764 }
2765
2766 sd_first_printk(KERN_ERR, sdkp, "No Caching mode page found\n");
2767 goto defaults;
2768
2769 Page_found:
2770 if (modepage == 8) {
2771 sdkp->WCE = ((buffer[offset + 2] & 0x04) != 0);
2772 sdkp->RCD = ((buffer[offset + 2] & 0x01) != 0);
2773 } else {
2774 sdkp->WCE = ((buffer[offset + 2] & 0x01) == 0);
2775 sdkp->RCD = 0;
2776 }
2777
2778 sdkp->DPOFUA = (data.device_specific & 0x10) != 0;
2779 if (sdp->broken_fua) {
2780 sd_first_printk(KERN_NOTICE, sdkp, "Disabling FUA\n");
2781 sdkp->DPOFUA = 0;
2782 } else if (sdkp->DPOFUA && !sdkp->device->use_10_for_rw &&
2783 !sdkp->device->use_16_for_rw) {
2784 sd_first_printk(KERN_NOTICE, sdkp,
2785 "Uses READ/WRITE(6), disabling FUA\n");
2786 sdkp->DPOFUA = 0;
2787 }
2788
2789 /* No cache flush allowed for write protected devices */
2790 if (sdkp->WCE && sdkp->write_prot)
2791 sdkp->WCE = 0;
2792
2793 if (sdkp->first_scan || old_wce != sdkp->WCE ||
2794 old_rcd != sdkp->RCD || old_dpofua != sdkp->DPOFUA)
2795 sd_printk(KERN_NOTICE, sdkp,
2796 "Write cache: %s, read cache: %s, %s\n",
2797 sdkp->WCE ? "enabled" : "disabled",
2798 sdkp->RCD ? "disabled" : "enabled",
2799 sdkp->DPOFUA ? "supports DPO and FUA"
2800 : "doesn't support DPO or FUA");
2801
2802 return;
2803 }
2804
2805bad_sense:
2806 if (scsi_sense_valid(&sshdr) &&
2807 sshdr.sense_key == ILLEGAL_REQUEST &&
2808 sshdr.asc == 0x24 && sshdr.ascq == 0x0)
2809 /* Invalid field in CDB */
2810 sd_first_printk(KERN_NOTICE, sdkp, "Cache data unavailable\n");
2811 else
2812 sd_first_printk(KERN_ERR, sdkp,
2813 "Asking for cache data failed\n");
2814
2815defaults:
2816 if (sdp->wce_default_on) {
2817 sd_first_printk(KERN_NOTICE, sdkp,
2818 "Assuming drive cache: write back\n");
2819 sdkp->WCE = 1;
2820 } else {
2821 sd_first_printk(KERN_ERR, sdkp,
2822 "Assuming drive cache: write through\n");
2823 sdkp->WCE = 0;
2824 }
2825 sdkp->RCD = 0;
2826 sdkp->DPOFUA = 0;
2827}
2828
2829/*
2830 * The ATO bit indicates whether the DIF application tag is available
2831 * for use by the operating system.
2832 */
2833static void sd_read_app_tag_own(struct scsi_disk *sdkp, unsigned char *buffer)
2834{
2835 int res, offset;
2836 struct scsi_device *sdp = sdkp->device;
2837 struct scsi_mode_data data;
2838 struct scsi_sense_hdr sshdr;
2839
2840 if (sdp->type != TYPE_DISK && sdp->type != TYPE_ZBC)
2841 return;
2842
2843 if (sdkp->protection_type == 0)
2844 return;
2845
2846 res = scsi_mode_sense(sdp, 1, 0x0a, buffer, 36, SD_TIMEOUT,
2847 SD_MAX_RETRIES, &data, &sshdr);
2848
2849 if (!scsi_status_is_good(res) || !data.header_length ||
2850 data.length < 6) {
2851 sd_first_printk(KERN_WARNING, sdkp,
2852 "getting Control mode page failed, assume no ATO\n");
2853
2854 if (scsi_sense_valid(&sshdr))
2855 sd_print_sense_hdr(sdkp, &sshdr);
2856
2857 return;
2858 }
2859
2860 offset = data.header_length + data.block_descriptor_length;
2861
2862 if ((buffer[offset] & 0x3f) != 0x0a) {
2863 sd_first_printk(KERN_ERR, sdkp, "ATO Got wrong page\n");
2864 return;
2865 }
2866
2867 if ((buffer[offset + 5] & 0x80) == 0)
2868 return;
2869
2870 sdkp->ATO = 1;
2871
2872 return;
2873}
2874
2875/**
2876 * sd_read_block_limits - Query disk device for preferred I/O sizes.
2877 * @sdkp: disk to query
2878 */
2879static void sd_read_block_limits(struct scsi_disk *sdkp)
2880{
2881 unsigned int sector_sz = sdkp->device->sector_size;
2882 const int vpd_len = 64;
2883 unsigned char *buffer = kmalloc(vpd_len, GFP_KERNEL);
2884
2885 if (!buffer ||
2886 /* Block Limits VPD */
2887 scsi_get_vpd_page(sdkp->device, 0xb0, buffer, vpd_len))
2888 goto out;
2889
2890 blk_queue_io_min(sdkp->disk->queue,
2891 get_unaligned_be16(&buffer[6]) * sector_sz);
2892
2893 sdkp->max_xfer_blocks = get_unaligned_be32(&buffer[8]);
2894 sdkp->opt_xfer_blocks = get_unaligned_be32(&buffer[12]);
2895
2896 if (buffer[3] == 0x3c) {
2897 unsigned int lba_count, desc_count;
2898
2899 sdkp->max_ws_blocks = (u32)get_unaligned_be64(&buffer[36]);
2900
2901 if (!sdkp->lbpme)
2902 goto out;
2903
2904 lba_count = get_unaligned_be32(&buffer[20]);
2905 desc_count = get_unaligned_be32(&buffer[24]);
2906
2907 if (lba_count && desc_count)
2908 sdkp->max_unmap_blocks = lba_count;
2909
2910 sdkp->unmap_granularity = get_unaligned_be32(&buffer[28]);
2911
2912 if (buffer[32] & 0x80)
2913 sdkp->unmap_alignment =
2914 get_unaligned_be32(&buffer[32]) & ~(1 << 31);
2915
2916 if (!sdkp->lbpvpd) { /* LBP VPD page not provided */
2917
2918 if (sdkp->max_unmap_blocks)
2919 sd_config_discard(sdkp, SD_LBP_UNMAP);
2920 else
2921 sd_config_discard(sdkp, SD_LBP_WS16);
2922
2923 } else { /* LBP VPD page tells us what to use */
2924 if (sdkp->lbpu && sdkp->max_unmap_blocks)
2925 sd_config_discard(sdkp, SD_LBP_UNMAP);
2926 else if (sdkp->lbpws)
2927 sd_config_discard(sdkp, SD_LBP_WS16);
2928 else if (sdkp->lbpws10)
2929 sd_config_discard(sdkp, SD_LBP_WS10);
2930 else
2931 sd_config_discard(sdkp, SD_LBP_DISABLE);
2932 }
2933 }
2934
2935 out:
2936 kfree(buffer);
2937}
2938
2939/**
2940 * sd_read_block_characteristics - Query block dev. characteristics
2941 * @sdkp: disk to query
2942 */
2943static void sd_read_block_characteristics(struct scsi_disk *sdkp)
2944{
2945 struct request_queue *q = sdkp->disk->queue;
2946 unsigned char *buffer;
2947 u16 rot;
2948 const int vpd_len = 64;
2949
2950 buffer = kmalloc(vpd_len, GFP_KERNEL);
2951
2952 if (!buffer ||
2953 /* Block Device Characteristics VPD */
2954 scsi_get_vpd_page(sdkp->device, 0xb1, buffer, vpd_len))
2955 goto out;
2956
2957 rot = get_unaligned_be16(&buffer[4]);
2958
2959 if (rot == 1) {
2960 blk_queue_flag_set(QUEUE_FLAG_NONROT, q);
2961 blk_queue_flag_clear(QUEUE_FLAG_ADD_RANDOM, q);
2962 }
2963
2964 if (sdkp->device->type == TYPE_ZBC) {
2965 /* Host-managed */
2966 q->limits.zoned = BLK_ZONED_HM;
2967 } else {
2968 sdkp->zoned = (buffer[8] >> 4) & 3;
2969 if (sdkp->zoned == 1)
2970 /* Host-aware */
2971 q->limits.zoned = BLK_ZONED_HA;
2972 else
2973 /*
2974 * Treat drive-managed devices as
2975 * regular block devices.
2976 */
2977 q->limits.zoned = BLK_ZONED_NONE;
2978 }
2979 if (blk_queue_is_zoned(q) && sdkp->first_scan)
2980 sd_printk(KERN_NOTICE, sdkp, "Host-%s zoned block device\n",
2981 q->limits.zoned == BLK_ZONED_HM ? "managed" : "aware");
2982
2983 out:
2984 kfree(buffer);
2985}
2986
2987/**
2988 * sd_read_block_provisioning - Query provisioning VPD page
2989 * @sdkp: disk to query
2990 */
2991static void sd_read_block_provisioning(struct scsi_disk *sdkp)
2992{
2993 unsigned char *buffer;
2994 const int vpd_len = 8;
2995
2996 if (sdkp->lbpme == 0)
2997 return;
2998
2999 buffer = kmalloc(vpd_len, GFP_KERNEL);
3000
3001 if (!buffer || scsi_get_vpd_page(sdkp->device, 0xb2, buffer, vpd_len))
3002 goto out;
3003
3004 sdkp->lbpvpd = 1;
3005 sdkp->lbpu = (buffer[5] >> 7) & 1; /* UNMAP */
3006 sdkp->lbpws = (buffer[5] >> 6) & 1; /* WRITE SAME(16) with UNMAP */
3007 sdkp->lbpws10 = (buffer[5] >> 5) & 1; /* WRITE SAME(10) with UNMAP */
3008
3009 out:
3010 kfree(buffer);
3011}
3012
3013static void sd_read_write_same(struct scsi_disk *sdkp, unsigned char *buffer)
3014{
3015 struct scsi_device *sdev = sdkp->device;
3016
3017 if (sdev->host->no_write_same) {
3018 sdev->no_write_same = 1;
3019
3020 return;
3021 }
3022
3023 if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE, INQUIRY) < 0) {
3024 /* too large values might cause issues with arcmsr */
3025 int vpd_buf_len = 64;
3026
3027 sdev->no_report_opcodes = 1;
3028
3029 /* Disable WRITE SAME if REPORT SUPPORTED OPERATION
3030 * CODES is unsupported and the device has an ATA
3031 * Information VPD page (SAT).
3032 */
3033 if (!scsi_get_vpd_page(sdev, 0x89, buffer, vpd_buf_len))
3034 sdev->no_write_same = 1;
3035 }
3036
3037 if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE, WRITE_SAME_16) == 1)
3038 sdkp->ws16 = 1;
3039
3040 if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE, WRITE_SAME) == 1)
3041 sdkp->ws10 = 1;
3042}
3043
3044static void sd_read_security(struct scsi_disk *sdkp, unsigned char *buffer)
3045{
3046 struct scsi_device *sdev = sdkp->device;
3047
3048 if (!sdev->security_supported)
3049 return;
3050
3051 if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE,
3052 SECURITY_PROTOCOL_IN) == 1 &&
3053 scsi_report_opcode(sdev, buffer, SD_BUF_SIZE,
3054 SECURITY_PROTOCOL_OUT) == 1)
3055 sdkp->security = 1;
3056}
3057
3058/**
3059 * sd_revalidate_disk - called the first time a new disk is seen,
3060 * performs disk spin up, read_capacity, etc.
3061 * @disk: struct gendisk we care about
3062 **/
3063static int sd_revalidate_disk(struct gendisk *disk)
3064{
3065 struct scsi_disk *sdkp = scsi_disk(disk);
3066 struct scsi_device *sdp = sdkp->device;
3067 struct request_queue *q = sdkp->disk->queue;
3068 sector_t old_capacity = sdkp->capacity;
3069 unsigned char *buffer;
3070 unsigned int dev_max, rw_max;
3071
3072 SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp,
3073 "sd_revalidate_disk\n"));
3074
3075 /*
3076 * If the device is offline, don't try and read capacity or any
3077 * of the other niceties.
3078 */
3079 if (!scsi_device_online(sdp))
3080 goto out;
3081
3082 buffer = kmalloc(SD_BUF_SIZE, GFP_KERNEL);
3083 if (!buffer) {
3084 sd_printk(KERN_WARNING, sdkp, "sd_revalidate_disk: Memory "
3085 "allocation failure.\n");
3086 goto out;
3087 }
3088
3089 sd_spinup_disk(sdkp);
3090
3091 /*
3092 * Without media there is no reason to ask; moreover, some devices
3093 * react badly if we do.
3094 */
3095 if (sdkp->media_present) {
3096 sd_read_capacity(sdkp, buffer);
3097
3098 if (scsi_device_supports_vpd(sdp)) {
3099 sd_read_block_provisioning(sdkp);
3100 sd_read_block_limits(sdkp);
3101 sd_read_block_characteristics(sdkp);
3102 sd_zbc_read_zones(sdkp, buffer);
3103 }
3104
3105 sd_print_capacity(sdkp, old_capacity);
3106
3107 sd_read_write_protect_flag(sdkp, buffer);
3108 sd_read_cache_type(sdkp, buffer);
3109 sd_read_app_tag_own(sdkp, buffer);
3110 sd_read_write_same(sdkp, buffer);
3111 sd_read_security(sdkp, buffer);
3112 }
3113
3114 /*
3115 * We now have all cache related info, determine how we deal
3116 * with flush requests.
3117 */
3118 sd_set_flush_flag(sdkp);
3119
3120 /* Initial block count limit based on CDB TRANSFER LENGTH field size. */
3121 dev_max = sdp->use_16_for_rw ? SD_MAX_XFER_BLOCKS : SD_DEF_XFER_BLOCKS;
3122
3123 /* Some devices report a maximum block count for READ/WRITE requests. */
3124 dev_max = min_not_zero(dev_max, sdkp->max_xfer_blocks);
3125 q->limits.max_dev_sectors = logical_to_sectors(sdp, dev_max);
3126
3127 /*
3128 * Determine the device's preferred I/O size for reads and writes
3129 * unless the reported value is unreasonably small, large, or
3130 * garbage.
3131 */
3132 if (sdkp->opt_xfer_blocks &&
3133 sdkp->opt_xfer_blocks <= dev_max &&
3134 sdkp->opt_xfer_blocks <= SD_DEF_XFER_BLOCKS &&
3135 logical_to_bytes(sdp, sdkp->opt_xfer_blocks) >= PAGE_SIZE) {
3136 q->limits.io_opt = logical_to_bytes(sdp, sdkp->opt_xfer_blocks);
3137 rw_max = logical_to_sectors(sdp, sdkp->opt_xfer_blocks);
3138 } else
3139 rw_max = min_not_zero(logical_to_sectors(sdp, dev_max),
3140 (sector_t)BLK_DEF_MAX_SECTORS);
3141
3142 /* Do not exceed controller limit */
3143 rw_max = min(rw_max, queue_max_hw_sectors(q));
3144
3145 /*
3146 * Only update max_sectors if previously unset or if the current value
3147 * exceeds the capabilities of the hardware.
3148 */
3149 if (sdkp->first_scan ||
3150 q->limits.max_sectors > q->limits.max_dev_sectors ||
3151 q->limits.max_sectors > q->limits.max_hw_sectors)
3152 q->limits.max_sectors = rw_max;
3153
3154 sdkp->first_scan = 0;
3155
3156 set_capacity(disk, logical_to_sectors(sdp, sdkp->capacity));
3157 sd_config_write_same(sdkp);
3158 kfree(buffer);
3159
3160 out:
3161 return 0;
3162}
3163
3164/**
3165 * sd_unlock_native_capacity - unlock native capacity
3166 * @disk: struct gendisk to set capacity for
3167 *
3168 * Block layer calls this function if it detects that partitions
3169 * on @disk reach beyond the end of the device. If the SCSI host
3170 * implements ->unlock_native_capacity() method, it's invoked to
3171 * give it a chance to adjust the device capacity.
3172 *
3173 * CONTEXT:
3174 * Defined by block layer. Might sleep.
3175 */
3176static void sd_unlock_native_capacity(struct gendisk *disk)
3177{
3178 struct scsi_device *sdev = scsi_disk(disk)->device;
3179
3180 if (sdev->host->hostt->unlock_native_capacity)
3181 sdev->host->hostt->unlock_native_capacity(sdev);
3182}
3183
3184/**
3185 * sd_format_disk_name - format disk name
3186 * @prefix: name prefix - ie. "sd" for SCSI disks
3187 * @index: index of the disk to format name for
3188 * @buf: output buffer
3189 * @buflen: length of the output buffer
3190 *
3191 * SCSI disk names starts at sda. The 26th device is sdz and the
3192 * 27th is sdaa. The last one for two lettered suffix is sdzz
3193 * which is followed by sdaaa.
3194 *
3195 * This is basically 26 base counting with one extra 'nil' entry
3196 * at the beginning from the second digit on and can be
3197 * determined using similar method as 26 base conversion with the
3198 * index shifted -1 after each digit is computed.
3199 *
3200 * CONTEXT:
3201 * Don't care.
3202 *
3203 * RETURNS:
3204 * 0 on success, -errno on failure.
3205 */
3206static int sd_format_disk_name(char *prefix, int index, char *buf, int buflen)
3207{
3208 const int base = 'z' - 'a' + 1;
3209 char *begin = buf + strlen(prefix);
3210 char *end = buf + buflen;
3211 char *p;
3212 int unit;
3213
3214 p = end - 1;
3215 *p = '\0';
3216 unit = base;
3217 do {
3218 if (p == begin)
3219 return -EINVAL;
3220 *--p = 'a' + (index % unit);
3221 index = (index / unit) - 1;
3222 } while (index >= 0);
3223
3224 memmove(begin, p, end - p);
3225 memcpy(buf, prefix, strlen(prefix));
3226
3227 return 0;
3228}
3229
3230/*
3231 * The asynchronous part of sd_probe
3232 */
3233static void sd_probe_async(void *data, async_cookie_t cookie)
3234{
3235 struct scsi_disk *sdkp = data;
3236 struct scsi_device *sdp;
3237 struct gendisk *gd;
3238 u32 index;
3239 struct device *dev;
3240
3241 sdp = sdkp->device;
3242 gd = sdkp->disk;
3243 index = sdkp->index;
3244 dev = &sdp->sdev_gendev;
3245
3246 gd->major = sd_major((index & 0xf0) >> 4);
3247 gd->first_minor = ((index & 0xf) << 4) | (index & 0xfff00);
3248
3249 gd->fops = &sd_fops;
3250 gd->private_data = &sdkp->driver;
3251 gd->queue = sdkp->device->request_queue;
3252
3253 /* defaults, until the device tells us otherwise */
3254 sdp->sector_size = 512;
3255 sdkp->capacity = 0;
3256 sdkp->media_present = 1;
3257 sdkp->write_prot = 0;
3258 sdkp->cache_override = 0;
3259 sdkp->WCE = 0;
3260 sdkp->RCD = 0;
3261 sdkp->ATO = 0;
3262 sdkp->first_scan = 1;
3263 sdkp->max_medium_access_timeouts = SD_MAX_MEDIUM_TIMEOUTS;
3264
3265 sd_revalidate_disk(gd);
3266
3267 gd->flags = GENHD_FL_EXT_DEVT;
3268 if (sdp->removable) {
3269 gd->flags |= GENHD_FL_REMOVABLE;
3270 gd->events |= DISK_EVENT_MEDIA_CHANGE;
3271 }
3272
3273 blk_pm_runtime_init(sdp->request_queue, dev);
3274 device_add_disk(dev, gd);
3275 if (sdkp->capacity)
3276 sd_dif_config_host(sdkp);
3277
3278 sd_revalidate_disk(gd);
3279
3280 if (sdkp->security) {
3281 sdkp->opal_dev = init_opal_dev(sdp, &sd_sec_submit);
3282 if (sdkp->opal_dev)
3283 sd_printk(KERN_NOTICE, sdkp, "supports TCG Opal\n");
3284 }
3285
3286 sd_printk(KERN_NOTICE, sdkp, "Attached SCSI %sdisk\n",
3287 sdp->removable ? "removable " : "");
3288 scsi_autopm_put_device(sdp);
3289 put_device(&sdkp->dev);
3290}
3291
3292/**
3293 * sd_probe - called during driver initialization and whenever a
3294 * new scsi device is attached to the system. It is called once
3295 * for each scsi device (not just disks) present.
3296 * @dev: pointer to device object
3297 *
3298 * Returns 0 if successful (or not interested in this scsi device
3299 * (e.g. scanner)); 1 when there is an error.
3300 *
3301 * Note: this function is invoked from the scsi mid-level.
3302 * This function sets up the mapping between a given
3303 * <host,channel,id,lun> (found in sdp) and new device name
3304 * (e.g. /dev/sda). More precisely it is the block device major
3305 * and minor number that is chosen here.
3306 *
3307 * Assume sd_probe is not re-entrant (for time being)
3308 * Also think about sd_probe() and sd_remove() running coincidentally.
3309 **/
3310static int sd_probe(struct device *dev)
3311{
3312 struct scsi_device *sdp = to_scsi_device(dev);
3313 struct scsi_disk *sdkp;
3314 struct gendisk *gd;
3315 int index;
3316 int error;
3317
3318 scsi_autopm_get_device(sdp);
3319 error = -ENODEV;
3320 if (sdp->type != TYPE_DISK &&
3321 sdp->type != TYPE_ZBC &&
3322 sdp->type != TYPE_MOD &&
3323 sdp->type != TYPE_RBC)
3324 goto out;
3325
3326#ifndef CONFIG_BLK_DEV_ZONED
3327 if (sdp->type == TYPE_ZBC)
3328 goto out;
3329#endif
3330 SCSI_LOG_HLQUEUE(3, sdev_printk(KERN_INFO, sdp,
3331 "sd_probe\n"));
3332
3333 error = -ENOMEM;
3334 sdkp = kzalloc(sizeof(*sdkp), GFP_KERNEL);
3335 if (!sdkp)
3336 goto out;
3337
3338 gd = alloc_disk(SD_MINORS);
3339 if (!gd)
3340 goto out_free;
3341
3342 do {
3343 if (!ida_pre_get(&sd_index_ida, GFP_KERNEL))
3344 goto out_put;
3345
3346 spin_lock(&sd_index_lock);
3347 error = ida_get_new(&sd_index_ida, &index);
3348 spin_unlock(&sd_index_lock);
3349 } while (error == -EAGAIN);
3350
3351 if (error) {
3352 sdev_printk(KERN_WARNING, sdp, "sd_probe: memory exhausted.\n");
3353 goto out_put;
3354 }
3355
3356 error = sd_format_disk_name("sd", index, gd->disk_name, DISK_NAME_LEN);
3357 if (error) {
3358 sdev_printk(KERN_WARNING, sdp, "SCSI disk (sd) name length exceeded.\n");
3359 goto out_free_index;
3360 }
3361
3362 sdkp->device = sdp;
3363 sdkp->driver = &sd_template;
3364 sdkp->disk = gd;
3365 sdkp->index = index;
3366 atomic_set(&sdkp->openers, 0);
3367 atomic_set(&sdkp->device->ioerr_cnt, 0);
3368
3369 if (!sdp->request_queue->rq_timeout) {
3370 if (sdp->type != TYPE_MOD)
3371 blk_queue_rq_timeout(sdp->request_queue, SD_TIMEOUT);
3372 else
3373 blk_queue_rq_timeout(sdp->request_queue,
3374 SD_MOD_TIMEOUT);
3375 }
3376
3377 device_initialize(&sdkp->dev);
3378 sdkp->dev.parent = dev;
3379 sdkp->dev.class = &sd_disk_class;
3380 dev_set_name(&sdkp->dev, "%s", dev_name(dev));
3381
3382 error = device_add(&sdkp->dev);
3383 if (error)
3384 goto out_free_index;
3385
3386 get_device(dev);
3387 dev_set_drvdata(dev, sdkp);
3388
3389 get_device(&sdkp->dev); /* prevent release before async_schedule */
3390 async_schedule_domain(sd_probe_async, sdkp, &scsi_sd_probe_domain);
3391
3392 return 0;
3393
3394 out_free_index:
3395 spin_lock(&sd_index_lock);
3396 ida_remove(&sd_index_ida, index);
3397 spin_unlock(&sd_index_lock);
3398 out_put:
3399 put_disk(gd);
3400 out_free:
3401 kfree(sdkp);
3402 out:
3403 scsi_autopm_put_device(sdp);
3404 return error;
3405}
3406
3407/**
3408 * sd_remove - called whenever a scsi disk (previously recognized by
3409 * sd_probe) is detached from the system. It is called (potentially
3410 * multiple times) during sd module unload.
3411 * @dev: pointer to device object
3412 *
3413 * Note: this function is invoked from the scsi mid-level.
3414 * This function potentially frees up a device name (e.g. /dev/sdc)
3415 * that could be re-used by a subsequent sd_probe().
3416 * This function is not called when the built-in sd driver is "exit-ed".
3417 **/
3418static int sd_remove(struct device *dev)
3419{
3420 struct scsi_disk *sdkp;
3421 dev_t devt;
3422
3423 sdkp = dev_get_drvdata(dev);
3424 devt = disk_devt(sdkp->disk);
3425 scsi_autopm_get_device(sdkp->device);
3426
3427 async_synchronize_full_domain(&scsi_sd_pm_domain);
3428 async_synchronize_full_domain(&scsi_sd_probe_domain);
3429 device_del(&sdkp->dev);
3430 del_gendisk(sdkp->disk);
3431 sd_shutdown(dev);
3432
3433 sd_zbc_remove(sdkp);
3434
3435 free_opal_dev(sdkp->opal_dev);
3436
3437 blk_register_region(devt, SD_MINORS, NULL,
3438 sd_default_probe, NULL, NULL);
3439
3440 mutex_lock(&sd_ref_mutex);
3441 dev_set_drvdata(dev, NULL);
3442 put_device(&sdkp->dev);
3443 mutex_unlock(&sd_ref_mutex);
3444
3445 return 0;
3446}
3447
3448/**
3449 * scsi_disk_release - Called to free the scsi_disk structure
3450 * @dev: pointer to embedded class device
3451 *
3452 * sd_ref_mutex must be held entering this routine. Because it is
3453 * called on last put, you should always use the scsi_disk_get()
3454 * scsi_disk_put() helpers which manipulate the semaphore directly
3455 * and never do a direct put_device.
3456 **/
3457static void scsi_disk_release(struct device *dev)
3458{
3459 struct scsi_disk *sdkp = to_scsi_disk(dev);
3460 struct gendisk *disk = sdkp->disk;
3461
3462 spin_lock(&sd_index_lock);
3463 ida_remove(&sd_index_ida, sdkp->index);
3464 spin_unlock(&sd_index_lock);
3465
3466 disk->private_data = NULL;
3467 put_disk(disk);
3468 put_device(&sdkp->device->sdev_gendev);
3469
3470 kfree(sdkp);
3471}
3472
3473static int sd_start_stop_device(struct scsi_disk *sdkp, int start)
3474{
3475 unsigned char cmd[6] = { START_STOP }; /* START_VALID */
3476 struct scsi_sense_hdr sshdr;
3477 struct scsi_device *sdp = sdkp->device;
3478 int res;
3479
3480 if (start)
3481 cmd[4] |= 1; /* START */
3482
3483 if (sdp->start_stop_pwr_cond)
3484 cmd[4] |= start ? 1 << 4 : 3 << 4; /* Active or Standby */
3485
3486 if (!scsi_device_online(sdp))
3487 return -ENODEV;
3488
3489 res = scsi_execute(sdp, cmd, DMA_NONE, NULL, 0, NULL, &sshdr,
3490 SD_TIMEOUT, SD_MAX_RETRIES, 0, RQF_PM, NULL);
3491 if (res) {
3492 sd_print_result(sdkp, "Start/Stop Unit failed", res);
3493 if (driver_byte(res) & DRIVER_SENSE)
3494 sd_print_sense_hdr(sdkp, &sshdr);
3495 if (scsi_sense_valid(&sshdr) &&
3496 /* 0x3a is medium not present */
3497 sshdr.asc == 0x3a)
3498 res = 0;
3499 }
3500
3501 /* SCSI error codes must not go to the generic layer */
3502 if (res)
3503 return -EIO;
3504
3505 return 0;
3506}
3507
3508/*
3509 * Send a SYNCHRONIZE CACHE instruction down to the device through
3510 * the normal SCSI command structure. Wait for the command to
3511 * complete.
3512 */
3513static void sd_shutdown(struct device *dev)
3514{
3515 struct scsi_disk *sdkp = dev_get_drvdata(dev);
3516
3517 if (!sdkp)
3518 return; /* this can happen */
3519
3520 if (pm_runtime_suspended(dev))
3521 return;
3522
3523 if (sdkp->WCE && sdkp->media_present) {
3524 sd_printk(KERN_NOTICE, sdkp, "Synchronizing SCSI cache\n");
3525 sd_sync_cache(sdkp, NULL);
3526 }
3527
3528 if (system_state != SYSTEM_RESTART && sdkp->device->manage_start_stop) {
3529 sd_printk(KERN_NOTICE, sdkp, "Stopping disk\n");
3530 sd_start_stop_device(sdkp, 0);
3531 }
3532}
3533
3534static int sd_suspend_common(struct device *dev, bool ignore_stop_errors)
3535{
3536 struct scsi_disk *sdkp = dev_get_drvdata(dev);
3537 struct scsi_sense_hdr sshdr;
3538 int ret = 0;
3539
3540 if (!sdkp) /* E.g.: runtime suspend following sd_remove() */
3541 return 0;
3542
3543 if (sdkp->WCE && sdkp->media_present) {
3544 sd_printk(KERN_NOTICE, sdkp, "Synchronizing SCSI cache\n");
3545 ret = sd_sync_cache(sdkp, &sshdr);
3546
3547 if (ret) {
3548 /* ignore OFFLINE device */
3549 if (ret == -ENODEV)
3550 return 0;
3551
3552 if (!scsi_sense_valid(&sshdr) ||
3553 sshdr.sense_key != ILLEGAL_REQUEST)
3554 return ret;
3555
3556 /*
3557 * sshdr.sense_key == ILLEGAL_REQUEST means this drive
3558 * doesn't support sync. There's not much to do and
3559 * suspend shouldn't fail.
3560 */
3561 ret = 0;
3562 }
3563 }
3564
3565 if (sdkp->device->manage_start_stop) {
3566 sd_printk(KERN_NOTICE, sdkp, "Stopping disk\n");
3567 /* an error is not worth aborting a system sleep */
3568 ret = sd_start_stop_device(sdkp, 0);
3569 if (ignore_stop_errors)
3570 ret = 0;
3571 }
3572
3573 return ret;
3574}
3575
3576static int sd_suspend_system(struct device *dev)
3577{
3578 return sd_suspend_common(dev, true);
3579}
3580
3581static int sd_suspend_runtime(struct device *dev)
3582{
3583 return sd_suspend_common(dev, false);
3584}
3585
3586static int sd_resume(struct device *dev)
3587{
3588 struct scsi_disk *sdkp = dev_get_drvdata(dev);
3589 int ret;
3590
3591 if (!sdkp) /* E.g.: runtime resume at the start of sd_probe() */
3592 return 0;
3593
3594 if (!sdkp->device->manage_start_stop)
3595 return 0;
3596
3597 sd_printk(KERN_NOTICE, sdkp, "Starting disk\n");
3598 ret = sd_start_stop_device(sdkp, 1);
3599 if (!ret)
3600 opal_unlock_from_suspend(sdkp->opal_dev);
3601 return ret;
3602}
3603
3604/**
3605 * init_sd - entry point for this driver (both when built in or when
3606 * a module).
3607 *
3608 * Note: this function registers this driver with the scsi mid-level.
3609 **/
3610static int __init init_sd(void)
3611{
3612 int majors = 0, i, err;
3613
3614 SCSI_LOG_HLQUEUE(3, printk("init_sd: sd driver entry point\n"));
3615
3616 for (i = 0; i < SD_MAJORS; i++) {
3617 if (register_blkdev(sd_major(i), "sd") != 0)
3618 continue;
3619 majors++;
3620 blk_register_region(sd_major(i), SD_MINORS, NULL,
3621 sd_default_probe, NULL, NULL);
3622 }
3623
3624 if (!majors)
3625 return -ENODEV;
3626
3627 err = class_register(&sd_disk_class);
3628 if (err)
3629 goto err_out;
3630
3631 sd_cdb_cache = kmem_cache_create("sd_ext_cdb", SD_EXT_CDB_SIZE,
3632 0, 0, NULL);
3633 if (!sd_cdb_cache) {
3634 printk(KERN_ERR "sd: can't init extended cdb cache\n");
3635 err = -ENOMEM;
3636 goto err_out_class;
3637 }
3638
3639 sd_cdb_pool = mempool_create_slab_pool(SD_MEMPOOL_SIZE, sd_cdb_cache);
3640 if (!sd_cdb_pool) {
3641 printk(KERN_ERR "sd: can't init extended cdb pool\n");
3642 err = -ENOMEM;
3643 goto err_out_cache;
3644 }
3645
3646 err = scsi_register_driver(&sd_template.gendrv);
3647 if (err)
3648 goto err_out_driver;
3649
3650 return 0;
3651
3652err_out_driver:
3653 mempool_destroy(sd_cdb_pool);
3654
3655err_out_cache:
3656 kmem_cache_destroy(sd_cdb_cache);
3657
3658err_out_class:
3659 class_unregister(&sd_disk_class);
3660err_out:
3661 for (i = 0; i < SD_MAJORS; i++)
3662 unregister_blkdev(sd_major(i), "sd");
3663 return err;
3664}
3665
3666/**
3667 * exit_sd - exit point for this driver (when it is a module).
3668 *
3669 * Note: this function unregisters this driver from the scsi mid-level.
3670 **/
3671static void __exit exit_sd(void)
3672{
3673 int i;
3674
3675 SCSI_LOG_HLQUEUE(3, printk("exit_sd: exiting sd driver\n"));
3676
3677 scsi_unregister_driver(&sd_template.gendrv);
3678 mempool_destroy(sd_cdb_pool);
3679 kmem_cache_destroy(sd_cdb_cache);
3680
3681 class_unregister(&sd_disk_class);
3682
3683 for (i = 0; i < SD_MAJORS; i++) {
3684 blk_unregister_region(sd_major(i), SD_MINORS);
3685 unregister_blkdev(sd_major(i), "sd");
3686 }
3687}
3688
3689module_init(init_sd);
3690module_exit(exit_sd);
3691
3692static void sd_print_sense_hdr(struct scsi_disk *sdkp,
3693 struct scsi_sense_hdr *sshdr)
3694{
3695 scsi_print_sense_hdr(sdkp->device,
3696 sdkp->disk ? sdkp->disk->disk_name : NULL, sshdr);
3697}
3698
3699static void sd_print_result(const struct scsi_disk *sdkp, const char *msg,
3700 int result)
3701{
3702 const char *hb_string = scsi_hostbyte_string(result);
3703 const char *db_string = scsi_driverbyte_string(result);
3704
3705 if (hb_string || db_string)
3706 sd_printk(KERN_INFO, sdkp,
3707 "%s: Result: hostbyte=%s driverbyte=%s\n", msg,
3708 hb_string ? hb_string : "invalid",
3709 db_string ? db_string : "invalid");
3710 else
3711 sd_printk(KERN_INFO, sdkp,
3712 "%s: Result: hostbyte=0x%02x driverbyte=0x%02x\n",
3713 msg, host_byte(result), driver_byte(result));
3714}
3715
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/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/blk-pm.h>
49#include <linux/delay.h>
50#include <linux/rw_hint.h>
51#include <linux/major.h>
52#include <linux/mutex.h>
53#include <linux/string_helpers.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#include <scsi/scsi_common.h>
72
73#include "sd.h"
74#include "scsi_priv.h"
75#include "scsi_logging.h"
76
77MODULE_AUTHOR("Eric Youngdale");
78MODULE_DESCRIPTION("SCSI disk (sd) driver");
79MODULE_LICENSE("GPL");
80
81MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK0_MAJOR);
82MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK1_MAJOR);
83MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK2_MAJOR);
84MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK3_MAJOR);
85MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK4_MAJOR);
86MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK5_MAJOR);
87MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK6_MAJOR);
88MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK7_MAJOR);
89MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK8_MAJOR);
90MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK9_MAJOR);
91MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK10_MAJOR);
92MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK11_MAJOR);
93MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK12_MAJOR);
94MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK13_MAJOR);
95MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK14_MAJOR);
96MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK15_MAJOR);
97MODULE_ALIAS_SCSI_DEVICE(TYPE_DISK);
98MODULE_ALIAS_SCSI_DEVICE(TYPE_MOD);
99MODULE_ALIAS_SCSI_DEVICE(TYPE_RBC);
100MODULE_ALIAS_SCSI_DEVICE(TYPE_ZBC);
101
102#define SD_MINORS 16
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 void sd_shutdown(struct device *);
109static void sd_read_capacity(struct scsi_disk *sdkp, unsigned char *buffer);
110static void scsi_disk_release(struct device *cdev);
111
112static DEFINE_IDA(sd_index_ida);
113
114static mempool_t *sd_page_pool;
115static struct lock_class_key sd_bio_compl_lkclass;
116
117static const char *sd_cache_types[] = {
118 "write through", "none", "write back",
119 "write back, no read (daft)"
120};
121
122static void sd_set_flush_flag(struct scsi_disk *sdkp)
123{
124 bool wc = false, fua = false;
125
126 if (sdkp->WCE) {
127 wc = true;
128 if (sdkp->DPOFUA)
129 fua = true;
130 }
131
132 blk_queue_write_cache(sdkp->disk->queue, wc, fua);
133}
134
135static ssize_t
136cache_type_store(struct device *dev, struct device_attribute *attr,
137 const char *buf, size_t count)
138{
139 int ct, rcd, wce, sp;
140 struct scsi_disk *sdkp = to_scsi_disk(dev);
141 struct scsi_device *sdp = sdkp->device;
142 char buffer[64];
143 char *buffer_data;
144 struct scsi_mode_data data;
145 struct scsi_sense_hdr sshdr;
146 static const char temp[] = "temporary ";
147 int len, ret;
148
149 if (sdp->type != TYPE_DISK && sdp->type != TYPE_ZBC)
150 /* no cache control on RBC devices; theoretically they
151 * can do it, but there's probably so many exceptions
152 * it's not worth the risk */
153 return -EINVAL;
154
155 if (strncmp(buf, temp, sizeof(temp) - 1) == 0) {
156 buf += sizeof(temp) - 1;
157 sdkp->cache_override = 1;
158 } else {
159 sdkp->cache_override = 0;
160 }
161
162 ct = sysfs_match_string(sd_cache_types, buf);
163 if (ct < 0)
164 return -EINVAL;
165
166 rcd = ct & 0x01 ? 1 : 0;
167 wce = (ct & 0x02) && !sdkp->write_prot ? 1 : 0;
168
169 if (sdkp->cache_override) {
170 sdkp->WCE = wce;
171 sdkp->RCD = rcd;
172 sd_set_flush_flag(sdkp);
173 return count;
174 }
175
176 if (scsi_mode_sense(sdp, 0x08, 8, 0, buffer, sizeof(buffer), SD_TIMEOUT,
177 sdkp->max_retries, &data, NULL))
178 return -EINVAL;
179 len = min_t(size_t, sizeof(buffer), data.length - data.header_length -
180 data.block_descriptor_length);
181 buffer_data = buffer + data.header_length +
182 data.block_descriptor_length;
183 buffer_data[2] &= ~0x05;
184 buffer_data[2] |= wce << 2 | rcd;
185 sp = buffer_data[0] & 0x80 ? 1 : 0;
186 buffer_data[0] &= ~0x80;
187
188 /*
189 * Ensure WP, DPOFUA, and RESERVED fields are cleared in
190 * received mode parameter buffer before doing MODE SELECT.
191 */
192 data.device_specific = 0;
193
194 ret = scsi_mode_select(sdp, 1, sp, buffer_data, len, SD_TIMEOUT,
195 sdkp->max_retries, &data, &sshdr);
196 if (ret) {
197 if (ret > 0 && scsi_sense_valid(&sshdr))
198 sd_print_sense_hdr(sdkp, &sshdr);
199 return -EINVAL;
200 }
201 sd_revalidate_disk(sdkp->disk);
202 return count;
203}
204
205static ssize_t
206manage_start_stop_show(struct device *dev,
207 struct device_attribute *attr, char *buf)
208{
209 struct scsi_disk *sdkp = to_scsi_disk(dev);
210 struct scsi_device *sdp = sdkp->device;
211
212 return sysfs_emit(buf, "%u\n",
213 sdp->manage_system_start_stop &&
214 sdp->manage_runtime_start_stop &&
215 sdp->manage_shutdown);
216}
217static DEVICE_ATTR_RO(manage_start_stop);
218
219static ssize_t
220manage_system_start_stop_show(struct device *dev,
221 struct device_attribute *attr, char *buf)
222{
223 struct scsi_disk *sdkp = to_scsi_disk(dev);
224 struct scsi_device *sdp = sdkp->device;
225
226 return sysfs_emit(buf, "%u\n", sdp->manage_system_start_stop);
227}
228
229static ssize_t
230manage_system_start_stop_store(struct device *dev,
231 struct device_attribute *attr,
232 const char *buf, size_t count)
233{
234 struct scsi_disk *sdkp = to_scsi_disk(dev);
235 struct scsi_device *sdp = sdkp->device;
236 bool v;
237
238 if (!capable(CAP_SYS_ADMIN))
239 return -EACCES;
240
241 if (kstrtobool(buf, &v))
242 return -EINVAL;
243
244 sdp->manage_system_start_stop = v;
245
246 return count;
247}
248static DEVICE_ATTR_RW(manage_system_start_stop);
249
250static ssize_t
251manage_runtime_start_stop_show(struct device *dev,
252 struct device_attribute *attr, char *buf)
253{
254 struct scsi_disk *sdkp = to_scsi_disk(dev);
255 struct scsi_device *sdp = sdkp->device;
256
257 return sysfs_emit(buf, "%u\n", sdp->manage_runtime_start_stop);
258}
259
260static ssize_t
261manage_runtime_start_stop_store(struct device *dev,
262 struct device_attribute *attr,
263 const char *buf, size_t count)
264{
265 struct scsi_disk *sdkp = to_scsi_disk(dev);
266 struct scsi_device *sdp = sdkp->device;
267 bool v;
268
269 if (!capable(CAP_SYS_ADMIN))
270 return -EACCES;
271
272 if (kstrtobool(buf, &v))
273 return -EINVAL;
274
275 sdp->manage_runtime_start_stop = v;
276
277 return count;
278}
279static DEVICE_ATTR_RW(manage_runtime_start_stop);
280
281static ssize_t manage_shutdown_show(struct device *dev,
282 struct device_attribute *attr, char *buf)
283{
284 struct scsi_disk *sdkp = to_scsi_disk(dev);
285 struct scsi_device *sdp = sdkp->device;
286
287 return sysfs_emit(buf, "%u\n", sdp->manage_shutdown);
288}
289
290static ssize_t manage_shutdown_store(struct device *dev,
291 struct device_attribute *attr,
292 const char *buf, size_t count)
293{
294 struct scsi_disk *sdkp = to_scsi_disk(dev);
295 struct scsi_device *sdp = sdkp->device;
296 bool v;
297
298 if (!capable(CAP_SYS_ADMIN))
299 return -EACCES;
300
301 if (kstrtobool(buf, &v))
302 return -EINVAL;
303
304 sdp->manage_shutdown = v;
305
306 return count;
307}
308static DEVICE_ATTR_RW(manage_shutdown);
309
310static ssize_t
311allow_restart_show(struct device *dev, struct device_attribute *attr, char *buf)
312{
313 struct scsi_disk *sdkp = to_scsi_disk(dev);
314
315 return sprintf(buf, "%u\n", sdkp->device->allow_restart);
316}
317
318static ssize_t
319allow_restart_store(struct device *dev, struct device_attribute *attr,
320 const char *buf, size_t count)
321{
322 bool v;
323 struct scsi_disk *sdkp = to_scsi_disk(dev);
324 struct scsi_device *sdp = sdkp->device;
325
326 if (!capable(CAP_SYS_ADMIN))
327 return -EACCES;
328
329 if (sdp->type != TYPE_DISK && sdp->type != TYPE_ZBC)
330 return -EINVAL;
331
332 if (kstrtobool(buf, &v))
333 return -EINVAL;
334
335 sdp->allow_restart = v;
336
337 return count;
338}
339static DEVICE_ATTR_RW(allow_restart);
340
341static ssize_t
342cache_type_show(struct device *dev, struct device_attribute *attr, char *buf)
343{
344 struct scsi_disk *sdkp = to_scsi_disk(dev);
345 int ct = sdkp->RCD + 2*sdkp->WCE;
346
347 return sprintf(buf, "%s\n", sd_cache_types[ct]);
348}
349static DEVICE_ATTR_RW(cache_type);
350
351static ssize_t
352FUA_show(struct device *dev, struct device_attribute *attr, char *buf)
353{
354 struct scsi_disk *sdkp = to_scsi_disk(dev);
355
356 return sprintf(buf, "%u\n", sdkp->DPOFUA);
357}
358static DEVICE_ATTR_RO(FUA);
359
360static ssize_t
361protection_type_show(struct device *dev, struct device_attribute *attr,
362 char *buf)
363{
364 struct scsi_disk *sdkp = to_scsi_disk(dev);
365
366 return sprintf(buf, "%u\n", sdkp->protection_type);
367}
368
369static ssize_t
370protection_type_store(struct device *dev, struct device_attribute *attr,
371 const char *buf, size_t count)
372{
373 struct scsi_disk *sdkp = to_scsi_disk(dev);
374 unsigned int val;
375 int err;
376
377 if (!capable(CAP_SYS_ADMIN))
378 return -EACCES;
379
380 err = kstrtouint(buf, 10, &val);
381
382 if (err)
383 return err;
384
385 if (val <= T10_PI_TYPE3_PROTECTION)
386 sdkp->protection_type = val;
387
388 return count;
389}
390static DEVICE_ATTR_RW(protection_type);
391
392static ssize_t
393protection_mode_show(struct device *dev, struct device_attribute *attr,
394 char *buf)
395{
396 struct scsi_disk *sdkp = to_scsi_disk(dev);
397 struct scsi_device *sdp = sdkp->device;
398 unsigned int dif, dix;
399
400 dif = scsi_host_dif_capable(sdp->host, sdkp->protection_type);
401 dix = scsi_host_dix_capable(sdp->host, sdkp->protection_type);
402
403 if (!dix && scsi_host_dix_capable(sdp->host, T10_PI_TYPE0_PROTECTION)) {
404 dif = 0;
405 dix = 1;
406 }
407
408 if (!dif && !dix)
409 return sprintf(buf, "none\n");
410
411 return sprintf(buf, "%s%u\n", dix ? "dix" : "dif", dif);
412}
413static DEVICE_ATTR_RO(protection_mode);
414
415static ssize_t
416app_tag_own_show(struct device *dev, struct device_attribute *attr, char *buf)
417{
418 struct scsi_disk *sdkp = to_scsi_disk(dev);
419
420 return sprintf(buf, "%u\n", sdkp->ATO);
421}
422static DEVICE_ATTR_RO(app_tag_own);
423
424static ssize_t
425thin_provisioning_show(struct device *dev, struct device_attribute *attr,
426 char *buf)
427{
428 struct scsi_disk *sdkp = to_scsi_disk(dev);
429
430 return sprintf(buf, "%u\n", sdkp->lbpme);
431}
432static DEVICE_ATTR_RO(thin_provisioning);
433
434/* sysfs_match_string() requires dense arrays */
435static const char *lbp_mode[] = {
436 [SD_LBP_FULL] = "full",
437 [SD_LBP_UNMAP] = "unmap",
438 [SD_LBP_WS16] = "writesame_16",
439 [SD_LBP_WS10] = "writesame_10",
440 [SD_LBP_ZERO] = "writesame_zero",
441 [SD_LBP_DISABLE] = "disabled",
442};
443
444static ssize_t
445provisioning_mode_show(struct device *dev, struct device_attribute *attr,
446 char *buf)
447{
448 struct scsi_disk *sdkp = to_scsi_disk(dev);
449
450 return sprintf(buf, "%s\n", lbp_mode[sdkp->provisioning_mode]);
451}
452
453static ssize_t
454provisioning_mode_store(struct device *dev, struct device_attribute *attr,
455 const char *buf, size_t count)
456{
457 struct scsi_disk *sdkp = to_scsi_disk(dev);
458 struct scsi_device *sdp = sdkp->device;
459 int mode;
460
461 if (!capable(CAP_SYS_ADMIN))
462 return -EACCES;
463
464 if (sd_is_zoned(sdkp)) {
465 sd_config_discard(sdkp, SD_LBP_DISABLE);
466 return count;
467 }
468
469 if (sdp->type != TYPE_DISK)
470 return -EINVAL;
471
472 mode = sysfs_match_string(lbp_mode, buf);
473 if (mode < 0)
474 return -EINVAL;
475
476 sd_config_discard(sdkp, mode);
477
478 return count;
479}
480static DEVICE_ATTR_RW(provisioning_mode);
481
482/* sysfs_match_string() requires dense arrays */
483static const char *zeroing_mode[] = {
484 [SD_ZERO_WRITE] = "write",
485 [SD_ZERO_WS] = "writesame",
486 [SD_ZERO_WS16_UNMAP] = "writesame_16_unmap",
487 [SD_ZERO_WS10_UNMAP] = "writesame_10_unmap",
488};
489
490static ssize_t
491zeroing_mode_show(struct device *dev, struct device_attribute *attr,
492 char *buf)
493{
494 struct scsi_disk *sdkp = to_scsi_disk(dev);
495
496 return sprintf(buf, "%s\n", zeroing_mode[sdkp->zeroing_mode]);
497}
498
499static ssize_t
500zeroing_mode_store(struct device *dev, struct device_attribute *attr,
501 const char *buf, size_t count)
502{
503 struct scsi_disk *sdkp = to_scsi_disk(dev);
504 int mode;
505
506 if (!capable(CAP_SYS_ADMIN))
507 return -EACCES;
508
509 mode = sysfs_match_string(zeroing_mode, buf);
510 if (mode < 0)
511 return -EINVAL;
512
513 sdkp->zeroing_mode = mode;
514
515 return count;
516}
517static DEVICE_ATTR_RW(zeroing_mode);
518
519static ssize_t
520max_medium_access_timeouts_show(struct device *dev,
521 struct device_attribute *attr, char *buf)
522{
523 struct scsi_disk *sdkp = to_scsi_disk(dev);
524
525 return sprintf(buf, "%u\n", sdkp->max_medium_access_timeouts);
526}
527
528static ssize_t
529max_medium_access_timeouts_store(struct device *dev,
530 struct device_attribute *attr, const char *buf,
531 size_t count)
532{
533 struct scsi_disk *sdkp = to_scsi_disk(dev);
534 int err;
535
536 if (!capable(CAP_SYS_ADMIN))
537 return -EACCES;
538
539 err = kstrtouint(buf, 10, &sdkp->max_medium_access_timeouts);
540
541 return err ? err : count;
542}
543static DEVICE_ATTR_RW(max_medium_access_timeouts);
544
545static ssize_t
546max_write_same_blocks_show(struct device *dev, struct device_attribute *attr,
547 char *buf)
548{
549 struct scsi_disk *sdkp = to_scsi_disk(dev);
550
551 return sprintf(buf, "%u\n", sdkp->max_ws_blocks);
552}
553
554static ssize_t
555max_write_same_blocks_store(struct device *dev, struct device_attribute *attr,
556 const char *buf, size_t count)
557{
558 struct scsi_disk *sdkp = to_scsi_disk(dev);
559 struct scsi_device *sdp = sdkp->device;
560 unsigned long max;
561 int err;
562
563 if (!capable(CAP_SYS_ADMIN))
564 return -EACCES;
565
566 if (sdp->type != TYPE_DISK && sdp->type != TYPE_ZBC)
567 return -EINVAL;
568
569 err = kstrtoul(buf, 10, &max);
570
571 if (err)
572 return err;
573
574 if (max == 0)
575 sdp->no_write_same = 1;
576 else if (max <= SD_MAX_WS16_BLOCKS) {
577 sdp->no_write_same = 0;
578 sdkp->max_ws_blocks = max;
579 }
580
581 sd_config_write_same(sdkp);
582
583 return count;
584}
585static DEVICE_ATTR_RW(max_write_same_blocks);
586
587static ssize_t
588zoned_cap_show(struct device *dev, struct device_attribute *attr, char *buf)
589{
590 struct scsi_disk *sdkp = to_scsi_disk(dev);
591
592 if (sdkp->device->type == TYPE_ZBC)
593 return sprintf(buf, "host-managed\n");
594 if (sdkp->zoned == 1)
595 return sprintf(buf, "host-aware\n");
596 if (sdkp->zoned == 2)
597 return sprintf(buf, "drive-managed\n");
598 return sprintf(buf, "none\n");
599}
600static DEVICE_ATTR_RO(zoned_cap);
601
602static ssize_t
603max_retries_store(struct device *dev, struct device_attribute *attr,
604 const char *buf, size_t count)
605{
606 struct scsi_disk *sdkp = to_scsi_disk(dev);
607 struct scsi_device *sdev = sdkp->device;
608 int retries, err;
609
610 err = kstrtoint(buf, 10, &retries);
611 if (err)
612 return err;
613
614 if (retries == SCSI_CMD_RETRIES_NO_LIMIT || retries <= SD_MAX_RETRIES) {
615 sdkp->max_retries = retries;
616 return count;
617 }
618
619 sdev_printk(KERN_ERR, sdev, "max_retries must be between -1 and %d\n",
620 SD_MAX_RETRIES);
621 return -EINVAL;
622}
623
624static ssize_t
625max_retries_show(struct device *dev, struct device_attribute *attr,
626 char *buf)
627{
628 struct scsi_disk *sdkp = to_scsi_disk(dev);
629
630 return sprintf(buf, "%d\n", sdkp->max_retries);
631}
632
633static DEVICE_ATTR_RW(max_retries);
634
635static struct attribute *sd_disk_attrs[] = {
636 &dev_attr_cache_type.attr,
637 &dev_attr_FUA.attr,
638 &dev_attr_allow_restart.attr,
639 &dev_attr_manage_start_stop.attr,
640 &dev_attr_manage_system_start_stop.attr,
641 &dev_attr_manage_runtime_start_stop.attr,
642 &dev_attr_manage_shutdown.attr,
643 &dev_attr_protection_type.attr,
644 &dev_attr_protection_mode.attr,
645 &dev_attr_app_tag_own.attr,
646 &dev_attr_thin_provisioning.attr,
647 &dev_attr_provisioning_mode.attr,
648 &dev_attr_zeroing_mode.attr,
649 &dev_attr_max_write_same_blocks.attr,
650 &dev_attr_max_medium_access_timeouts.attr,
651 &dev_attr_zoned_cap.attr,
652 &dev_attr_max_retries.attr,
653 NULL,
654};
655ATTRIBUTE_GROUPS(sd_disk);
656
657static struct class sd_disk_class = {
658 .name = "scsi_disk",
659 .dev_release = scsi_disk_release,
660 .dev_groups = sd_disk_groups,
661};
662
663/*
664 * Don't request a new module, as that could deadlock in multipath
665 * environment.
666 */
667static void sd_default_probe(dev_t devt)
668{
669}
670
671/*
672 * Device no to disk mapping:
673 *
674 * major disc2 disc p1
675 * |............|.............|....|....| <- dev_t
676 * 31 20 19 8 7 4 3 0
677 *
678 * Inside a major, we have 16k disks, however mapped non-
679 * contiguously. The first 16 disks are for major0, the next
680 * ones with major1, ... Disk 256 is for major0 again, disk 272
681 * for major1, ...
682 * As we stay compatible with our numbering scheme, we can reuse
683 * the well-know SCSI majors 8, 65--71, 136--143.
684 */
685static int sd_major(int major_idx)
686{
687 switch (major_idx) {
688 case 0:
689 return SCSI_DISK0_MAJOR;
690 case 1 ... 7:
691 return SCSI_DISK1_MAJOR + major_idx - 1;
692 case 8 ... 15:
693 return SCSI_DISK8_MAJOR + major_idx - 8;
694 default:
695 BUG();
696 return 0; /* shut up gcc */
697 }
698}
699
700#ifdef CONFIG_BLK_SED_OPAL
701static int sd_sec_submit(void *data, u16 spsp, u8 secp, void *buffer,
702 size_t len, bool send)
703{
704 struct scsi_disk *sdkp = data;
705 struct scsi_device *sdev = sdkp->device;
706 u8 cdb[12] = { 0, };
707 const struct scsi_exec_args exec_args = {
708 .req_flags = BLK_MQ_REQ_PM,
709 };
710 int ret;
711
712 cdb[0] = send ? SECURITY_PROTOCOL_OUT : SECURITY_PROTOCOL_IN;
713 cdb[1] = secp;
714 put_unaligned_be16(spsp, &cdb[2]);
715 put_unaligned_be32(len, &cdb[6]);
716
717 ret = scsi_execute_cmd(sdev, cdb, send ? REQ_OP_DRV_OUT : REQ_OP_DRV_IN,
718 buffer, len, SD_TIMEOUT, sdkp->max_retries,
719 &exec_args);
720 return ret <= 0 ? ret : -EIO;
721}
722#endif /* CONFIG_BLK_SED_OPAL */
723
724/*
725 * Look up the DIX operation based on whether the command is read or
726 * write and whether dix and dif are enabled.
727 */
728static unsigned int sd_prot_op(bool write, bool dix, bool dif)
729{
730 /* Lookup table: bit 2 (write), bit 1 (dix), bit 0 (dif) */
731 static const unsigned int ops[] = { /* wrt dix dif */
732 SCSI_PROT_NORMAL, /* 0 0 0 */
733 SCSI_PROT_READ_STRIP, /* 0 0 1 */
734 SCSI_PROT_READ_INSERT, /* 0 1 0 */
735 SCSI_PROT_READ_PASS, /* 0 1 1 */
736 SCSI_PROT_NORMAL, /* 1 0 0 */
737 SCSI_PROT_WRITE_INSERT, /* 1 0 1 */
738 SCSI_PROT_WRITE_STRIP, /* 1 1 0 */
739 SCSI_PROT_WRITE_PASS, /* 1 1 1 */
740 };
741
742 return ops[write << 2 | dix << 1 | dif];
743}
744
745/*
746 * Returns a mask of the protection flags that are valid for a given DIX
747 * operation.
748 */
749static unsigned int sd_prot_flag_mask(unsigned int prot_op)
750{
751 static const unsigned int flag_mask[] = {
752 [SCSI_PROT_NORMAL] = 0,
753
754 [SCSI_PROT_READ_STRIP] = SCSI_PROT_TRANSFER_PI |
755 SCSI_PROT_GUARD_CHECK |
756 SCSI_PROT_REF_CHECK |
757 SCSI_PROT_REF_INCREMENT,
758
759 [SCSI_PROT_READ_INSERT] = SCSI_PROT_REF_INCREMENT |
760 SCSI_PROT_IP_CHECKSUM,
761
762 [SCSI_PROT_READ_PASS] = SCSI_PROT_TRANSFER_PI |
763 SCSI_PROT_GUARD_CHECK |
764 SCSI_PROT_REF_CHECK |
765 SCSI_PROT_REF_INCREMENT |
766 SCSI_PROT_IP_CHECKSUM,
767
768 [SCSI_PROT_WRITE_INSERT] = SCSI_PROT_TRANSFER_PI |
769 SCSI_PROT_REF_INCREMENT,
770
771 [SCSI_PROT_WRITE_STRIP] = SCSI_PROT_GUARD_CHECK |
772 SCSI_PROT_REF_CHECK |
773 SCSI_PROT_REF_INCREMENT |
774 SCSI_PROT_IP_CHECKSUM,
775
776 [SCSI_PROT_WRITE_PASS] = SCSI_PROT_TRANSFER_PI |
777 SCSI_PROT_GUARD_CHECK |
778 SCSI_PROT_REF_CHECK |
779 SCSI_PROT_REF_INCREMENT |
780 SCSI_PROT_IP_CHECKSUM,
781 };
782
783 return flag_mask[prot_op];
784}
785
786static unsigned char sd_setup_protect_cmnd(struct scsi_cmnd *scmd,
787 unsigned int dix, unsigned int dif)
788{
789 struct request *rq = scsi_cmd_to_rq(scmd);
790 struct bio *bio = rq->bio;
791 unsigned int prot_op = sd_prot_op(rq_data_dir(rq), dix, dif);
792 unsigned int protect = 0;
793
794 if (dix) { /* DIX Type 0, 1, 2, 3 */
795 if (bio_integrity_flagged(bio, BIP_IP_CHECKSUM))
796 scmd->prot_flags |= SCSI_PROT_IP_CHECKSUM;
797
798 if (bio_integrity_flagged(bio, BIP_CTRL_NOCHECK) == false)
799 scmd->prot_flags |= SCSI_PROT_GUARD_CHECK;
800 }
801
802 if (dif != T10_PI_TYPE3_PROTECTION) { /* DIX/DIF Type 0, 1, 2 */
803 scmd->prot_flags |= SCSI_PROT_REF_INCREMENT;
804
805 if (bio_integrity_flagged(bio, BIP_CTRL_NOCHECK) == false)
806 scmd->prot_flags |= SCSI_PROT_REF_CHECK;
807 }
808
809 if (dif) { /* DIX/DIF Type 1, 2, 3 */
810 scmd->prot_flags |= SCSI_PROT_TRANSFER_PI;
811
812 if (bio_integrity_flagged(bio, BIP_DISK_NOCHECK))
813 protect = 3 << 5; /* Disable target PI checking */
814 else
815 protect = 1 << 5; /* Enable target PI checking */
816 }
817
818 scsi_set_prot_op(scmd, prot_op);
819 scsi_set_prot_type(scmd, dif);
820 scmd->prot_flags &= sd_prot_flag_mask(prot_op);
821
822 return protect;
823}
824
825static void sd_config_discard(struct scsi_disk *sdkp, unsigned int mode)
826{
827 struct request_queue *q = sdkp->disk->queue;
828 unsigned int logical_block_size = sdkp->device->sector_size;
829 unsigned int max_blocks = 0;
830
831 q->limits.discard_alignment =
832 sdkp->unmap_alignment * logical_block_size;
833 q->limits.discard_granularity =
834 max(sdkp->physical_block_size,
835 sdkp->unmap_granularity * logical_block_size);
836 sdkp->provisioning_mode = mode;
837
838 switch (mode) {
839
840 case SD_LBP_FULL:
841 case SD_LBP_DISABLE:
842 blk_queue_max_discard_sectors(q, 0);
843 return;
844
845 case SD_LBP_UNMAP:
846 max_blocks = min_not_zero(sdkp->max_unmap_blocks,
847 (u32)SD_MAX_WS16_BLOCKS);
848 break;
849
850 case SD_LBP_WS16:
851 if (sdkp->device->unmap_limit_for_ws)
852 max_blocks = sdkp->max_unmap_blocks;
853 else
854 max_blocks = sdkp->max_ws_blocks;
855
856 max_blocks = min_not_zero(max_blocks, (u32)SD_MAX_WS16_BLOCKS);
857 break;
858
859 case SD_LBP_WS10:
860 if (sdkp->device->unmap_limit_for_ws)
861 max_blocks = sdkp->max_unmap_blocks;
862 else
863 max_blocks = sdkp->max_ws_blocks;
864
865 max_blocks = min_not_zero(max_blocks, (u32)SD_MAX_WS10_BLOCKS);
866 break;
867
868 case SD_LBP_ZERO:
869 max_blocks = min_not_zero(sdkp->max_ws_blocks,
870 (u32)SD_MAX_WS10_BLOCKS);
871 break;
872 }
873
874 blk_queue_max_discard_sectors(q, max_blocks * (logical_block_size >> 9));
875}
876
877static void *sd_set_special_bvec(struct request *rq, unsigned int data_len)
878{
879 struct page *page;
880
881 page = mempool_alloc(sd_page_pool, GFP_ATOMIC);
882 if (!page)
883 return NULL;
884 clear_highpage(page);
885 bvec_set_page(&rq->special_vec, page, data_len, 0);
886 rq->rq_flags |= RQF_SPECIAL_PAYLOAD;
887 return bvec_virt(&rq->special_vec);
888}
889
890static blk_status_t sd_setup_unmap_cmnd(struct scsi_cmnd *cmd)
891{
892 struct scsi_device *sdp = cmd->device;
893 struct request *rq = scsi_cmd_to_rq(cmd);
894 struct scsi_disk *sdkp = scsi_disk(rq->q->disk);
895 u64 lba = sectors_to_logical(sdp, blk_rq_pos(rq));
896 u32 nr_blocks = sectors_to_logical(sdp, blk_rq_sectors(rq));
897 unsigned int data_len = 24;
898 char *buf;
899
900 buf = sd_set_special_bvec(rq, data_len);
901 if (!buf)
902 return BLK_STS_RESOURCE;
903
904 cmd->cmd_len = 10;
905 cmd->cmnd[0] = UNMAP;
906 cmd->cmnd[8] = 24;
907
908 put_unaligned_be16(6 + 16, &buf[0]);
909 put_unaligned_be16(16, &buf[2]);
910 put_unaligned_be64(lba, &buf[8]);
911 put_unaligned_be32(nr_blocks, &buf[16]);
912
913 cmd->allowed = sdkp->max_retries;
914 cmd->transfersize = data_len;
915 rq->timeout = SD_TIMEOUT;
916
917 return scsi_alloc_sgtables(cmd);
918}
919
920static blk_status_t sd_setup_write_same16_cmnd(struct scsi_cmnd *cmd,
921 bool unmap)
922{
923 struct scsi_device *sdp = cmd->device;
924 struct request *rq = scsi_cmd_to_rq(cmd);
925 struct scsi_disk *sdkp = scsi_disk(rq->q->disk);
926 u64 lba = sectors_to_logical(sdp, blk_rq_pos(rq));
927 u32 nr_blocks = sectors_to_logical(sdp, blk_rq_sectors(rq));
928 u32 data_len = sdp->sector_size;
929
930 if (!sd_set_special_bvec(rq, data_len))
931 return BLK_STS_RESOURCE;
932
933 cmd->cmd_len = 16;
934 cmd->cmnd[0] = WRITE_SAME_16;
935 if (unmap)
936 cmd->cmnd[1] = 0x8; /* UNMAP */
937 put_unaligned_be64(lba, &cmd->cmnd[2]);
938 put_unaligned_be32(nr_blocks, &cmd->cmnd[10]);
939
940 cmd->allowed = sdkp->max_retries;
941 cmd->transfersize = data_len;
942 rq->timeout = unmap ? SD_TIMEOUT : SD_WRITE_SAME_TIMEOUT;
943
944 return scsi_alloc_sgtables(cmd);
945}
946
947static blk_status_t sd_setup_write_same10_cmnd(struct scsi_cmnd *cmd,
948 bool unmap)
949{
950 struct scsi_device *sdp = cmd->device;
951 struct request *rq = scsi_cmd_to_rq(cmd);
952 struct scsi_disk *sdkp = scsi_disk(rq->q->disk);
953 u64 lba = sectors_to_logical(sdp, blk_rq_pos(rq));
954 u32 nr_blocks = sectors_to_logical(sdp, blk_rq_sectors(rq));
955 u32 data_len = sdp->sector_size;
956
957 if (!sd_set_special_bvec(rq, data_len))
958 return BLK_STS_RESOURCE;
959
960 cmd->cmd_len = 10;
961 cmd->cmnd[0] = WRITE_SAME;
962 if (unmap)
963 cmd->cmnd[1] = 0x8; /* UNMAP */
964 put_unaligned_be32(lba, &cmd->cmnd[2]);
965 put_unaligned_be16(nr_blocks, &cmd->cmnd[7]);
966
967 cmd->allowed = sdkp->max_retries;
968 cmd->transfersize = data_len;
969 rq->timeout = unmap ? SD_TIMEOUT : SD_WRITE_SAME_TIMEOUT;
970
971 return scsi_alloc_sgtables(cmd);
972}
973
974static blk_status_t sd_setup_write_zeroes_cmnd(struct scsi_cmnd *cmd)
975{
976 struct request *rq = scsi_cmd_to_rq(cmd);
977 struct scsi_device *sdp = cmd->device;
978 struct scsi_disk *sdkp = scsi_disk(rq->q->disk);
979 u64 lba = sectors_to_logical(sdp, blk_rq_pos(rq));
980 u32 nr_blocks = sectors_to_logical(sdp, blk_rq_sectors(rq));
981
982 if (!(rq->cmd_flags & REQ_NOUNMAP)) {
983 switch (sdkp->zeroing_mode) {
984 case SD_ZERO_WS16_UNMAP:
985 return sd_setup_write_same16_cmnd(cmd, true);
986 case SD_ZERO_WS10_UNMAP:
987 return sd_setup_write_same10_cmnd(cmd, true);
988 }
989 }
990
991 if (sdp->no_write_same) {
992 rq->rq_flags |= RQF_QUIET;
993 return BLK_STS_TARGET;
994 }
995
996 if (sdkp->ws16 || lba > 0xffffffff || nr_blocks > 0xffff)
997 return sd_setup_write_same16_cmnd(cmd, false);
998
999 return sd_setup_write_same10_cmnd(cmd, false);
1000}
1001
1002static void sd_config_write_same(struct scsi_disk *sdkp)
1003{
1004 struct request_queue *q = sdkp->disk->queue;
1005 unsigned int logical_block_size = sdkp->device->sector_size;
1006
1007 if (sdkp->device->no_write_same) {
1008 sdkp->max_ws_blocks = 0;
1009 goto out;
1010 }
1011
1012 /* Some devices can not handle block counts above 0xffff despite
1013 * supporting WRITE SAME(16). Consequently we default to 64k
1014 * blocks per I/O unless the device explicitly advertises a
1015 * bigger limit.
1016 */
1017 if (sdkp->max_ws_blocks > SD_MAX_WS10_BLOCKS)
1018 sdkp->max_ws_blocks = min_not_zero(sdkp->max_ws_blocks,
1019 (u32)SD_MAX_WS16_BLOCKS);
1020 else if (sdkp->ws16 || sdkp->ws10 || sdkp->device->no_report_opcodes)
1021 sdkp->max_ws_blocks = min_not_zero(sdkp->max_ws_blocks,
1022 (u32)SD_MAX_WS10_BLOCKS);
1023 else {
1024 sdkp->device->no_write_same = 1;
1025 sdkp->max_ws_blocks = 0;
1026 }
1027
1028 if (sdkp->lbprz && sdkp->lbpws)
1029 sdkp->zeroing_mode = SD_ZERO_WS16_UNMAP;
1030 else if (sdkp->lbprz && sdkp->lbpws10)
1031 sdkp->zeroing_mode = SD_ZERO_WS10_UNMAP;
1032 else if (sdkp->max_ws_blocks)
1033 sdkp->zeroing_mode = SD_ZERO_WS;
1034 else
1035 sdkp->zeroing_mode = SD_ZERO_WRITE;
1036
1037 if (sdkp->max_ws_blocks &&
1038 sdkp->physical_block_size > logical_block_size) {
1039 /*
1040 * Reporting a maximum number of blocks that is not aligned
1041 * on the device physical size would cause a large write same
1042 * request to be split into physically unaligned chunks by
1043 * __blkdev_issue_write_zeroes() even if the caller of this
1044 * functions took care to align the large request. So make sure
1045 * the maximum reported is aligned to the device physical block
1046 * size. This is only an optional optimization for regular
1047 * disks, but this is mandatory to avoid failure of large write
1048 * same requests directed at sequential write required zones of
1049 * host-managed ZBC disks.
1050 */
1051 sdkp->max_ws_blocks =
1052 round_down(sdkp->max_ws_blocks,
1053 bytes_to_logical(sdkp->device,
1054 sdkp->physical_block_size));
1055 }
1056
1057out:
1058 blk_queue_max_write_zeroes_sectors(q, sdkp->max_ws_blocks *
1059 (logical_block_size >> 9));
1060}
1061
1062static blk_status_t sd_setup_flush_cmnd(struct scsi_cmnd *cmd)
1063{
1064 struct request *rq = scsi_cmd_to_rq(cmd);
1065 struct scsi_disk *sdkp = scsi_disk(rq->q->disk);
1066
1067 /* flush requests don't perform I/O, zero the S/G table */
1068 memset(&cmd->sdb, 0, sizeof(cmd->sdb));
1069
1070 if (cmd->device->use_16_for_sync) {
1071 cmd->cmnd[0] = SYNCHRONIZE_CACHE_16;
1072 cmd->cmd_len = 16;
1073 } else {
1074 cmd->cmnd[0] = SYNCHRONIZE_CACHE;
1075 cmd->cmd_len = 10;
1076 }
1077 cmd->transfersize = 0;
1078 cmd->allowed = sdkp->max_retries;
1079
1080 rq->timeout = rq->q->rq_timeout * SD_FLUSH_TIMEOUT_MULTIPLIER;
1081 return BLK_STS_OK;
1082}
1083
1084/**
1085 * sd_group_number() - Compute the GROUP NUMBER field
1086 * @cmd: SCSI command for which to compute the value of the six-bit GROUP NUMBER
1087 * field.
1088 *
1089 * From SBC-5 r05 (https://www.t10.org/cgi-bin/ac.pl?t=f&f=sbc5r05.pdf):
1090 * 0: no relative lifetime.
1091 * 1: shortest relative lifetime.
1092 * 2: second shortest relative lifetime.
1093 * 3 - 0x3d: intermediate relative lifetimes.
1094 * 0x3e: second longest relative lifetime.
1095 * 0x3f: longest relative lifetime.
1096 */
1097static u8 sd_group_number(struct scsi_cmnd *cmd)
1098{
1099 const struct request *rq = scsi_cmd_to_rq(cmd);
1100 struct scsi_disk *sdkp = scsi_disk(rq->q->disk);
1101
1102 if (!sdkp->rscs)
1103 return 0;
1104
1105 return min3((u32)rq->write_hint, (u32)sdkp->permanent_stream_count,
1106 0x3fu);
1107}
1108
1109static blk_status_t sd_setup_rw32_cmnd(struct scsi_cmnd *cmd, bool write,
1110 sector_t lba, unsigned int nr_blocks,
1111 unsigned char flags, unsigned int dld)
1112{
1113 cmd->cmd_len = SD_EXT_CDB_SIZE;
1114 cmd->cmnd[0] = VARIABLE_LENGTH_CMD;
1115 cmd->cmnd[6] = sd_group_number(cmd);
1116 cmd->cmnd[7] = 0x18; /* Additional CDB len */
1117 cmd->cmnd[9] = write ? WRITE_32 : READ_32;
1118 cmd->cmnd[10] = flags;
1119 cmd->cmnd[11] = dld & 0x07;
1120 put_unaligned_be64(lba, &cmd->cmnd[12]);
1121 put_unaligned_be32(lba, &cmd->cmnd[20]); /* Expected Indirect LBA */
1122 put_unaligned_be32(nr_blocks, &cmd->cmnd[28]);
1123
1124 return BLK_STS_OK;
1125}
1126
1127static blk_status_t sd_setup_rw16_cmnd(struct scsi_cmnd *cmd, bool write,
1128 sector_t lba, unsigned int nr_blocks,
1129 unsigned char flags, unsigned int dld)
1130{
1131 cmd->cmd_len = 16;
1132 cmd->cmnd[0] = write ? WRITE_16 : READ_16;
1133 cmd->cmnd[1] = flags | ((dld >> 2) & 0x01);
1134 cmd->cmnd[14] = ((dld & 0x03) << 6) | sd_group_number(cmd);
1135 cmd->cmnd[15] = 0;
1136 put_unaligned_be64(lba, &cmd->cmnd[2]);
1137 put_unaligned_be32(nr_blocks, &cmd->cmnd[10]);
1138
1139 return BLK_STS_OK;
1140}
1141
1142static blk_status_t sd_setup_rw10_cmnd(struct scsi_cmnd *cmd, bool write,
1143 sector_t lba, unsigned int nr_blocks,
1144 unsigned char flags)
1145{
1146 cmd->cmd_len = 10;
1147 cmd->cmnd[0] = write ? WRITE_10 : READ_10;
1148 cmd->cmnd[1] = flags;
1149 cmd->cmnd[6] = sd_group_number(cmd);
1150 cmd->cmnd[9] = 0;
1151 put_unaligned_be32(lba, &cmd->cmnd[2]);
1152 put_unaligned_be16(nr_blocks, &cmd->cmnd[7]);
1153
1154 return BLK_STS_OK;
1155}
1156
1157static blk_status_t sd_setup_rw6_cmnd(struct scsi_cmnd *cmd, bool write,
1158 sector_t lba, unsigned int nr_blocks,
1159 unsigned char flags)
1160{
1161 /* Avoid that 0 blocks gets translated into 256 blocks. */
1162 if (WARN_ON_ONCE(nr_blocks == 0))
1163 return BLK_STS_IOERR;
1164
1165 if (unlikely(flags & 0x8)) {
1166 /*
1167 * This happens only if this drive failed 10byte rw
1168 * command with ILLEGAL_REQUEST during operation and
1169 * thus turned off use_10_for_rw.
1170 */
1171 scmd_printk(KERN_ERR, cmd, "FUA write on READ/WRITE(6) drive\n");
1172 return BLK_STS_IOERR;
1173 }
1174
1175 cmd->cmd_len = 6;
1176 cmd->cmnd[0] = write ? WRITE_6 : READ_6;
1177 cmd->cmnd[1] = (lba >> 16) & 0x1f;
1178 cmd->cmnd[2] = (lba >> 8) & 0xff;
1179 cmd->cmnd[3] = lba & 0xff;
1180 cmd->cmnd[4] = nr_blocks;
1181 cmd->cmnd[5] = 0;
1182
1183 return BLK_STS_OK;
1184}
1185
1186/*
1187 * Check if a command has a duration limit set. If it does, and the target
1188 * device supports CDL and the feature is enabled, return the limit
1189 * descriptor index to use. Return 0 (no limit) otherwise.
1190 */
1191static int sd_cdl_dld(struct scsi_disk *sdkp, struct scsi_cmnd *scmd)
1192{
1193 struct scsi_device *sdp = sdkp->device;
1194 int hint;
1195
1196 if (!sdp->cdl_supported || !sdp->cdl_enable)
1197 return 0;
1198
1199 /*
1200 * Use "no limit" if the request ioprio does not specify a duration
1201 * limit hint.
1202 */
1203 hint = IOPRIO_PRIO_HINT(req_get_ioprio(scsi_cmd_to_rq(scmd)));
1204 if (hint < IOPRIO_HINT_DEV_DURATION_LIMIT_1 ||
1205 hint > IOPRIO_HINT_DEV_DURATION_LIMIT_7)
1206 return 0;
1207
1208 return (hint - IOPRIO_HINT_DEV_DURATION_LIMIT_1) + 1;
1209}
1210
1211static blk_status_t sd_setup_read_write_cmnd(struct scsi_cmnd *cmd)
1212{
1213 struct request *rq = scsi_cmd_to_rq(cmd);
1214 struct scsi_device *sdp = cmd->device;
1215 struct scsi_disk *sdkp = scsi_disk(rq->q->disk);
1216 sector_t lba = sectors_to_logical(sdp, blk_rq_pos(rq));
1217 sector_t threshold;
1218 unsigned int nr_blocks = sectors_to_logical(sdp, blk_rq_sectors(rq));
1219 unsigned int mask = logical_to_sectors(sdp, 1) - 1;
1220 bool write = rq_data_dir(rq) == WRITE;
1221 unsigned char protect, fua;
1222 unsigned int dld;
1223 blk_status_t ret;
1224 unsigned int dif;
1225 bool dix;
1226
1227 ret = scsi_alloc_sgtables(cmd);
1228 if (ret != BLK_STS_OK)
1229 return ret;
1230
1231 ret = BLK_STS_IOERR;
1232 if (!scsi_device_online(sdp) || sdp->changed) {
1233 scmd_printk(KERN_ERR, cmd, "device offline or changed\n");
1234 goto fail;
1235 }
1236
1237 if (blk_rq_pos(rq) + blk_rq_sectors(rq) > get_capacity(rq->q->disk)) {
1238 scmd_printk(KERN_ERR, cmd, "access beyond end of device\n");
1239 goto fail;
1240 }
1241
1242 if ((blk_rq_pos(rq) & mask) || (blk_rq_sectors(rq) & mask)) {
1243 scmd_printk(KERN_ERR, cmd, "request not aligned to the logical block size\n");
1244 goto fail;
1245 }
1246
1247 /*
1248 * Some SD card readers can't handle accesses which touch the
1249 * last one or two logical blocks. Split accesses as needed.
1250 */
1251 threshold = sdkp->capacity - SD_LAST_BUGGY_SECTORS;
1252
1253 if (unlikely(sdp->last_sector_bug && lba + nr_blocks > threshold)) {
1254 if (lba < threshold) {
1255 /* Access up to the threshold but not beyond */
1256 nr_blocks = threshold - lba;
1257 } else {
1258 /* Access only a single logical block */
1259 nr_blocks = 1;
1260 }
1261 }
1262
1263 if (req_op(rq) == REQ_OP_ZONE_APPEND) {
1264 ret = sd_zbc_prepare_zone_append(cmd, &lba, nr_blocks);
1265 if (ret)
1266 goto fail;
1267 }
1268
1269 fua = rq->cmd_flags & REQ_FUA ? 0x8 : 0;
1270 dix = scsi_prot_sg_count(cmd);
1271 dif = scsi_host_dif_capable(cmd->device->host, sdkp->protection_type);
1272 dld = sd_cdl_dld(sdkp, cmd);
1273
1274 if (dif || dix)
1275 protect = sd_setup_protect_cmnd(cmd, dix, dif);
1276 else
1277 protect = 0;
1278
1279 if (protect && sdkp->protection_type == T10_PI_TYPE2_PROTECTION) {
1280 ret = sd_setup_rw32_cmnd(cmd, write, lba, nr_blocks,
1281 protect | fua, dld);
1282 } else if (sdp->use_16_for_rw || (nr_blocks > 0xffff)) {
1283 ret = sd_setup_rw16_cmnd(cmd, write, lba, nr_blocks,
1284 protect | fua, dld);
1285 } else if ((nr_blocks > 0xff) || (lba > 0x1fffff) ||
1286 sdp->use_10_for_rw || protect || rq->write_hint) {
1287 ret = sd_setup_rw10_cmnd(cmd, write, lba, nr_blocks,
1288 protect | fua);
1289 } else {
1290 ret = sd_setup_rw6_cmnd(cmd, write, lba, nr_blocks,
1291 protect | fua);
1292 }
1293
1294 if (unlikely(ret != BLK_STS_OK))
1295 goto fail;
1296
1297 /*
1298 * We shouldn't disconnect in the middle of a sector, so with a dumb
1299 * host adapter, it's safe to assume that we can at least transfer
1300 * this many bytes between each connect / disconnect.
1301 */
1302 cmd->transfersize = sdp->sector_size;
1303 cmd->underflow = nr_blocks << 9;
1304 cmd->allowed = sdkp->max_retries;
1305 cmd->sdb.length = nr_blocks * sdp->sector_size;
1306
1307 SCSI_LOG_HLQUEUE(1,
1308 scmd_printk(KERN_INFO, cmd,
1309 "%s: block=%llu, count=%d\n", __func__,
1310 (unsigned long long)blk_rq_pos(rq),
1311 blk_rq_sectors(rq)));
1312 SCSI_LOG_HLQUEUE(2,
1313 scmd_printk(KERN_INFO, cmd,
1314 "%s %d/%u 512 byte blocks.\n",
1315 write ? "writing" : "reading", nr_blocks,
1316 blk_rq_sectors(rq)));
1317
1318 /*
1319 * This indicates that the command is ready from our end to be queued.
1320 */
1321 return BLK_STS_OK;
1322fail:
1323 scsi_free_sgtables(cmd);
1324 return ret;
1325}
1326
1327static blk_status_t sd_init_command(struct scsi_cmnd *cmd)
1328{
1329 struct request *rq = scsi_cmd_to_rq(cmd);
1330
1331 switch (req_op(rq)) {
1332 case REQ_OP_DISCARD:
1333 switch (scsi_disk(rq->q->disk)->provisioning_mode) {
1334 case SD_LBP_UNMAP:
1335 return sd_setup_unmap_cmnd(cmd);
1336 case SD_LBP_WS16:
1337 return sd_setup_write_same16_cmnd(cmd, true);
1338 case SD_LBP_WS10:
1339 return sd_setup_write_same10_cmnd(cmd, true);
1340 case SD_LBP_ZERO:
1341 return sd_setup_write_same10_cmnd(cmd, false);
1342 default:
1343 return BLK_STS_TARGET;
1344 }
1345 case REQ_OP_WRITE_ZEROES:
1346 return sd_setup_write_zeroes_cmnd(cmd);
1347 case REQ_OP_FLUSH:
1348 return sd_setup_flush_cmnd(cmd);
1349 case REQ_OP_READ:
1350 case REQ_OP_WRITE:
1351 case REQ_OP_ZONE_APPEND:
1352 return sd_setup_read_write_cmnd(cmd);
1353 case REQ_OP_ZONE_RESET:
1354 return sd_zbc_setup_zone_mgmt_cmnd(cmd, ZO_RESET_WRITE_POINTER,
1355 false);
1356 case REQ_OP_ZONE_RESET_ALL:
1357 return sd_zbc_setup_zone_mgmt_cmnd(cmd, ZO_RESET_WRITE_POINTER,
1358 true);
1359 case REQ_OP_ZONE_OPEN:
1360 return sd_zbc_setup_zone_mgmt_cmnd(cmd, ZO_OPEN_ZONE, false);
1361 case REQ_OP_ZONE_CLOSE:
1362 return sd_zbc_setup_zone_mgmt_cmnd(cmd, ZO_CLOSE_ZONE, false);
1363 case REQ_OP_ZONE_FINISH:
1364 return sd_zbc_setup_zone_mgmt_cmnd(cmd, ZO_FINISH_ZONE, false);
1365 default:
1366 WARN_ON_ONCE(1);
1367 return BLK_STS_NOTSUPP;
1368 }
1369}
1370
1371static void sd_uninit_command(struct scsi_cmnd *SCpnt)
1372{
1373 struct request *rq = scsi_cmd_to_rq(SCpnt);
1374
1375 if (rq->rq_flags & RQF_SPECIAL_PAYLOAD)
1376 mempool_free(rq->special_vec.bv_page, sd_page_pool);
1377}
1378
1379static bool sd_need_revalidate(struct gendisk *disk, struct scsi_disk *sdkp)
1380{
1381 if (sdkp->device->removable || sdkp->write_prot) {
1382 if (disk_check_media_change(disk))
1383 return true;
1384 }
1385
1386 /*
1387 * Force a full rescan after ioctl(BLKRRPART). While the disk state has
1388 * nothing to do with partitions, BLKRRPART is used to force a full
1389 * revalidate after things like a format for historical reasons.
1390 */
1391 return test_bit(GD_NEED_PART_SCAN, &disk->state);
1392}
1393
1394/**
1395 * sd_open - open a scsi disk device
1396 * @disk: disk to open
1397 * @mode: open mode
1398 *
1399 * Returns 0 if successful. Returns a negated errno value in case
1400 * of error.
1401 *
1402 * Note: This can be called from a user context (e.g. fsck(1) )
1403 * or from within the kernel (e.g. as a result of a mount(1) ).
1404 * In the latter case @inode and @filp carry an abridged amount
1405 * of information as noted above.
1406 *
1407 * Locking: called with disk->open_mutex held.
1408 **/
1409static int sd_open(struct gendisk *disk, blk_mode_t mode)
1410{
1411 struct scsi_disk *sdkp = scsi_disk(disk);
1412 struct scsi_device *sdev = sdkp->device;
1413 int retval;
1414
1415 if (scsi_device_get(sdev))
1416 return -ENXIO;
1417
1418 SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp, "sd_open\n"));
1419
1420 /*
1421 * If the device is in error recovery, wait until it is done.
1422 * If the device is offline, then disallow any access to it.
1423 */
1424 retval = -ENXIO;
1425 if (!scsi_block_when_processing_errors(sdev))
1426 goto error_out;
1427
1428 if (sd_need_revalidate(disk, sdkp))
1429 sd_revalidate_disk(disk);
1430
1431 /*
1432 * If the drive is empty, just let the open fail.
1433 */
1434 retval = -ENOMEDIUM;
1435 if (sdev->removable && !sdkp->media_present &&
1436 !(mode & BLK_OPEN_NDELAY))
1437 goto error_out;
1438
1439 /*
1440 * If the device has the write protect tab set, have the open fail
1441 * if the user expects to be able to write to the thing.
1442 */
1443 retval = -EROFS;
1444 if (sdkp->write_prot && (mode & BLK_OPEN_WRITE))
1445 goto error_out;
1446
1447 /*
1448 * It is possible that the disk changing stuff resulted in
1449 * the device being taken offline. If this is the case,
1450 * report this to the user, and don't pretend that the
1451 * open actually succeeded.
1452 */
1453 retval = -ENXIO;
1454 if (!scsi_device_online(sdev))
1455 goto error_out;
1456
1457 if ((atomic_inc_return(&sdkp->openers) == 1) && sdev->removable) {
1458 if (scsi_block_when_processing_errors(sdev))
1459 scsi_set_medium_removal(sdev, SCSI_REMOVAL_PREVENT);
1460 }
1461
1462 return 0;
1463
1464error_out:
1465 scsi_device_put(sdev);
1466 return retval;
1467}
1468
1469/**
1470 * sd_release - invoked when the (last) close(2) is called on this
1471 * scsi disk.
1472 * @disk: disk to release
1473 *
1474 * Returns 0.
1475 *
1476 * Note: may block (uninterruptible) if error recovery is underway
1477 * on this disk.
1478 *
1479 * Locking: called with disk->open_mutex held.
1480 **/
1481static void sd_release(struct gendisk *disk)
1482{
1483 struct scsi_disk *sdkp = scsi_disk(disk);
1484 struct scsi_device *sdev = sdkp->device;
1485
1486 SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp, "sd_release\n"));
1487
1488 if (atomic_dec_return(&sdkp->openers) == 0 && sdev->removable) {
1489 if (scsi_block_when_processing_errors(sdev))
1490 scsi_set_medium_removal(sdev, SCSI_REMOVAL_ALLOW);
1491 }
1492
1493 scsi_device_put(sdev);
1494}
1495
1496static int sd_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1497{
1498 struct scsi_disk *sdkp = scsi_disk(bdev->bd_disk);
1499 struct scsi_device *sdp = sdkp->device;
1500 struct Scsi_Host *host = sdp->host;
1501 sector_t capacity = logical_to_sectors(sdp, sdkp->capacity);
1502 int diskinfo[4];
1503
1504 /* default to most commonly used values */
1505 diskinfo[0] = 0x40; /* 1 << 6 */
1506 diskinfo[1] = 0x20; /* 1 << 5 */
1507 diskinfo[2] = capacity >> 11;
1508
1509 /* override with calculated, extended default, or driver values */
1510 if (host->hostt->bios_param)
1511 host->hostt->bios_param(sdp, bdev, capacity, diskinfo);
1512 else
1513 scsicam_bios_param(bdev, capacity, diskinfo);
1514
1515 geo->heads = diskinfo[0];
1516 geo->sectors = diskinfo[1];
1517 geo->cylinders = diskinfo[2];
1518 return 0;
1519}
1520
1521/**
1522 * sd_ioctl - process an ioctl
1523 * @bdev: target block device
1524 * @mode: open mode
1525 * @cmd: ioctl command number
1526 * @arg: this is third argument given to ioctl(2) system call.
1527 * Often contains a pointer.
1528 *
1529 * Returns 0 if successful (some ioctls return positive numbers on
1530 * success as well). Returns a negated errno value in case of error.
1531 *
1532 * Note: most ioctls are forward onto the block subsystem or further
1533 * down in the scsi subsystem.
1534 **/
1535static int sd_ioctl(struct block_device *bdev, blk_mode_t mode,
1536 unsigned int cmd, unsigned long arg)
1537{
1538 struct gendisk *disk = bdev->bd_disk;
1539 struct scsi_disk *sdkp = scsi_disk(disk);
1540 struct scsi_device *sdp = sdkp->device;
1541 void __user *p = (void __user *)arg;
1542 int error;
1543
1544 SCSI_LOG_IOCTL(1, sd_printk(KERN_INFO, sdkp, "sd_ioctl: disk=%s, "
1545 "cmd=0x%x\n", disk->disk_name, cmd));
1546
1547 if (bdev_is_partition(bdev) && !capable(CAP_SYS_RAWIO))
1548 return -ENOIOCTLCMD;
1549
1550 /*
1551 * If we are in the middle of error recovery, don't let anyone
1552 * else try and use this device. Also, if error recovery fails, it
1553 * may try and take the device offline, in which case all further
1554 * access to the device is prohibited.
1555 */
1556 error = scsi_ioctl_block_when_processing_errors(sdp, cmd,
1557 (mode & BLK_OPEN_NDELAY));
1558 if (error)
1559 return error;
1560
1561 if (is_sed_ioctl(cmd))
1562 return sed_ioctl(sdkp->opal_dev, cmd, p);
1563 return scsi_ioctl(sdp, mode & BLK_OPEN_WRITE, cmd, p);
1564}
1565
1566static void set_media_not_present(struct scsi_disk *sdkp)
1567{
1568 if (sdkp->media_present)
1569 sdkp->device->changed = 1;
1570
1571 if (sdkp->device->removable) {
1572 sdkp->media_present = 0;
1573 sdkp->capacity = 0;
1574 }
1575}
1576
1577static int media_not_present(struct scsi_disk *sdkp,
1578 struct scsi_sense_hdr *sshdr)
1579{
1580 if (!scsi_sense_valid(sshdr))
1581 return 0;
1582
1583 /* not invoked for commands that could return deferred errors */
1584 switch (sshdr->sense_key) {
1585 case UNIT_ATTENTION:
1586 case NOT_READY:
1587 /* medium not present */
1588 if (sshdr->asc == 0x3A) {
1589 set_media_not_present(sdkp);
1590 return 1;
1591 }
1592 }
1593 return 0;
1594}
1595
1596/**
1597 * sd_check_events - check media events
1598 * @disk: kernel device descriptor
1599 * @clearing: disk events currently being cleared
1600 *
1601 * Returns mask of DISK_EVENT_*.
1602 *
1603 * Note: this function is invoked from the block subsystem.
1604 **/
1605static unsigned int sd_check_events(struct gendisk *disk, unsigned int clearing)
1606{
1607 struct scsi_disk *sdkp = disk->private_data;
1608 struct scsi_device *sdp;
1609 int retval;
1610 bool disk_changed;
1611
1612 if (!sdkp)
1613 return 0;
1614
1615 sdp = sdkp->device;
1616 SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp, "sd_check_events\n"));
1617
1618 /*
1619 * If the device is offline, don't send any commands - just pretend as
1620 * if the command failed. If the device ever comes back online, we
1621 * can deal with it then. It is only because of unrecoverable errors
1622 * that we would ever take a device offline in the first place.
1623 */
1624 if (!scsi_device_online(sdp)) {
1625 set_media_not_present(sdkp);
1626 goto out;
1627 }
1628
1629 /*
1630 * Using TEST_UNIT_READY enables differentiation between drive with
1631 * no cartridge loaded - NOT READY, drive with changed cartridge -
1632 * UNIT ATTENTION, or with same cartridge - GOOD STATUS.
1633 *
1634 * Drives that auto spin down. eg iomega jaz 1G, will be started
1635 * by sd_spinup_disk() from sd_revalidate_disk(), which happens whenever
1636 * sd_revalidate() is called.
1637 */
1638 if (scsi_block_when_processing_errors(sdp)) {
1639 struct scsi_sense_hdr sshdr = { 0, };
1640
1641 retval = scsi_test_unit_ready(sdp, SD_TIMEOUT, sdkp->max_retries,
1642 &sshdr);
1643
1644 /* failed to execute TUR, assume media not present */
1645 if (retval < 0 || host_byte(retval)) {
1646 set_media_not_present(sdkp);
1647 goto out;
1648 }
1649
1650 if (media_not_present(sdkp, &sshdr))
1651 goto out;
1652 }
1653
1654 /*
1655 * For removable scsi disk we have to recognise the presence
1656 * of a disk in the drive.
1657 */
1658 if (!sdkp->media_present)
1659 sdp->changed = 1;
1660 sdkp->media_present = 1;
1661out:
1662 /*
1663 * sdp->changed is set under the following conditions:
1664 *
1665 * Medium present state has changed in either direction.
1666 * Device has indicated UNIT_ATTENTION.
1667 */
1668 disk_changed = sdp->changed;
1669 sdp->changed = 0;
1670 return disk_changed ? DISK_EVENT_MEDIA_CHANGE : 0;
1671}
1672
1673static int sd_sync_cache(struct scsi_disk *sdkp)
1674{
1675 int res;
1676 struct scsi_device *sdp = sdkp->device;
1677 const int timeout = sdp->request_queue->rq_timeout
1678 * SD_FLUSH_TIMEOUT_MULTIPLIER;
1679 /* Leave the rest of the command zero to indicate flush everything. */
1680 const unsigned char cmd[16] = { sdp->use_16_for_sync ?
1681 SYNCHRONIZE_CACHE_16 : SYNCHRONIZE_CACHE };
1682 struct scsi_sense_hdr sshdr;
1683 struct scsi_failure failure_defs[] = {
1684 {
1685 .allowed = 3,
1686 .result = SCMD_FAILURE_RESULT_ANY,
1687 },
1688 {}
1689 };
1690 struct scsi_failures failures = {
1691 .failure_definitions = failure_defs,
1692 };
1693 const struct scsi_exec_args exec_args = {
1694 .req_flags = BLK_MQ_REQ_PM,
1695 .sshdr = &sshdr,
1696 .failures = &failures,
1697 };
1698
1699 if (!scsi_device_online(sdp))
1700 return -ENODEV;
1701
1702 res = scsi_execute_cmd(sdp, cmd, REQ_OP_DRV_IN, NULL, 0, timeout,
1703 sdkp->max_retries, &exec_args);
1704 if (res) {
1705 sd_print_result(sdkp, "Synchronize Cache(10) failed", res);
1706
1707 if (res < 0)
1708 return res;
1709
1710 if (scsi_status_is_check_condition(res) &&
1711 scsi_sense_valid(&sshdr)) {
1712 sd_print_sense_hdr(sdkp, &sshdr);
1713
1714 /* we need to evaluate the error return */
1715 if (sshdr.asc == 0x3a || /* medium not present */
1716 sshdr.asc == 0x20 || /* invalid command */
1717 (sshdr.asc == 0x74 && sshdr.ascq == 0x71)) /* drive is password locked */
1718 /* this is no error here */
1719 return 0;
1720 /*
1721 * This drive doesn't support sync and there's not much
1722 * we can do because this is called during shutdown
1723 * or suspend so just return success so those operations
1724 * can proceed.
1725 */
1726 if (sshdr.sense_key == ILLEGAL_REQUEST)
1727 return 0;
1728 }
1729
1730 switch (host_byte(res)) {
1731 /* ignore errors due to racing a disconnection */
1732 case DID_BAD_TARGET:
1733 case DID_NO_CONNECT:
1734 return 0;
1735 /* signal the upper layer it might try again */
1736 case DID_BUS_BUSY:
1737 case DID_IMM_RETRY:
1738 case DID_REQUEUE:
1739 case DID_SOFT_ERROR:
1740 return -EBUSY;
1741 default:
1742 return -EIO;
1743 }
1744 }
1745 return 0;
1746}
1747
1748static void sd_rescan(struct device *dev)
1749{
1750 struct scsi_disk *sdkp = dev_get_drvdata(dev);
1751
1752 sd_revalidate_disk(sdkp->disk);
1753}
1754
1755static int sd_get_unique_id(struct gendisk *disk, u8 id[16],
1756 enum blk_unique_id type)
1757{
1758 struct scsi_device *sdev = scsi_disk(disk)->device;
1759 const struct scsi_vpd *vpd;
1760 const unsigned char *d;
1761 int ret = -ENXIO, len;
1762
1763 rcu_read_lock();
1764 vpd = rcu_dereference(sdev->vpd_pg83);
1765 if (!vpd)
1766 goto out_unlock;
1767
1768 ret = -EINVAL;
1769 for (d = vpd->data + 4; d < vpd->data + vpd->len; d += d[3] + 4) {
1770 /* we only care about designators with LU association */
1771 if (((d[1] >> 4) & 0x3) != 0x00)
1772 continue;
1773 if ((d[1] & 0xf) != type)
1774 continue;
1775
1776 /*
1777 * Only exit early if a 16-byte descriptor was found. Otherwise
1778 * keep looking as one with more entropy might still show up.
1779 */
1780 len = d[3];
1781 if (len != 8 && len != 12 && len != 16)
1782 continue;
1783 ret = len;
1784 memcpy(id, d + 4, len);
1785 if (len == 16)
1786 break;
1787 }
1788out_unlock:
1789 rcu_read_unlock();
1790 return ret;
1791}
1792
1793static int sd_scsi_to_pr_err(struct scsi_sense_hdr *sshdr, int result)
1794{
1795 switch (host_byte(result)) {
1796 case DID_TRANSPORT_MARGINAL:
1797 case DID_TRANSPORT_DISRUPTED:
1798 case DID_BUS_BUSY:
1799 return PR_STS_RETRY_PATH_FAILURE;
1800 case DID_NO_CONNECT:
1801 return PR_STS_PATH_FAILED;
1802 case DID_TRANSPORT_FAILFAST:
1803 return PR_STS_PATH_FAST_FAILED;
1804 }
1805
1806 switch (status_byte(result)) {
1807 case SAM_STAT_RESERVATION_CONFLICT:
1808 return PR_STS_RESERVATION_CONFLICT;
1809 case SAM_STAT_CHECK_CONDITION:
1810 if (!scsi_sense_valid(sshdr))
1811 return PR_STS_IOERR;
1812
1813 if (sshdr->sense_key == ILLEGAL_REQUEST &&
1814 (sshdr->asc == 0x26 || sshdr->asc == 0x24))
1815 return -EINVAL;
1816
1817 fallthrough;
1818 default:
1819 return PR_STS_IOERR;
1820 }
1821}
1822
1823static int sd_pr_in_command(struct block_device *bdev, u8 sa,
1824 unsigned char *data, int data_len)
1825{
1826 struct scsi_disk *sdkp = scsi_disk(bdev->bd_disk);
1827 struct scsi_device *sdev = sdkp->device;
1828 struct scsi_sense_hdr sshdr;
1829 u8 cmd[10] = { PERSISTENT_RESERVE_IN, sa };
1830 struct scsi_failure failure_defs[] = {
1831 {
1832 .sense = UNIT_ATTENTION,
1833 .asc = SCMD_FAILURE_ASC_ANY,
1834 .ascq = SCMD_FAILURE_ASCQ_ANY,
1835 .allowed = 5,
1836 .result = SAM_STAT_CHECK_CONDITION,
1837 },
1838 {}
1839 };
1840 struct scsi_failures failures = {
1841 .failure_definitions = failure_defs,
1842 };
1843 const struct scsi_exec_args exec_args = {
1844 .sshdr = &sshdr,
1845 .failures = &failures,
1846 };
1847 int result;
1848
1849 put_unaligned_be16(data_len, &cmd[7]);
1850
1851 result = scsi_execute_cmd(sdev, cmd, REQ_OP_DRV_IN, data, data_len,
1852 SD_TIMEOUT, sdkp->max_retries, &exec_args);
1853 if (scsi_status_is_check_condition(result) &&
1854 scsi_sense_valid(&sshdr)) {
1855 sdev_printk(KERN_INFO, sdev, "PR command failed: %d\n", result);
1856 scsi_print_sense_hdr(sdev, NULL, &sshdr);
1857 }
1858
1859 if (result <= 0)
1860 return result;
1861
1862 return sd_scsi_to_pr_err(&sshdr, result);
1863}
1864
1865static int sd_pr_read_keys(struct block_device *bdev, struct pr_keys *keys_info)
1866{
1867 int result, i, data_offset, num_copy_keys;
1868 u32 num_keys = keys_info->num_keys;
1869 int data_len = num_keys * 8 + 8;
1870 u8 *data;
1871
1872 data = kzalloc(data_len, GFP_KERNEL);
1873 if (!data)
1874 return -ENOMEM;
1875
1876 result = sd_pr_in_command(bdev, READ_KEYS, data, data_len);
1877 if (result)
1878 goto free_data;
1879
1880 keys_info->generation = get_unaligned_be32(&data[0]);
1881 keys_info->num_keys = get_unaligned_be32(&data[4]) / 8;
1882
1883 data_offset = 8;
1884 num_copy_keys = min(num_keys, keys_info->num_keys);
1885
1886 for (i = 0; i < num_copy_keys; i++) {
1887 keys_info->keys[i] = get_unaligned_be64(&data[data_offset]);
1888 data_offset += 8;
1889 }
1890
1891free_data:
1892 kfree(data);
1893 return result;
1894}
1895
1896static int sd_pr_read_reservation(struct block_device *bdev,
1897 struct pr_held_reservation *rsv)
1898{
1899 struct scsi_disk *sdkp = scsi_disk(bdev->bd_disk);
1900 struct scsi_device *sdev = sdkp->device;
1901 u8 data[24] = { };
1902 int result, len;
1903
1904 result = sd_pr_in_command(bdev, READ_RESERVATION, data, sizeof(data));
1905 if (result)
1906 return result;
1907
1908 len = get_unaligned_be32(&data[4]);
1909 if (!len)
1910 return 0;
1911
1912 /* Make sure we have at least the key and type */
1913 if (len < 14) {
1914 sdev_printk(KERN_INFO, sdev,
1915 "READ RESERVATION failed due to short return buffer of %d bytes\n",
1916 len);
1917 return -EINVAL;
1918 }
1919
1920 rsv->generation = get_unaligned_be32(&data[0]);
1921 rsv->key = get_unaligned_be64(&data[8]);
1922 rsv->type = scsi_pr_type_to_block(data[21] & 0x0f);
1923 return 0;
1924}
1925
1926static int sd_pr_out_command(struct block_device *bdev, u8 sa, u64 key,
1927 u64 sa_key, enum scsi_pr_type type, u8 flags)
1928{
1929 struct scsi_disk *sdkp = scsi_disk(bdev->bd_disk);
1930 struct scsi_device *sdev = sdkp->device;
1931 struct scsi_sense_hdr sshdr;
1932 struct scsi_failure failure_defs[] = {
1933 {
1934 .sense = UNIT_ATTENTION,
1935 .asc = SCMD_FAILURE_ASC_ANY,
1936 .ascq = SCMD_FAILURE_ASCQ_ANY,
1937 .allowed = 5,
1938 .result = SAM_STAT_CHECK_CONDITION,
1939 },
1940 {}
1941 };
1942 struct scsi_failures failures = {
1943 .failure_definitions = failure_defs,
1944 };
1945 const struct scsi_exec_args exec_args = {
1946 .sshdr = &sshdr,
1947 .failures = &failures,
1948 };
1949 int result;
1950 u8 cmd[16] = { 0, };
1951 u8 data[24] = { 0, };
1952
1953 cmd[0] = PERSISTENT_RESERVE_OUT;
1954 cmd[1] = sa;
1955 cmd[2] = type;
1956 put_unaligned_be32(sizeof(data), &cmd[5]);
1957
1958 put_unaligned_be64(key, &data[0]);
1959 put_unaligned_be64(sa_key, &data[8]);
1960 data[20] = flags;
1961
1962 result = scsi_execute_cmd(sdev, cmd, REQ_OP_DRV_OUT, &data,
1963 sizeof(data), SD_TIMEOUT, sdkp->max_retries,
1964 &exec_args);
1965
1966 if (scsi_status_is_check_condition(result) &&
1967 scsi_sense_valid(&sshdr)) {
1968 sdev_printk(KERN_INFO, sdev, "PR command failed: %d\n", result);
1969 scsi_print_sense_hdr(sdev, NULL, &sshdr);
1970 }
1971
1972 if (result <= 0)
1973 return result;
1974
1975 return sd_scsi_to_pr_err(&sshdr, result);
1976}
1977
1978static int sd_pr_register(struct block_device *bdev, u64 old_key, u64 new_key,
1979 u32 flags)
1980{
1981 if (flags & ~PR_FL_IGNORE_KEY)
1982 return -EOPNOTSUPP;
1983 return sd_pr_out_command(bdev, (flags & PR_FL_IGNORE_KEY) ? 0x06 : 0x00,
1984 old_key, new_key, 0,
1985 (1 << 0) /* APTPL */);
1986}
1987
1988static int sd_pr_reserve(struct block_device *bdev, u64 key, enum pr_type type,
1989 u32 flags)
1990{
1991 if (flags)
1992 return -EOPNOTSUPP;
1993 return sd_pr_out_command(bdev, 0x01, key, 0,
1994 block_pr_type_to_scsi(type), 0);
1995}
1996
1997static int sd_pr_release(struct block_device *bdev, u64 key, enum pr_type type)
1998{
1999 return sd_pr_out_command(bdev, 0x02, key, 0,
2000 block_pr_type_to_scsi(type), 0);
2001}
2002
2003static int sd_pr_preempt(struct block_device *bdev, u64 old_key, u64 new_key,
2004 enum pr_type type, bool abort)
2005{
2006 return sd_pr_out_command(bdev, abort ? 0x05 : 0x04, old_key, new_key,
2007 block_pr_type_to_scsi(type), 0);
2008}
2009
2010static int sd_pr_clear(struct block_device *bdev, u64 key)
2011{
2012 return sd_pr_out_command(bdev, 0x03, key, 0, 0, 0);
2013}
2014
2015static const struct pr_ops sd_pr_ops = {
2016 .pr_register = sd_pr_register,
2017 .pr_reserve = sd_pr_reserve,
2018 .pr_release = sd_pr_release,
2019 .pr_preempt = sd_pr_preempt,
2020 .pr_clear = sd_pr_clear,
2021 .pr_read_keys = sd_pr_read_keys,
2022 .pr_read_reservation = sd_pr_read_reservation,
2023};
2024
2025static void scsi_disk_free_disk(struct gendisk *disk)
2026{
2027 struct scsi_disk *sdkp = scsi_disk(disk);
2028
2029 put_device(&sdkp->disk_dev);
2030}
2031
2032static const struct block_device_operations sd_fops = {
2033 .owner = THIS_MODULE,
2034 .open = sd_open,
2035 .release = sd_release,
2036 .ioctl = sd_ioctl,
2037 .getgeo = sd_getgeo,
2038 .compat_ioctl = blkdev_compat_ptr_ioctl,
2039 .check_events = sd_check_events,
2040 .unlock_native_capacity = sd_unlock_native_capacity,
2041 .report_zones = sd_zbc_report_zones,
2042 .get_unique_id = sd_get_unique_id,
2043 .free_disk = scsi_disk_free_disk,
2044 .pr_ops = &sd_pr_ops,
2045};
2046
2047/**
2048 * sd_eh_reset - reset error handling callback
2049 * @scmd: sd-issued command that has failed
2050 *
2051 * This function is called by the SCSI midlayer before starting
2052 * SCSI EH. When counting medium access failures we have to be
2053 * careful to register it only only once per device and SCSI EH run;
2054 * there might be several timed out commands which will cause the
2055 * 'max_medium_access_timeouts' counter to trigger after the first
2056 * SCSI EH run already and set the device to offline.
2057 * So this function resets the internal counter before starting SCSI EH.
2058 **/
2059static void sd_eh_reset(struct scsi_cmnd *scmd)
2060{
2061 struct scsi_disk *sdkp = scsi_disk(scsi_cmd_to_rq(scmd)->q->disk);
2062
2063 /* New SCSI EH run, reset gate variable */
2064 sdkp->ignore_medium_access_errors = false;
2065}
2066
2067/**
2068 * sd_eh_action - error handling callback
2069 * @scmd: sd-issued command that has failed
2070 * @eh_disp: The recovery disposition suggested by the midlayer
2071 *
2072 * This function is called by the SCSI midlayer upon completion of an
2073 * error test command (currently TEST UNIT READY). The result of sending
2074 * the eh command is passed in eh_disp. We're looking for devices that
2075 * fail medium access commands but are OK with non access commands like
2076 * test unit ready (so wrongly see the device as having a successful
2077 * recovery)
2078 **/
2079static int sd_eh_action(struct scsi_cmnd *scmd, int eh_disp)
2080{
2081 struct scsi_disk *sdkp = scsi_disk(scsi_cmd_to_rq(scmd)->q->disk);
2082 struct scsi_device *sdev = scmd->device;
2083
2084 if (!scsi_device_online(sdev) ||
2085 !scsi_medium_access_command(scmd) ||
2086 host_byte(scmd->result) != DID_TIME_OUT ||
2087 eh_disp != SUCCESS)
2088 return eh_disp;
2089
2090 /*
2091 * The device has timed out executing a medium access command.
2092 * However, the TEST UNIT READY command sent during error
2093 * handling completed successfully. Either the device is in the
2094 * process of recovering or has it suffered an internal failure
2095 * that prevents access to the storage medium.
2096 */
2097 if (!sdkp->ignore_medium_access_errors) {
2098 sdkp->medium_access_timed_out++;
2099 sdkp->ignore_medium_access_errors = true;
2100 }
2101
2102 /*
2103 * If the device keeps failing read/write commands but TEST UNIT
2104 * READY always completes successfully we assume that medium
2105 * access is no longer possible and take the device offline.
2106 */
2107 if (sdkp->medium_access_timed_out >= sdkp->max_medium_access_timeouts) {
2108 scmd_printk(KERN_ERR, scmd,
2109 "Medium access timeout failure. Offlining disk!\n");
2110 mutex_lock(&sdev->state_mutex);
2111 scsi_device_set_state(sdev, SDEV_OFFLINE);
2112 mutex_unlock(&sdev->state_mutex);
2113
2114 return SUCCESS;
2115 }
2116
2117 return eh_disp;
2118}
2119
2120static unsigned int sd_completed_bytes(struct scsi_cmnd *scmd)
2121{
2122 struct request *req = scsi_cmd_to_rq(scmd);
2123 struct scsi_device *sdev = scmd->device;
2124 unsigned int transferred, good_bytes;
2125 u64 start_lba, end_lba, bad_lba;
2126
2127 /*
2128 * Some commands have a payload smaller than the device logical
2129 * block size (e.g. INQUIRY on a 4K disk).
2130 */
2131 if (scsi_bufflen(scmd) <= sdev->sector_size)
2132 return 0;
2133
2134 /* Check if we have a 'bad_lba' information */
2135 if (!scsi_get_sense_info_fld(scmd->sense_buffer,
2136 SCSI_SENSE_BUFFERSIZE,
2137 &bad_lba))
2138 return 0;
2139
2140 /*
2141 * If the bad lba was reported incorrectly, we have no idea where
2142 * the error is.
2143 */
2144 start_lba = sectors_to_logical(sdev, blk_rq_pos(req));
2145 end_lba = start_lba + bytes_to_logical(sdev, scsi_bufflen(scmd));
2146 if (bad_lba < start_lba || bad_lba >= end_lba)
2147 return 0;
2148
2149 /*
2150 * resid is optional but mostly filled in. When it's unused,
2151 * its value is zero, so we assume the whole buffer transferred
2152 */
2153 transferred = scsi_bufflen(scmd) - scsi_get_resid(scmd);
2154
2155 /* This computation should always be done in terms of the
2156 * resolution of the device's medium.
2157 */
2158 good_bytes = logical_to_bytes(sdev, bad_lba - start_lba);
2159
2160 return min(good_bytes, transferred);
2161}
2162
2163/**
2164 * sd_done - bottom half handler: called when the lower level
2165 * driver has completed (successfully or otherwise) a scsi command.
2166 * @SCpnt: mid-level's per command structure.
2167 *
2168 * Note: potentially run from within an ISR. Must not block.
2169 **/
2170static int sd_done(struct scsi_cmnd *SCpnt)
2171{
2172 int result = SCpnt->result;
2173 unsigned int good_bytes = result ? 0 : scsi_bufflen(SCpnt);
2174 unsigned int sector_size = SCpnt->device->sector_size;
2175 unsigned int resid;
2176 struct scsi_sense_hdr sshdr;
2177 struct request *req = scsi_cmd_to_rq(SCpnt);
2178 struct scsi_disk *sdkp = scsi_disk(req->q->disk);
2179 int sense_valid = 0;
2180 int sense_deferred = 0;
2181
2182 switch (req_op(req)) {
2183 case REQ_OP_DISCARD:
2184 case REQ_OP_WRITE_ZEROES:
2185 case REQ_OP_ZONE_RESET:
2186 case REQ_OP_ZONE_RESET_ALL:
2187 case REQ_OP_ZONE_OPEN:
2188 case REQ_OP_ZONE_CLOSE:
2189 case REQ_OP_ZONE_FINISH:
2190 if (!result) {
2191 good_bytes = blk_rq_bytes(req);
2192 scsi_set_resid(SCpnt, 0);
2193 } else {
2194 good_bytes = 0;
2195 scsi_set_resid(SCpnt, blk_rq_bytes(req));
2196 }
2197 break;
2198 default:
2199 /*
2200 * In case of bogus fw or device, we could end up having
2201 * an unaligned partial completion. Check this here and force
2202 * alignment.
2203 */
2204 resid = scsi_get_resid(SCpnt);
2205 if (resid & (sector_size - 1)) {
2206 sd_printk(KERN_INFO, sdkp,
2207 "Unaligned partial completion (resid=%u, sector_sz=%u)\n",
2208 resid, sector_size);
2209 scsi_print_command(SCpnt);
2210 resid = min(scsi_bufflen(SCpnt),
2211 round_up(resid, sector_size));
2212 scsi_set_resid(SCpnt, resid);
2213 }
2214 }
2215
2216 if (result) {
2217 sense_valid = scsi_command_normalize_sense(SCpnt, &sshdr);
2218 if (sense_valid)
2219 sense_deferred = scsi_sense_is_deferred(&sshdr);
2220 }
2221 sdkp->medium_access_timed_out = 0;
2222
2223 if (!scsi_status_is_check_condition(result) &&
2224 (!sense_valid || sense_deferred))
2225 goto out;
2226
2227 switch (sshdr.sense_key) {
2228 case HARDWARE_ERROR:
2229 case MEDIUM_ERROR:
2230 good_bytes = sd_completed_bytes(SCpnt);
2231 break;
2232 case RECOVERED_ERROR:
2233 good_bytes = scsi_bufflen(SCpnt);
2234 break;
2235 case NO_SENSE:
2236 /* This indicates a false check condition, so ignore it. An
2237 * unknown amount of data was transferred so treat it as an
2238 * error.
2239 */
2240 SCpnt->result = 0;
2241 memset(SCpnt->sense_buffer, 0, SCSI_SENSE_BUFFERSIZE);
2242 break;
2243 case ABORTED_COMMAND:
2244 if (sshdr.asc == 0x10) /* DIF: Target detected corruption */
2245 good_bytes = sd_completed_bytes(SCpnt);
2246 break;
2247 case ILLEGAL_REQUEST:
2248 switch (sshdr.asc) {
2249 case 0x10: /* DIX: Host detected corruption */
2250 good_bytes = sd_completed_bytes(SCpnt);
2251 break;
2252 case 0x20: /* INVALID COMMAND OPCODE */
2253 case 0x24: /* INVALID FIELD IN CDB */
2254 switch (SCpnt->cmnd[0]) {
2255 case UNMAP:
2256 sd_config_discard(sdkp, SD_LBP_DISABLE);
2257 break;
2258 case WRITE_SAME_16:
2259 case WRITE_SAME:
2260 if (SCpnt->cmnd[1] & 8) { /* UNMAP */
2261 sd_config_discard(sdkp, SD_LBP_DISABLE);
2262 } else {
2263 sdkp->device->no_write_same = 1;
2264 sd_config_write_same(sdkp);
2265 req->rq_flags |= RQF_QUIET;
2266 }
2267 break;
2268 }
2269 }
2270 break;
2271 default:
2272 break;
2273 }
2274
2275 out:
2276 if (sd_is_zoned(sdkp))
2277 good_bytes = sd_zbc_complete(SCpnt, good_bytes, &sshdr);
2278
2279 SCSI_LOG_HLCOMPLETE(1, scmd_printk(KERN_INFO, SCpnt,
2280 "sd_done: completed %d of %d bytes\n",
2281 good_bytes, scsi_bufflen(SCpnt)));
2282
2283 return good_bytes;
2284}
2285
2286/*
2287 * spinup disk - called only in sd_revalidate_disk()
2288 */
2289static void
2290sd_spinup_disk(struct scsi_disk *sdkp)
2291{
2292 static const u8 cmd[10] = { TEST_UNIT_READY };
2293 unsigned long spintime_expire = 0;
2294 int spintime, sense_valid = 0;
2295 unsigned int the_result;
2296 struct scsi_sense_hdr sshdr;
2297 struct scsi_failure failure_defs[] = {
2298 /* Do not retry Medium Not Present */
2299 {
2300 .sense = UNIT_ATTENTION,
2301 .asc = 0x3A,
2302 .ascq = SCMD_FAILURE_ASCQ_ANY,
2303 .result = SAM_STAT_CHECK_CONDITION,
2304 },
2305 {
2306 .sense = NOT_READY,
2307 .asc = 0x3A,
2308 .ascq = SCMD_FAILURE_ASCQ_ANY,
2309 .result = SAM_STAT_CHECK_CONDITION,
2310 },
2311 /* Retry when scsi_status_is_good would return false 3 times */
2312 {
2313 .result = SCMD_FAILURE_STAT_ANY,
2314 .allowed = 3,
2315 },
2316 {}
2317 };
2318 struct scsi_failures failures = {
2319 .failure_definitions = failure_defs,
2320 };
2321 const struct scsi_exec_args exec_args = {
2322 .sshdr = &sshdr,
2323 .failures = &failures,
2324 };
2325
2326 spintime = 0;
2327
2328 /* Spin up drives, as required. Only do this at boot time */
2329 /* Spinup needs to be done for module loads too. */
2330 do {
2331 bool media_was_present = sdkp->media_present;
2332
2333 scsi_failures_reset_retries(&failures);
2334
2335 the_result = scsi_execute_cmd(sdkp->device, cmd, REQ_OP_DRV_IN,
2336 NULL, 0, SD_TIMEOUT,
2337 sdkp->max_retries, &exec_args);
2338
2339
2340 if (the_result > 0) {
2341 /*
2342 * If the drive has indicated to us that it doesn't
2343 * have any media in it, don't bother with any more
2344 * polling.
2345 */
2346 if (media_not_present(sdkp, &sshdr)) {
2347 if (media_was_present)
2348 sd_printk(KERN_NOTICE, sdkp,
2349 "Media removed, stopped polling\n");
2350 return;
2351 }
2352 sense_valid = scsi_sense_valid(&sshdr);
2353 }
2354
2355 if (!scsi_status_is_check_condition(the_result)) {
2356 /* no sense, TUR either succeeded or failed
2357 * with a status error */
2358 if(!spintime && !scsi_status_is_good(the_result)) {
2359 sd_print_result(sdkp, "Test Unit Ready failed",
2360 the_result);
2361 }
2362 break;
2363 }
2364
2365 /*
2366 * The device does not want the automatic start to be issued.
2367 */
2368 if (sdkp->device->no_start_on_add)
2369 break;
2370
2371 if (sense_valid && sshdr.sense_key == NOT_READY) {
2372 if (sshdr.asc == 4 && sshdr.ascq == 3)
2373 break; /* manual intervention required */
2374 if (sshdr.asc == 4 && sshdr.ascq == 0xb)
2375 break; /* standby */
2376 if (sshdr.asc == 4 && sshdr.ascq == 0xc)
2377 break; /* unavailable */
2378 if (sshdr.asc == 4 && sshdr.ascq == 0x1b)
2379 break; /* sanitize in progress */
2380 if (sshdr.asc == 4 && sshdr.ascq == 0x24)
2381 break; /* depopulation in progress */
2382 if (sshdr.asc == 4 && sshdr.ascq == 0x25)
2383 break; /* depopulation restoration in progress */
2384 /*
2385 * Issue command to spin up drive when not ready
2386 */
2387 if (!spintime) {
2388 /* Return immediately and start spin cycle */
2389 const u8 start_cmd[10] = {
2390 [0] = START_STOP,
2391 [1] = 1,
2392 [4] = sdkp->device->start_stop_pwr_cond ?
2393 0x11 : 1,
2394 };
2395
2396 sd_printk(KERN_NOTICE, sdkp, "Spinning up disk...");
2397 scsi_execute_cmd(sdkp->device, start_cmd,
2398 REQ_OP_DRV_IN, NULL, 0,
2399 SD_TIMEOUT, sdkp->max_retries,
2400 &exec_args);
2401 spintime_expire = jiffies + 100 * HZ;
2402 spintime = 1;
2403 }
2404 /* Wait 1 second for next try */
2405 msleep(1000);
2406 printk(KERN_CONT ".");
2407
2408 /*
2409 * Wait for USB flash devices with slow firmware.
2410 * Yes, this sense key/ASC combination shouldn't
2411 * occur here. It's characteristic of these devices.
2412 */
2413 } else if (sense_valid &&
2414 sshdr.sense_key == UNIT_ATTENTION &&
2415 sshdr.asc == 0x28) {
2416 if (!spintime) {
2417 spintime_expire = jiffies + 5 * HZ;
2418 spintime = 1;
2419 }
2420 /* Wait 1 second for next try */
2421 msleep(1000);
2422 } else {
2423 /* we don't understand the sense code, so it's
2424 * probably pointless to loop */
2425 if(!spintime) {
2426 sd_printk(KERN_NOTICE, sdkp, "Unit Not Ready\n");
2427 sd_print_sense_hdr(sdkp, &sshdr);
2428 }
2429 break;
2430 }
2431
2432 } while (spintime && time_before_eq(jiffies, spintime_expire));
2433
2434 if (spintime) {
2435 if (scsi_status_is_good(the_result))
2436 printk(KERN_CONT "ready\n");
2437 else
2438 printk(KERN_CONT "not responding...\n");
2439 }
2440}
2441
2442/*
2443 * Determine whether disk supports Data Integrity Field.
2444 */
2445static int sd_read_protection_type(struct scsi_disk *sdkp, unsigned char *buffer)
2446{
2447 struct scsi_device *sdp = sdkp->device;
2448 u8 type;
2449
2450 if (scsi_device_protection(sdp) == 0 || (buffer[12] & 1) == 0) {
2451 sdkp->protection_type = 0;
2452 return 0;
2453 }
2454
2455 type = ((buffer[12] >> 1) & 7) + 1; /* P_TYPE 0 = Type 1 */
2456
2457 if (type > T10_PI_TYPE3_PROTECTION) {
2458 sd_printk(KERN_ERR, sdkp, "formatted with unsupported" \
2459 " protection type %u. Disabling disk!\n",
2460 type);
2461 sdkp->protection_type = 0;
2462 return -ENODEV;
2463 }
2464
2465 sdkp->protection_type = type;
2466
2467 return 0;
2468}
2469
2470static void sd_config_protection(struct scsi_disk *sdkp)
2471{
2472 struct scsi_device *sdp = sdkp->device;
2473
2474 sd_dif_config_host(sdkp);
2475
2476 if (!sdkp->protection_type)
2477 return;
2478
2479 if (!scsi_host_dif_capable(sdp->host, sdkp->protection_type)) {
2480 sd_first_printk(KERN_NOTICE, sdkp,
2481 "Disabling DIF Type %u protection\n",
2482 sdkp->protection_type);
2483 sdkp->protection_type = 0;
2484 }
2485
2486 sd_first_printk(KERN_NOTICE, sdkp, "Enabling DIF Type %u protection\n",
2487 sdkp->protection_type);
2488}
2489
2490static void read_capacity_error(struct scsi_disk *sdkp, struct scsi_device *sdp,
2491 struct scsi_sense_hdr *sshdr, int sense_valid,
2492 int the_result)
2493{
2494 if (sense_valid)
2495 sd_print_sense_hdr(sdkp, sshdr);
2496 else
2497 sd_printk(KERN_NOTICE, sdkp, "Sense not available.\n");
2498
2499 /*
2500 * Set dirty bit for removable devices if not ready -
2501 * sometimes drives will not report this properly.
2502 */
2503 if (sdp->removable &&
2504 sense_valid && sshdr->sense_key == NOT_READY)
2505 set_media_not_present(sdkp);
2506
2507 /*
2508 * We used to set media_present to 0 here to indicate no media
2509 * in the drive, but some drives fail read capacity even with
2510 * media present, so we can't do that.
2511 */
2512 sdkp->capacity = 0; /* unknown mapped to zero - as usual */
2513}
2514
2515#define RC16_LEN 32
2516#if RC16_LEN > SD_BUF_SIZE
2517#error RC16_LEN must not be more than SD_BUF_SIZE
2518#endif
2519
2520#define READ_CAPACITY_RETRIES_ON_RESET 10
2521
2522static int read_capacity_16(struct scsi_disk *sdkp, struct scsi_device *sdp,
2523 unsigned char *buffer)
2524{
2525 unsigned char cmd[16];
2526 struct scsi_sense_hdr sshdr;
2527 const struct scsi_exec_args exec_args = {
2528 .sshdr = &sshdr,
2529 };
2530 int sense_valid = 0;
2531 int the_result;
2532 int retries = 3, reset_retries = READ_CAPACITY_RETRIES_ON_RESET;
2533 unsigned int alignment;
2534 unsigned long long lba;
2535 unsigned sector_size;
2536
2537 if (sdp->no_read_capacity_16)
2538 return -EINVAL;
2539
2540 do {
2541 memset(cmd, 0, 16);
2542 cmd[0] = SERVICE_ACTION_IN_16;
2543 cmd[1] = SAI_READ_CAPACITY_16;
2544 cmd[13] = RC16_LEN;
2545 memset(buffer, 0, RC16_LEN);
2546
2547 the_result = scsi_execute_cmd(sdp, cmd, REQ_OP_DRV_IN,
2548 buffer, RC16_LEN, SD_TIMEOUT,
2549 sdkp->max_retries, &exec_args);
2550 if (the_result > 0) {
2551 if (media_not_present(sdkp, &sshdr))
2552 return -ENODEV;
2553
2554 sense_valid = scsi_sense_valid(&sshdr);
2555 if (sense_valid &&
2556 sshdr.sense_key == ILLEGAL_REQUEST &&
2557 (sshdr.asc == 0x20 || sshdr.asc == 0x24) &&
2558 sshdr.ascq == 0x00)
2559 /* Invalid Command Operation Code or
2560 * Invalid Field in CDB, just retry
2561 * silently with RC10 */
2562 return -EINVAL;
2563 if (sense_valid &&
2564 sshdr.sense_key == UNIT_ATTENTION &&
2565 sshdr.asc == 0x29 && sshdr.ascq == 0x00)
2566 /* Device reset might occur several times,
2567 * give it one more chance */
2568 if (--reset_retries > 0)
2569 continue;
2570 }
2571 retries--;
2572
2573 } while (the_result && retries);
2574
2575 if (the_result) {
2576 sd_print_result(sdkp, "Read Capacity(16) failed", the_result);
2577 read_capacity_error(sdkp, sdp, &sshdr, sense_valid, the_result);
2578 return -EINVAL;
2579 }
2580
2581 sector_size = get_unaligned_be32(&buffer[8]);
2582 lba = get_unaligned_be64(&buffer[0]);
2583
2584 if (sd_read_protection_type(sdkp, buffer) < 0) {
2585 sdkp->capacity = 0;
2586 return -ENODEV;
2587 }
2588
2589 /* Logical blocks per physical block exponent */
2590 sdkp->physical_block_size = (1 << (buffer[13] & 0xf)) * sector_size;
2591
2592 /* RC basis */
2593 sdkp->rc_basis = (buffer[12] >> 4) & 0x3;
2594
2595 /* Lowest aligned logical block */
2596 alignment = ((buffer[14] & 0x3f) << 8 | buffer[15]) * sector_size;
2597 blk_queue_alignment_offset(sdp->request_queue, alignment);
2598 if (alignment && sdkp->first_scan)
2599 sd_printk(KERN_NOTICE, sdkp,
2600 "physical block alignment offset: %u\n", alignment);
2601
2602 if (buffer[14] & 0x80) { /* LBPME */
2603 sdkp->lbpme = 1;
2604
2605 if (buffer[14] & 0x40) /* LBPRZ */
2606 sdkp->lbprz = 1;
2607
2608 sd_config_discard(sdkp, SD_LBP_WS16);
2609 }
2610
2611 sdkp->capacity = lba + 1;
2612 return sector_size;
2613}
2614
2615static int read_capacity_10(struct scsi_disk *sdkp, struct scsi_device *sdp,
2616 unsigned char *buffer)
2617{
2618 static const u8 cmd[10] = { READ_CAPACITY };
2619 struct scsi_sense_hdr sshdr;
2620 struct scsi_failure failure_defs[] = {
2621 /* Do not retry Medium Not Present */
2622 {
2623 .sense = UNIT_ATTENTION,
2624 .asc = 0x3A,
2625 .result = SAM_STAT_CHECK_CONDITION,
2626 },
2627 {
2628 .sense = NOT_READY,
2629 .asc = 0x3A,
2630 .result = SAM_STAT_CHECK_CONDITION,
2631 },
2632 /* Device reset might occur several times so retry a lot */
2633 {
2634 .sense = UNIT_ATTENTION,
2635 .asc = 0x29,
2636 .allowed = READ_CAPACITY_RETRIES_ON_RESET,
2637 .result = SAM_STAT_CHECK_CONDITION,
2638 },
2639 /* Any other error not listed above retry 3 times */
2640 {
2641 .result = SCMD_FAILURE_RESULT_ANY,
2642 .allowed = 3,
2643 },
2644 {}
2645 };
2646 struct scsi_failures failures = {
2647 .failure_definitions = failure_defs,
2648 };
2649 const struct scsi_exec_args exec_args = {
2650 .sshdr = &sshdr,
2651 .failures = &failures,
2652 };
2653 int sense_valid = 0;
2654 int the_result;
2655 sector_t lba;
2656 unsigned sector_size;
2657
2658 memset(buffer, 0, 8);
2659
2660 the_result = scsi_execute_cmd(sdp, cmd, REQ_OP_DRV_IN, buffer,
2661 8, SD_TIMEOUT, sdkp->max_retries,
2662 &exec_args);
2663
2664 if (the_result > 0) {
2665 sense_valid = scsi_sense_valid(&sshdr);
2666
2667 if (media_not_present(sdkp, &sshdr))
2668 return -ENODEV;
2669 }
2670
2671 if (the_result) {
2672 sd_print_result(sdkp, "Read Capacity(10) failed", the_result);
2673 read_capacity_error(sdkp, sdp, &sshdr, sense_valid, the_result);
2674 return -EINVAL;
2675 }
2676
2677 sector_size = get_unaligned_be32(&buffer[4]);
2678 lba = get_unaligned_be32(&buffer[0]);
2679
2680 if (sdp->no_read_capacity_16 && (lba == 0xffffffff)) {
2681 /* Some buggy (usb cardreader) devices return an lba of
2682 0xffffffff when the want to report a size of 0 (with
2683 which they really mean no media is present) */
2684 sdkp->capacity = 0;
2685 sdkp->physical_block_size = sector_size;
2686 return sector_size;
2687 }
2688
2689 sdkp->capacity = lba + 1;
2690 sdkp->physical_block_size = sector_size;
2691 return sector_size;
2692}
2693
2694static int sd_try_rc16_first(struct scsi_device *sdp)
2695{
2696 if (sdp->host->max_cmd_len < 16)
2697 return 0;
2698 if (sdp->try_rc_10_first)
2699 return 0;
2700 if (sdp->scsi_level > SCSI_SPC_2)
2701 return 1;
2702 if (scsi_device_protection(sdp))
2703 return 1;
2704 return 0;
2705}
2706
2707/*
2708 * read disk capacity
2709 */
2710static void
2711sd_read_capacity(struct scsi_disk *sdkp, unsigned char *buffer)
2712{
2713 int sector_size;
2714 struct scsi_device *sdp = sdkp->device;
2715
2716 if (sd_try_rc16_first(sdp)) {
2717 sector_size = read_capacity_16(sdkp, sdp, buffer);
2718 if (sector_size == -EOVERFLOW)
2719 goto got_data;
2720 if (sector_size == -ENODEV)
2721 return;
2722 if (sector_size < 0)
2723 sector_size = read_capacity_10(sdkp, sdp, buffer);
2724 if (sector_size < 0)
2725 return;
2726 } else {
2727 sector_size = read_capacity_10(sdkp, sdp, buffer);
2728 if (sector_size == -EOVERFLOW)
2729 goto got_data;
2730 if (sector_size < 0)
2731 return;
2732 if ((sizeof(sdkp->capacity) > 4) &&
2733 (sdkp->capacity > 0xffffffffULL)) {
2734 int old_sector_size = sector_size;
2735 sd_printk(KERN_NOTICE, sdkp, "Very big device. "
2736 "Trying to use READ CAPACITY(16).\n");
2737 sector_size = read_capacity_16(sdkp, sdp, buffer);
2738 if (sector_size < 0) {
2739 sd_printk(KERN_NOTICE, sdkp,
2740 "Using 0xffffffff as device size\n");
2741 sdkp->capacity = 1 + (sector_t) 0xffffffff;
2742 sector_size = old_sector_size;
2743 goto got_data;
2744 }
2745 /* Remember that READ CAPACITY(16) succeeded */
2746 sdp->try_rc_10_first = 0;
2747 }
2748 }
2749
2750 /* Some devices are known to return the total number of blocks,
2751 * not the highest block number. Some devices have versions
2752 * which do this and others which do not. Some devices we might
2753 * suspect of doing this but we don't know for certain.
2754 *
2755 * If we know the reported capacity is wrong, decrement it. If
2756 * we can only guess, then assume the number of blocks is even
2757 * (usually true but not always) and err on the side of lowering
2758 * the capacity.
2759 */
2760 if (sdp->fix_capacity ||
2761 (sdp->guess_capacity && (sdkp->capacity & 0x01))) {
2762 sd_printk(KERN_INFO, sdkp, "Adjusting the sector count "
2763 "from its reported value: %llu\n",
2764 (unsigned long long) sdkp->capacity);
2765 --sdkp->capacity;
2766 }
2767
2768got_data:
2769 if (sector_size == 0) {
2770 sector_size = 512;
2771 sd_printk(KERN_NOTICE, sdkp, "Sector size 0 reported, "
2772 "assuming 512.\n");
2773 }
2774
2775 if (sector_size != 512 &&
2776 sector_size != 1024 &&
2777 sector_size != 2048 &&
2778 sector_size != 4096) {
2779 sd_printk(KERN_NOTICE, sdkp, "Unsupported sector size %d.\n",
2780 sector_size);
2781 /*
2782 * The user might want to re-format the drive with
2783 * a supported sectorsize. Once this happens, it
2784 * would be relatively trivial to set the thing up.
2785 * For this reason, we leave the thing in the table.
2786 */
2787 sdkp->capacity = 0;
2788 /*
2789 * set a bogus sector size so the normal read/write
2790 * logic in the block layer will eventually refuse any
2791 * request on this device without tripping over power
2792 * of two sector size assumptions
2793 */
2794 sector_size = 512;
2795 }
2796 blk_queue_logical_block_size(sdp->request_queue, sector_size);
2797 blk_queue_physical_block_size(sdp->request_queue,
2798 sdkp->physical_block_size);
2799 sdkp->device->sector_size = sector_size;
2800
2801 if (sdkp->capacity > 0xffffffff)
2802 sdp->use_16_for_rw = 1;
2803
2804}
2805
2806/*
2807 * Print disk capacity
2808 */
2809static void
2810sd_print_capacity(struct scsi_disk *sdkp,
2811 sector_t old_capacity)
2812{
2813 int sector_size = sdkp->device->sector_size;
2814 char cap_str_2[10], cap_str_10[10];
2815
2816 if (!sdkp->first_scan && old_capacity == sdkp->capacity)
2817 return;
2818
2819 string_get_size(sdkp->capacity, sector_size,
2820 STRING_UNITS_2, cap_str_2, sizeof(cap_str_2));
2821 string_get_size(sdkp->capacity, sector_size,
2822 STRING_UNITS_10, cap_str_10, sizeof(cap_str_10));
2823
2824 sd_printk(KERN_NOTICE, sdkp,
2825 "%llu %d-byte logical blocks: (%s/%s)\n",
2826 (unsigned long long)sdkp->capacity,
2827 sector_size, cap_str_10, cap_str_2);
2828
2829 if (sdkp->physical_block_size != sector_size)
2830 sd_printk(KERN_NOTICE, sdkp,
2831 "%u-byte physical blocks\n",
2832 sdkp->physical_block_size);
2833}
2834
2835/* called with buffer of length 512 */
2836static inline int
2837sd_do_mode_sense(struct scsi_disk *sdkp, int dbd, int modepage,
2838 unsigned char *buffer, int len, struct scsi_mode_data *data,
2839 struct scsi_sense_hdr *sshdr)
2840{
2841 /*
2842 * If we must use MODE SENSE(10), make sure that the buffer length
2843 * is at least 8 bytes so that the mode sense header fits.
2844 */
2845 if (sdkp->device->use_10_for_ms && len < 8)
2846 len = 8;
2847
2848 return scsi_mode_sense(sdkp->device, dbd, modepage, 0, buffer, len,
2849 SD_TIMEOUT, sdkp->max_retries, data, sshdr);
2850}
2851
2852/*
2853 * read write protect setting, if possible - called only in sd_revalidate_disk()
2854 * called with buffer of length SD_BUF_SIZE
2855 */
2856static void
2857sd_read_write_protect_flag(struct scsi_disk *sdkp, unsigned char *buffer)
2858{
2859 int res;
2860 struct scsi_device *sdp = sdkp->device;
2861 struct scsi_mode_data data;
2862 int old_wp = sdkp->write_prot;
2863
2864 set_disk_ro(sdkp->disk, 0);
2865 if (sdp->skip_ms_page_3f) {
2866 sd_first_printk(KERN_NOTICE, sdkp, "Assuming Write Enabled\n");
2867 return;
2868 }
2869
2870 if (sdp->use_192_bytes_for_3f) {
2871 res = sd_do_mode_sense(sdkp, 0, 0x3F, buffer, 192, &data, NULL);
2872 } else {
2873 /*
2874 * First attempt: ask for all pages (0x3F), but only 4 bytes.
2875 * We have to start carefully: some devices hang if we ask
2876 * for more than is available.
2877 */
2878 res = sd_do_mode_sense(sdkp, 0, 0x3F, buffer, 4, &data, NULL);
2879
2880 /*
2881 * Second attempt: ask for page 0 When only page 0 is
2882 * implemented, a request for page 3F may return Sense Key
2883 * 5: Illegal Request, Sense Code 24: Invalid field in
2884 * CDB.
2885 */
2886 if (res < 0)
2887 res = sd_do_mode_sense(sdkp, 0, 0, buffer, 4, &data, NULL);
2888
2889 /*
2890 * Third attempt: ask 255 bytes, as we did earlier.
2891 */
2892 if (res < 0)
2893 res = sd_do_mode_sense(sdkp, 0, 0x3F, buffer, 255,
2894 &data, NULL);
2895 }
2896
2897 if (res < 0) {
2898 sd_first_printk(KERN_WARNING, sdkp,
2899 "Test WP failed, assume Write Enabled\n");
2900 } else {
2901 sdkp->write_prot = ((data.device_specific & 0x80) != 0);
2902 set_disk_ro(sdkp->disk, sdkp->write_prot);
2903 if (sdkp->first_scan || old_wp != sdkp->write_prot) {
2904 sd_printk(KERN_NOTICE, sdkp, "Write Protect is %s\n",
2905 sdkp->write_prot ? "on" : "off");
2906 sd_printk(KERN_DEBUG, sdkp, "Mode Sense: %4ph\n", buffer);
2907 }
2908 }
2909}
2910
2911/*
2912 * sd_read_cache_type - called only from sd_revalidate_disk()
2913 * called with buffer of length SD_BUF_SIZE
2914 */
2915static void
2916sd_read_cache_type(struct scsi_disk *sdkp, unsigned char *buffer)
2917{
2918 int len = 0, res;
2919 struct scsi_device *sdp = sdkp->device;
2920
2921 int dbd;
2922 int modepage;
2923 int first_len;
2924 struct scsi_mode_data data;
2925 struct scsi_sense_hdr sshdr;
2926 int old_wce = sdkp->WCE;
2927 int old_rcd = sdkp->RCD;
2928 int old_dpofua = sdkp->DPOFUA;
2929
2930
2931 if (sdkp->cache_override)
2932 return;
2933
2934 first_len = 4;
2935 if (sdp->skip_ms_page_8) {
2936 if (sdp->type == TYPE_RBC)
2937 goto defaults;
2938 else {
2939 if (sdp->skip_ms_page_3f)
2940 goto defaults;
2941 modepage = 0x3F;
2942 if (sdp->use_192_bytes_for_3f)
2943 first_len = 192;
2944 dbd = 0;
2945 }
2946 } else if (sdp->type == TYPE_RBC) {
2947 modepage = 6;
2948 dbd = 8;
2949 } else {
2950 modepage = 8;
2951 dbd = 0;
2952 }
2953
2954 /* cautiously ask */
2955 res = sd_do_mode_sense(sdkp, dbd, modepage, buffer, first_len,
2956 &data, &sshdr);
2957
2958 if (res < 0)
2959 goto bad_sense;
2960
2961 if (!data.header_length) {
2962 modepage = 6;
2963 first_len = 0;
2964 sd_first_printk(KERN_ERR, sdkp,
2965 "Missing header in MODE_SENSE response\n");
2966 }
2967
2968 /* that went OK, now ask for the proper length */
2969 len = data.length;
2970
2971 /*
2972 * We're only interested in the first three bytes, actually.
2973 * But the data cache page is defined for the first 20.
2974 */
2975 if (len < 3)
2976 goto bad_sense;
2977 else if (len > SD_BUF_SIZE) {
2978 sd_first_printk(KERN_NOTICE, sdkp, "Truncating mode parameter "
2979 "data from %d to %d bytes\n", len, SD_BUF_SIZE);
2980 len = SD_BUF_SIZE;
2981 }
2982 if (modepage == 0x3F && sdp->use_192_bytes_for_3f)
2983 len = 192;
2984
2985 /* Get the data */
2986 if (len > first_len)
2987 res = sd_do_mode_sense(sdkp, dbd, modepage, buffer, len,
2988 &data, &sshdr);
2989
2990 if (!res) {
2991 int offset = data.header_length + data.block_descriptor_length;
2992
2993 while (offset < len) {
2994 u8 page_code = buffer[offset] & 0x3F;
2995 u8 spf = buffer[offset] & 0x40;
2996
2997 if (page_code == 8 || page_code == 6) {
2998 /* We're interested only in the first 3 bytes.
2999 */
3000 if (len - offset <= 2) {
3001 sd_first_printk(KERN_ERR, sdkp,
3002 "Incomplete mode parameter "
3003 "data\n");
3004 goto defaults;
3005 } else {
3006 modepage = page_code;
3007 goto Page_found;
3008 }
3009 } else {
3010 /* Go to the next page */
3011 if (spf && len - offset > 3)
3012 offset += 4 + (buffer[offset+2] << 8) +
3013 buffer[offset+3];
3014 else if (!spf && len - offset > 1)
3015 offset += 2 + buffer[offset+1];
3016 else {
3017 sd_first_printk(KERN_ERR, sdkp,
3018 "Incomplete mode "
3019 "parameter data\n");
3020 goto defaults;
3021 }
3022 }
3023 }
3024
3025 sd_first_printk(KERN_WARNING, sdkp,
3026 "No Caching mode page found\n");
3027 goto defaults;
3028
3029 Page_found:
3030 if (modepage == 8) {
3031 sdkp->WCE = ((buffer[offset + 2] & 0x04) != 0);
3032 sdkp->RCD = ((buffer[offset + 2] & 0x01) != 0);
3033 } else {
3034 sdkp->WCE = ((buffer[offset + 2] & 0x01) == 0);
3035 sdkp->RCD = 0;
3036 }
3037
3038 sdkp->DPOFUA = (data.device_specific & 0x10) != 0;
3039 if (sdp->broken_fua) {
3040 sd_first_printk(KERN_NOTICE, sdkp, "Disabling FUA\n");
3041 sdkp->DPOFUA = 0;
3042 } else if (sdkp->DPOFUA && !sdkp->device->use_10_for_rw &&
3043 !sdkp->device->use_16_for_rw) {
3044 sd_first_printk(KERN_NOTICE, sdkp,
3045 "Uses READ/WRITE(6), disabling FUA\n");
3046 sdkp->DPOFUA = 0;
3047 }
3048
3049 /* No cache flush allowed for write protected devices */
3050 if (sdkp->WCE && sdkp->write_prot)
3051 sdkp->WCE = 0;
3052
3053 if (sdkp->first_scan || old_wce != sdkp->WCE ||
3054 old_rcd != sdkp->RCD || old_dpofua != sdkp->DPOFUA)
3055 sd_printk(KERN_NOTICE, sdkp,
3056 "Write cache: %s, read cache: %s, %s\n",
3057 sdkp->WCE ? "enabled" : "disabled",
3058 sdkp->RCD ? "disabled" : "enabled",
3059 sdkp->DPOFUA ? "supports DPO and FUA"
3060 : "doesn't support DPO or FUA");
3061
3062 return;
3063 }
3064
3065bad_sense:
3066 if (res == -EIO && scsi_sense_valid(&sshdr) &&
3067 sshdr.sense_key == ILLEGAL_REQUEST &&
3068 sshdr.asc == 0x24 && sshdr.ascq == 0x0)
3069 /* Invalid field in CDB */
3070 sd_first_printk(KERN_NOTICE, sdkp, "Cache data unavailable\n");
3071 else
3072 sd_first_printk(KERN_ERR, sdkp,
3073 "Asking for cache data failed\n");
3074
3075defaults:
3076 if (sdp->wce_default_on) {
3077 sd_first_printk(KERN_NOTICE, sdkp,
3078 "Assuming drive cache: write back\n");
3079 sdkp->WCE = 1;
3080 } else {
3081 sd_first_printk(KERN_WARNING, sdkp,
3082 "Assuming drive cache: write through\n");
3083 sdkp->WCE = 0;
3084 }
3085 sdkp->RCD = 0;
3086 sdkp->DPOFUA = 0;
3087}
3088
3089static bool sd_is_perm_stream(struct scsi_disk *sdkp, unsigned int stream_id)
3090{
3091 u8 cdb[16] = { SERVICE_ACTION_IN_16, SAI_GET_STREAM_STATUS };
3092 struct {
3093 struct scsi_stream_status_header h;
3094 struct scsi_stream_status s;
3095 } buf;
3096 struct scsi_device *sdev = sdkp->device;
3097 struct scsi_sense_hdr sshdr;
3098 const struct scsi_exec_args exec_args = {
3099 .sshdr = &sshdr,
3100 };
3101 int res;
3102
3103 put_unaligned_be16(stream_id, &cdb[4]);
3104 put_unaligned_be32(sizeof(buf), &cdb[10]);
3105
3106 res = scsi_execute_cmd(sdev, cdb, REQ_OP_DRV_IN, &buf, sizeof(buf),
3107 SD_TIMEOUT, sdkp->max_retries, &exec_args);
3108 if (res < 0)
3109 return false;
3110 if (scsi_status_is_check_condition(res) && scsi_sense_valid(&sshdr))
3111 sd_print_sense_hdr(sdkp, &sshdr);
3112 if (res)
3113 return false;
3114 if (get_unaligned_be32(&buf.h.len) < sizeof(struct scsi_stream_status))
3115 return false;
3116 return buf.h.stream_status[0].perm;
3117}
3118
3119static void sd_read_io_hints(struct scsi_disk *sdkp, unsigned char *buffer)
3120{
3121 struct scsi_device *sdp = sdkp->device;
3122 const struct scsi_io_group_descriptor *desc, *start, *end;
3123 u16 permanent_stream_count_old;
3124 struct scsi_sense_hdr sshdr;
3125 struct scsi_mode_data data;
3126 int res;
3127
3128 res = scsi_mode_sense(sdp, /*dbd=*/0x8, /*modepage=*/0x0a,
3129 /*subpage=*/0x05, buffer, SD_BUF_SIZE, SD_TIMEOUT,
3130 sdkp->max_retries, &data, &sshdr);
3131 if (res < 0)
3132 return;
3133 start = (void *)buffer + data.header_length + 16;
3134 end = (void *)buffer + ALIGN_DOWN(data.header_length + data.length,
3135 sizeof(*end));
3136 /*
3137 * From "SBC-5 Constrained Streams with Data Lifetimes": Device severs
3138 * should assign the lowest numbered stream identifiers to permanent
3139 * streams.
3140 */
3141 for (desc = start; desc < end; desc++)
3142 if (!desc->st_enble || !sd_is_perm_stream(sdkp, desc - start))
3143 break;
3144 permanent_stream_count_old = sdkp->permanent_stream_count;
3145 sdkp->permanent_stream_count = desc - start;
3146 if (sdkp->rscs && sdkp->permanent_stream_count < 2)
3147 sd_printk(KERN_INFO, sdkp,
3148 "Unexpected: RSCS has been set and the permanent stream count is %u\n",
3149 sdkp->permanent_stream_count);
3150 else if (sdkp->permanent_stream_count != permanent_stream_count_old)
3151 sd_printk(KERN_INFO, sdkp, "permanent stream count = %d\n",
3152 sdkp->permanent_stream_count);
3153}
3154
3155/*
3156 * The ATO bit indicates whether the DIF application tag is available
3157 * for use by the operating system.
3158 */
3159static void sd_read_app_tag_own(struct scsi_disk *sdkp, unsigned char *buffer)
3160{
3161 int res, offset;
3162 struct scsi_device *sdp = sdkp->device;
3163 struct scsi_mode_data data;
3164 struct scsi_sense_hdr sshdr;
3165
3166 if (sdp->type != TYPE_DISK && sdp->type != TYPE_ZBC)
3167 return;
3168
3169 if (sdkp->protection_type == 0)
3170 return;
3171
3172 res = scsi_mode_sense(sdp, 1, 0x0a, 0, buffer, 36, SD_TIMEOUT,
3173 sdkp->max_retries, &data, &sshdr);
3174
3175 if (res < 0 || !data.header_length ||
3176 data.length < 6) {
3177 sd_first_printk(KERN_WARNING, sdkp,
3178 "getting Control mode page failed, assume no ATO\n");
3179
3180 if (res == -EIO && scsi_sense_valid(&sshdr))
3181 sd_print_sense_hdr(sdkp, &sshdr);
3182
3183 return;
3184 }
3185
3186 offset = data.header_length + data.block_descriptor_length;
3187
3188 if ((buffer[offset] & 0x3f) != 0x0a) {
3189 sd_first_printk(KERN_ERR, sdkp, "ATO Got wrong page\n");
3190 return;
3191 }
3192
3193 if ((buffer[offset + 5] & 0x80) == 0)
3194 return;
3195
3196 sdkp->ATO = 1;
3197
3198 return;
3199}
3200
3201/**
3202 * sd_read_block_limits - Query disk device for preferred I/O sizes.
3203 * @sdkp: disk to query
3204 */
3205static void sd_read_block_limits(struct scsi_disk *sdkp)
3206{
3207 struct scsi_vpd *vpd;
3208
3209 rcu_read_lock();
3210
3211 vpd = rcu_dereference(sdkp->device->vpd_pgb0);
3212 if (!vpd || vpd->len < 16)
3213 goto out;
3214
3215 sdkp->min_xfer_blocks = get_unaligned_be16(&vpd->data[6]);
3216 sdkp->max_xfer_blocks = get_unaligned_be32(&vpd->data[8]);
3217 sdkp->opt_xfer_blocks = get_unaligned_be32(&vpd->data[12]);
3218
3219 if (vpd->len >= 64) {
3220 unsigned int lba_count, desc_count;
3221
3222 sdkp->max_ws_blocks = (u32)get_unaligned_be64(&vpd->data[36]);
3223
3224 if (!sdkp->lbpme)
3225 goto out;
3226
3227 lba_count = get_unaligned_be32(&vpd->data[20]);
3228 desc_count = get_unaligned_be32(&vpd->data[24]);
3229
3230 if (lba_count && desc_count)
3231 sdkp->max_unmap_blocks = lba_count;
3232
3233 sdkp->unmap_granularity = get_unaligned_be32(&vpd->data[28]);
3234
3235 if (vpd->data[32] & 0x80)
3236 sdkp->unmap_alignment =
3237 get_unaligned_be32(&vpd->data[32]) & ~(1 << 31);
3238
3239 if (!sdkp->lbpvpd) { /* LBP VPD page not provided */
3240
3241 if (sdkp->max_unmap_blocks)
3242 sd_config_discard(sdkp, SD_LBP_UNMAP);
3243 else
3244 sd_config_discard(sdkp, SD_LBP_WS16);
3245
3246 } else { /* LBP VPD page tells us what to use */
3247 if (sdkp->lbpu && sdkp->max_unmap_blocks)
3248 sd_config_discard(sdkp, SD_LBP_UNMAP);
3249 else if (sdkp->lbpws)
3250 sd_config_discard(sdkp, SD_LBP_WS16);
3251 else if (sdkp->lbpws10)
3252 sd_config_discard(sdkp, SD_LBP_WS10);
3253 else
3254 sd_config_discard(sdkp, SD_LBP_DISABLE);
3255 }
3256 }
3257
3258 out:
3259 rcu_read_unlock();
3260}
3261
3262/* Parse the Block Limits Extension VPD page (0xb7) */
3263static void sd_read_block_limits_ext(struct scsi_disk *sdkp)
3264{
3265 struct scsi_vpd *vpd;
3266
3267 rcu_read_lock();
3268 vpd = rcu_dereference(sdkp->device->vpd_pgb7);
3269 if (vpd && vpd->len >= 2)
3270 sdkp->rscs = vpd->data[5] & 1;
3271 rcu_read_unlock();
3272}
3273
3274/**
3275 * sd_read_block_characteristics - Query block dev. characteristics
3276 * @sdkp: disk to query
3277 */
3278static void sd_read_block_characteristics(struct scsi_disk *sdkp)
3279{
3280 struct request_queue *q = sdkp->disk->queue;
3281 struct scsi_vpd *vpd;
3282 u16 rot;
3283
3284 rcu_read_lock();
3285 vpd = rcu_dereference(sdkp->device->vpd_pgb1);
3286
3287 if (!vpd || vpd->len < 8) {
3288 rcu_read_unlock();
3289 return;
3290 }
3291
3292 rot = get_unaligned_be16(&vpd->data[4]);
3293 sdkp->zoned = (vpd->data[8] >> 4) & 3;
3294 rcu_read_unlock();
3295
3296 if (rot == 1) {
3297 blk_queue_flag_set(QUEUE_FLAG_NONROT, q);
3298 blk_queue_flag_clear(QUEUE_FLAG_ADD_RANDOM, q);
3299 }
3300
3301
3302#ifdef CONFIG_BLK_DEV_ZONED /* sd_probe rejects ZBD devices early otherwise */
3303 if (sdkp->device->type == TYPE_ZBC) {
3304 /*
3305 * Host-managed.
3306 */
3307 disk_set_zoned(sdkp->disk);
3308
3309 /*
3310 * Per ZBC and ZAC specifications, writes in sequential write
3311 * required zones of host-managed devices must be aligned to
3312 * the device physical block size.
3313 */
3314 blk_queue_zone_write_granularity(q, sdkp->physical_block_size);
3315 } else {
3316 /*
3317 * Host-aware devices are treated as conventional.
3318 */
3319 WARN_ON_ONCE(blk_queue_is_zoned(q));
3320 }
3321#endif /* CONFIG_BLK_DEV_ZONED */
3322
3323 if (!sdkp->first_scan)
3324 return;
3325
3326 if (blk_queue_is_zoned(q))
3327 sd_printk(KERN_NOTICE, sdkp, "Host-managed zoned block device\n");
3328 else if (sdkp->zoned == 1)
3329 sd_printk(KERN_NOTICE, sdkp, "Host-aware SMR disk used as regular disk\n");
3330 else if (sdkp->zoned == 2)
3331 sd_printk(KERN_NOTICE, sdkp, "Drive-managed SMR disk\n");
3332}
3333
3334/**
3335 * sd_read_block_provisioning - Query provisioning VPD page
3336 * @sdkp: disk to query
3337 */
3338static void sd_read_block_provisioning(struct scsi_disk *sdkp)
3339{
3340 struct scsi_vpd *vpd;
3341
3342 if (sdkp->lbpme == 0)
3343 return;
3344
3345 rcu_read_lock();
3346 vpd = rcu_dereference(sdkp->device->vpd_pgb2);
3347
3348 if (!vpd || vpd->len < 8) {
3349 rcu_read_unlock();
3350 return;
3351 }
3352
3353 sdkp->lbpvpd = 1;
3354 sdkp->lbpu = (vpd->data[5] >> 7) & 1; /* UNMAP */
3355 sdkp->lbpws = (vpd->data[5] >> 6) & 1; /* WRITE SAME(16) w/ UNMAP */
3356 sdkp->lbpws10 = (vpd->data[5] >> 5) & 1; /* WRITE SAME(10) w/ UNMAP */
3357 rcu_read_unlock();
3358}
3359
3360static void sd_read_write_same(struct scsi_disk *sdkp, unsigned char *buffer)
3361{
3362 struct scsi_device *sdev = sdkp->device;
3363
3364 if (sdev->host->no_write_same) {
3365 sdev->no_write_same = 1;
3366
3367 return;
3368 }
3369
3370 if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE, INQUIRY, 0) < 0) {
3371 struct scsi_vpd *vpd;
3372
3373 sdev->no_report_opcodes = 1;
3374
3375 /* Disable WRITE SAME if REPORT SUPPORTED OPERATION
3376 * CODES is unsupported and the device has an ATA
3377 * Information VPD page (SAT).
3378 */
3379 rcu_read_lock();
3380 vpd = rcu_dereference(sdev->vpd_pg89);
3381 if (vpd)
3382 sdev->no_write_same = 1;
3383 rcu_read_unlock();
3384 }
3385
3386 if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE, WRITE_SAME_16, 0) == 1)
3387 sdkp->ws16 = 1;
3388
3389 if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE, WRITE_SAME, 0) == 1)
3390 sdkp->ws10 = 1;
3391}
3392
3393static void sd_read_security(struct scsi_disk *sdkp, unsigned char *buffer)
3394{
3395 struct scsi_device *sdev = sdkp->device;
3396
3397 if (!sdev->security_supported)
3398 return;
3399
3400 if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE,
3401 SECURITY_PROTOCOL_IN, 0) == 1 &&
3402 scsi_report_opcode(sdev, buffer, SD_BUF_SIZE,
3403 SECURITY_PROTOCOL_OUT, 0) == 1)
3404 sdkp->security = 1;
3405}
3406
3407static inline sector_t sd64_to_sectors(struct scsi_disk *sdkp, u8 *buf)
3408{
3409 return logical_to_sectors(sdkp->device, get_unaligned_be64(buf));
3410}
3411
3412/**
3413 * sd_read_cpr - Query concurrent positioning ranges
3414 * @sdkp: disk to query
3415 */
3416static void sd_read_cpr(struct scsi_disk *sdkp)
3417{
3418 struct blk_independent_access_ranges *iars = NULL;
3419 unsigned char *buffer = NULL;
3420 unsigned int nr_cpr = 0;
3421 int i, vpd_len, buf_len = SD_BUF_SIZE;
3422 u8 *desc;
3423
3424 /*
3425 * We need to have the capacity set first for the block layer to be
3426 * able to check the ranges.
3427 */
3428 if (sdkp->first_scan)
3429 return;
3430
3431 if (!sdkp->capacity)
3432 goto out;
3433
3434 /*
3435 * Concurrent Positioning Ranges VPD: there can be at most 256 ranges,
3436 * leading to a maximum page size of 64 + 256*32 bytes.
3437 */
3438 buf_len = 64 + 256*32;
3439 buffer = kmalloc(buf_len, GFP_KERNEL);
3440 if (!buffer || scsi_get_vpd_page(sdkp->device, 0xb9, buffer, buf_len))
3441 goto out;
3442
3443 /* We must have at least a 64B header and one 32B range descriptor */
3444 vpd_len = get_unaligned_be16(&buffer[2]) + 4;
3445 if (vpd_len > buf_len || vpd_len < 64 + 32 || (vpd_len & 31)) {
3446 sd_printk(KERN_ERR, sdkp,
3447 "Invalid Concurrent Positioning Ranges VPD page\n");
3448 goto out;
3449 }
3450
3451 nr_cpr = (vpd_len - 64) / 32;
3452 if (nr_cpr == 1) {
3453 nr_cpr = 0;
3454 goto out;
3455 }
3456
3457 iars = disk_alloc_independent_access_ranges(sdkp->disk, nr_cpr);
3458 if (!iars) {
3459 nr_cpr = 0;
3460 goto out;
3461 }
3462
3463 desc = &buffer[64];
3464 for (i = 0; i < nr_cpr; i++, desc += 32) {
3465 if (desc[0] != i) {
3466 sd_printk(KERN_ERR, sdkp,
3467 "Invalid Concurrent Positioning Range number\n");
3468 nr_cpr = 0;
3469 break;
3470 }
3471
3472 iars->ia_range[i].sector = sd64_to_sectors(sdkp, desc + 8);
3473 iars->ia_range[i].nr_sectors = sd64_to_sectors(sdkp, desc + 16);
3474 }
3475
3476out:
3477 disk_set_independent_access_ranges(sdkp->disk, iars);
3478 if (nr_cpr && sdkp->nr_actuators != nr_cpr) {
3479 sd_printk(KERN_NOTICE, sdkp,
3480 "%u concurrent positioning ranges\n", nr_cpr);
3481 sdkp->nr_actuators = nr_cpr;
3482 }
3483
3484 kfree(buffer);
3485}
3486
3487static bool sd_validate_min_xfer_size(struct scsi_disk *sdkp)
3488{
3489 struct scsi_device *sdp = sdkp->device;
3490 unsigned int min_xfer_bytes =
3491 logical_to_bytes(sdp, sdkp->min_xfer_blocks);
3492
3493 if (sdkp->min_xfer_blocks == 0)
3494 return false;
3495
3496 if (min_xfer_bytes & (sdkp->physical_block_size - 1)) {
3497 sd_first_printk(KERN_WARNING, sdkp,
3498 "Preferred minimum I/O size %u bytes not a " \
3499 "multiple of physical block size (%u bytes)\n",
3500 min_xfer_bytes, sdkp->physical_block_size);
3501 sdkp->min_xfer_blocks = 0;
3502 return false;
3503 }
3504
3505 sd_first_printk(KERN_INFO, sdkp, "Preferred minimum I/O size %u bytes\n",
3506 min_xfer_bytes);
3507 return true;
3508}
3509
3510/*
3511 * Determine the device's preferred I/O size for reads and writes
3512 * unless the reported value is unreasonably small, large, not a
3513 * multiple of the physical block size, or simply garbage.
3514 */
3515static bool sd_validate_opt_xfer_size(struct scsi_disk *sdkp,
3516 unsigned int dev_max)
3517{
3518 struct scsi_device *sdp = sdkp->device;
3519 unsigned int opt_xfer_bytes =
3520 logical_to_bytes(sdp, sdkp->opt_xfer_blocks);
3521 unsigned int min_xfer_bytes =
3522 logical_to_bytes(sdp, sdkp->min_xfer_blocks);
3523
3524 if (sdkp->opt_xfer_blocks == 0)
3525 return false;
3526
3527 if (sdkp->opt_xfer_blocks > dev_max) {
3528 sd_first_printk(KERN_WARNING, sdkp,
3529 "Optimal transfer size %u logical blocks " \
3530 "> dev_max (%u logical blocks)\n",
3531 sdkp->opt_xfer_blocks, dev_max);
3532 return false;
3533 }
3534
3535 if (sdkp->opt_xfer_blocks > SD_DEF_XFER_BLOCKS) {
3536 sd_first_printk(KERN_WARNING, sdkp,
3537 "Optimal transfer size %u logical blocks " \
3538 "> sd driver limit (%u logical blocks)\n",
3539 sdkp->opt_xfer_blocks, SD_DEF_XFER_BLOCKS);
3540 return false;
3541 }
3542
3543 if (opt_xfer_bytes < PAGE_SIZE) {
3544 sd_first_printk(KERN_WARNING, sdkp,
3545 "Optimal transfer size %u bytes < " \
3546 "PAGE_SIZE (%u bytes)\n",
3547 opt_xfer_bytes, (unsigned int)PAGE_SIZE);
3548 return false;
3549 }
3550
3551 if (min_xfer_bytes && opt_xfer_bytes % min_xfer_bytes) {
3552 sd_first_printk(KERN_WARNING, sdkp,
3553 "Optimal transfer size %u bytes not a " \
3554 "multiple of preferred minimum block " \
3555 "size (%u bytes)\n",
3556 opt_xfer_bytes, min_xfer_bytes);
3557 return false;
3558 }
3559
3560 if (opt_xfer_bytes & (sdkp->physical_block_size - 1)) {
3561 sd_first_printk(KERN_WARNING, sdkp,
3562 "Optimal transfer size %u bytes not a " \
3563 "multiple of physical block size (%u bytes)\n",
3564 opt_xfer_bytes, sdkp->physical_block_size);
3565 return false;
3566 }
3567
3568 sd_first_printk(KERN_INFO, sdkp, "Optimal transfer size %u bytes\n",
3569 opt_xfer_bytes);
3570 return true;
3571}
3572
3573static void sd_read_block_zero(struct scsi_disk *sdkp)
3574{
3575 unsigned int buf_len = sdkp->device->sector_size;
3576 char *buffer, cmd[10] = { };
3577
3578 buffer = kmalloc(buf_len, GFP_KERNEL);
3579 if (!buffer)
3580 return;
3581
3582 cmd[0] = READ_10;
3583 put_unaligned_be32(0, &cmd[2]); /* Logical block address 0 */
3584 put_unaligned_be16(1, &cmd[7]); /* Transfer 1 logical block */
3585
3586 scsi_execute_cmd(sdkp->device, cmd, REQ_OP_DRV_IN, buffer, buf_len,
3587 SD_TIMEOUT, sdkp->max_retries, NULL);
3588 kfree(buffer);
3589}
3590
3591/**
3592 * sd_revalidate_disk - called the first time a new disk is seen,
3593 * performs disk spin up, read_capacity, etc.
3594 * @disk: struct gendisk we care about
3595 **/
3596static int sd_revalidate_disk(struct gendisk *disk)
3597{
3598 struct scsi_disk *sdkp = scsi_disk(disk);
3599 struct scsi_device *sdp = sdkp->device;
3600 struct request_queue *q = sdkp->disk->queue;
3601 sector_t old_capacity = sdkp->capacity;
3602 unsigned char *buffer;
3603 unsigned int dev_max, rw_max;
3604
3605 SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp,
3606 "sd_revalidate_disk\n"));
3607
3608 /*
3609 * If the device is offline, don't try and read capacity or any
3610 * of the other niceties.
3611 */
3612 if (!scsi_device_online(sdp))
3613 goto out;
3614
3615 buffer = kmalloc(SD_BUF_SIZE, GFP_KERNEL);
3616 if (!buffer) {
3617 sd_printk(KERN_WARNING, sdkp, "sd_revalidate_disk: Memory "
3618 "allocation failure.\n");
3619 goto out;
3620 }
3621
3622 sd_spinup_disk(sdkp);
3623
3624 /*
3625 * Without media there is no reason to ask; moreover, some devices
3626 * react badly if we do.
3627 */
3628 if (sdkp->media_present) {
3629 sd_read_capacity(sdkp, buffer);
3630 /*
3631 * Some USB/UAS devices return generic values for mode pages
3632 * until the media has been accessed. Trigger a READ operation
3633 * to force the device to populate mode pages.
3634 */
3635 if (sdp->read_before_ms)
3636 sd_read_block_zero(sdkp);
3637 /*
3638 * set the default to rotational. All non-rotational devices
3639 * support the block characteristics VPD page, which will
3640 * cause this to be updated correctly and any device which
3641 * doesn't support it should be treated as rotational.
3642 */
3643 blk_queue_flag_clear(QUEUE_FLAG_NONROT, q);
3644 blk_queue_flag_set(QUEUE_FLAG_ADD_RANDOM, q);
3645
3646 if (scsi_device_supports_vpd(sdp)) {
3647 sd_read_block_provisioning(sdkp);
3648 sd_read_block_limits(sdkp);
3649 sd_read_block_limits_ext(sdkp);
3650 sd_read_block_characteristics(sdkp);
3651 sd_zbc_read_zones(sdkp, buffer);
3652 sd_read_cpr(sdkp);
3653 }
3654
3655 sd_print_capacity(sdkp, old_capacity);
3656
3657 sd_read_write_protect_flag(sdkp, buffer);
3658 sd_read_cache_type(sdkp, buffer);
3659 sd_read_io_hints(sdkp, buffer);
3660 sd_read_app_tag_own(sdkp, buffer);
3661 sd_read_write_same(sdkp, buffer);
3662 sd_read_security(sdkp, buffer);
3663 sd_config_protection(sdkp);
3664 }
3665
3666 /*
3667 * We now have all cache related info, determine how we deal
3668 * with flush requests.
3669 */
3670 sd_set_flush_flag(sdkp);
3671
3672 /* Initial block count limit based on CDB TRANSFER LENGTH field size. */
3673 dev_max = sdp->use_16_for_rw ? SD_MAX_XFER_BLOCKS : SD_DEF_XFER_BLOCKS;
3674
3675 /* Some devices report a maximum block count for READ/WRITE requests. */
3676 dev_max = min_not_zero(dev_max, sdkp->max_xfer_blocks);
3677 q->limits.max_dev_sectors = logical_to_sectors(sdp, dev_max);
3678
3679 if (sd_validate_min_xfer_size(sdkp))
3680 blk_queue_io_min(sdkp->disk->queue,
3681 logical_to_bytes(sdp, sdkp->min_xfer_blocks));
3682 else
3683 blk_queue_io_min(sdkp->disk->queue, 0);
3684
3685 if (sd_validate_opt_xfer_size(sdkp, dev_max)) {
3686 q->limits.io_opt = logical_to_bytes(sdp, sdkp->opt_xfer_blocks);
3687 rw_max = logical_to_sectors(sdp, sdkp->opt_xfer_blocks);
3688 } else {
3689 q->limits.io_opt = 0;
3690 rw_max = min_not_zero(logical_to_sectors(sdp, dev_max),
3691 (sector_t)BLK_DEF_MAX_SECTORS_CAP);
3692 }
3693
3694 /*
3695 * Limit default to SCSI host optimal sector limit if set. There may be
3696 * an impact on performance for when the size of a request exceeds this
3697 * host limit.
3698 */
3699 rw_max = min_not_zero(rw_max, sdp->host->opt_sectors);
3700
3701 /* Do not exceed controller limit */
3702 rw_max = min(rw_max, queue_max_hw_sectors(q));
3703
3704 /*
3705 * Only update max_sectors if previously unset or if the current value
3706 * exceeds the capabilities of the hardware.
3707 */
3708 if (sdkp->first_scan ||
3709 q->limits.max_sectors > q->limits.max_dev_sectors ||
3710 q->limits.max_sectors > q->limits.max_hw_sectors) {
3711 q->limits.max_sectors = rw_max;
3712 q->limits.max_user_sectors = rw_max;
3713 }
3714
3715 sdkp->first_scan = 0;
3716
3717 set_capacity_and_notify(disk, logical_to_sectors(sdp, sdkp->capacity));
3718 sd_config_write_same(sdkp);
3719 kfree(buffer);
3720
3721 /*
3722 * For a zoned drive, revalidating the zones can be done only once
3723 * the gendisk capacity is set. So if this fails, set back the gendisk
3724 * capacity to 0.
3725 */
3726 if (sd_zbc_revalidate_zones(sdkp))
3727 set_capacity_and_notify(disk, 0);
3728
3729 out:
3730 return 0;
3731}
3732
3733/**
3734 * sd_unlock_native_capacity - unlock native capacity
3735 * @disk: struct gendisk to set capacity for
3736 *
3737 * Block layer calls this function if it detects that partitions
3738 * on @disk reach beyond the end of the device. If the SCSI host
3739 * implements ->unlock_native_capacity() method, it's invoked to
3740 * give it a chance to adjust the device capacity.
3741 *
3742 * CONTEXT:
3743 * Defined by block layer. Might sleep.
3744 */
3745static void sd_unlock_native_capacity(struct gendisk *disk)
3746{
3747 struct scsi_device *sdev = scsi_disk(disk)->device;
3748
3749 if (sdev->host->hostt->unlock_native_capacity)
3750 sdev->host->hostt->unlock_native_capacity(sdev);
3751}
3752
3753/**
3754 * sd_format_disk_name - format disk name
3755 * @prefix: name prefix - ie. "sd" for SCSI disks
3756 * @index: index of the disk to format name for
3757 * @buf: output buffer
3758 * @buflen: length of the output buffer
3759 *
3760 * SCSI disk names starts at sda. The 26th device is sdz and the
3761 * 27th is sdaa. The last one for two lettered suffix is sdzz
3762 * which is followed by sdaaa.
3763 *
3764 * This is basically 26 base counting with one extra 'nil' entry
3765 * at the beginning from the second digit on and can be
3766 * determined using similar method as 26 base conversion with the
3767 * index shifted -1 after each digit is computed.
3768 *
3769 * CONTEXT:
3770 * Don't care.
3771 *
3772 * RETURNS:
3773 * 0 on success, -errno on failure.
3774 */
3775static int sd_format_disk_name(char *prefix, int index, char *buf, int buflen)
3776{
3777 const int base = 'z' - 'a' + 1;
3778 char *begin = buf + strlen(prefix);
3779 char *end = buf + buflen;
3780 char *p;
3781 int unit;
3782
3783 p = end - 1;
3784 *p = '\0';
3785 unit = base;
3786 do {
3787 if (p == begin)
3788 return -EINVAL;
3789 *--p = 'a' + (index % unit);
3790 index = (index / unit) - 1;
3791 } while (index >= 0);
3792
3793 memmove(begin, p, end - p);
3794 memcpy(buf, prefix, strlen(prefix));
3795
3796 return 0;
3797}
3798
3799/**
3800 * sd_probe - called during driver initialization and whenever a
3801 * new scsi device is attached to the system. It is called once
3802 * for each scsi device (not just disks) present.
3803 * @dev: pointer to device object
3804 *
3805 * Returns 0 if successful (or not interested in this scsi device
3806 * (e.g. scanner)); 1 when there is an error.
3807 *
3808 * Note: this function is invoked from the scsi mid-level.
3809 * This function sets up the mapping between a given
3810 * <host,channel,id,lun> (found in sdp) and new device name
3811 * (e.g. /dev/sda). More precisely it is the block device major
3812 * and minor number that is chosen here.
3813 *
3814 * Assume sd_probe is not re-entrant (for time being)
3815 * Also think about sd_probe() and sd_remove() running coincidentally.
3816 **/
3817static int sd_probe(struct device *dev)
3818{
3819 struct scsi_device *sdp = to_scsi_device(dev);
3820 struct scsi_disk *sdkp;
3821 struct gendisk *gd;
3822 int index;
3823 int error;
3824
3825 scsi_autopm_get_device(sdp);
3826 error = -ENODEV;
3827 if (sdp->type != TYPE_DISK &&
3828 sdp->type != TYPE_ZBC &&
3829 sdp->type != TYPE_MOD &&
3830 sdp->type != TYPE_RBC)
3831 goto out;
3832
3833 if (!IS_ENABLED(CONFIG_BLK_DEV_ZONED) && sdp->type == TYPE_ZBC) {
3834 sdev_printk(KERN_WARNING, sdp,
3835 "Unsupported ZBC host-managed device.\n");
3836 goto out;
3837 }
3838
3839 SCSI_LOG_HLQUEUE(3, sdev_printk(KERN_INFO, sdp,
3840 "sd_probe\n"));
3841
3842 error = -ENOMEM;
3843 sdkp = kzalloc(sizeof(*sdkp), GFP_KERNEL);
3844 if (!sdkp)
3845 goto out;
3846
3847 gd = blk_mq_alloc_disk_for_queue(sdp->request_queue,
3848 &sd_bio_compl_lkclass);
3849 if (!gd)
3850 goto out_free;
3851
3852 index = ida_alloc(&sd_index_ida, GFP_KERNEL);
3853 if (index < 0) {
3854 sdev_printk(KERN_WARNING, sdp, "sd_probe: memory exhausted.\n");
3855 goto out_put;
3856 }
3857
3858 error = sd_format_disk_name("sd", index, gd->disk_name, DISK_NAME_LEN);
3859 if (error) {
3860 sdev_printk(KERN_WARNING, sdp, "SCSI disk (sd) name length exceeded.\n");
3861 goto out_free_index;
3862 }
3863
3864 sdkp->device = sdp;
3865 sdkp->disk = gd;
3866 sdkp->index = index;
3867 sdkp->max_retries = SD_MAX_RETRIES;
3868 atomic_set(&sdkp->openers, 0);
3869 atomic_set(&sdkp->device->ioerr_cnt, 0);
3870
3871 if (!sdp->request_queue->rq_timeout) {
3872 if (sdp->type != TYPE_MOD)
3873 blk_queue_rq_timeout(sdp->request_queue, SD_TIMEOUT);
3874 else
3875 blk_queue_rq_timeout(sdp->request_queue,
3876 SD_MOD_TIMEOUT);
3877 }
3878
3879 device_initialize(&sdkp->disk_dev);
3880 sdkp->disk_dev.parent = get_device(dev);
3881 sdkp->disk_dev.class = &sd_disk_class;
3882 dev_set_name(&sdkp->disk_dev, "%s", dev_name(dev));
3883
3884 error = device_add(&sdkp->disk_dev);
3885 if (error) {
3886 put_device(&sdkp->disk_dev);
3887 goto out;
3888 }
3889
3890 dev_set_drvdata(dev, sdkp);
3891
3892 gd->major = sd_major((index & 0xf0) >> 4);
3893 gd->first_minor = ((index & 0xf) << 4) | (index & 0xfff00);
3894 gd->minors = SD_MINORS;
3895
3896 gd->fops = &sd_fops;
3897 gd->private_data = sdkp;
3898
3899 /* defaults, until the device tells us otherwise */
3900 sdp->sector_size = 512;
3901 sdkp->capacity = 0;
3902 sdkp->media_present = 1;
3903 sdkp->write_prot = 0;
3904 sdkp->cache_override = 0;
3905 sdkp->WCE = 0;
3906 sdkp->RCD = 0;
3907 sdkp->ATO = 0;
3908 sdkp->first_scan = 1;
3909 sdkp->max_medium_access_timeouts = SD_MAX_MEDIUM_TIMEOUTS;
3910
3911 sd_revalidate_disk(gd);
3912
3913 if (sdp->removable) {
3914 gd->flags |= GENHD_FL_REMOVABLE;
3915 gd->events |= DISK_EVENT_MEDIA_CHANGE;
3916 gd->event_flags = DISK_EVENT_FLAG_POLL | DISK_EVENT_FLAG_UEVENT;
3917 }
3918
3919 blk_pm_runtime_init(sdp->request_queue, dev);
3920 if (sdp->rpm_autosuspend) {
3921 pm_runtime_set_autosuspend_delay(dev,
3922 sdp->host->rpm_autosuspend_delay);
3923 }
3924
3925 error = device_add_disk(dev, gd, NULL);
3926 if (error) {
3927 device_unregister(&sdkp->disk_dev);
3928 put_disk(gd);
3929 goto out;
3930 }
3931
3932 if (sdkp->security) {
3933 sdkp->opal_dev = init_opal_dev(sdkp, &sd_sec_submit);
3934 if (sdkp->opal_dev)
3935 sd_printk(KERN_NOTICE, sdkp, "supports TCG Opal\n");
3936 }
3937
3938 sd_printk(KERN_NOTICE, sdkp, "Attached SCSI %sdisk\n",
3939 sdp->removable ? "removable " : "");
3940 scsi_autopm_put_device(sdp);
3941
3942 return 0;
3943
3944 out_free_index:
3945 ida_free(&sd_index_ida, index);
3946 out_put:
3947 put_disk(gd);
3948 out_free:
3949 kfree(sdkp);
3950 out:
3951 scsi_autopm_put_device(sdp);
3952 return error;
3953}
3954
3955/**
3956 * sd_remove - called whenever a scsi disk (previously recognized by
3957 * sd_probe) is detached from the system. It is called (potentially
3958 * multiple times) during sd module unload.
3959 * @dev: pointer to device object
3960 *
3961 * Note: this function is invoked from the scsi mid-level.
3962 * This function potentially frees up a device name (e.g. /dev/sdc)
3963 * that could be re-used by a subsequent sd_probe().
3964 * This function is not called when the built-in sd driver is "exit-ed".
3965 **/
3966static int sd_remove(struct device *dev)
3967{
3968 struct scsi_disk *sdkp = dev_get_drvdata(dev);
3969
3970 scsi_autopm_get_device(sdkp->device);
3971
3972 device_del(&sdkp->disk_dev);
3973 del_gendisk(sdkp->disk);
3974 if (!sdkp->suspended)
3975 sd_shutdown(dev);
3976
3977 put_disk(sdkp->disk);
3978 return 0;
3979}
3980
3981static void scsi_disk_release(struct device *dev)
3982{
3983 struct scsi_disk *sdkp = to_scsi_disk(dev);
3984
3985 ida_free(&sd_index_ida, sdkp->index);
3986 sd_zbc_free_zone_info(sdkp);
3987 put_device(&sdkp->device->sdev_gendev);
3988 free_opal_dev(sdkp->opal_dev);
3989
3990 kfree(sdkp);
3991}
3992
3993static int sd_start_stop_device(struct scsi_disk *sdkp, int start)
3994{
3995 unsigned char cmd[6] = { START_STOP }; /* START_VALID */
3996 struct scsi_sense_hdr sshdr;
3997 const struct scsi_exec_args exec_args = {
3998 .sshdr = &sshdr,
3999 .req_flags = BLK_MQ_REQ_PM,
4000 };
4001 struct scsi_device *sdp = sdkp->device;
4002 int res;
4003
4004 if (start)
4005 cmd[4] |= 1; /* START */
4006
4007 if (sdp->start_stop_pwr_cond)
4008 cmd[4] |= start ? 1 << 4 : 3 << 4; /* Active or Standby */
4009
4010 if (!scsi_device_online(sdp))
4011 return -ENODEV;
4012
4013 res = scsi_execute_cmd(sdp, cmd, REQ_OP_DRV_IN, NULL, 0, SD_TIMEOUT,
4014 sdkp->max_retries, &exec_args);
4015 if (res) {
4016 sd_print_result(sdkp, "Start/Stop Unit failed", res);
4017 if (res > 0 && scsi_sense_valid(&sshdr)) {
4018 sd_print_sense_hdr(sdkp, &sshdr);
4019 /* 0x3a is medium not present */
4020 if (sshdr.asc == 0x3a)
4021 res = 0;
4022 }
4023 }
4024
4025 /* SCSI error codes must not go to the generic layer */
4026 if (res)
4027 return -EIO;
4028
4029 return 0;
4030}
4031
4032/*
4033 * Send a SYNCHRONIZE CACHE instruction down to the device through
4034 * the normal SCSI command structure. Wait for the command to
4035 * complete.
4036 */
4037static void sd_shutdown(struct device *dev)
4038{
4039 struct scsi_disk *sdkp = dev_get_drvdata(dev);
4040
4041 if (!sdkp)
4042 return; /* this can happen */
4043
4044 if (pm_runtime_suspended(dev))
4045 return;
4046
4047 if (sdkp->WCE && sdkp->media_present) {
4048 sd_printk(KERN_NOTICE, sdkp, "Synchronizing SCSI cache\n");
4049 sd_sync_cache(sdkp);
4050 }
4051
4052 if ((system_state != SYSTEM_RESTART &&
4053 sdkp->device->manage_system_start_stop) ||
4054 (system_state == SYSTEM_POWER_OFF &&
4055 sdkp->device->manage_shutdown)) {
4056 sd_printk(KERN_NOTICE, sdkp, "Stopping disk\n");
4057 sd_start_stop_device(sdkp, 0);
4058 }
4059}
4060
4061static inline bool sd_do_start_stop(struct scsi_device *sdev, bool runtime)
4062{
4063 return (sdev->manage_system_start_stop && !runtime) ||
4064 (sdev->manage_runtime_start_stop && runtime);
4065}
4066
4067static int sd_suspend_common(struct device *dev, bool runtime)
4068{
4069 struct scsi_disk *sdkp = dev_get_drvdata(dev);
4070 int ret = 0;
4071
4072 if (!sdkp) /* E.g.: runtime suspend following sd_remove() */
4073 return 0;
4074
4075 if (sdkp->WCE && sdkp->media_present) {
4076 if (!sdkp->device->silence_suspend)
4077 sd_printk(KERN_NOTICE, sdkp, "Synchronizing SCSI cache\n");
4078 ret = sd_sync_cache(sdkp);
4079 /* ignore OFFLINE device */
4080 if (ret == -ENODEV)
4081 return 0;
4082
4083 if (ret)
4084 return ret;
4085 }
4086
4087 if (sd_do_start_stop(sdkp->device, runtime)) {
4088 if (!sdkp->device->silence_suspend)
4089 sd_printk(KERN_NOTICE, sdkp, "Stopping disk\n");
4090 /* an error is not worth aborting a system sleep */
4091 ret = sd_start_stop_device(sdkp, 0);
4092 if (!runtime)
4093 ret = 0;
4094 }
4095
4096 if (!ret)
4097 sdkp->suspended = true;
4098
4099 return ret;
4100}
4101
4102static int sd_suspend_system(struct device *dev)
4103{
4104 if (pm_runtime_suspended(dev))
4105 return 0;
4106
4107 return sd_suspend_common(dev, false);
4108}
4109
4110static int sd_suspend_runtime(struct device *dev)
4111{
4112 return sd_suspend_common(dev, true);
4113}
4114
4115static int sd_resume(struct device *dev)
4116{
4117 struct scsi_disk *sdkp = dev_get_drvdata(dev);
4118
4119 sd_printk(KERN_NOTICE, sdkp, "Starting disk\n");
4120
4121 if (opal_unlock_from_suspend(sdkp->opal_dev)) {
4122 sd_printk(KERN_NOTICE, sdkp, "OPAL unlock failed\n");
4123 return -EIO;
4124 }
4125
4126 return 0;
4127}
4128
4129static int sd_resume_common(struct device *dev, bool runtime)
4130{
4131 struct scsi_disk *sdkp = dev_get_drvdata(dev);
4132 int ret;
4133
4134 if (!sdkp) /* E.g.: runtime resume at the start of sd_probe() */
4135 return 0;
4136
4137 if (!sd_do_start_stop(sdkp->device, runtime)) {
4138 sdkp->suspended = false;
4139 return 0;
4140 }
4141
4142 sd_printk(KERN_NOTICE, sdkp, "Starting disk\n");
4143 ret = sd_start_stop_device(sdkp, 1);
4144 if (!ret) {
4145 sd_resume(dev);
4146 sdkp->suspended = false;
4147 }
4148
4149 return ret;
4150}
4151
4152static int sd_resume_system(struct device *dev)
4153{
4154 if (pm_runtime_suspended(dev)) {
4155 struct scsi_disk *sdkp = dev_get_drvdata(dev);
4156 struct scsi_device *sdp = sdkp ? sdkp->device : NULL;
4157
4158 if (sdp && sdp->force_runtime_start_on_system_start)
4159 pm_request_resume(dev);
4160
4161 return 0;
4162 }
4163
4164 return sd_resume_common(dev, false);
4165}
4166
4167static int sd_resume_runtime(struct device *dev)
4168{
4169 struct scsi_disk *sdkp = dev_get_drvdata(dev);
4170 struct scsi_device *sdp;
4171
4172 if (!sdkp) /* E.g.: runtime resume at the start of sd_probe() */
4173 return 0;
4174
4175 sdp = sdkp->device;
4176
4177 if (sdp->ignore_media_change) {
4178 /* clear the device's sense data */
4179 static const u8 cmd[10] = { REQUEST_SENSE };
4180 const struct scsi_exec_args exec_args = {
4181 .req_flags = BLK_MQ_REQ_PM,
4182 };
4183
4184 if (scsi_execute_cmd(sdp, cmd, REQ_OP_DRV_IN, NULL, 0,
4185 sdp->request_queue->rq_timeout, 1,
4186 &exec_args))
4187 sd_printk(KERN_NOTICE, sdkp,
4188 "Failed to clear sense data\n");
4189 }
4190
4191 return sd_resume_common(dev, true);
4192}
4193
4194static const struct dev_pm_ops sd_pm_ops = {
4195 .suspend = sd_suspend_system,
4196 .resume = sd_resume_system,
4197 .poweroff = sd_suspend_system,
4198 .restore = sd_resume_system,
4199 .runtime_suspend = sd_suspend_runtime,
4200 .runtime_resume = sd_resume_runtime,
4201};
4202
4203static struct scsi_driver sd_template = {
4204 .gendrv = {
4205 .name = "sd",
4206 .owner = THIS_MODULE,
4207 .probe = sd_probe,
4208 .probe_type = PROBE_PREFER_ASYNCHRONOUS,
4209 .remove = sd_remove,
4210 .shutdown = sd_shutdown,
4211 .pm = &sd_pm_ops,
4212 },
4213 .rescan = sd_rescan,
4214 .resume = sd_resume,
4215 .init_command = sd_init_command,
4216 .uninit_command = sd_uninit_command,
4217 .done = sd_done,
4218 .eh_action = sd_eh_action,
4219 .eh_reset = sd_eh_reset,
4220};
4221
4222/**
4223 * init_sd - entry point for this driver (both when built in or when
4224 * a module).
4225 *
4226 * Note: this function registers this driver with the scsi mid-level.
4227 **/
4228static int __init init_sd(void)
4229{
4230 int majors = 0, i, err;
4231
4232 SCSI_LOG_HLQUEUE(3, printk("init_sd: sd driver entry point\n"));
4233
4234 for (i = 0; i < SD_MAJORS; i++) {
4235 if (__register_blkdev(sd_major(i), "sd", sd_default_probe))
4236 continue;
4237 majors++;
4238 }
4239
4240 if (!majors)
4241 return -ENODEV;
4242
4243 err = class_register(&sd_disk_class);
4244 if (err)
4245 goto err_out;
4246
4247 sd_page_pool = mempool_create_page_pool(SD_MEMPOOL_SIZE, 0);
4248 if (!sd_page_pool) {
4249 printk(KERN_ERR "sd: can't init discard page pool\n");
4250 err = -ENOMEM;
4251 goto err_out_class;
4252 }
4253
4254 err = scsi_register_driver(&sd_template.gendrv);
4255 if (err)
4256 goto err_out_driver;
4257
4258 return 0;
4259
4260err_out_driver:
4261 mempool_destroy(sd_page_pool);
4262err_out_class:
4263 class_unregister(&sd_disk_class);
4264err_out:
4265 for (i = 0; i < SD_MAJORS; i++)
4266 unregister_blkdev(sd_major(i), "sd");
4267 return err;
4268}
4269
4270/**
4271 * exit_sd - exit point for this driver (when it is a module).
4272 *
4273 * Note: this function unregisters this driver from the scsi mid-level.
4274 **/
4275static void __exit exit_sd(void)
4276{
4277 int i;
4278
4279 SCSI_LOG_HLQUEUE(3, printk("exit_sd: exiting sd driver\n"));
4280
4281 scsi_unregister_driver(&sd_template.gendrv);
4282 mempool_destroy(sd_page_pool);
4283
4284 class_unregister(&sd_disk_class);
4285
4286 for (i = 0; i < SD_MAJORS; i++)
4287 unregister_blkdev(sd_major(i), "sd");
4288}
4289
4290module_init(init_sd);
4291module_exit(exit_sd);
4292
4293void sd_print_sense_hdr(struct scsi_disk *sdkp, struct scsi_sense_hdr *sshdr)
4294{
4295 scsi_print_sense_hdr(sdkp->device,
4296 sdkp->disk ? sdkp->disk->disk_name : NULL, sshdr);
4297}
4298
4299void sd_print_result(const struct scsi_disk *sdkp, const char *msg, int result)
4300{
4301 const char *hb_string = scsi_hostbyte_string(result);
4302
4303 if (hb_string)
4304 sd_printk(KERN_INFO, sdkp,
4305 "%s: Result: hostbyte=%s driverbyte=%s\n", msg,
4306 hb_string ? hb_string : "invalid",
4307 "DRIVER_OK");
4308 else
4309 sd_printk(KERN_INFO, sdkp,
4310 "%s: Result: hostbyte=0x%02x driverbyte=%s\n",
4311 msg, host_byte(result), "DRIVER_OK");
4312}