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 */
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>
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>
70 #include "sd.h"
71 #include "scsi_priv.h"
72 #include "scsi_logging.h"
74 MODULE_AUTHOR("Eric Youngdale");
75 MODULE_DESCRIPTION("SCSI disk (sd) driver");
76 MODULE_LICENSE("GPL");
78 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK0_MAJOR);
79 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK1_MAJOR);
80 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK2_MAJOR);
81 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK3_MAJOR);
82 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK4_MAJOR);
83 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK5_MAJOR);
84 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK6_MAJOR);
85 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK7_MAJOR);
86 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK8_MAJOR);
87 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK9_MAJOR);
88 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK10_MAJOR);
89 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK11_MAJOR);
90 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK12_MAJOR);
91 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK13_MAJOR);
92 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK14_MAJOR);
93 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK15_MAJOR);
94 MODULE_ALIAS_SCSI_DEVICE(TYPE_DISK);
95 MODULE_ALIAS_SCSI_DEVICE(TYPE_MOD);
96 MODULE_ALIAS_SCSI_DEVICE(TYPE_RBC);
97 MODULE_ALIAS_SCSI_DEVICE(TYPE_ZBC);
99 #if !defined(CONFIG_DEBUG_BLOCK_EXT_DEVT)
100 #define SD_MINORS 16
101 #else
102 #define SD_MINORS 0
103 #endif
105 static void sd_config_discard(struct scsi_disk *, unsigned int);
106 static void sd_config_write_same(struct scsi_disk *);
107 static int sd_revalidate_disk(struct gendisk *);
108 static void sd_unlock_native_capacity(struct gendisk *disk);
109 static int sd_probe(struct device *);
110 static int sd_remove(struct device *);
111 static void sd_shutdown(struct device *);
112 static int sd_suspend_system(struct device *);
113 static int sd_suspend_runtime(struct device *);
114 static int sd_resume(struct device *);
115 static void sd_rescan(struct device *);
116 static int sd_init_command(struct scsi_cmnd *SCpnt);
117 static void sd_uninit_command(struct scsi_cmnd *SCpnt);
118 static int sd_done(struct scsi_cmnd *);
119 static void sd_eh_reset(struct scsi_cmnd *);
120 static int sd_eh_action(struct scsi_cmnd *, int);
121 static void sd_read_capacity(struct scsi_disk *sdkp, unsigned char *buffer);
122 static void scsi_disk_release(struct device *cdev);
123 static void sd_print_sense_hdr(struct scsi_disk *, struct scsi_sense_hdr *);
124 static void sd_print_result(const struct scsi_disk *, const char *, int);
126 static DEFINE_IDA(sd_index_ida);
128 /* This semaphore is used to mediate the 0->1 reference get in the
129 * face of object destruction (i.e. we can't allow a get on an
130 * object after last put) */
131 static DEFINE_MUTEX(sd_ref_mutex);
133 static struct kmem_cache *sd_cdb_cache;
134 static mempool_t *sd_cdb_pool;
136 static const char *sd_cache_types[] = {
137 "write through", "none", "write back",
138 "write back, no read (daft)"
139 };
141 static void sd_set_flush_flag(struct scsi_disk *sdkp)
142 {
143 bool wc = false, fua = false;
145 if (sdkp->WCE) {
146 wc = true;
147 if (sdkp->DPOFUA)
148 fua = true;
149 }
151 blk_queue_write_cache(sdkp->disk->queue, wc, fua);
152 }
154 static ssize_t
155 cache_type_store(struct device *dev, struct device_attribute *attr,
156 const char *buf, size_t count)
157 {
158 int ct, rcd, wce, sp;
159 struct scsi_disk *sdkp = to_scsi_disk(dev);
160 struct scsi_device *sdp = sdkp->device;
161 char buffer[64];
162 char *buffer_data;
163 struct scsi_mode_data data;
164 struct scsi_sense_hdr sshdr;
165 static const char temp[] = "temporary ";
166 int len;
168 if (sdp->type != TYPE_DISK && sdp->type != TYPE_ZBC)
169 /* no cache control on RBC devices; theoretically they
170 * can do it, but there's probably so many exceptions
171 * it's not worth the risk */
172 return -EINVAL;
174 if (strncmp(buf, temp, sizeof(temp) - 1) == 0) {
175 buf += sizeof(temp) - 1;
176 sdkp->cache_override = 1;
177 } else {
178 sdkp->cache_override = 0;
179 }
181 ct = sysfs_match_string(sd_cache_types, buf);
182 if (ct < 0)
183 return -EINVAL;
185 rcd = ct & 0x01 ? 1 : 0;
186 wce = (ct & 0x02) && !sdkp->write_prot ? 1 : 0;
188 if (sdkp->cache_override) {
189 sdkp->WCE = wce;
190 sdkp->RCD = rcd;
191 sd_set_flush_flag(sdkp);
192 return count;
193 }
195 if (scsi_mode_sense(sdp, 0x08, 8, buffer, sizeof(buffer), SD_TIMEOUT,
196 SD_MAX_RETRIES, &data, NULL))
197 return -EINVAL;
198 len = min_t(size_t, sizeof(buffer), data.length - data.header_length -
199 data.block_descriptor_length);
200 buffer_data = buffer + data.header_length +
201 data.block_descriptor_length;
202 buffer_data[2] &= ~0x05;
203 buffer_data[2] |= wce << 2 | rcd;
204 sp = buffer_data[0] & 0x80 ? 1 : 0;
205 buffer_data[0] &= ~0x80;
207 if (scsi_mode_select(sdp, 1, sp, 8, buffer_data, len, SD_TIMEOUT,
208 SD_MAX_RETRIES, &data, &sshdr)) {
209 if (scsi_sense_valid(&sshdr))
210 sd_print_sense_hdr(sdkp, &sshdr);
211 return -EINVAL;
212 }
213 revalidate_disk(sdkp->disk);
214 return count;
215 }
217 static ssize_t
218 manage_start_stop_show(struct device *dev, struct device_attribute *attr,
219 char *buf)
220 {
221 struct scsi_disk *sdkp = to_scsi_disk(dev);
222 struct scsi_device *sdp = sdkp->device;
224 return sprintf(buf, "%u\n", sdp->manage_start_stop);
225 }
227 static ssize_t
228 manage_start_stop_store(struct device *dev, struct device_attribute *attr,
229 const char *buf, size_t count)
230 {
231 struct scsi_disk *sdkp = to_scsi_disk(dev);
232 struct scsi_device *sdp = sdkp->device;
233 bool v;
235 if (!capable(CAP_SYS_ADMIN))
236 return -EACCES;
238 if (kstrtobool(buf, &v))
239 return -EINVAL;
241 sdp->manage_start_stop = v;
243 return count;
244 }
245 static DEVICE_ATTR_RW(manage_start_stop);
247 static ssize_t
248 allow_restart_show(struct device *dev, struct device_attribute *attr, char *buf)
249 {
250 struct scsi_disk *sdkp = to_scsi_disk(dev);
252 return sprintf(buf, "%u\n", sdkp->device->allow_restart);
253 }
255 static ssize_t
256 allow_restart_store(struct device *dev, struct device_attribute *attr,
257 const char *buf, size_t count)
258 {
259 bool v;
260 struct scsi_disk *sdkp = to_scsi_disk(dev);
261 struct scsi_device *sdp = sdkp->device;
263 if (!capable(CAP_SYS_ADMIN))
264 return -EACCES;
266 if (sdp->type != TYPE_DISK && sdp->type != TYPE_ZBC)
267 return -EINVAL;
269 if (kstrtobool(buf, &v))
270 return -EINVAL;
272 sdp->allow_restart = v;
274 return count;
275 }
276 static DEVICE_ATTR_RW(allow_restart);
278 static ssize_t
279 cache_type_show(struct device *dev, struct device_attribute *attr, char *buf)
280 {
281 struct scsi_disk *sdkp = to_scsi_disk(dev);
282 int ct = sdkp->RCD + 2*sdkp->WCE;
284 return sprintf(buf, "%s\n", sd_cache_types[ct]);
285 }
286 static DEVICE_ATTR_RW(cache_type);
288 static ssize_t
289 FUA_show(struct device *dev, struct device_attribute *attr, char *buf)
290 {
291 struct scsi_disk *sdkp = to_scsi_disk(dev);
293 return sprintf(buf, "%u\n", sdkp->DPOFUA);
294 }
295 static DEVICE_ATTR_RO(FUA);
297 static ssize_t
298 protection_type_show(struct device *dev, struct device_attribute *attr,
299 char *buf)
300 {
301 struct scsi_disk *sdkp = to_scsi_disk(dev);
303 return sprintf(buf, "%u\n", sdkp->protection_type);
304 }
306 static ssize_t
307 protection_type_store(struct device *dev, struct device_attribute *attr,
308 const char *buf, size_t count)
309 {
310 struct scsi_disk *sdkp = to_scsi_disk(dev);
311 unsigned int val;
312 int err;
314 if (!capable(CAP_SYS_ADMIN))
315 return -EACCES;
317 err = kstrtouint(buf, 10, &val);
319 if (err)
320 return err;
322 if (val <= T10_PI_TYPE3_PROTECTION)
323 sdkp->protection_type = val;
325 return count;
326 }
327 static DEVICE_ATTR_RW(protection_type);
329 static ssize_t
330 protection_mode_show(struct device *dev, struct device_attribute *attr,
331 char *buf)
332 {
333 struct scsi_disk *sdkp = to_scsi_disk(dev);
334 struct scsi_device *sdp = sdkp->device;
335 unsigned int dif, dix;
337 dif = scsi_host_dif_capable(sdp->host, sdkp->protection_type);
338 dix = scsi_host_dix_capable(sdp->host, sdkp->protection_type);
340 if (!dix && scsi_host_dix_capable(sdp->host, T10_PI_TYPE0_PROTECTION)) {
341 dif = 0;
342 dix = 1;
343 }
345 if (!dif && !dix)
346 return sprintf(buf, "none\n");
348 return sprintf(buf, "%s%u\n", dix ? "dix" : "dif", dif);
349 }
350 static DEVICE_ATTR_RO(protection_mode);
352 static ssize_t
353 app_tag_own_show(struct device *dev, struct device_attribute *attr, char *buf)
354 {
355 struct scsi_disk *sdkp = to_scsi_disk(dev);
357 return sprintf(buf, "%u\n", sdkp->ATO);
358 }
359 static DEVICE_ATTR_RO(app_tag_own);
361 static ssize_t
362 thin_provisioning_show(struct device *dev, struct device_attribute *attr,
363 char *buf)
364 {
365 struct scsi_disk *sdkp = to_scsi_disk(dev);
367 return sprintf(buf, "%u\n", sdkp->lbpme);
368 }
369 static DEVICE_ATTR_RO(thin_provisioning);
371 /* sysfs_match_string() requires dense arrays */
372 static const char *lbp_mode[] = {
373 [SD_LBP_FULL] = "full",
374 [SD_LBP_UNMAP] = "unmap",
375 [SD_LBP_WS16] = "writesame_16",
376 [SD_LBP_WS10] = "writesame_10",
377 [SD_LBP_ZERO] = "writesame_zero",
378 [SD_LBP_DISABLE] = "disabled",
379 };
381 static ssize_t
382 provisioning_mode_show(struct device *dev, struct device_attribute *attr,
383 char *buf)
384 {
385 struct scsi_disk *sdkp = to_scsi_disk(dev);
387 return sprintf(buf, "%s\n", lbp_mode[sdkp->provisioning_mode]);
388 }
390 static ssize_t
391 provisioning_mode_store(struct device *dev, struct device_attribute *attr,
392 const char *buf, size_t count)
393 {
394 struct scsi_disk *sdkp = to_scsi_disk(dev);
395 struct scsi_device *sdp = sdkp->device;
396 int mode;
398 if (!capable(CAP_SYS_ADMIN))
399 return -EACCES;
401 if (sd_is_zoned(sdkp)) {
402 sd_config_discard(sdkp, SD_LBP_DISABLE);
403 return count;
404 }
406 if (sdp->type != TYPE_DISK)
407 return -EINVAL;
409 mode = sysfs_match_string(lbp_mode, buf);
410 if (mode < 0)
411 return -EINVAL;
413 sd_config_discard(sdkp, mode);
415 return count;
416 }
417 static DEVICE_ATTR_RW(provisioning_mode);
419 /* sysfs_match_string() requires dense arrays */
420 static const char *zeroing_mode[] = {
421 [SD_ZERO_WRITE] = "write",
422 [SD_ZERO_WS] = "writesame",
423 [SD_ZERO_WS16_UNMAP] = "writesame_16_unmap",
424 [SD_ZERO_WS10_UNMAP] = "writesame_10_unmap",
425 };
427 static ssize_t
428 zeroing_mode_show(struct device *dev, struct device_attribute *attr,
429 char *buf)
430 {
431 struct scsi_disk *sdkp = to_scsi_disk(dev);
433 return sprintf(buf, "%s\n", zeroing_mode[sdkp->zeroing_mode]);
434 }
436 static ssize_t
437 zeroing_mode_store(struct device *dev, struct device_attribute *attr,
438 const char *buf, size_t count)
439 {
440 struct scsi_disk *sdkp = to_scsi_disk(dev);
441 int mode;
443 if (!capable(CAP_SYS_ADMIN))
444 return -EACCES;
446 mode = sysfs_match_string(zeroing_mode, buf);
447 if (mode < 0)
448 return -EINVAL;
450 sdkp->zeroing_mode = mode;
452 return count;
453 }
454 static DEVICE_ATTR_RW(zeroing_mode);
456 static ssize_t
457 max_medium_access_timeouts_show(struct device *dev,
458 struct device_attribute *attr, char *buf)
459 {
460 struct scsi_disk *sdkp = to_scsi_disk(dev);
462 return sprintf(buf, "%u\n", sdkp->max_medium_access_timeouts);
463 }
465 static ssize_t
466 max_medium_access_timeouts_store(struct device *dev,
467 struct device_attribute *attr, const char *buf,
468 size_t count)
469 {
470 struct scsi_disk *sdkp = to_scsi_disk(dev);
471 int err;
473 if (!capable(CAP_SYS_ADMIN))
474 return -EACCES;
476 err = kstrtouint(buf, 10, &sdkp->max_medium_access_timeouts);
478 return err ? err : count;
479 }
480 static DEVICE_ATTR_RW(max_medium_access_timeouts);
482 static ssize_t
483 max_write_same_blocks_show(struct device *dev, struct device_attribute *attr,
484 char *buf)
485 {
486 struct scsi_disk *sdkp = to_scsi_disk(dev);
488 return sprintf(buf, "%u\n", sdkp->max_ws_blocks);
489 }
491 static ssize_t
492 max_write_same_blocks_store(struct device *dev, struct device_attribute *attr,
493 const char *buf, size_t count)
494 {
495 struct scsi_disk *sdkp = to_scsi_disk(dev);
496 struct scsi_device *sdp = sdkp->device;
497 unsigned long max;
498 int err;
500 if (!capable(CAP_SYS_ADMIN))
501 return -EACCES;
503 if (sdp->type != TYPE_DISK && sdp->type != TYPE_ZBC)
504 return -EINVAL;
506 err = kstrtoul(buf, 10, &max);
508 if (err)
509 return err;
511 if (max == 0)
512 sdp->no_write_same = 1;
513 else if (max <= SD_MAX_WS16_BLOCKS) {
514 sdp->no_write_same = 0;
515 sdkp->max_ws_blocks = max;
516 }
518 sd_config_write_same(sdkp);
520 return count;
521 }
522 static DEVICE_ATTR_RW(max_write_same_blocks);
524 static struct attribute *sd_disk_attrs[] = {
525 &dev_attr_cache_type.attr,
526 &dev_attr_FUA.attr,
527 &dev_attr_allow_restart.attr,
528 &dev_attr_manage_start_stop.attr,
529 &dev_attr_protection_type.attr,
530 &dev_attr_protection_mode.attr,
531 &dev_attr_app_tag_own.attr,
532 &dev_attr_thin_provisioning.attr,
533 &dev_attr_provisioning_mode.attr,
534 &dev_attr_zeroing_mode.attr,
535 &dev_attr_max_write_same_blocks.attr,
536 &dev_attr_max_medium_access_timeouts.attr,
537 NULL,
538 };
539 ATTRIBUTE_GROUPS(sd_disk);
541 static struct class sd_disk_class = {
542 .name = "scsi_disk",
543 .owner = THIS_MODULE,
544 .dev_release = scsi_disk_release,
545 .dev_groups = sd_disk_groups,
546 };
548 static const struct dev_pm_ops sd_pm_ops = {
549 .suspend = sd_suspend_system,
550 .resume = sd_resume,
551 .poweroff = sd_suspend_system,
552 .restore = sd_resume,
553 .runtime_suspend = sd_suspend_runtime,
554 .runtime_resume = sd_resume,
555 };
557 static struct scsi_driver sd_template = {
558 .gendrv = {
559 .name = "sd",
560 .owner = THIS_MODULE,
561 .probe = sd_probe,
562 .remove = sd_remove,
563 .shutdown = sd_shutdown,
564 .pm = &sd_pm_ops,
565 },
566 .rescan = sd_rescan,
567 .init_command = sd_init_command,
568 .uninit_command = sd_uninit_command,
569 .done = sd_done,
570 .eh_action = sd_eh_action,
571 .eh_reset = sd_eh_reset,
572 };
574 /*
575 * Dummy kobj_map->probe function.
576 * The default ->probe function will call modprobe, which is
577 * pointless as this module is already loaded.
578 */
579 static struct kobject *sd_default_probe(dev_t devt, int *partno, void *data)
580 {
581 return NULL;
582 }
584 /*
585 * Device no to disk mapping:
586 *
587 * major disc2 disc p1
588 * |............|.............|....|....| <- dev_t
589 * 31 20 19 8 7 4 3 0
590 *
591 * Inside a major, we have 16k disks, however mapped non-
592 * contiguously. The first 16 disks are for major0, the next
593 * ones with major1, ... Disk 256 is for major0 again, disk 272
594 * for major1, ...
595 * As we stay compatible with our numbering scheme, we can reuse
596 * the well-know SCSI majors 8, 65--71, 136--143.
597 */
598 static int sd_major(int major_idx)
599 {
600 switch (major_idx) {
601 case 0:
602 return SCSI_DISK0_MAJOR;
603 case 1 ... 7:
604 return SCSI_DISK1_MAJOR + major_idx - 1;
605 case 8 ... 15:
606 return SCSI_DISK8_MAJOR + major_idx - 8;
607 default:
608 BUG();
609 return 0; /* shut up gcc */
610 }
611 }
613 static struct scsi_disk *scsi_disk_get(struct gendisk *disk)
614 {
615 struct scsi_disk *sdkp = NULL;
617 mutex_lock(&sd_ref_mutex);
619 if (disk->private_data) {
620 sdkp = scsi_disk(disk);
621 if (scsi_device_get(sdkp->device) == 0)
622 get_device(&sdkp->dev);
623 else
624 sdkp = NULL;
625 }
626 mutex_unlock(&sd_ref_mutex);
627 return sdkp;
628 }
630 static void scsi_disk_put(struct scsi_disk *sdkp)
631 {
632 struct scsi_device *sdev = sdkp->device;
634 mutex_lock(&sd_ref_mutex);
635 put_device(&sdkp->dev);
636 scsi_device_put(sdev);
637 mutex_unlock(&sd_ref_mutex);
638 }
640 #ifdef CONFIG_BLK_SED_OPAL
641 static int sd_sec_submit(void *data, u16 spsp, u8 secp, void *buffer,
642 size_t len, bool send)
643 {
644 struct scsi_device *sdev = data;
645 u8 cdb[12] = { 0, };
646 int ret;
648 cdb[0] = send ? SECURITY_PROTOCOL_OUT : SECURITY_PROTOCOL_IN;
649 cdb[1] = secp;
650 put_unaligned_be16(spsp, &cdb[2]);
651 put_unaligned_be32(len, &cdb[6]);
653 ret = scsi_execute_req(sdev, cdb,
654 send ? DMA_TO_DEVICE : DMA_FROM_DEVICE,
655 buffer, len, NULL, SD_TIMEOUT, SD_MAX_RETRIES, NULL);
656 return ret <= 0 ? ret : -EIO;
657 }
658 #endif /* CONFIG_BLK_SED_OPAL */
660 static unsigned char sd_setup_protect_cmnd(struct scsi_cmnd *scmd,
661 unsigned int dix, unsigned int dif)
662 {
663 struct bio *bio = scmd->request->bio;
664 unsigned int prot_op = sd_prot_op(rq_data_dir(scmd->request), dix, dif);
665 unsigned int protect = 0;
667 if (dix) { /* DIX Type 0, 1, 2, 3 */
668 if (bio_integrity_flagged(bio, BIP_IP_CHECKSUM))
669 scmd->prot_flags |= SCSI_PROT_IP_CHECKSUM;
671 if (bio_integrity_flagged(bio, BIP_CTRL_NOCHECK) == false)
672 scmd->prot_flags |= SCSI_PROT_GUARD_CHECK;
673 }
675 if (dif != T10_PI_TYPE3_PROTECTION) { /* DIX/DIF Type 0, 1, 2 */
676 scmd->prot_flags |= SCSI_PROT_REF_INCREMENT;
678 if (bio_integrity_flagged(bio, BIP_CTRL_NOCHECK) == false)
679 scmd->prot_flags |= SCSI_PROT_REF_CHECK;
680 }
682 if (dif) { /* DIX/DIF Type 1, 2, 3 */
683 scmd->prot_flags |= SCSI_PROT_TRANSFER_PI;
685 if (bio_integrity_flagged(bio, BIP_DISK_NOCHECK))
686 protect = 3 << 5; /* Disable target PI checking */
687 else
688 protect = 1 << 5; /* Enable target PI checking */
689 }
691 scsi_set_prot_op(scmd, prot_op);
692 scsi_set_prot_type(scmd, dif);
693 scmd->prot_flags &= sd_prot_flag_mask(prot_op);
695 return protect;
696 }
698 static void sd_config_discard(struct scsi_disk *sdkp, unsigned int mode)
699 {
700 struct request_queue *q = sdkp->disk->queue;
701 unsigned int logical_block_size = sdkp->device->sector_size;
702 unsigned int max_blocks = 0;
704 q->limits.discard_alignment =
705 sdkp->unmap_alignment * logical_block_size;
706 q->limits.discard_granularity =
707 max(sdkp->physical_block_size,
708 sdkp->unmap_granularity * logical_block_size);
709 sdkp->provisioning_mode = mode;
711 switch (mode) {
713 case SD_LBP_FULL:
714 case SD_LBP_DISABLE:
715 blk_queue_max_discard_sectors(q, 0);
716 blk_queue_flag_clear(QUEUE_FLAG_DISCARD, q);
717 return;
719 case SD_LBP_UNMAP:
720 max_blocks = min_not_zero(sdkp->max_unmap_blocks,
721 (u32)SD_MAX_WS16_BLOCKS);
722 break;
724 case SD_LBP_WS16:
725 if (sdkp->device->unmap_limit_for_ws)
726 max_blocks = sdkp->max_unmap_blocks;
727 else
728 max_blocks = sdkp->max_ws_blocks;
730 max_blocks = min_not_zero(max_blocks, (u32)SD_MAX_WS16_BLOCKS);
731 break;
733 case SD_LBP_WS10:
734 if (sdkp->device->unmap_limit_for_ws)
735 max_blocks = sdkp->max_unmap_blocks;
736 else
737 max_blocks = sdkp->max_ws_blocks;
739 max_blocks = min_not_zero(max_blocks, (u32)SD_MAX_WS10_BLOCKS);
740 break;
742 case SD_LBP_ZERO:
743 max_blocks = min_not_zero(sdkp->max_ws_blocks,
744 (u32)SD_MAX_WS10_BLOCKS);
745 break;
746 }
748 blk_queue_max_discard_sectors(q, max_blocks * (logical_block_size >> 9));
749 blk_queue_flag_set(QUEUE_FLAG_DISCARD, q);
750 }
752 static int sd_setup_unmap_cmnd(struct scsi_cmnd *cmd)
753 {
754 struct scsi_device *sdp = cmd->device;
755 struct request *rq = cmd->request;
756 u64 sector = blk_rq_pos(rq) >> (ilog2(sdp->sector_size) - 9);
757 u32 nr_sectors = blk_rq_sectors(rq) >> (ilog2(sdp->sector_size) - 9);
758 unsigned int data_len = 24;
759 char *buf;
761 rq->special_vec.bv_page = alloc_page(GFP_ATOMIC | __GFP_ZERO);
762 if (!rq->special_vec.bv_page)
763 return BLKPREP_DEFER;
764 rq->special_vec.bv_offset = 0;
765 rq->special_vec.bv_len = data_len;
766 rq->rq_flags |= RQF_SPECIAL_PAYLOAD;
768 cmd->cmd_len = 10;
769 cmd->cmnd[0] = UNMAP;
770 cmd->cmnd[8] = 24;
772 buf = page_address(rq->special_vec.bv_page);
773 put_unaligned_be16(6 + 16, &buf[0]);
774 put_unaligned_be16(16, &buf[2]);
775 put_unaligned_be64(sector, &buf[8]);
776 put_unaligned_be32(nr_sectors, &buf[16]);
778 cmd->allowed = SD_MAX_RETRIES;
779 cmd->transfersize = data_len;
780 rq->timeout = SD_TIMEOUT;
781 scsi_req(rq)->resid_len = data_len;
783 return scsi_init_io(cmd);
784 }
786 static int sd_setup_write_same16_cmnd(struct scsi_cmnd *cmd, bool unmap)
787 {
788 struct scsi_device *sdp = cmd->device;
789 struct request *rq = cmd->request;
790 u64 sector = blk_rq_pos(rq) >> (ilog2(sdp->sector_size) - 9);
791 u32 nr_sectors = blk_rq_sectors(rq) >> (ilog2(sdp->sector_size) - 9);
792 u32 data_len = sdp->sector_size;
794 rq->special_vec.bv_page = alloc_page(GFP_ATOMIC | __GFP_ZERO);
795 if (!rq->special_vec.bv_page)
796 return BLKPREP_DEFER;
797 rq->special_vec.bv_offset = 0;
798 rq->special_vec.bv_len = data_len;
799 rq->rq_flags |= RQF_SPECIAL_PAYLOAD;
801 cmd->cmd_len = 16;
802 cmd->cmnd[0] = WRITE_SAME_16;
803 if (unmap)
804 cmd->cmnd[1] = 0x8; /* UNMAP */
805 put_unaligned_be64(sector, &cmd->cmnd[2]);
806 put_unaligned_be32(nr_sectors, &cmd->cmnd[10]);
808 cmd->allowed = SD_MAX_RETRIES;
809 cmd->transfersize = data_len;
810 rq->timeout = unmap ? SD_TIMEOUT : SD_WRITE_SAME_TIMEOUT;
811 scsi_req(rq)->resid_len = data_len;
813 return scsi_init_io(cmd);
814 }
816 static int sd_setup_write_same10_cmnd(struct scsi_cmnd *cmd, bool unmap)
817 {
818 struct scsi_device *sdp = cmd->device;
819 struct request *rq = cmd->request;
820 u64 sector = blk_rq_pos(rq) >> (ilog2(sdp->sector_size) - 9);
821 u32 nr_sectors = blk_rq_sectors(rq) >> (ilog2(sdp->sector_size) - 9);
822 u32 data_len = sdp->sector_size;
824 rq->special_vec.bv_page = alloc_page(GFP_ATOMIC | __GFP_ZERO);
825 if (!rq->special_vec.bv_page)
826 return BLKPREP_DEFER;
827 rq->special_vec.bv_offset = 0;
828 rq->special_vec.bv_len = data_len;
829 rq->rq_flags |= RQF_SPECIAL_PAYLOAD;
831 cmd->cmd_len = 10;
832 cmd->cmnd[0] = WRITE_SAME;
833 if (unmap)
834 cmd->cmnd[1] = 0x8; /* UNMAP */
835 put_unaligned_be32(sector, &cmd->cmnd[2]);
836 put_unaligned_be16(nr_sectors, &cmd->cmnd[7]);
838 cmd->allowed = SD_MAX_RETRIES;
839 cmd->transfersize = data_len;
840 rq->timeout = unmap ? SD_TIMEOUT : SD_WRITE_SAME_TIMEOUT;
841 scsi_req(rq)->resid_len = data_len;
843 return scsi_init_io(cmd);
844 }
846 static int sd_setup_write_zeroes_cmnd(struct scsi_cmnd *cmd)
847 {
848 struct request *rq = cmd->request;
849 struct scsi_device *sdp = cmd->device;
850 struct scsi_disk *sdkp = scsi_disk(rq->rq_disk);
851 u64 sector = blk_rq_pos(rq) >> (ilog2(sdp->sector_size) - 9);
852 u32 nr_sectors = blk_rq_sectors(rq) >> (ilog2(sdp->sector_size) - 9);
854 if (!(rq->cmd_flags & REQ_NOUNMAP)) {
855 switch (sdkp->zeroing_mode) {
856 case SD_ZERO_WS16_UNMAP:
857 return sd_setup_write_same16_cmnd(cmd, true);
858 case SD_ZERO_WS10_UNMAP:
859 return sd_setup_write_same10_cmnd(cmd, true);
860 }
861 }
863 if (sdp->no_write_same)
864 return BLKPREP_INVALID;
866 if (sdkp->ws16 || sector > 0xffffffff || nr_sectors > 0xffff)
867 return sd_setup_write_same16_cmnd(cmd, false);
869 return sd_setup_write_same10_cmnd(cmd, false);
870 }
872 static void sd_config_write_same(struct scsi_disk *sdkp)
873 {
874 struct request_queue *q = sdkp->disk->queue;
875 unsigned int logical_block_size = sdkp->device->sector_size;
877 if (sdkp->device->no_write_same) {
878 sdkp->max_ws_blocks = 0;
879 goto out;
880 }
882 /* Some devices can not handle block counts above 0xffff despite
883 * supporting WRITE SAME(16). Consequently we default to 64k
884 * blocks per I/O unless the device explicitly advertises a
885 * bigger limit.
886 */
887 if (sdkp->max_ws_blocks > SD_MAX_WS10_BLOCKS)
888 sdkp->max_ws_blocks = min_not_zero(sdkp->max_ws_blocks,
889 (u32)SD_MAX_WS16_BLOCKS);
890 else if (sdkp->ws16 || sdkp->ws10 || sdkp->device->no_report_opcodes)
891 sdkp->max_ws_blocks = min_not_zero(sdkp->max_ws_blocks,
892 (u32)SD_MAX_WS10_BLOCKS);
893 else {
894 sdkp->device->no_write_same = 1;
895 sdkp->max_ws_blocks = 0;
896 }
898 if (sdkp->lbprz && sdkp->lbpws)
899 sdkp->zeroing_mode = SD_ZERO_WS16_UNMAP;
900 else if (sdkp->lbprz && sdkp->lbpws10)
901 sdkp->zeroing_mode = SD_ZERO_WS10_UNMAP;
902 else if (sdkp->max_ws_blocks)
903 sdkp->zeroing_mode = SD_ZERO_WS;
904 else
905 sdkp->zeroing_mode = SD_ZERO_WRITE;
907 if (sdkp->max_ws_blocks &&
908 sdkp->physical_block_size > logical_block_size) {
909 /*
910 * Reporting a maximum number of blocks that is not aligned
911 * on the device physical size would cause a large write same
912 * request to be split into physically unaligned chunks by
913 * __blkdev_issue_write_zeroes() and __blkdev_issue_write_same()
914 * even if the caller of these functions took care to align the
915 * large request. So make sure the maximum reported is aligned
916 * to the device physical block size. This is only an optional
917 * optimization for regular disks, but this is mandatory to
918 * avoid failure of large write same requests directed at
919 * sequential write required zones of host-managed ZBC disks.
920 */
921 sdkp->max_ws_blocks =
922 round_down(sdkp->max_ws_blocks,
923 bytes_to_logical(sdkp->device,
924 sdkp->physical_block_size));
925 }
927 out:
928 blk_queue_max_write_same_sectors(q, sdkp->max_ws_blocks *
929 (logical_block_size >> 9));
930 blk_queue_max_write_zeroes_sectors(q, sdkp->max_ws_blocks *
931 (logical_block_size >> 9));
932 }
934 /**
935 * sd_setup_write_same_cmnd - write the same data to multiple blocks
936 * @cmd: command to prepare
937 *
938 * Will set up either WRITE SAME(10) or WRITE SAME(16) depending on
939 * the preference indicated by the target device.
940 **/
941 static int sd_setup_write_same_cmnd(struct scsi_cmnd *cmd)
942 {
943 struct request *rq = cmd->request;
944 struct scsi_device *sdp = cmd->device;
945 struct scsi_disk *sdkp = scsi_disk(rq->rq_disk);
946 struct bio *bio = rq->bio;
947 sector_t sector = blk_rq_pos(rq);
948 unsigned int nr_sectors = blk_rq_sectors(rq);
949 unsigned int nr_bytes = blk_rq_bytes(rq);
950 int ret;
952 if (sdkp->device->no_write_same)
953 return BLKPREP_INVALID;
955 BUG_ON(bio_offset(bio) || bio_iovec(bio).bv_len != sdp->sector_size);
957 sector >>= ilog2(sdp->sector_size) - 9;
958 nr_sectors >>= ilog2(sdp->sector_size) - 9;
960 rq->timeout = SD_WRITE_SAME_TIMEOUT;
962 if (sdkp->ws16 || sector > 0xffffffff || nr_sectors > 0xffff) {
963 cmd->cmd_len = 16;
964 cmd->cmnd[0] = WRITE_SAME_16;
965 put_unaligned_be64(sector, &cmd->cmnd[2]);
966 put_unaligned_be32(nr_sectors, &cmd->cmnd[10]);
967 } else {
968 cmd->cmd_len = 10;
969 cmd->cmnd[0] = WRITE_SAME;
970 put_unaligned_be32(sector, &cmd->cmnd[2]);
971 put_unaligned_be16(nr_sectors, &cmd->cmnd[7]);
972 }
974 cmd->transfersize = sdp->sector_size;
975 cmd->allowed = SD_MAX_RETRIES;
977 /*
978 * For WRITE SAME the data transferred via the DATA OUT buffer is
979 * different from the amount of data actually written to the target.
980 *
981 * We set up __data_len to the amount of data transferred via the
982 * DATA OUT buffer so that blk_rq_map_sg sets up the proper S/G list
983 * to transfer a single sector of data first, but then reset it to
984 * the amount of data to be written right after so that the I/O path
985 * knows how much to actually write.
986 */
987 rq->__data_len = sdp->sector_size;
988 ret = scsi_init_io(cmd);
989 rq->__data_len = nr_bytes;
991 return ret;
992 }
994 static int sd_setup_flush_cmnd(struct scsi_cmnd *cmd)
995 {
996 struct request *rq = cmd->request;
998 /* flush requests don't perform I/O, zero the S/G table */
999 memset(&cmd->sdb, 0, sizeof(cmd->sdb));
1001 cmd->cmnd[0] = SYNCHRONIZE_CACHE;
1002 cmd->cmd_len = 10;
1003 cmd->transfersize = 0;
1004 cmd->allowed = SD_MAX_RETRIES;
1006 rq->timeout = rq->q->rq_timeout * SD_FLUSH_TIMEOUT_MULTIPLIER;
1007 return BLKPREP_OK;
1008 }
1010 static int sd_setup_read_write_cmnd(struct scsi_cmnd *SCpnt)
1011 {
1012 struct request *rq = SCpnt->request;
1013 struct scsi_device *sdp = SCpnt->device;
1014 struct gendisk *disk = rq->rq_disk;
1015 struct scsi_disk *sdkp = scsi_disk(disk);
1016 sector_t block = blk_rq_pos(rq);
1017 sector_t threshold;
1018 unsigned int this_count = blk_rq_sectors(rq);
1019 unsigned int dif, dix;
1020 int ret;
1021 unsigned char protect;
1023 ret = scsi_init_io(SCpnt);
1024 if (ret != BLKPREP_OK)
1025 return ret;
1026 WARN_ON_ONCE(SCpnt != rq->special);
1028 /* from here on until we're complete, any goto out
1029 * is used for a killable error condition */
1030 ret = BLKPREP_KILL;
1032 SCSI_LOG_HLQUEUE(1,
1033 scmd_printk(KERN_INFO, SCpnt,
1034 "%s: block=%llu, count=%d\n",
1035 __func__, (unsigned long long)block, this_count));
1037 if (!sdp || !scsi_device_online(sdp) ||
1038 block + blk_rq_sectors(rq) > get_capacity(disk)) {
1039 SCSI_LOG_HLQUEUE(2, scmd_printk(KERN_INFO, SCpnt,
1040 "Finishing %u sectors\n",
1041 blk_rq_sectors(rq)));
1042 SCSI_LOG_HLQUEUE(2, scmd_printk(KERN_INFO, SCpnt,
1043 "Retry with 0x%p\n", SCpnt));
1044 goto out;
1045 }
1047 if (sdp->changed) {
1048 /*
1049 * quietly refuse to do anything to a changed disc until
1050 * the changed bit has been reset
1051 */
1052 /* printk("SCSI disk has been changed or is not present. Prohibiting further I/O.\n"); */
1053 goto out;
1054 }
1056 /*
1057 * Some SD card readers can't handle multi-sector accesses which touch
1058 * the last one or two hardware sectors. Split accesses as needed.
1059 */
1060 threshold = get_capacity(disk) - SD_LAST_BUGGY_SECTORS *
1061 (sdp->sector_size / 512);
1063 if (unlikely(sdp->last_sector_bug && block + this_count > threshold)) {
1064 if (block < threshold) {
1065 /* Access up to the threshold but not beyond */
1066 this_count = threshold - block;
1067 } else {
1068 /* Access only a single hardware sector */
1069 this_count = sdp->sector_size / 512;
1070 }
1071 }
1073 SCSI_LOG_HLQUEUE(2, scmd_printk(KERN_INFO, SCpnt, "block=%llu\n",
1074 (unsigned long long)block));
1076 /*
1077 * If we have a 1K hardware sectorsize, prevent access to single
1078 * 512 byte sectors. In theory we could handle this - in fact
1079 * the scsi cdrom driver must be able to handle this because
1080 * we typically use 1K blocksizes, and cdroms typically have
1081 * 2K hardware sectorsizes. Of course, things are simpler
1082 * with the cdrom, since it is read-only. For performance
1083 * reasons, the filesystems should be able to handle this
1084 * and not force the scsi disk driver to use bounce buffers
1085 * for this.
1086 */
1087 if (sdp->sector_size == 1024) {
1088 if ((block & 1) || (blk_rq_sectors(rq) & 1)) {
1089 scmd_printk(KERN_ERR, SCpnt,
1090 "Bad block number requested\n");
1091 goto out;
1092 } else {
1093 block = block >> 1;
1094 this_count = this_count >> 1;
1095 }
1096 }
1097 if (sdp->sector_size == 2048) {
1098 if ((block & 3) || (blk_rq_sectors(rq) & 3)) {
1099 scmd_printk(KERN_ERR, SCpnt,
1100 "Bad block number requested\n");
1101 goto out;
1102 } else {
1103 block = block >> 2;
1104 this_count = this_count >> 2;
1105 }
1106 }
1107 if (sdp->sector_size == 4096) {
1108 if ((block & 7) || (blk_rq_sectors(rq) & 7)) {
1109 scmd_printk(KERN_ERR, SCpnt,
1110 "Bad block number requested\n");
1111 goto out;
1112 } else {
1113 block = block >> 3;
1114 this_count = this_count >> 3;
1115 }
1116 }
1117 if (rq_data_dir(rq) == WRITE) {
1118 SCpnt->cmnd[0] = WRITE_6;
1120 if (blk_integrity_rq(rq))
1121 t10_pi_prepare(SCpnt->request, sdkp->protection_type);
1123 } else if (rq_data_dir(rq) == READ) {
1124 SCpnt->cmnd[0] = READ_6;
1125 } else {
1126 scmd_printk(KERN_ERR, SCpnt, "Unknown command %d\n", req_op(rq));
1127 goto out;
1128 }
1130 SCSI_LOG_HLQUEUE(2, scmd_printk(KERN_INFO, SCpnt,
1131 "%s %d/%u 512 byte blocks.\n",
1132 (rq_data_dir(rq) == WRITE) ?
1133 "writing" : "reading", this_count,
1134 blk_rq_sectors(rq)));
1136 dix = scsi_prot_sg_count(SCpnt);
1137 dif = scsi_host_dif_capable(SCpnt->device->host, sdkp->protection_type);
1139 if (dif || dix)
1140 protect = sd_setup_protect_cmnd(SCpnt, dix, dif);
1141 else
1142 protect = 0;
1144 if (protect && sdkp->protection_type == T10_PI_TYPE2_PROTECTION) {
1145 SCpnt->cmnd = mempool_alloc(sd_cdb_pool, GFP_ATOMIC);
1147 if (unlikely(SCpnt->cmnd == NULL)) {
1148 ret = BLKPREP_DEFER;
1149 goto out;
1150 }
1152 SCpnt->cmd_len = SD_EXT_CDB_SIZE;
1153 memset(SCpnt->cmnd, 0, SCpnt->cmd_len);
1154 SCpnt->cmnd[0] = VARIABLE_LENGTH_CMD;
1155 SCpnt->cmnd[7] = 0x18;
1156 SCpnt->cmnd[9] = (rq_data_dir(rq) == READ) ? READ_32 : WRITE_32;
1157 SCpnt->cmnd[10] = protect | ((rq->cmd_flags & REQ_FUA) ? 0x8 : 0);
1159 /* LBA */
1160 SCpnt->cmnd[12] = sizeof(block) > 4 ? (unsigned char) (block >> 56) & 0xff : 0;
1161 SCpnt->cmnd[13] = sizeof(block) > 4 ? (unsigned char) (block >> 48) & 0xff : 0;
1162 SCpnt->cmnd[14] = sizeof(block) > 4 ? (unsigned char) (block >> 40) & 0xff : 0;
1163 SCpnt->cmnd[15] = sizeof(block) > 4 ? (unsigned char) (block >> 32) & 0xff : 0;
1164 SCpnt->cmnd[16] = (unsigned char) (block >> 24) & 0xff;
1165 SCpnt->cmnd[17] = (unsigned char) (block >> 16) & 0xff;
1166 SCpnt->cmnd[18] = (unsigned char) (block >> 8) & 0xff;
1167 SCpnt->cmnd[19] = (unsigned char) block & 0xff;
1169 /* Expected Indirect LBA */
1170 SCpnt->cmnd[20] = (unsigned char) (block >> 24) & 0xff;
1171 SCpnt->cmnd[21] = (unsigned char) (block >> 16) & 0xff;
1172 SCpnt->cmnd[22] = (unsigned char) (block >> 8) & 0xff;
1173 SCpnt->cmnd[23] = (unsigned char) block & 0xff;
1175 /* Transfer length */
1176 SCpnt->cmnd[28] = (unsigned char) (this_count >> 24) & 0xff;
1177 SCpnt->cmnd[29] = (unsigned char) (this_count >> 16) & 0xff;
1178 SCpnt->cmnd[30] = (unsigned char) (this_count >> 8) & 0xff;
1179 SCpnt->cmnd[31] = (unsigned char) this_count & 0xff;
1180 } else if (sdp->use_16_for_rw || (this_count > 0xffff)) {
1181 SCpnt->cmnd[0] += READ_16 - READ_6;
1182 SCpnt->cmnd[1] = protect | ((rq->cmd_flags & REQ_FUA) ? 0x8 : 0);
1183 SCpnt->cmnd[2] = sizeof(block) > 4 ? (unsigned char) (block >> 56) & 0xff : 0;
1184 SCpnt->cmnd[3] = sizeof(block) > 4 ? (unsigned char) (block >> 48) & 0xff : 0;
1185 SCpnt->cmnd[4] = sizeof(block) > 4 ? (unsigned char) (block >> 40) & 0xff : 0;
1186 SCpnt->cmnd[5] = sizeof(block) > 4 ? (unsigned char) (block >> 32) & 0xff : 0;
1187 SCpnt->cmnd[6] = (unsigned char) (block >> 24) & 0xff;
1188 SCpnt->cmnd[7] = (unsigned char) (block >> 16) & 0xff;
1189 SCpnt->cmnd[8] = (unsigned char) (block >> 8) & 0xff;
1190 SCpnt->cmnd[9] = (unsigned char) block & 0xff;
1191 SCpnt->cmnd[10] = (unsigned char) (this_count >> 24) & 0xff;
1192 SCpnt->cmnd[11] = (unsigned char) (this_count >> 16) & 0xff;
1193 SCpnt->cmnd[12] = (unsigned char) (this_count >> 8) & 0xff;
1194 SCpnt->cmnd[13] = (unsigned char) this_count & 0xff;
1195 SCpnt->cmnd[14] = SCpnt->cmnd[15] = 0;
1196 } else if ((this_count > 0xff) || (block > 0x1fffff) ||
1197 scsi_device_protection(SCpnt->device) ||
1198 SCpnt->device->use_10_for_rw) {
1199 SCpnt->cmnd[0] += READ_10 - READ_6;
1200 SCpnt->cmnd[1] = protect | ((rq->cmd_flags & REQ_FUA) ? 0x8 : 0);
1201 SCpnt->cmnd[2] = (unsigned char) (block >> 24) & 0xff;
1202 SCpnt->cmnd[3] = (unsigned char) (block >> 16) & 0xff;
1203 SCpnt->cmnd[4] = (unsigned char) (block >> 8) & 0xff;
1204 SCpnt->cmnd[5] = (unsigned char) block & 0xff;
1205 SCpnt->cmnd[6] = SCpnt->cmnd[9] = 0;
1206 SCpnt->cmnd[7] = (unsigned char) (this_count >> 8) & 0xff;
1207 SCpnt->cmnd[8] = (unsigned char) this_count & 0xff;
1208 } else {
1209 if (unlikely(rq->cmd_flags & REQ_FUA)) {
1210 /*
1211 * This happens only if this drive failed
1212 * 10byte rw command with ILLEGAL_REQUEST
1213 * during operation and thus turned off
1214 * use_10_for_rw.
1215 */
1216 scmd_printk(KERN_ERR, SCpnt,
1217 "FUA write on READ/WRITE(6) drive\n");
1218 goto out;
1219 }
1221 SCpnt->cmnd[1] |= (unsigned char) ((block >> 16) & 0x1f);
1222 SCpnt->cmnd[2] = (unsigned char) ((block >> 8) & 0xff);
1223 SCpnt->cmnd[3] = (unsigned char) block & 0xff;
1224 SCpnt->cmnd[4] = (unsigned char) this_count;
1225 SCpnt->cmnd[5] = 0;
1226 }
1227 SCpnt->sdb.length = this_count * sdp->sector_size;
1229 /*
1230 * We shouldn't disconnect in the middle of a sector, so with a dumb
1231 * host adapter, it's safe to assume that we can at least transfer
1232 * this many bytes between each connect / disconnect.
1233 */
1234 SCpnt->transfersize = sdp->sector_size;
1235 SCpnt->underflow = this_count << 9;
1236 SCpnt->allowed = SD_MAX_RETRIES;
1238 /*
1239 * This indicates that the command is ready from our end to be
1240 * queued.
1241 */
1242 ret = BLKPREP_OK;
1243 out:
1244 return ret;
1245 }
1247 static int sd_init_command(struct scsi_cmnd *cmd)
1248 {
1249 struct request *rq = cmd->request;
1251 switch (req_op(rq)) {
1252 case REQ_OP_DISCARD:
1253 switch (scsi_disk(rq->rq_disk)->provisioning_mode) {
1254 case SD_LBP_UNMAP:
1255 return sd_setup_unmap_cmnd(cmd);
1256 case SD_LBP_WS16:
1257 return sd_setup_write_same16_cmnd(cmd, true);
1258 case SD_LBP_WS10:
1259 return sd_setup_write_same10_cmnd(cmd, true);
1260 case SD_LBP_ZERO:
1261 return sd_setup_write_same10_cmnd(cmd, false);
1262 default:
1263 return BLKPREP_INVALID;
1264 }
1265 case REQ_OP_WRITE_ZEROES:
1266 return sd_setup_write_zeroes_cmnd(cmd);
1267 case REQ_OP_WRITE_SAME:
1268 return sd_setup_write_same_cmnd(cmd);
1269 case REQ_OP_FLUSH:
1270 return sd_setup_flush_cmnd(cmd);
1271 case REQ_OP_READ:
1272 case REQ_OP_WRITE:
1273 return sd_setup_read_write_cmnd(cmd);
1274 case REQ_OP_ZONE_REPORT:
1275 return sd_zbc_setup_report_cmnd(cmd);
1276 case REQ_OP_ZONE_RESET:
1277 return sd_zbc_setup_reset_cmnd(cmd);
1278 default:
1279 WARN_ON_ONCE(1);
1280 return BLKPREP_KILL;
1281 }
1282 }
1284 static void sd_uninit_command(struct scsi_cmnd *SCpnt)
1285 {
1286 struct request *rq = SCpnt->request;
1287 u8 *cmnd;
1289 if (rq->rq_flags & RQF_SPECIAL_PAYLOAD)
1290 __free_page(rq->special_vec.bv_page);
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 }
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 **/
1315 static 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;
1321 if (!sdkp)
1322 return -ENXIO;
1324 SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp, "sd_open\n"));
1326 sdev = sdkp->device;
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;
1336 if (sdev->removable || sdkp->write_prot)
1337 check_disk_change(bdev);
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;
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;
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;
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 }
1369 return 0;
1371 error_out:
1372 scsi_disk_put(sdkp);
1373 return retval;
1374 }
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 **/
1389 static 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;
1394 SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp, "sd_release\n"));
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 }
1401 /*
1402 * XXX and what if there are packets in flight and this close()
1403 * XXX is followed by a "rmmod sd_mod"?
1404 */
1406 scsi_disk_put(sdkp);
1407 }
1409 static 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];
1417 /* default to most commonly used values */
1418 diskinfo[0] = 0x40; /* 1 << 6 */
1419 diskinfo[1] = 0x20; /* 1 << 5 */
1420 diskinfo[2] = capacity >> 11;
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);
1428 geo->heads = diskinfo[0];
1429 geo->sectors = diskinfo[1];
1430 geo->cylinders = diskinfo[2];
1431 return 0;
1432 }
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 **/
1448 static 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;
1457 SCSI_LOG_IOCTL(1, sd_printk(KERN_INFO, sdkp, "sd_ioctl: disk=%s, "
1458 "cmd=0x%x\n", disk->disk_name, cmd));
1460 error = scsi_verify_blk_ioctl(bdev, cmd);
1461 if (error < 0)
1462 return error;
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;
1475 if (is_sed_ioctl(cmd))
1476 return sed_ioctl(sdkp->opal_dev, cmd, p);
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 }
1495 out:
1496 return error;
1497 }
1499 static void set_media_not_present(struct scsi_disk *sdkp)
1500 {
1501 if (sdkp->media_present)
1502 sdkp->device->changed = 1;
1504 if (sdkp->device->removable) {
1505 sdkp->media_present = 0;
1506 sdkp->capacity = 0;
1507 }
1508 }
1510 static int media_not_present(struct scsi_disk *sdkp,
1511 struct scsi_sense_hdr *sshdr)
1512 {
1513 if (!scsi_sense_valid(sshdr))
1514 return 0;
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 }
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 **/
1538 static 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;
1544 if (!sdkp)
1545 return 0;
1547 sdp = sdkp->device;
1548 SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp, "sd_check_events\n"));
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 }
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, };
1573 retval = scsi_test_unit_ready(sdp, SD_TIMEOUT, SD_MAX_RETRIES,
1574 &sshdr);
1576 /* failed to execute TUR, assume media not present */
1577 if (host_byte(retval)) {
1578 set_media_not_present(sdkp);
1579 goto out;
1580 }
1582 if (media_not_present(sdkp, &sshdr))
1583 goto out;
1584 }
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;
1593 out:
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 }
1606 static 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;
1614 if (!scsi_device_online(sdp))
1615 return -ENODEV;
1617 /* caller might not be interested in sense, but we need it */
1618 if (!sshdr)
1619 sshdr = &my_sshdr;
1621 for (retries = 3; retries > 0; --retries) {
1622 unsigned char cmd[10] = { 0 };
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 }
1635 if (res) {
1636 sd_print_result(sdkp, "Synchronize Cache(10) failed", res);
1638 if (driver_byte(res) == DRIVER_SENSE)
1639 sd_print_sense_hdr(sdkp, sshdr);
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;
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 }
1666 static void sd_rescan(struct device *dev)
1667 {
1668 struct scsi_disk *sdkp = dev_get_drvdata(dev);
1670 revalidate_disk(sdkp->disk);
1671 }
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 */
1679 static 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;
1685 error = scsi_ioctl_block_when_processing_errors(sdev, cmd,
1686 (mode & FMODE_NDELAY) != 0);
1687 if (error)
1688 return error;
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
1699 static 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 };
1719 static 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, };
1728 cmd[0] = PERSISTENT_RESERVE_OUT;
1729 cmd[1] = sa;
1730 cmd[2] = type;
1731 put_unaligned_be32(sizeof(data), &cmd[5]);
1733 put_unaligned_be64(key, &data[0]);
1734 put_unaligned_be64(sa_key, &data[8]);
1735 data[20] = flags;
1737 result = scsi_execute_req(sdev, cmd, DMA_TO_DEVICE, &data, sizeof(data),
1738 &sshdr, SD_TIMEOUT, SD_MAX_RETRIES, NULL);
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 }
1746 return result;
1747 }
1749 static 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 }
1759 static 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 }
1767 static 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 }
1772 static 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 }
1779 static int sd_pr_clear(struct block_device *bdev, u64 key)
1780 {
1781 return sd_pr_command(bdev, 0x03, key, 0, 0, 0);
1782 }
1784 static 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 };
1792 static 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 };
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 **/
1819 static void sd_eh_reset(struct scsi_cmnd *scmd)
1820 {
1821 struct scsi_disk *sdkp = scsi_disk(scmd->request->rq_disk);
1823 /* New SCSI EH run, reset gate variable */
1824 sdkp->ignore_medium_access_errors = false;
1825 }
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 **/
1839 static 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;
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;
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 }
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);
1874 return SUCCESS;
1875 }
1877 return eh_disp;
1878 }
1880 static 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;
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;
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;
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;
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);
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);
1920 return min(good_bytes, transferred);
1921 }
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 **/
1930 static 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;
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 }
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;
1989 if (driver_byte(result) != DRIVER_SENSE &&
1990 (!sense_valid || sense_deferred))
1991 goto out;
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->rq_flags |= RQF_QUIET;
2032 }
2033 break;
2034 }
2035 }
2036 break;
2037 default:
2038 break;
2039 }
2041 out:
2042 if (sd_is_zoned(sdkp))
2043 sd_zbc_complete(SCpnt, good_bytes, &sshdr);
2045 SCSI_LOG_HLCOMPLETE(1, scmd_printk(KERN_INFO, SCpnt,
2046 "sd_done: completed %d of %d bytes\n",
2047 good_bytes, scsi_bufflen(SCpnt)));
2049 if (rq_data_dir(SCpnt->request) == READ && scsi_prot_sg_count(SCpnt) &&
2050 good_bytes)
2051 t10_pi_complete(SCpnt->request, sdkp->protection_type,
2052 good_bytes / scsi_prot_interval(SCpnt));
2054 return good_bytes;
2055 }
2057 /*
2058 * spinup disk - called only in sd_revalidate_disk()
2059 */
2060 static void
2061 sd_spinup_disk(struct scsi_disk *sdkp)
2062 {
2063 unsigned char cmd[10];
2064 unsigned long spintime_expire = 0;
2065 int retries, spintime;
2066 unsigned int the_result;
2067 struct scsi_sense_hdr sshdr;
2068 int sense_valid = 0;
2070 spintime = 0;
2072 /* Spin up drives, as required. Only do this at boot time */
2073 /* Spinup needs to be done for module loads too. */
2074 do {
2075 retries = 0;
2077 do {
2078 cmd[0] = TEST_UNIT_READY;
2079 memset((void *) &cmd[1], 0, 9);
2081 the_result = scsi_execute_req(sdkp->device, cmd,
2082 DMA_NONE, NULL, 0,
2083 &sshdr, SD_TIMEOUT,
2084 SD_MAX_RETRIES, NULL);
2086 /*
2087 * If the drive has indicated to us that it
2088 * doesn't have any media in it, don't bother
2089 * with any more polling.
2090 */
2091 if (media_not_present(sdkp, &sshdr))
2092 return;
2094 if (the_result)
2095 sense_valid = scsi_sense_valid(&sshdr);
2096 retries++;
2097 } while (retries < 3 &&
2098 (!scsi_status_is_good(the_result) ||
2099 ((driver_byte(the_result) == DRIVER_SENSE) &&
2100 sense_valid && sshdr.sense_key == UNIT_ATTENTION)));
2102 if (driver_byte(the_result) != DRIVER_SENSE) {
2103 /* no sense, TUR either succeeded or failed
2104 * with a status error */
2105 if(!spintime && !scsi_status_is_good(the_result)) {
2106 sd_print_result(sdkp, "Test Unit Ready failed",
2107 the_result);
2108 }
2109 break;
2110 }
2112 /*
2113 * The device does not want the automatic start to be issued.
2114 */
2115 if (sdkp->device->no_start_on_add)
2116 break;
2118 if (sense_valid && sshdr.sense_key == NOT_READY) {
2119 if (sshdr.asc == 4 && sshdr.ascq == 3)
2120 break; /* manual intervention required */
2121 if (sshdr.asc == 4 && sshdr.ascq == 0xb)
2122 break; /* standby */
2123 if (sshdr.asc == 4 && sshdr.ascq == 0xc)
2124 break; /* unavailable */
2125 if (sshdr.asc == 4 && sshdr.ascq == 0x1b)
2126 break; /* sanitize in progress */
2127 /*
2128 * Issue command to spin up drive when not ready
2129 */
2130 if (!spintime) {
2131 sd_printk(KERN_NOTICE, sdkp, "Spinning up disk...");
2132 cmd[0] = START_STOP;
2133 cmd[1] = 1; /* Return immediately */
2134 memset((void *) &cmd[2], 0, 8);
2135 cmd[4] = 1; /* Start spin cycle */
2136 if (sdkp->device->start_stop_pwr_cond)
2137 cmd[4] |= 1 << 4;
2138 scsi_execute_req(sdkp->device, cmd, DMA_NONE,
2139 NULL, 0, &sshdr,
2140 SD_TIMEOUT, SD_MAX_RETRIES,
2141 NULL);
2142 spintime_expire = jiffies + 100 * HZ;
2143 spintime = 1;
2144 }
2145 /* Wait 1 second for next try */
2146 msleep(1000);
2147 printk(KERN_CONT ".");
2149 /*
2150 * Wait for USB flash devices with slow firmware.
2151 * Yes, this sense key/ASC combination shouldn't
2152 * occur here. It's characteristic of these devices.
2153 */
2154 } else if (sense_valid &&
2155 sshdr.sense_key == UNIT_ATTENTION &&
2156 sshdr.asc == 0x28) {
2157 if (!spintime) {
2158 spintime_expire = jiffies + 5 * HZ;
2159 spintime = 1;
2160 }
2161 /* Wait 1 second for next try */
2162 msleep(1000);
2163 } else {
2164 /* we don't understand the sense code, so it's
2165 * probably pointless to loop */
2166 if(!spintime) {
2167 sd_printk(KERN_NOTICE, sdkp, "Unit Not Ready\n");
2168 sd_print_sense_hdr(sdkp, &sshdr);
2169 }
2170 break;
2171 }
2173 } while (spintime && time_before_eq(jiffies, spintime_expire));
2175 if (spintime) {
2176 if (scsi_status_is_good(the_result))
2177 printk(KERN_CONT "ready\n");
2178 else
2179 printk(KERN_CONT "not responding...\n");
2180 }
2181 }
2183 /*
2184 * Determine whether disk supports Data Integrity Field.
2185 */
2186 static int sd_read_protection_type(struct scsi_disk *sdkp, unsigned char *buffer)
2187 {
2188 struct scsi_device *sdp = sdkp->device;
2189 u8 type;
2190 int ret = 0;
2192 if (scsi_device_protection(sdp) == 0 || (buffer[12] & 1) == 0)
2193 return ret;
2195 type = ((buffer[12] >> 1) & 7) + 1; /* P_TYPE 0 = Type 1 */
2197 if (type > T10_PI_TYPE3_PROTECTION)
2198 ret = -ENODEV;
2199 else if (scsi_host_dif_capable(sdp->host, type))
2200 ret = 1;
2202 if (sdkp->first_scan || type != sdkp->protection_type)
2203 switch (ret) {
2204 case -ENODEV:
2205 sd_printk(KERN_ERR, sdkp, "formatted with unsupported" \
2206 " protection type %u. Disabling disk!\n",
2207 type);
2208 break;
2209 case 1:
2210 sd_printk(KERN_NOTICE, sdkp,
2211 "Enabling DIF Type %u protection\n", type);
2212 break;
2213 case 0:
2214 sd_printk(KERN_NOTICE, sdkp,
2215 "Disabling DIF Type %u protection\n", type);
2216 break;
2217 }
2219 sdkp->protection_type = type;
2221 return ret;
2222 }
2224 static void read_capacity_error(struct scsi_disk *sdkp, struct scsi_device *sdp,
2225 struct scsi_sense_hdr *sshdr, int sense_valid,
2226 int the_result)
2227 {
2228 if (driver_byte(the_result) == DRIVER_SENSE)
2229 sd_print_sense_hdr(sdkp, sshdr);
2230 else
2231 sd_printk(KERN_NOTICE, sdkp, "Sense not available.\n");
2233 /*
2234 * Set dirty bit for removable devices if not ready -
2235 * sometimes drives will not report this properly.
2236 */
2237 if (sdp->removable &&
2238 sense_valid && sshdr->sense_key == NOT_READY)
2239 set_media_not_present(sdkp);
2241 /*
2242 * We used to set media_present to 0 here to indicate no media
2243 * in the drive, but some drives fail read capacity even with
2244 * media present, so we can't do that.
2245 */
2246 sdkp->capacity = 0; /* unknown mapped to zero - as usual */
2247 }
2249 #define RC16_LEN 32
2250 #if RC16_LEN > SD_BUF_SIZE
2251 #error RC16_LEN must not be more than SD_BUF_SIZE
2252 #endif
2254 #define READ_CAPACITY_RETRIES_ON_RESET 10
2256 /*
2257 * Ensure that we don't overflow sector_t when CONFIG_LBDAF is not set
2258 * and the reported logical block size is bigger than 512 bytes. Note
2259 * that last_sector is a u64 and therefore logical_to_sectors() is not
2260 * applicable.
2261 */
2262 static bool sd_addressable_capacity(u64 lba, unsigned int sector_size)
2263 {
2264 u64 last_sector = (lba + 1ULL) << (ilog2(sector_size) - 9);
2266 if (sizeof(sector_t) == 4 && last_sector > U32_MAX)
2267 return false;
2269 return true;
2270 }
2272 static int read_capacity_16(struct scsi_disk *sdkp, struct scsi_device *sdp,
2273 unsigned char *buffer)
2274 {
2275 unsigned char cmd[16];
2276 struct scsi_sense_hdr sshdr;
2277 int sense_valid = 0;
2278 int the_result;
2279 int retries = 3, reset_retries = READ_CAPACITY_RETRIES_ON_RESET;
2280 unsigned int alignment;
2281 unsigned long long lba;
2282 unsigned sector_size;
2284 if (sdp->no_read_capacity_16)
2285 return -EINVAL;
2287 do {
2288 memset(cmd, 0, 16);
2289 cmd[0] = SERVICE_ACTION_IN_16;
2290 cmd[1] = SAI_READ_CAPACITY_16;
2291 cmd[13] = RC16_LEN;
2292 memset(buffer, 0, RC16_LEN);
2294 the_result = scsi_execute_req(sdp, cmd, DMA_FROM_DEVICE,
2295 buffer, RC16_LEN, &sshdr,
2296 SD_TIMEOUT, SD_MAX_RETRIES, NULL);
2298 if (media_not_present(sdkp, &sshdr))
2299 return -ENODEV;
2301 if (the_result) {
2302 sense_valid = scsi_sense_valid(&sshdr);
2303 if (sense_valid &&
2304 sshdr.sense_key == ILLEGAL_REQUEST &&
2305 (sshdr.asc == 0x20 || sshdr.asc == 0x24) &&
2306 sshdr.ascq == 0x00)
2307 /* Invalid Command Operation Code or
2308 * Invalid Field in CDB, just retry
2309 * silently with RC10 */
2310 return -EINVAL;
2311 if (sense_valid &&
2312 sshdr.sense_key == UNIT_ATTENTION &&
2313 sshdr.asc == 0x29 && sshdr.ascq == 0x00)
2314 /* Device reset might occur several times,
2315 * give it one more chance */
2316 if (--reset_retries > 0)
2317 continue;
2318 }
2319 retries--;
2321 } while (the_result && retries);
2323 if (the_result) {
2324 sd_print_result(sdkp, "Read Capacity(16) failed", the_result);
2325 read_capacity_error(sdkp, sdp, &sshdr, sense_valid, the_result);
2326 return -EINVAL;
2327 }
2329 sector_size = get_unaligned_be32(&buffer[8]);
2330 lba = get_unaligned_be64(&buffer[0]);
2332 if (sd_read_protection_type(sdkp, buffer) < 0) {
2333 sdkp->capacity = 0;
2334 return -ENODEV;
2335 }
2337 if (!sd_addressable_capacity(lba, sector_size)) {
2338 sd_printk(KERN_ERR, sdkp, "Too big for this kernel. Use a "
2339 "kernel compiled with support for large block "
2340 "devices.\n");
2341 sdkp->capacity = 0;
2342 return -EOVERFLOW;
2343 }
2345 /* Logical blocks per physical block exponent */
2346 sdkp->physical_block_size = (1 << (buffer[13] & 0xf)) * sector_size;
2348 /* RC basis */
2349 sdkp->rc_basis = (buffer[12] >> 4) & 0x3;
2351 /* Lowest aligned logical block */
2352 alignment = ((buffer[14] & 0x3f) << 8 | buffer[15]) * sector_size;
2353 blk_queue_alignment_offset(sdp->request_queue, alignment);
2354 if (alignment && sdkp->first_scan)
2355 sd_printk(KERN_NOTICE, sdkp,
2356 "physical block alignment offset: %u\n", alignment);
2358 if (buffer[14] & 0x80) { /* LBPME */
2359 sdkp->lbpme = 1;
2361 if (buffer[14] & 0x40) /* LBPRZ */
2362 sdkp->lbprz = 1;
2364 sd_config_discard(sdkp, SD_LBP_WS16);
2365 }
2367 sdkp->capacity = lba + 1;
2368 return sector_size;
2369 }
2371 static int read_capacity_10(struct scsi_disk *sdkp, struct scsi_device *sdp,
2372 unsigned char *buffer)
2373 {
2374 unsigned char cmd[16];
2375 struct scsi_sense_hdr sshdr;
2376 int sense_valid = 0;
2377 int the_result;
2378 int retries = 3, reset_retries = READ_CAPACITY_RETRIES_ON_RESET;
2379 sector_t lba;
2380 unsigned sector_size;
2382 do {
2383 cmd[0] = READ_CAPACITY;
2384 memset(&cmd[1], 0, 9);
2385 memset(buffer, 0, 8);
2387 the_result = scsi_execute_req(sdp, cmd, DMA_FROM_DEVICE,
2388 buffer, 8, &sshdr,
2389 SD_TIMEOUT, SD_MAX_RETRIES, NULL);
2391 if (media_not_present(sdkp, &sshdr))
2392 return -ENODEV;
2394 if (the_result) {
2395 sense_valid = scsi_sense_valid(&sshdr);
2396 if (sense_valid &&
2397 sshdr.sense_key == UNIT_ATTENTION &&
2398 sshdr.asc == 0x29 && sshdr.ascq == 0x00)
2399 /* Device reset might occur several times,
2400 * give it one more chance */
2401 if (--reset_retries > 0)
2402 continue;
2403 }
2404 retries--;
2406 } while (the_result && retries);
2408 if (the_result) {
2409 sd_print_result(sdkp, "Read Capacity(10) failed", the_result);
2410 read_capacity_error(sdkp, sdp, &sshdr, sense_valid, the_result);
2411 return -EINVAL;
2412 }
2414 sector_size = get_unaligned_be32(&buffer[4]);
2415 lba = get_unaligned_be32(&buffer[0]);
2417 if (sdp->no_read_capacity_16 && (lba == 0xffffffff)) {
2418 /* Some buggy (usb cardreader) devices return an lba of
2419 0xffffffff when the want to report a size of 0 (with
2420 which they really mean no media is present) */
2421 sdkp->capacity = 0;
2422 sdkp->physical_block_size = sector_size;
2423 return sector_size;
2424 }
2426 if (!sd_addressable_capacity(lba, sector_size)) {
2427 sd_printk(KERN_ERR, sdkp, "Too big for this kernel. Use a "
2428 "kernel compiled with support for large block "
2429 "devices.\n");
2430 sdkp->capacity = 0;
2431 return -EOVERFLOW;
2432 }
2434 sdkp->capacity = lba + 1;
2435 sdkp->physical_block_size = sector_size;
2436 return sector_size;
2437 }
2439 static int sd_try_rc16_first(struct scsi_device *sdp)
2440 {
2441 if (sdp->host->max_cmd_len < 16)
2442 return 0;
2443 if (sdp->try_rc_10_first)
2444 return 0;
2445 if (sdp->scsi_level > SCSI_SPC_2)
2446 return 1;
2447 if (scsi_device_protection(sdp))
2448 return 1;
2449 return 0;
2450 }
2452 /*
2453 * read disk capacity
2454 */
2455 static void
2456 sd_read_capacity(struct scsi_disk *sdkp, unsigned char *buffer)
2457 {
2458 int sector_size;
2459 struct scsi_device *sdp = sdkp->device;
2461 if (sd_try_rc16_first(sdp)) {
2462 sector_size = read_capacity_16(sdkp, sdp, buffer);
2463 if (sector_size == -EOVERFLOW)
2464 goto got_data;
2465 if (sector_size == -ENODEV)
2466 return;
2467 if (sector_size < 0)
2468 sector_size = read_capacity_10(sdkp, sdp, buffer);
2469 if (sector_size < 0)
2470 return;
2471 } else {
2472 sector_size = read_capacity_10(sdkp, sdp, buffer);
2473 if (sector_size == -EOVERFLOW)
2474 goto got_data;
2475 if (sector_size < 0)
2476 return;
2477 if ((sizeof(sdkp->capacity) > 4) &&
2478 (sdkp->capacity > 0xffffffffULL)) {
2479 int old_sector_size = sector_size;
2480 sd_printk(KERN_NOTICE, sdkp, "Very big device. "
2481 "Trying to use READ CAPACITY(16).\n");
2482 sector_size = read_capacity_16(sdkp, sdp, buffer);
2483 if (sector_size < 0) {
2484 sd_printk(KERN_NOTICE, sdkp,
2485 "Using 0xffffffff as device size\n");
2486 sdkp->capacity = 1 + (sector_t) 0xffffffff;
2487 sector_size = old_sector_size;
2488 goto got_data;
2489 }
2490 /* Remember that READ CAPACITY(16) succeeded */
2491 sdp->try_rc_10_first = 0;
2492 }
2493 }
2495 /* Some devices are known to return the total number of blocks,
2496 * not the highest block number. Some devices have versions
2497 * which do this and others which do not. Some devices we might
2498 * suspect of doing this but we don't know for certain.
2499 *
2500 * If we know the reported capacity is wrong, decrement it. If
2501 * we can only guess, then assume the number of blocks is even
2502 * (usually true but not always) and err on the side of lowering
2503 * the capacity.
2504 */
2505 if (sdp->fix_capacity ||
2506 (sdp->guess_capacity && (sdkp->capacity & 0x01))) {
2507 sd_printk(KERN_INFO, sdkp, "Adjusting the sector count "
2508 "from its reported value: %llu\n",
2509 (unsigned long long) sdkp->capacity);
2510 --sdkp->capacity;
2511 }
2513 got_data:
2514 if (sector_size == 0) {
2515 sector_size = 512;
2516 sd_printk(KERN_NOTICE, sdkp, "Sector size 0 reported, "
2517 "assuming 512.\n");
2518 }
2520 if (sector_size != 512 &&
2521 sector_size != 1024 &&
2522 sector_size != 2048 &&
2523 sector_size != 4096) {
2524 sd_printk(KERN_NOTICE, sdkp, "Unsupported sector size %d.\n",
2525 sector_size);
2526 /*
2527 * The user might want to re-format the drive with
2528 * a supported sectorsize. Once this happens, it
2529 * would be relatively trivial to set the thing up.
2530 * For this reason, we leave the thing in the table.
2531 */
2532 sdkp->capacity = 0;
2533 /*
2534 * set a bogus sector size so the normal read/write
2535 * logic in the block layer will eventually refuse any
2536 * request on this device without tripping over power
2537 * of two sector size assumptions
2538 */
2539 sector_size = 512;
2540 }
2541 blk_queue_logical_block_size(sdp->request_queue, sector_size);
2542 blk_queue_physical_block_size(sdp->request_queue,
2543 sdkp->physical_block_size);
2544 sdkp->device->sector_size = sector_size;
2546 if (sdkp->capacity > 0xffffffff)
2547 sdp->use_16_for_rw = 1;
2549 }
2551 /*
2552 * Print disk capacity
2553 */
2554 static void
2555 sd_print_capacity(struct scsi_disk *sdkp,
2556 sector_t old_capacity)
2557 {
2558 int sector_size = sdkp->device->sector_size;
2559 char cap_str_2[10], cap_str_10[10];
2561 string_get_size(sdkp->capacity, sector_size,
2562 STRING_UNITS_2, cap_str_2, sizeof(cap_str_2));
2563 string_get_size(sdkp->capacity, sector_size,
2564 STRING_UNITS_10, cap_str_10,
2565 sizeof(cap_str_10));
2567 if (sdkp->first_scan || old_capacity != sdkp->capacity) {
2568 sd_printk(KERN_NOTICE, sdkp,
2569 "%llu %d-byte logical blocks: (%s/%s)\n",
2570 (unsigned long long)sdkp->capacity,
2571 sector_size, cap_str_10, cap_str_2);
2573 if (sdkp->physical_block_size != sector_size)
2574 sd_printk(KERN_NOTICE, sdkp,
2575 "%u-byte physical blocks\n",
2576 sdkp->physical_block_size);
2578 sd_zbc_print_zones(sdkp);
2579 }
2580 }
2582 /* called with buffer of length 512 */
2583 static inline int
2584 sd_do_mode_sense(struct scsi_device *sdp, int dbd, int modepage,
2585 unsigned char *buffer, int len, struct scsi_mode_data *data,
2586 struct scsi_sense_hdr *sshdr)
2587 {
2588 return scsi_mode_sense(sdp, dbd, modepage, buffer, len,
2589 SD_TIMEOUT, SD_MAX_RETRIES, data,
2590 sshdr);
2591 }
2593 /*
2594 * read write protect setting, if possible - called only in sd_revalidate_disk()
2595 * called with buffer of length SD_BUF_SIZE
2596 */
2597 static void
2598 sd_read_write_protect_flag(struct scsi_disk *sdkp, unsigned char *buffer)
2599 {
2600 int res;
2601 struct scsi_device *sdp = sdkp->device;
2602 struct scsi_mode_data data;
2603 int disk_ro = get_disk_ro(sdkp->disk);
2604 int old_wp = sdkp->write_prot;
2606 set_disk_ro(sdkp->disk, 0);
2607 if (sdp->skip_ms_page_3f) {
2608 sd_first_printk(KERN_NOTICE, sdkp, "Assuming Write Enabled\n");
2609 return;
2610 }
2612 if (sdp->use_192_bytes_for_3f) {
2613 res = sd_do_mode_sense(sdp, 0, 0x3F, buffer, 192, &data, NULL);
2614 } else {
2615 /*
2616 * First attempt: ask for all pages (0x3F), but only 4 bytes.
2617 * We have to start carefully: some devices hang if we ask
2618 * for more than is available.
2619 */
2620 res = sd_do_mode_sense(sdp, 0, 0x3F, buffer, 4, &data, NULL);
2622 /*
2623 * Second attempt: ask for page 0 When only page 0 is
2624 * implemented, a request for page 3F may return Sense Key
2625 * 5: Illegal Request, Sense Code 24: Invalid field in
2626 * CDB.
2627 */
2628 if (!scsi_status_is_good(res))
2629 res = sd_do_mode_sense(sdp, 0, 0, buffer, 4, &data, NULL);
2631 /*
2632 * Third attempt: ask 255 bytes, as we did earlier.
2633 */
2634 if (!scsi_status_is_good(res))
2635 res = sd_do_mode_sense(sdp, 0, 0x3F, buffer, 255,
2636 &data, NULL);
2637 }
2639 if (!scsi_status_is_good(res)) {
2640 sd_first_printk(KERN_WARNING, sdkp,
2641 "Test WP failed, assume Write Enabled\n");
2642 } else {
2643 sdkp->write_prot = ((data.device_specific & 0x80) != 0);
2644 set_disk_ro(sdkp->disk, sdkp->write_prot || disk_ro);
2645 if (sdkp->first_scan || old_wp != sdkp->write_prot) {
2646 sd_printk(KERN_NOTICE, sdkp, "Write Protect is %s\n",
2647 sdkp->write_prot ? "on" : "off");
2648 sd_printk(KERN_DEBUG, sdkp, "Mode Sense: %4ph\n", buffer);
2649 }
2650 }
2651 }
2653 /*
2654 * sd_read_cache_type - called only from sd_revalidate_disk()
2655 * called with buffer of length SD_BUF_SIZE
2656 */
2657 static void
2658 sd_read_cache_type(struct scsi_disk *sdkp, unsigned char *buffer)
2659 {
2660 int len = 0, res;
2661 struct scsi_device *sdp = sdkp->device;
2663 int dbd;
2664 int modepage;
2665 int first_len;
2666 struct scsi_mode_data data;
2667 struct scsi_sense_hdr sshdr;
2668 int old_wce = sdkp->WCE;
2669 int old_rcd = sdkp->RCD;
2670 int old_dpofua = sdkp->DPOFUA;
2673 if (sdkp->cache_override)
2674 return;
2676 first_len = 4;
2677 if (sdp->skip_ms_page_8) {
2678 if (sdp->type == TYPE_RBC)
2679 goto defaults;
2680 else {
2681 if (sdp->skip_ms_page_3f)
2682 goto defaults;
2683 modepage = 0x3F;
2684 if (sdp->use_192_bytes_for_3f)
2685 first_len = 192;
2686 dbd = 0;
2687 }
2688 } else if (sdp->type == TYPE_RBC) {
2689 modepage = 6;
2690 dbd = 8;
2691 } else {
2692 modepage = 8;
2693 dbd = 0;
2694 }
2696 /* cautiously ask */
2697 res = sd_do_mode_sense(sdp, dbd, modepage, buffer, first_len,
2698 &data, &sshdr);
2700 if (!scsi_status_is_good(res))
2701 goto bad_sense;
2703 if (!data.header_length) {
2704 modepage = 6;
2705 first_len = 0;
2706 sd_first_printk(KERN_ERR, sdkp,
2707 "Missing header in MODE_SENSE response\n");
2708 }
2710 /* that went OK, now ask for the proper length */
2711 len = data.length;
2713 /*
2714 * We're only interested in the first three bytes, actually.
2715 * But the data cache page is defined for the first 20.
2716 */
2717 if (len < 3)
2718 goto bad_sense;
2719 else if (len > SD_BUF_SIZE) {
2720 sd_first_printk(KERN_NOTICE, sdkp, "Truncating mode parameter "
2721 "data from %d to %d bytes\n", len, SD_BUF_SIZE);
2722 len = SD_BUF_SIZE;
2723 }
2724 if (modepage == 0x3F && sdp->use_192_bytes_for_3f)
2725 len = 192;
2727 /* Get the data */
2728 if (len > first_len)
2729 res = sd_do_mode_sense(sdp, dbd, modepage, buffer, len,
2730 &data, &sshdr);
2732 if (scsi_status_is_good(res)) {
2733 int offset = data.header_length + data.block_descriptor_length;
2735 while (offset < len) {
2736 u8 page_code = buffer[offset] & 0x3F;
2737 u8 spf = buffer[offset] & 0x40;
2739 if (page_code == 8 || page_code == 6) {
2740 /* We're interested only in the first 3 bytes.
2741 */
2742 if (len - offset <= 2) {
2743 sd_first_printk(KERN_ERR, sdkp,
2744 "Incomplete mode parameter "
2745 "data\n");
2746 goto defaults;
2747 } else {
2748 modepage = page_code;
2749 goto Page_found;
2750 }
2751 } else {
2752 /* Go to the next page */
2753 if (spf && len - offset > 3)
2754 offset += 4 + (buffer[offset+2] << 8) +
2755 buffer[offset+3];
2756 else if (!spf && len - offset > 1)
2757 offset += 2 + buffer[offset+1];
2758 else {
2759 sd_first_printk(KERN_ERR, sdkp,
2760 "Incomplete mode "
2761 "parameter data\n");
2762 goto defaults;
2763 }
2764 }
2765 }
2767 sd_first_printk(KERN_ERR, sdkp, "No Caching mode page found\n");
2768 goto defaults;
2770 Page_found:
2771 if (modepage == 8) {
2772 sdkp->WCE = ((buffer[offset + 2] & 0x04) != 0);
2773 sdkp->RCD = ((buffer[offset + 2] & 0x01) != 0);
2774 } else {
2775 sdkp->WCE = ((buffer[offset + 2] & 0x01) == 0);
2776 sdkp->RCD = 0;
2777 }
2779 sdkp->DPOFUA = (data.device_specific & 0x10) != 0;
2780 if (sdp->broken_fua) {
2781 sd_first_printk(KERN_NOTICE, sdkp, "Disabling FUA\n");
2782 sdkp->DPOFUA = 0;
2783 } else if (sdkp->DPOFUA && !sdkp->device->use_10_for_rw &&
2784 !sdkp->device->use_16_for_rw) {
2785 sd_first_printk(KERN_NOTICE, sdkp,
2786 "Uses READ/WRITE(6), disabling FUA\n");
2787 sdkp->DPOFUA = 0;
2788 }
2790 /* No cache flush allowed for write protected devices */
2791 if (sdkp->WCE && sdkp->write_prot)
2792 sdkp->WCE = 0;
2794 if (sdkp->first_scan || old_wce != sdkp->WCE ||
2795 old_rcd != sdkp->RCD || old_dpofua != sdkp->DPOFUA)
2796 sd_printk(KERN_NOTICE, sdkp,
2797 "Write cache: %s, read cache: %s, %s\n",
2798 sdkp->WCE ? "enabled" : "disabled",
2799 sdkp->RCD ? "disabled" : "enabled",
2800 sdkp->DPOFUA ? "supports DPO and FUA"
2801 : "doesn't support DPO or FUA");
2803 return;
2804 }
2806 bad_sense:
2807 if (scsi_sense_valid(&sshdr) &&
2808 sshdr.sense_key == ILLEGAL_REQUEST &&
2809 sshdr.asc == 0x24 && sshdr.ascq == 0x0)
2810 /* Invalid field in CDB */
2811 sd_first_printk(KERN_NOTICE, sdkp, "Cache data unavailable\n");
2812 else
2813 sd_first_printk(KERN_ERR, sdkp,
2814 "Asking for cache data failed\n");
2816 defaults:
2817 if (sdp->wce_default_on) {
2818 sd_first_printk(KERN_NOTICE, sdkp,
2819 "Assuming drive cache: write back\n");
2820 sdkp->WCE = 1;
2821 } else {
2822 sd_first_printk(KERN_ERR, sdkp,
2823 "Assuming drive cache: write through\n");
2824 sdkp->WCE = 0;
2825 }
2826 sdkp->RCD = 0;
2827 sdkp->DPOFUA = 0;
2828 }
2830 /*
2831 * The ATO bit indicates whether the DIF application tag is available
2832 * for use by the operating system.
2833 */
2834 static void sd_read_app_tag_own(struct scsi_disk *sdkp, unsigned char *buffer)
2835 {
2836 int res, offset;
2837 struct scsi_device *sdp = sdkp->device;
2838 struct scsi_mode_data data;
2839 struct scsi_sense_hdr sshdr;
2841 if (sdp->type != TYPE_DISK && sdp->type != TYPE_ZBC)
2842 return;
2844 if (sdkp->protection_type == 0)
2845 return;
2847 res = scsi_mode_sense(sdp, 1, 0x0a, buffer, 36, SD_TIMEOUT,
2848 SD_MAX_RETRIES, &data, &sshdr);
2850 if (!scsi_status_is_good(res) || !data.header_length ||
2851 data.length < 6) {
2852 sd_first_printk(KERN_WARNING, sdkp,
2853 "getting Control mode page failed, assume no ATO\n");
2855 if (scsi_sense_valid(&sshdr))
2856 sd_print_sense_hdr(sdkp, &sshdr);
2858 return;
2859 }
2861 offset = data.header_length + data.block_descriptor_length;
2863 if ((buffer[offset] & 0x3f) != 0x0a) {
2864 sd_first_printk(KERN_ERR, sdkp, "ATO Got wrong page\n");
2865 return;
2866 }
2868 if ((buffer[offset + 5] & 0x80) == 0)
2869 return;
2871 sdkp->ATO = 1;
2873 return;
2874 }
2876 /**
2877 * sd_read_block_limits - Query disk device for preferred I/O sizes.
2878 * @sdkp: disk to query
2879 */
2880 static void sd_read_block_limits(struct scsi_disk *sdkp)
2881 {
2882 unsigned int sector_sz = sdkp->device->sector_size;
2883 const int vpd_len = 64;
2884 unsigned char *buffer = kmalloc(vpd_len, GFP_KERNEL);
2886 if (!buffer ||
2887 /* Block Limits VPD */
2888 scsi_get_vpd_page(sdkp->device, 0xb0, buffer, vpd_len))
2889 goto out;
2891 blk_queue_io_min(sdkp->disk->queue,
2892 get_unaligned_be16(&buffer[6]) * sector_sz);
2894 sdkp->max_xfer_blocks = get_unaligned_be32(&buffer[8]);
2895 sdkp->opt_xfer_blocks = get_unaligned_be32(&buffer[12]);
2897 if (buffer[3] == 0x3c) {
2898 unsigned int lba_count, desc_count;
2900 sdkp->max_ws_blocks = (u32)get_unaligned_be64(&buffer[36]);
2902 if (!sdkp->lbpme)
2903 goto out;
2905 lba_count = get_unaligned_be32(&buffer[20]);
2906 desc_count = get_unaligned_be32(&buffer[24]);
2908 if (lba_count && desc_count)
2909 sdkp->max_unmap_blocks = lba_count;
2911 sdkp->unmap_granularity = get_unaligned_be32(&buffer[28]);
2913 if (buffer[32] & 0x80)
2914 sdkp->unmap_alignment =
2915 get_unaligned_be32(&buffer[32]) & ~(1 << 31);
2917 if (!sdkp->lbpvpd) { /* LBP VPD page not provided */
2919 if (sdkp->max_unmap_blocks)
2920 sd_config_discard(sdkp, SD_LBP_UNMAP);
2921 else
2922 sd_config_discard(sdkp, SD_LBP_WS16);
2924 } else { /* LBP VPD page tells us what to use */
2925 if (sdkp->lbpu && sdkp->max_unmap_blocks)
2926 sd_config_discard(sdkp, SD_LBP_UNMAP);
2927 else if (sdkp->lbpws)
2928 sd_config_discard(sdkp, SD_LBP_WS16);
2929 else if (sdkp->lbpws10)
2930 sd_config_discard(sdkp, SD_LBP_WS10);
2931 else
2932 sd_config_discard(sdkp, SD_LBP_DISABLE);
2933 }
2934 }
2936 out:
2937 kfree(buffer);
2938 }
2940 /**
2941 * sd_read_block_characteristics - Query block dev. characteristics
2942 * @sdkp: disk to query
2943 */
2944 static void sd_read_block_characteristics(struct scsi_disk *sdkp)
2945 {
2946 struct request_queue *q = sdkp->disk->queue;
2947 unsigned char *buffer;
2948 u16 rot;
2949 const int vpd_len = 64;
2951 buffer = kmalloc(vpd_len, GFP_KERNEL);
2953 if (!buffer ||
2954 /* Block Device Characteristics VPD */
2955 scsi_get_vpd_page(sdkp->device, 0xb1, buffer, vpd_len))
2956 goto out;
2958 rot = get_unaligned_be16(&buffer[4]);
2960 if (rot == 1) {
2961 blk_queue_flag_set(QUEUE_FLAG_NONROT, q);
2962 blk_queue_flag_clear(QUEUE_FLAG_ADD_RANDOM, q);
2963 } else {
2964 blk_queue_flag_clear(QUEUE_FLAG_NONROT, q);
2965 blk_queue_flag_set(QUEUE_FLAG_ADD_RANDOM, q);
2966 }
2968 if (sdkp->device->type == TYPE_ZBC) {
2969 /* Host-managed */
2970 q->limits.zoned = BLK_ZONED_HM;
2971 } else {
2972 sdkp->zoned = (buffer[8] >> 4) & 3;
2973 if (sdkp->zoned == 1)
2974 /* Host-aware */
2975 q->limits.zoned = BLK_ZONED_HA;
2976 else
2977 /*
2978 * Treat drive-managed devices as
2979 * regular block devices.
2980 */
2981 q->limits.zoned = BLK_ZONED_NONE;
2982 }
2983 if (blk_queue_is_zoned(q) && sdkp->first_scan)
2984 sd_printk(KERN_NOTICE, sdkp, "Host-%s zoned block device\n",
2985 q->limits.zoned == BLK_ZONED_HM ? "managed" : "aware");
2987 out:
2988 kfree(buffer);
2989 }
2991 /**
2992 * sd_read_block_provisioning - Query provisioning VPD page
2993 * @sdkp: disk to query
2994 */
2995 static void sd_read_block_provisioning(struct scsi_disk *sdkp)
2996 {
2997 unsigned char *buffer;
2998 const int vpd_len = 8;
3000 if (sdkp->lbpme == 0)
3001 return;
3003 buffer = kmalloc(vpd_len, GFP_KERNEL);
3005 if (!buffer || scsi_get_vpd_page(sdkp->device, 0xb2, buffer, vpd_len))
3006 goto out;
3008 sdkp->lbpvpd = 1;
3009 sdkp->lbpu = (buffer[5] >> 7) & 1; /* UNMAP */
3010 sdkp->lbpws = (buffer[5] >> 6) & 1; /* WRITE SAME(16) with UNMAP */
3011 sdkp->lbpws10 = (buffer[5] >> 5) & 1; /* WRITE SAME(10) with UNMAP */
3013 out:
3014 kfree(buffer);
3015 }
3017 static void sd_read_write_same(struct scsi_disk *sdkp, unsigned char *buffer)
3018 {
3019 struct scsi_device *sdev = sdkp->device;
3021 if (sdev->host->no_write_same) {
3022 sdev->no_write_same = 1;
3024 return;
3025 }
3027 if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE, INQUIRY) < 0) {
3028 /* too large values might cause issues with arcmsr */
3029 int vpd_buf_len = 64;
3031 sdev->no_report_opcodes = 1;
3033 /* Disable WRITE SAME if REPORT SUPPORTED OPERATION
3034 * CODES is unsupported and the device has an ATA
3035 * Information VPD page (SAT).
3036 */
3037 if (!scsi_get_vpd_page(sdev, 0x89, buffer, vpd_buf_len))
3038 sdev->no_write_same = 1;
3039 }
3041 if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE, WRITE_SAME_16) == 1)
3042 sdkp->ws16 = 1;
3044 if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE, WRITE_SAME) == 1)
3045 sdkp->ws10 = 1;
3046 }
3048 static void sd_read_security(struct scsi_disk *sdkp, unsigned char *buffer)
3049 {
3050 struct scsi_device *sdev = sdkp->device;
3052 if (!sdev->security_supported)
3053 return;
3055 if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE,
3056 SECURITY_PROTOCOL_IN) == 1 &&
3057 scsi_report_opcode(sdev, buffer, SD_BUF_SIZE,
3058 SECURITY_PROTOCOL_OUT) == 1)
3059 sdkp->security = 1;
3060 }
3062 /**
3063 * sd_revalidate_disk - called the first time a new disk is seen,
3064 * performs disk spin up, read_capacity, etc.
3065 * @disk: struct gendisk we care about
3066 **/
3067 static int sd_revalidate_disk(struct gendisk *disk)
3068 {
3069 struct scsi_disk *sdkp = scsi_disk(disk);
3070 struct scsi_device *sdp = sdkp->device;
3071 struct request_queue *q = sdkp->disk->queue;
3072 sector_t old_capacity = sdkp->capacity;
3073 unsigned char *buffer;
3074 unsigned int dev_max, rw_max;
3076 SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp,
3077 "sd_revalidate_disk\n"));
3079 /*
3080 * If the device is offline, don't try and read capacity or any
3081 * of the other niceties.
3082 */
3083 if (!scsi_device_online(sdp))
3084 goto out;
3086 buffer = kmalloc(SD_BUF_SIZE, GFP_KERNEL);
3087 if (!buffer) {
3088 sd_printk(KERN_WARNING, sdkp, "sd_revalidate_disk: Memory "
3089 "allocation failure.\n");
3090 goto out;
3091 }
3093 sd_spinup_disk(sdkp);
3095 /*
3096 * Without media there is no reason to ask; moreover, some devices
3097 * react badly if we do.
3098 */
3099 if (sdkp->media_present) {
3100 sd_read_capacity(sdkp, buffer);
3102 if (scsi_device_supports_vpd(sdp)) {
3103 sd_read_block_provisioning(sdkp);
3104 sd_read_block_limits(sdkp);
3105 sd_read_block_characteristics(sdkp);
3106 sd_zbc_read_zones(sdkp, buffer);
3107 }
3109 sd_print_capacity(sdkp, old_capacity);
3111 sd_read_write_protect_flag(sdkp, buffer);
3112 sd_read_cache_type(sdkp, buffer);
3113 sd_read_app_tag_own(sdkp, buffer);
3114 sd_read_write_same(sdkp, buffer);
3115 sd_read_security(sdkp, buffer);
3116 }
3118 /*
3119 * We now have all cache related info, determine how we deal
3120 * with flush requests.
3121 */
3122 sd_set_flush_flag(sdkp);
3124 /* Initial block count limit based on CDB TRANSFER LENGTH field size. */
3125 dev_max = sdp->use_16_for_rw ? SD_MAX_XFER_BLOCKS : SD_DEF_XFER_BLOCKS;
3127 /* Some devices report a maximum block count for READ/WRITE requests. */
3128 dev_max = min_not_zero(dev_max, sdkp->max_xfer_blocks);
3129 q->limits.max_dev_sectors = logical_to_sectors(sdp, dev_max);
3131 /*
3132 * Determine the device's preferred I/O size for reads and writes
3133 * unless the reported value is unreasonably small, large, or
3134 * garbage.
3135 */
3136 if (sdkp->opt_xfer_blocks &&
3137 sdkp->opt_xfer_blocks <= dev_max &&
3138 sdkp->opt_xfer_blocks <= SD_DEF_XFER_BLOCKS &&
3139 logical_to_bytes(sdp, sdkp->opt_xfer_blocks) >= PAGE_SIZE) {
3140 q->limits.io_opt = logical_to_bytes(sdp, sdkp->opt_xfer_blocks);
3141 rw_max = logical_to_sectors(sdp, sdkp->opt_xfer_blocks);
3142 } else
3143 rw_max = min_not_zero(logical_to_sectors(sdp, dev_max),
3144 (sector_t)BLK_DEF_MAX_SECTORS);
3146 /* Do not exceed controller limit */
3147 rw_max = min(rw_max, queue_max_hw_sectors(q));
3149 /*
3150 * Only update max_sectors if previously unset or if the current value
3151 * exceeds the capabilities of the hardware.
3152 */
3153 if (sdkp->first_scan ||
3154 q->limits.max_sectors > q->limits.max_dev_sectors ||
3155 q->limits.max_sectors > q->limits.max_hw_sectors)
3156 q->limits.max_sectors = rw_max;
3158 sdkp->first_scan = 0;
3160 set_capacity(disk, logical_to_sectors(sdp, sdkp->capacity));
3161 sd_config_write_same(sdkp);
3162 kfree(buffer);
3164 out:
3165 return 0;
3166 }
3168 /**
3169 * sd_unlock_native_capacity - unlock native capacity
3170 * @disk: struct gendisk to set capacity for
3171 *
3172 * Block layer calls this function if it detects that partitions
3173 * on @disk reach beyond the end of the device. If the SCSI host
3174 * implements ->unlock_native_capacity() method, it's invoked to
3175 * give it a chance to adjust the device capacity.
3176 *
3177 * CONTEXT:
3178 * Defined by block layer. Might sleep.
3179 */
3180 static void sd_unlock_native_capacity(struct gendisk *disk)
3181 {
3182 struct scsi_device *sdev = scsi_disk(disk)->device;
3184 if (sdev->host->hostt->unlock_native_capacity)
3185 sdev->host->hostt->unlock_native_capacity(sdev);
3186 }
3188 /**
3189 * sd_format_disk_name - format disk name
3190 * @prefix: name prefix - ie. "sd" for SCSI disks
3191 * @index: index of the disk to format name for
3192 * @buf: output buffer
3193 * @buflen: length of the output buffer
3194 *
3195 * SCSI disk names starts at sda. The 26th device is sdz and the
3196 * 27th is sdaa. The last one for two lettered suffix is sdzz
3197 * which is followed by sdaaa.
3198 *
3199 * This is basically 26 base counting with one extra 'nil' entry
3200 * at the beginning from the second digit on and can be
3201 * determined using similar method as 26 base conversion with the
3202 * index shifted -1 after each digit is computed.
3203 *
3204 * CONTEXT:
3205 * Don't care.
3206 *
3207 * RETURNS:
3208 * 0 on success, -errno on failure.
3209 */
3210 static int sd_format_disk_name(char *prefix, int index, char *buf, int buflen)
3211 {
3212 const int base = 'z' - 'a' + 1;
3213 char *begin = buf + strlen(prefix);
3214 char *end = buf + buflen;
3215 char *p;
3216 int unit;
3218 p = end - 1;
3219 *p = '\0';
3220 unit = base;
3221 do {
3222 if (p == begin)
3223 return -EINVAL;
3224 *--p = 'a' + (index % unit);
3225 index = (index / unit) - 1;
3226 } while (index >= 0);
3228 memmove(begin, p, end - p);
3229 memcpy(buf, prefix, strlen(prefix));
3231 return 0;
3232 }
3234 /*
3235 * The asynchronous part of sd_probe
3236 */
3237 static void sd_probe_async(void *data, async_cookie_t cookie)
3238 {
3239 struct scsi_disk *sdkp = data;
3240 struct scsi_device *sdp;
3241 struct gendisk *gd;
3242 u32 index;
3243 struct device *dev;
3245 sdp = sdkp->device;
3246 gd = sdkp->disk;
3247 index = sdkp->index;
3248 dev = &sdp->sdev_gendev;
3250 gd->major = sd_major((index & 0xf0) >> 4);
3251 gd->first_minor = ((index & 0xf) << 4) | (index & 0xfff00);
3253 gd->fops = &sd_fops;
3254 gd->private_data = &sdkp->driver;
3255 gd->queue = sdkp->device->request_queue;
3257 /* defaults, until the device tells us otherwise */
3258 sdp->sector_size = 512;
3259 sdkp->capacity = 0;
3260 sdkp->media_present = 1;
3261 sdkp->write_prot = 0;
3262 sdkp->cache_override = 0;
3263 sdkp->WCE = 0;
3264 sdkp->RCD = 0;
3265 sdkp->ATO = 0;
3266 sdkp->first_scan = 1;
3267 sdkp->max_medium_access_timeouts = SD_MAX_MEDIUM_TIMEOUTS;
3269 sd_revalidate_disk(gd);
3271 gd->flags = GENHD_FL_EXT_DEVT;
3272 if (sdp->removable) {
3273 gd->flags |= GENHD_FL_REMOVABLE;
3274 gd->events |= DISK_EVENT_MEDIA_CHANGE;
3275 }
3277 blk_pm_runtime_init(sdp->request_queue, dev);
3278 device_add_disk(dev, gd);
3279 if (sdkp->capacity)
3280 sd_dif_config_host(sdkp);
3282 sd_revalidate_disk(gd);
3284 if (sdkp->security) {
3285 sdkp->opal_dev = init_opal_dev(sdp, &sd_sec_submit);
3286 if (sdkp->opal_dev)
3287 sd_printk(KERN_NOTICE, sdkp, "supports TCG Opal\n");
3288 }
3290 sd_printk(KERN_NOTICE, sdkp, "Attached SCSI %sdisk\n",
3291 sdp->removable ? "removable " : "");
3292 scsi_autopm_put_device(sdp);
3293 put_device(&sdkp->dev);
3294 }
3296 /**
3297 * sd_probe - called during driver initialization and whenever a
3298 * new scsi device is attached to the system. It is called once
3299 * for each scsi device (not just disks) present.
3300 * @dev: pointer to device object
3301 *
3302 * Returns 0 if successful (or not interested in this scsi device
3303 * (e.g. scanner)); 1 when there is an error.
3304 *
3305 * Note: this function is invoked from the scsi mid-level.
3306 * This function sets up the mapping between a given
3307 * <host,channel,id,lun> (found in sdp) and new device name
3308 * (e.g. /dev/sda). More precisely it is the block device major
3309 * and minor number that is chosen here.
3310 *
3311 * Assume sd_probe is not re-entrant (for time being)
3312 * Also think about sd_probe() and sd_remove() running coincidentally.
3313 **/
3314 static int sd_probe(struct device *dev)
3315 {
3316 struct scsi_device *sdp = to_scsi_device(dev);
3317 struct scsi_disk *sdkp;
3318 struct gendisk *gd;
3319 int index;
3320 int error;
3322 scsi_autopm_get_device(sdp);
3323 error = -ENODEV;
3324 if (sdp->type != TYPE_DISK &&
3325 sdp->type != TYPE_ZBC &&
3326 sdp->type != TYPE_MOD &&
3327 sdp->type != TYPE_RBC)
3328 goto out;
3330 #ifndef CONFIG_BLK_DEV_ZONED
3331 if (sdp->type == TYPE_ZBC)
3332 goto out;
3333 #endif
3334 SCSI_LOG_HLQUEUE(3, sdev_printk(KERN_INFO, sdp,
3335 "sd_probe\n"));
3337 error = -ENOMEM;
3338 sdkp = kzalloc(sizeof(*sdkp), GFP_KERNEL);
3339 if (!sdkp)
3340 goto out;
3342 gd = alloc_disk(SD_MINORS);
3343 if (!gd)
3344 goto out_free;
3346 index = ida_alloc(&sd_index_ida, GFP_KERNEL);
3347 if (index < 0) {
3348 sdev_printk(KERN_WARNING, sdp, "sd_probe: memory exhausted.\n");
3349 goto out_put;
3350 }
3352 error = sd_format_disk_name("sd", index, gd->disk_name, DISK_NAME_LEN);
3353 if (error) {
3354 sdev_printk(KERN_WARNING, sdp, "SCSI disk (sd) name length exceeded.\n");
3355 goto out_free_index;
3356 }
3358 sdkp->device = sdp;
3359 sdkp->driver = &sd_template;
3360 sdkp->disk = gd;
3361 sdkp->index = index;
3362 atomic_set(&sdkp->openers, 0);
3363 atomic_set(&sdkp->device->ioerr_cnt, 0);
3365 if (!sdp->request_queue->rq_timeout) {
3366 if (sdp->type != TYPE_MOD)
3367 blk_queue_rq_timeout(sdp->request_queue, SD_TIMEOUT);
3368 else
3369 blk_queue_rq_timeout(sdp->request_queue,
3370 SD_MOD_TIMEOUT);
3371 }
3373 device_initialize(&sdkp->dev);
3374 sdkp->dev.parent = dev;
3375 sdkp->dev.class = &sd_disk_class;
3376 dev_set_name(&sdkp->dev, "%s", dev_name(dev));
3378 error = device_add(&sdkp->dev);
3379 if (error)
3380 goto out_free_index;
3382 get_device(dev);
3383 dev_set_drvdata(dev, sdkp);
3385 get_device(&sdkp->dev); /* prevent release before async_schedule */
3386 async_schedule_domain(sd_probe_async, sdkp, &scsi_sd_probe_domain);
3388 return 0;
3390 out_free_index:
3391 ida_free(&sd_index_ida, index);
3392 out_put:
3393 put_disk(gd);
3394 out_free:
3395 kfree(sdkp);
3396 out:
3397 scsi_autopm_put_device(sdp);
3398 return error;
3399 }
3401 /**
3402 * sd_remove - called whenever a scsi disk (previously recognized by
3403 * sd_probe) is detached from the system. It is called (potentially
3404 * multiple times) during sd module unload.
3405 * @dev: pointer to device object
3406 *
3407 * Note: this function is invoked from the scsi mid-level.
3408 * This function potentially frees up a device name (e.g. /dev/sdc)
3409 * that could be re-used by a subsequent sd_probe().
3410 * This function is not called when the built-in sd driver is "exit-ed".
3411 **/
3412 static int sd_remove(struct device *dev)
3413 {
3414 struct scsi_disk *sdkp;
3415 dev_t devt;
3417 sdkp = dev_get_drvdata(dev);
3418 devt = disk_devt(sdkp->disk);
3419 scsi_autopm_get_device(sdkp->device);
3421 async_synchronize_full_domain(&scsi_sd_pm_domain);
3422 async_synchronize_full_domain(&scsi_sd_probe_domain);
3423 device_del(&sdkp->dev);
3424 del_gendisk(sdkp->disk);
3425 sd_shutdown(dev);
3427 sd_zbc_remove(sdkp);
3429 free_opal_dev(sdkp->opal_dev);
3431 blk_register_region(devt, SD_MINORS, NULL,
3432 sd_default_probe, NULL, NULL);
3434 mutex_lock(&sd_ref_mutex);
3435 dev_set_drvdata(dev, NULL);
3436 put_device(&sdkp->dev);
3437 mutex_unlock(&sd_ref_mutex);
3439 return 0;
3440 }
3442 /**
3443 * scsi_disk_release - Called to free the scsi_disk structure
3444 * @dev: pointer to embedded class device
3445 *
3446 * sd_ref_mutex must be held entering this routine. Because it is
3447 * called on last put, you should always use the scsi_disk_get()
3448 * scsi_disk_put() helpers which manipulate the semaphore directly
3449 * and never do a direct put_device.
3450 **/
3451 static void scsi_disk_release(struct device *dev)
3452 {
3453 struct scsi_disk *sdkp = to_scsi_disk(dev);
3454 struct gendisk *disk = sdkp->disk;
3456 ida_free(&sd_index_ida, sdkp->index);
3458 disk->private_data = NULL;
3459 put_disk(disk);
3460 put_device(&sdkp->device->sdev_gendev);
3462 kfree(sdkp);
3463 }
3465 static int sd_start_stop_device(struct scsi_disk *sdkp, int start)
3466 {
3467 unsigned char cmd[6] = { START_STOP }; /* START_VALID */
3468 struct scsi_sense_hdr sshdr;
3469 struct scsi_device *sdp = sdkp->device;
3470 int res;
3472 if (start)
3473 cmd[4] |= 1; /* START */
3475 if (sdp->start_stop_pwr_cond)
3476 cmd[4] |= start ? 1 << 4 : 3 << 4; /* Active or Standby */
3478 if (!scsi_device_online(sdp))
3479 return -ENODEV;
3481 res = scsi_execute(sdp, cmd, DMA_NONE, NULL, 0, NULL, &sshdr,
3482 SD_TIMEOUT, SD_MAX_RETRIES, 0, RQF_PM, NULL);
3483 if (res) {
3484 sd_print_result(sdkp, "Start/Stop Unit failed", res);
3485 if (driver_byte(res) == DRIVER_SENSE)
3486 sd_print_sense_hdr(sdkp, &sshdr);
3487 if (scsi_sense_valid(&sshdr) &&
3488 /* 0x3a is medium not present */
3489 sshdr.asc == 0x3a)
3490 res = 0;
3491 }
3493 /* SCSI error codes must not go to the generic layer */
3494 if (res)
3495 return -EIO;
3497 return 0;
3498 }
3500 /*
3501 * Send a SYNCHRONIZE CACHE instruction down to the device through
3502 * the normal SCSI command structure. Wait for the command to
3503 * complete.
3504 */
3505 static void sd_shutdown(struct device *dev)
3506 {
3507 struct scsi_disk *sdkp = dev_get_drvdata(dev);
3509 if (!sdkp)
3510 return; /* this can happen */
3512 if (pm_runtime_suspended(dev))
3513 return;
3515 if (sdkp->WCE && sdkp->media_present) {
3516 sd_printk(KERN_NOTICE, sdkp, "Synchronizing SCSI cache\n");
3517 sd_sync_cache(sdkp, NULL);
3518 }
3520 if (system_state != SYSTEM_RESTART && sdkp->device->manage_start_stop) {
3521 sd_printk(KERN_NOTICE, sdkp, "Stopping disk\n");
3522 sd_start_stop_device(sdkp, 0);
3523 }
3524 }
3526 static int sd_suspend_common(struct device *dev, bool ignore_stop_errors)
3527 {
3528 struct scsi_disk *sdkp = dev_get_drvdata(dev);
3529 struct scsi_sense_hdr sshdr;
3530 int ret = 0;
3532 if (!sdkp) /* E.g.: runtime suspend following sd_remove() */
3533 return 0;
3535 if (sdkp->WCE && sdkp->media_present) {
3536 sd_printk(KERN_NOTICE, sdkp, "Synchronizing SCSI cache\n");
3537 ret = sd_sync_cache(sdkp, &sshdr);
3539 if (ret) {
3540 /* ignore OFFLINE device */
3541 if (ret == -ENODEV)
3542 return 0;
3544 if (!scsi_sense_valid(&sshdr) ||
3545 sshdr.sense_key != ILLEGAL_REQUEST)
3546 return ret;
3548 /*
3549 * sshdr.sense_key == ILLEGAL_REQUEST means this drive
3550 * doesn't support sync. There's not much to do and
3551 * suspend shouldn't fail.
3552 */
3553 ret = 0;
3554 }
3555 }
3557 if (sdkp->device->manage_start_stop) {
3558 sd_printk(KERN_NOTICE, sdkp, "Stopping disk\n");
3559 /* an error is not worth aborting a system sleep */
3560 ret = sd_start_stop_device(sdkp, 0);
3561 if (ignore_stop_errors)
3562 ret = 0;
3563 }
3565 return ret;
3566 }
3568 static int sd_suspend_system(struct device *dev)
3569 {
3570 return sd_suspend_common(dev, true);
3571 }
3573 static int sd_suspend_runtime(struct device *dev)
3574 {
3575 return sd_suspend_common(dev, false);
3576 }
3578 static int sd_resume(struct device *dev)
3579 {
3580 struct scsi_disk *sdkp = dev_get_drvdata(dev);
3581 int ret;
3583 if (!sdkp) /* E.g.: runtime resume at the start of sd_probe() */
3584 return 0;
3586 if (!sdkp->device->manage_start_stop)
3587 return 0;
3589 sd_printk(KERN_NOTICE, sdkp, "Starting disk\n");
3590 ret = sd_start_stop_device(sdkp, 1);
3591 if (!ret)
3592 opal_unlock_from_suspend(sdkp->opal_dev);
3593 return ret;
3594 }
3596 /**
3597 * init_sd - entry point for this driver (both when built in or when
3598 * a module).
3599 *
3600 * Note: this function registers this driver with the scsi mid-level.
3601 **/
3602 static int __init init_sd(void)
3603 {
3604 int majors = 0, i, err;
3606 SCSI_LOG_HLQUEUE(3, printk("init_sd: sd driver entry point\n"));
3608 for (i = 0; i < SD_MAJORS; i++) {
3609 if (register_blkdev(sd_major(i), "sd") != 0)
3610 continue;
3611 majors++;
3612 blk_register_region(sd_major(i), SD_MINORS, NULL,
3613 sd_default_probe, NULL, NULL);
3614 }
3616 if (!majors)
3617 return -ENODEV;
3619 err = class_register(&sd_disk_class);
3620 if (err)
3621 goto err_out;
3623 sd_cdb_cache = kmem_cache_create("sd_ext_cdb", SD_EXT_CDB_SIZE,
3624 0, 0, NULL);
3625 if (!sd_cdb_cache) {
3626 printk(KERN_ERR "sd: can't init extended cdb cache\n");
3627 err = -ENOMEM;
3628 goto err_out_class;
3629 }
3631 sd_cdb_pool = mempool_create_slab_pool(SD_MEMPOOL_SIZE, sd_cdb_cache);
3632 if (!sd_cdb_pool) {
3633 printk(KERN_ERR "sd: can't init extended cdb pool\n");
3634 err = -ENOMEM;
3635 goto err_out_cache;
3636 }
3638 err = scsi_register_driver(&sd_template.gendrv);
3639 if (err)
3640 goto err_out_driver;
3642 return 0;
3644 err_out_driver:
3645 mempool_destroy(sd_cdb_pool);
3647 err_out_cache:
3648 kmem_cache_destroy(sd_cdb_cache);
3650 err_out_class:
3651 class_unregister(&sd_disk_class);
3652 err_out:
3653 for (i = 0; i < SD_MAJORS; i++)
3654 unregister_blkdev(sd_major(i), "sd");
3655 return err;
3656 }
3658 /**
3659 * exit_sd - exit point for this driver (when it is a module).
3660 *
3661 * Note: this function unregisters this driver from the scsi mid-level.
3662 **/
3663 static void __exit exit_sd(void)
3664 {
3665 int i;
3667 SCSI_LOG_HLQUEUE(3, printk("exit_sd: exiting sd driver\n"));
3669 scsi_unregister_driver(&sd_template.gendrv);
3670 mempool_destroy(sd_cdb_pool);
3671 kmem_cache_destroy(sd_cdb_cache);
3673 class_unregister(&sd_disk_class);
3675 for (i = 0; i < SD_MAJORS; i++) {
3676 blk_unregister_region(sd_major(i), SD_MINORS);
3677 unregister_blkdev(sd_major(i), "sd");
3678 }
3679 }
3681 module_init(init_sd);
3682 module_exit(exit_sd);
3684 static void sd_print_sense_hdr(struct scsi_disk *sdkp,
3685 struct scsi_sense_hdr *sshdr)
3686 {
3687 scsi_print_sense_hdr(sdkp->device,
3688 sdkp->disk ? sdkp->disk->disk_name : NULL, sshdr);
3689 }
3691 static void sd_print_result(const struct scsi_disk *sdkp, const char *msg,
3692 int result)
3693 {
3694 const char *hb_string = scsi_hostbyte_string(result);
3695 const char *db_string = scsi_driverbyte_string(result);
3697 if (hb_string || db_string)
3698 sd_printk(KERN_INFO, sdkp,
3699 "%s: Result: hostbyte=%s driverbyte=%s\n", msg,
3700 hb_string ? hb_string : "invalid",
3701 db_string ? db_string : "invalid");
3702 else
3703 sd_printk(KERN_INFO, sdkp,
3704 "%s: Result: hostbyte=0x%02x driverbyte=0x%02x\n",
3705 msg, host_byte(result), driver_byte(result));
3706 }