xref: /linux/drivers/scsi/sd.c (revision 0074281bb6316108e0cff094bd4db78ab3eee236)
1  // SPDX-License-Identifier: GPL-2.0-only
2  /*
3   *      sd.c Copyright (C) 1992 Drew Eckhardt
4   *           Copyright (C) 1993, 1994, 1995, 1999 Eric Youngdale
5   *
6   *      Linux scsi disk driver
7   *              Initial versions: Drew Eckhardt
8   *              Subsequent revisions: Eric Youngdale
9   *	Modification history:
10   *       - Drew Eckhardt <drew@colorado.edu> original
11   *       - Eric Youngdale <eric@andante.org> add scatter-gather, multiple
12   *         outstanding request, and other enhancements.
13   *         Support loadable low-level scsi drivers.
14   *       - Jirka Hanika <geo@ff.cuni.cz> support more scsi disks using
15   *         eight major numbers.
16   *       - Richard Gooch <rgooch@atnf.csiro.au> support devfs.
17   *	 - Torben Mathiasen <tmm@image.dk> Resource allocation fixes in
18   *	   sd_init and cleanups.
19   *	 - Alex Davis <letmein@erols.com> Fix problem where partition info
20   *	   not being read in sd_open. Fix problem where removable media
21   *	   could be ejected after sd_open.
22   *	 - Douglas Gilbert <dgilbert@interlog.com> cleanup for lk 2.5.x
23   *	 - Badari Pulavarty <pbadari@us.ibm.com>, Matthew Wilcox
24   *	   <willy@debian.org>, Kurt Garloff <garloff@suse.de>:
25   *	   Support 32k/1M disks.
26   *
27   *	Logging policy (needs CONFIG_SCSI_LOGGING defined):
28   *	 - setting up transfer: SCSI_LOG_HLQUEUE levels 1 and 2
29   *	 - end of transfer (bh + scsi_lib): SCSI_LOG_HLCOMPLETE level 1
30   *	 - entering sd_ioctl: SCSI_LOG_IOCTL level 1
31   *	 - entering other commands: SCSI_LOG_HLQUEUE level 3
32   *	Note: when the logging level is set by the user, it must be greater
33   *	than the level indicated above to trigger output.
34   */
35  
36  #include <linux/bio-integrity.h>
37  #include <linux/module.h>
38  #include <linux/fs.h>
39  #include <linux/kernel.h>
40  #include <linux/mm.h>
41  #include <linux/hdreg.h>
42  #include <linux/errno.h>
43  #include <linux/idr.h>
44  #include <linux/interrupt.h>
45  #include <linux/init.h>
46  #include <linux/blkdev.h>
47  #include <linux/blkpg.h>
48  #include <linux/blk-pm.h>
49  #include <linux/delay.h>
50  #include <linux/rw_hint.h>
51  #include <linux/major.h>
52  #include <linux/mutex.h>
53  #include <linux/string_helpers.h>
54  #include <linux/slab.h>
55  #include <linux/sed-opal.h>
56  #include <linux/pm_runtime.h>
57  #include <linux/pr.h>
58  #include <linux/t10-pi.h>
59  #include <linux/uaccess.h>
60  #include <linux/unaligned.h>
61  
62  #include <scsi/scsi.h>
63  #include <scsi/scsi_cmnd.h>
64  #include <scsi/scsi_dbg.h>
65  #include <scsi/scsi_device.h>
66  #include <scsi/scsi_devinfo.h>
67  #include <scsi/scsi_driver.h>
68  #include <scsi/scsi_eh.h>
69  #include <scsi/scsi_host.h>
70  #include <scsi/scsi_ioctl.h>
71  #include <scsi/scsicam.h>
72  #include <scsi/scsi_common.h>
73  
74  #include "sd.h"
75  #include "scsi_priv.h"
76  #include "scsi_logging.h"
77  
78  MODULE_AUTHOR("Eric Youngdale");
79  MODULE_DESCRIPTION("SCSI disk (sd) driver");
80  MODULE_LICENSE("GPL");
81  
82  MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK0_MAJOR);
83  MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK1_MAJOR);
84  MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK2_MAJOR);
85  MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK3_MAJOR);
86  MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK4_MAJOR);
87  MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK5_MAJOR);
88  MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK6_MAJOR);
89  MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK7_MAJOR);
90  MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK8_MAJOR);
91  MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK9_MAJOR);
92  MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK10_MAJOR);
93  MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK11_MAJOR);
94  MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK12_MAJOR);
95  MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK13_MAJOR);
96  MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK14_MAJOR);
97  MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK15_MAJOR);
98  MODULE_ALIAS_SCSI_DEVICE(TYPE_DISK);
99  MODULE_ALIAS_SCSI_DEVICE(TYPE_MOD);
100  MODULE_ALIAS_SCSI_DEVICE(TYPE_RBC);
101  MODULE_ALIAS_SCSI_DEVICE(TYPE_ZBC);
102  
103  #define SD_MINORS	16
104  
105  static void sd_config_discard(struct scsi_disk *sdkp, struct queue_limits *lim,
106  		unsigned int mode);
107  static void sd_config_write_same(struct scsi_disk *sdkp,
108  		struct queue_limits *lim);
109  static int  sd_revalidate_disk(struct gendisk *);
110  static void sd_unlock_native_capacity(struct gendisk *disk);
111  static void sd_shutdown(struct device *);
112  static void scsi_disk_release(struct device *cdev);
113  
114  static DEFINE_IDA(sd_index_ida);
115  
116  static mempool_t *sd_page_pool;
117  static struct lock_class_key sd_bio_compl_lkclass;
118  
119  static const char *sd_cache_types[] = {
120  	"write through", "none", "write back",
121  	"write back, no read (daft)"
122  };
123  
sd_set_flush_flag(struct scsi_disk * sdkp,struct queue_limits * lim)124  static void sd_set_flush_flag(struct scsi_disk *sdkp,
125  		struct queue_limits *lim)
126  {
127  	if (sdkp->WCE) {
128  		lim->features |= BLK_FEAT_WRITE_CACHE;
129  		if (sdkp->DPOFUA)
130  			lim->features |= BLK_FEAT_FUA;
131  		else
132  			lim->features &= ~BLK_FEAT_FUA;
133  	} else {
134  		lim->features &= ~(BLK_FEAT_WRITE_CACHE | BLK_FEAT_FUA);
135  	}
136  }
137  
138  static ssize_t
cache_type_store(struct device * dev,struct device_attribute * attr,const char * buf,size_t count)139  cache_type_store(struct device *dev, struct device_attribute *attr,
140  		 const char *buf, size_t count)
141  {
142  	int ct, rcd, wce, sp;
143  	struct scsi_disk *sdkp = to_scsi_disk(dev);
144  	struct scsi_device *sdp = sdkp->device;
145  	char buffer[64];
146  	char *buffer_data;
147  	struct scsi_mode_data data;
148  	struct scsi_sense_hdr sshdr;
149  	static const char temp[] = "temporary ";
150  	int len, ret;
151  
152  	if (sdp->type != TYPE_DISK && sdp->type != TYPE_ZBC)
153  		/* no cache control on RBC devices; theoretically they
154  		 * can do it, but there's probably so many exceptions
155  		 * it's not worth the risk */
156  		return -EINVAL;
157  
158  	if (strncmp(buf, temp, sizeof(temp) - 1) == 0) {
159  		buf += sizeof(temp) - 1;
160  		sdkp->cache_override = 1;
161  	} else {
162  		sdkp->cache_override = 0;
163  	}
164  
165  	ct = sysfs_match_string(sd_cache_types, buf);
166  	if (ct < 0)
167  		return -EINVAL;
168  
169  	rcd = ct & 0x01 ? 1 : 0;
170  	wce = (ct & 0x02) && !sdkp->write_prot ? 1 : 0;
171  
172  	if (sdkp->cache_override) {
173  		struct queue_limits lim;
174  
175  		sdkp->WCE = wce;
176  		sdkp->RCD = rcd;
177  
178  		lim = queue_limits_start_update(sdkp->disk->queue);
179  		sd_set_flush_flag(sdkp, &lim);
180  		ret = queue_limits_commit_update_frozen(sdkp->disk->queue,
181  				&lim);
182  		if (ret)
183  			return ret;
184  		return count;
185  	}
186  
187  	if (scsi_mode_sense(sdp, 0x08, 8, 0, buffer, sizeof(buffer), SD_TIMEOUT,
188  			    sdkp->max_retries, &data, NULL))
189  		return -EINVAL;
190  	len = min_t(size_t, sizeof(buffer), data.length - data.header_length -
191  		  data.block_descriptor_length);
192  	buffer_data = buffer + data.header_length +
193  		data.block_descriptor_length;
194  	buffer_data[2] &= ~0x05;
195  	buffer_data[2] |= wce << 2 | rcd;
196  	sp = buffer_data[0] & 0x80 ? 1 : 0;
197  	buffer_data[0] &= ~0x80;
198  
199  	/*
200  	 * Ensure WP, DPOFUA, and RESERVED fields are cleared in
201  	 * received mode parameter buffer before doing MODE SELECT.
202  	 */
203  	data.device_specific = 0;
204  
205  	ret = scsi_mode_select(sdp, 1, sp, buffer_data, len, SD_TIMEOUT,
206  			       sdkp->max_retries, &data, &sshdr);
207  	if (ret) {
208  		if (ret > 0 && scsi_sense_valid(&sshdr))
209  			sd_print_sense_hdr(sdkp, &sshdr);
210  		return -EINVAL;
211  	}
212  	sd_revalidate_disk(sdkp->disk);
213  	return count;
214  }
215  
216  static ssize_t
manage_start_stop_show(struct device * dev,struct device_attribute * attr,char * buf)217  manage_start_stop_show(struct device *dev,
218  		       struct device_attribute *attr, char *buf)
219  {
220  	struct scsi_disk *sdkp = to_scsi_disk(dev);
221  	struct scsi_device *sdp = sdkp->device;
222  
223  	return sysfs_emit(buf, "%u\n",
224  			  sdp->manage_system_start_stop &&
225  			  sdp->manage_runtime_start_stop &&
226  			  sdp->manage_shutdown);
227  }
228  static DEVICE_ATTR_RO(manage_start_stop);
229  
230  static ssize_t
manage_system_start_stop_show(struct device * dev,struct device_attribute * attr,char * buf)231  manage_system_start_stop_show(struct device *dev,
232  			      struct device_attribute *attr, char *buf)
233  {
234  	struct scsi_disk *sdkp = to_scsi_disk(dev);
235  	struct scsi_device *sdp = sdkp->device;
236  
237  	return sysfs_emit(buf, "%u\n", sdp->manage_system_start_stop);
238  }
239  
240  static ssize_t
manage_system_start_stop_store(struct device * dev,struct device_attribute * attr,const char * buf,size_t count)241  manage_system_start_stop_store(struct device *dev,
242  			       struct device_attribute *attr,
243  			       const char *buf, size_t count)
244  {
245  	struct scsi_disk *sdkp = to_scsi_disk(dev);
246  	struct scsi_device *sdp = sdkp->device;
247  	bool v;
248  
249  	if (!capable(CAP_SYS_ADMIN))
250  		return -EACCES;
251  
252  	if (kstrtobool(buf, &v))
253  		return -EINVAL;
254  
255  	sdp->manage_system_start_stop = v;
256  
257  	return count;
258  }
259  static DEVICE_ATTR_RW(manage_system_start_stop);
260  
261  static ssize_t
manage_runtime_start_stop_show(struct device * dev,struct device_attribute * attr,char * buf)262  manage_runtime_start_stop_show(struct device *dev,
263  			       struct device_attribute *attr, char *buf)
264  {
265  	struct scsi_disk *sdkp = to_scsi_disk(dev);
266  	struct scsi_device *sdp = sdkp->device;
267  
268  	return sysfs_emit(buf, "%u\n", sdp->manage_runtime_start_stop);
269  }
270  
271  static ssize_t
manage_runtime_start_stop_store(struct device * dev,struct device_attribute * attr,const char * buf,size_t count)272  manage_runtime_start_stop_store(struct device *dev,
273  				struct device_attribute *attr,
274  				const char *buf, size_t count)
275  {
276  	struct scsi_disk *sdkp = to_scsi_disk(dev);
277  	struct scsi_device *sdp = sdkp->device;
278  	bool v;
279  
280  	if (!capable(CAP_SYS_ADMIN))
281  		return -EACCES;
282  
283  	if (kstrtobool(buf, &v))
284  		return -EINVAL;
285  
286  	sdp->manage_runtime_start_stop = v;
287  
288  	return count;
289  }
290  static DEVICE_ATTR_RW(manage_runtime_start_stop);
291  
manage_shutdown_show(struct device * dev,struct device_attribute * attr,char * buf)292  static ssize_t manage_shutdown_show(struct device *dev,
293  				    struct device_attribute *attr, char *buf)
294  {
295  	struct scsi_disk *sdkp = to_scsi_disk(dev);
296  	struct scsi_device *sdp = sdkp->device;
297  
298  	return sysfs_emit(buf, "%u\n", sdp->manage_shutdown);
299  }
300  
manage_shutdown_store(struct device * dev,struct device_attribute * attr,const char * buf,size_t count)301  static ssize_t manage_shutdown_store(struct device *dev,
302  				     struct device_attribute *attr,
303  				     const char *buf, size_t count)
304  {
305  	struct scsi_disk *sdkp = to_scsi_disk(dev);
306  	struct scsi_device *sdp = sdkp->device;
307  	bool v;
308  
309  	if (!capable(CAP_SYS_ADMIN))
310  		return -EACCES;
311  
312  	if (kstrtobool(buf, &v))
313  		return -EINVAL;
314  
315  	sdp->manage_shutdown = v;
316  
317  	return count;
318  }
319  static DEVICE_ATTR_RW(manage_shutdown);
320  
321  static ssize_t
allow_restart_show(struct device * dev,struct device_attribute * attr,char * buf)322  allow_restart_show(struct device *dev, struct device_attribute *attr, char *buf)
323  {
324  	struct scsi_disk *sdkp = to_scsi_disk(dev);
325  
326  	return sprintf(buf, "%u\n", sdkp->device->allow_restart);
327  }
328  
329  static ssize_t
allow_restart_store(struct device * dev,struct device_attribute * attr,const char * buf,size_t count)330  allow_restart_store(struct device *dev, struct device_attribute *attr,
331  		    const char *buf, size_t count)
332  {
333  	bool v;
334  	struct scsi_disk *sdkp = to_scsi_disk(dev);
335  	struct scsi_device *sdp = sdkp->device;
336  
337  	if (!capable(CAP_SYS_ADMIN))
338  		return -EACCES;
339  
340  	if (sdp->type != TYPE_DISK && sdp->type != TYPE_ZBC)
341  		return -EINVAL;
342  
343  	if (kstrtobool(buf, &v))
344  		return -EINVAL;
345  
346  	sdp->allow_restart = v;
347  
348  	return count;
349  }
350  static DEVICE_ATTR_RW(allow_restart);
351  
352  static ssize_t
cache_type_show(struct device * dev,struct device_attribute * attr,char * buf)353  cache_type_show(struct device *dev, struct device_attribute *attr, char *buf)
354  {
355  	struct scsi_disk *sdkp = to_scsi_disk(dev);
356  	int ct = sdkp->RCD + 2*sdkp->WCE;
357  
358  	return sprintf(buf, "%s\n", sd_cache_types[ct]);
359  }
360  static DEVICE_ATTR_RW(cache_type);
361  
362  static ssize_t
FUA_show(struct device * dev,struct device_attribute * attr,char * buf)363  FUA_show(struct device *dev, struct device_attribute *attr, char *buf)
364  {
365  	struct scsi_disk *sdkp = to_scsi_disk(dev);
366  
367  	return sprintf(buf, "%u\n", sdkp->DPOFUA);
368  }
369  static DEVICE_ATTR_RO(FUA);
370  
371  static ssize_t
protection_type_show(struct device * dev,struct device_attribute * attr,char * buf)372  protection_type_show(struct device *dev, struct device_attribute *attr,
373  		     char *buf)
374  {
375  	struct scsi_disk *sdkp = to_scsi_disk(dev);
376  
377  	return sprintf(buf, "%u\n", sdkp->protection_type);
378  }
379  
380  static ssize_t
protection_type_store(struct device * dev,struct device_attribute * attr,const char * buf,size_t count)381  protection_type_store(struct device *dev, struct device_attribute *attr,
382  		      const char *buf, size_t count)
383  {
384  	struct scsi_disk *sdkp = to_scsi_disk(dev);
385  	unsigned int val;
386  	int err;
387  
388  	if (!capable(CAP_SYS_ADMIN))
389  		return -EACCES;
390  
391  	err = kstrtouint(buf, 10, &val);
392  
393  	if (err)
394  		return err;
395  
396  	if (val <= T10_PI_TYPE3_PROTECTION)
397  		sdkp->protection_type = val;
398  
399  	return count;
400  }
401  static DEVICE_ATTR_RW(protection_type);
402  
403  static ssize_t
protection_mode_show(struct device * dev,struct device_attribute * attr,char * buf)404  protection_mode_show(struct device *dev, struct device_attribute *attr,
405  		     char *buf)
406  {
407  	struct scsi_disk *sdkp = to_scsi_disk(dev);
408  	struct scsi_device *sdp = sdkp->device;
409  	unsigned int dif, dix;
410  
411  	dif = scsi_host_dif_capable(sdp->host, sdkp->protection_type);
412  	dix = scsi_host_dix_capable(sdp->host, sdkp->protection_type);
413  
414  	if (!dix && scsi_host_dix_capable(sdp->host, T10_PI_TYPE0_PROTECTION)) {
415  		dif = 0;
416  		dix = 1;
417  	}
418  
419  	if (!dif && !dix)
420  		return sprintf(buf, "none\n");
421  
422  	return sprintf(buf, "%s%u\n", dix ? "dix" : "dif", dif);
423  }
424  static DEVICE_ATTR_RO(protection_mode);
425  
426  static ssize_t
app_tag_own_show(struct device * dev,struct device_attribute * attr,char * buf)427  app_tag_own_show(struct device *dev, struct device_attribute *attr, char *buf)
428  {
429  	struct scsi_disk *sdkp = to_scsi_disk(dev);
430  
431  	return sprintf(buf, "%u\n", sdkp->ATO);
432  }
433  static DEVICE_ATTR_RO(app_tag_own);
434  
435  static ssize_t
thin_provisioning_show(struct device * dev,struct device_attribute * attr,char * buf)436  thin_provisioning_show(struct device *dev, struct device_attribute *attr,
437  		       char *buf)
438  {
439  	struct scsi_disk *sdkp = to_scsi_disk(dev);
440  
441  	return sprintf(buf, "%u\n", sdkp->lbpme);
442  }
443  static DEVICE_ATTR_RO(thin_provisioning);
444  
445  /* sysfs_match_string() requires dense arrays */
446  static const char *lbp_mode[] = {
447  	[SD_LBP_FULL]		= "full",
448  	[SD_LBP_UNMAP]		= "unmap",
449  	[SD_LBP_WS16]		= "writesame_16",
450  	[SD_LBP_WS10]		= "writesame_10",
451  	[SD_LBP_ZERO]		= "writesame_zero",
452  	[SD_LBP_DISABLE]	= "disabled",
453  };
454  
455  static ssize_t
provisioning_mode_show(struct device * dev,struct device_attribute * attr,char * buf)456  provisioning_mode_show(struct device *dev, struct device_attribute *attr,
457  		       char *buf)
458  {
459  	struct scsi_disk *sdkp = to_scsi_disk(dev);
460  
461  	return sprintf(buf, "%s\n", lbp_mode[sdkp->provisioning_mode]);
462  }
463  
464  static ssize_t
provisioning_mode_store(struct device * dev,struct device_attribute * attr,const char * buf,size_t count)465  provisioning_mode_store(struct device *dev, struct device_attribute *attr,
466  			const char *buf, size_t count)
467  {
468  	struct scsi_disk *sdkp = to_scsi_disk(dev);
469  	struct scsi_device *sdp = sdkp->device;
470  	struct queue_limits lim;
471  	int mode, err;
472  
473  	if (!capable(CAP_SYS_ADMIN))
474  		return -EACCES;
475  
476  	if (sdp->type != TYPE_DISK)
477  		return -EINVAL;
478  
479  	mode = sysfs_match_string(lbp_mode, buf);
480  	if (mode < 0)
481  		return -EINVAL;
482  
483  	lim = queue_limits_start_update(sdkp->disk->queue);
484  	sd_config_discard(sdkp, &lim, mode);
485  	err = queue_limits_commit_update_frozen(sdkp->disk->queue, &lim);
486  	if (err)
487  		return err;
488  	return count;
489  }
490  static DEVICE_ATTR_RW(provisioning_mode);
491  
492  /* sysfs_match_string() requires dense arrays */
493  static const char *zeroing_mode[] = {
494  	[SD_ZERO_WRITE]		= "write",
495  	[SD_ZERO_WS]		= "writesame",
496  	[SD_ZERO_WS16_UNMAP]	= "writesame_16_unmap",
497  	[SD_ZERO_WS10_UNMAP]	= "writesame_10_unmap",
498  };
499  
500  static ssize_t
zeroing_mode_show(struct device * dev,struct device_attribute * attr,char * buf)501  zeroing_mode_show(struct device *dev, struct device_attribute *attr,
502  		  char *buf)
503  {
504  	struct scsi_disk *sdkp = to_scsi_disk(dev);
505  
506  	return sprintf(buf, "%s\n", zeroing_mode[sdkp->zeroing_mode]);
507  }
508  
509  static ssize_t
zeroing_mode_store(struct device * dev,struct device_attribute * attr,const char * buf,size_t count)510  zeroing_mode_store(struct device *dev, struct device_attribute *attr,
511  		   const char *buf, size_t count)
512  {
513  	struct scsi_disk *sdkp = to_scsi_disk(dev);
514  	int mode;
515  
516  	if (!capable(CAP_SYS_ADMIN))
517  		return -EACCES;
518  
519  	mode = sysfs_match_string(zeroing_mode, buf);
520  	if (mode < 0)
521  		return -EINVAL;
522  
523  	sdkp->zeroing_mode = mode;
524  
525  	return count;
526  }
527  static DEVICE_ATTR_RW(zeroing_mode);
528  
529  static ssize_t
max_medium_access_timeouts_show(struct device * dev,struct device_attribute * attr,char * buf)530  max_medium_access_timeouts_show(struct device *dev,
531  				struct device_attribute *attr, char *buf)
532  {
533  	struct scsi_disk *sdkp = to_scsi_disk(dev);
534  
535  	return sprintf(buf, "%u\n", sdkp->max_medium_access_timeouts);
536  }
537  
538  static ssize_t
max_medium_access_timeouts_store(struct device * dev,struct device_attribute * attr,const char * buf,size_t count)539  max_medium_access_timeouts_store(struct device *dev,
540  				 struct device_attribute *attr, const char *buf,
541  				 size_t count)
542  {
543  	struct scsi_disk *sdkp = to_scsi_disk(dev);
544  	int err;
545  
546  	if (!capable(CAP_SYS_ADMIN))
547  		return -EACCES;
548  
549  	err = kstrtouint(buf, 10, &sdkp->max_medium_access_timeouts);
550  
551  	return err ? err : count;
552  }
553  static DEVICE_ATTR_RW(max_medium_access_timeouts);
554  
555  static ssize_t
max_write_same_blocks_show(struct device * dev,struct device_attribute * attr,char * buf)556  max_write_same_blocks_show(struct device *dev, struct device_attribute *attr,
557  			   char *buf)
558  {
559  	struct scsi_disk *sdkp = to_scsi_disk(dev);
560  
561  	return sprintf(buf, "%u\n", sdkp->max_ws_blocks);
562  }
563  
564  static ssize_t
max_write_same_blocks_store(struct device * dev,struct device_attribute * attr,const char * buf,size_t count)565  max_write_same_blocks_store(struct device *dev, struct device_attribute *attr,
566  			    const char *buf, size_t count)
567  {
568  	struct scsi_disk *sdkp = to_scsi_disk(dev);
569  	struct scsi_device *sdp = sdkp->device;
570  	struct queue_limits lim;
571  	unsigned long max;
572  	int err;
573  
574  	if (!capable(CAP_SYS_ADMIN))
575  		return -EACCES;
576  
577  	if (sdp->type != TYPE_DISK && sdp->type != TYPE_ZBC)
578  		return -EINVAL;
579  
580  	err = kstrtoul(buf, 10, &max);
581  
582  	if (err)
583  		return err;
584  
585  	if (max == 0)
586  		sdp->no_write_same = 1;
587  	else if (max <= SD_MAX_WS16_BLOCKS) {
588  		sdp->no_write_same = 0;
589  		sdkp->max_ws_blocks = max;
590  	}
591  
592  	lim = queue_limits_start_update(sdkp->disk->queue);
593  	sd_config_write_same(sdkp, &lim);
594  	err = queue_limits_commit_update_frozen(sdkp->disk->queue, &lim);
595  	if (err)
596  		return err;
597  	return count;
598  }
599  static DEVICE_ATTR_RW(max_write_same_blocks);
600  
601  static ssize_t
zoned_cap_show(struct device * dev,struct device_attribute * attr,char * buf)602  zoned_cap_show(struct device *dev, struct device_attribute *attr, char *buf)
603  {
604  	struct scsi_disk *sdkp = to_scsi_disk(dev);
605  
606  	if (sdkp->device->type == TYPE_ZBC)
607  		return sprintf(buf, "host-managed\n");
608  	if (sdkp->zoned == 1)
609  		return sprintf(buf, "host-aware\n");
610  	if (sdkp->zoned == 2)
611  		return sprintf(buf, "drive-managed\n");
612  	return sprintf(buf, "none\n");
613  }
614  static DEVICE_ATTR_RO(zoned_cap);
615  
616  static ssize_t
max_retries_store(struct device * dev,struct device_attribute * attr,const char * buf,size_t count)617  max_retries_store(struct device *dev, struct device_attribute *attr,
618  		  const char *buf, size_t count)
619  {
620  	struct scsi_disk *sdkp = to_scsi_disk(dev);
621  	struct scsi_device *sdev = sdkp->device;
622  	int retries, err;
623  
624  	err = kstrtoint(buf, 10, &retries);
625  	if (err)
626  		return err;
627  
628  	if (retries == SCSI_CMD_RETRIES_NO_LIMIT || retries <= SD_MAX_RETRIES) {
629  		sdkp->max_retries = retries;
630  		return count;
631  	}
632  
633  	sdev_printk(KERN_ERR, sdev, "max_retries must be between -1 and %d\n",
634  		    SD_MAX_RETRIES);
635  	return -EINVAL;
636  }
637  
638  static ssize_t
max_retries_show(struct device * dev,struct device_attribute * attr,char * buf)639  max_retries_show(struct device *dev, struct device_attribute *attr,
640  		 char *buf)
641  {
642  	struct scsi_disk *sdkp = to_scsi_disk(dev);
643  
644  	return sprintf(buf, "%d\n", sdkp->max_retries);
645  }
646  
647  static DEVICE_ATTR_RW(max_retries);
648  
649  static struct attribute *sd_disk_attrs[] = {
650  	&dev_attr_cache_type.attr,
651  	&dev_attr_FUA.attr,
652  	&dev_attr_allow_restart.attr,
653  	&dev_attr_manage_start_stop.attr,
654  	&dev_attr_manage_system_start_stop.attr,
655  	&dev_attr_manage_runtime_start_stop.attr,
656  	&dev_attr_manage_shutdown.attr,
657  	&dev_attr_protection_type.attr,
658  	&dev_attr_protection_mode.attr,
659  	&dev_attr_app_tag_own.attr,
660  	&dev_attr_thin_provisioning.attr,
661  	&dev_attr_provisioning_mode.attr,
662  	&dev_attr_zeroing_mode.attr,
663  	&dev_attr_max_write_same_blocks.attr,
664  	&dev_attr_max_medium_access_timeouts.attr,
665  	&dev_attr_zoned_cap.attr,
666  	&dev_attr_max_retries.attr,
667  	NULL,
668  };
669  ATTRIBUTE_GROUPS(sd_disk);
670  
671  static struct class sd_disk_class = {
672  	.name		= "scsi_disk",
673  	.dev_release	= scsi_disk_release,
674  	.dev_groups	= sd_disk_groups,
675  };
676  
677  /*
678   * Don't request a new module, as that could deadlock in multipath
679   * environment.
680   */
sd_default_probe(dev_t devt)681  static void sd_default_probe(dev_t devt)
682  {
683  }
684  
685  /*
686   * Device no to disk mapping:
687   *
688   *       major         disc2     disc  p1
689   *   |............|.............|....|....| <- dev_t
690   *    31        20 19          8 7  4 3  0
691   *
692   * Inside a major, we have 16k disks, however mapped non-
693   * contiguously. The first 16 disks are for major0, the next
694   * ones with major1, ... Disk 256 is for major0 again, disk 272
695   * for major1, ...
696   * As we stay compatible with our numbering scheme, we can reuse
697   * the well-know SCSI majors 8, 65--71, 136--143.
698   */
sd_major(int major_idx)699  static int sd_major(int major_idx)
700  {
701  	switch (major_idx) {
702  	case 0:
703  		return SCSI_DISK0_MAJOR;
704  	case 1 ... 7:
705  		return SCSI_DISK1_MAJOR + major_idx - 1;
706  	case 8 ... 15:
707  		return SCSI_DISK8_MAJOR + major_idx - 8;
708  	default:
709  		BUG();
710  		return 0;	/* shut up gcc */
711  	}
712  }
713  
714  #ifdef CONFIG_BLK_SED_OPAL
sd_sec_submit(void * data,u16 spsp,u8 secp,void * buffer,size_t len,bool send)715  static int sd_sec_submit(void *data, u16 spsp, u8 secp, void *buffer,
716  		size_t len, bool send)
717  {
718  	struct scsi_disk *sdkp = data;
719  	struct scsi_device *sdev = sdkp->device;
720  	u8 cdb[12] = { 0, };
721  	const struct scsi_exec_args exec_args = {
722  		.req_flags = BLK_MQ_REQ_PM,
723  	};
724  	int ret;
725  
726  	cdb[0] = send ? SECURITY_PROTOCOL_OUT : SECURITY_PROTOCOL_IN;
727  	cdb[1] = secp;
728  	put_unaligned_be16(spsp, &cdb[2]);
729  	put_unaligned_be32(len, &cdb[6]);
730  
731  	ret = scsi_execute_cmd(sdev, cdb, send ? REQ_OP_DRV_OUT : REQ_OP_DRV_IN,
732  			       buffer, len, SD_TIMEOUT, sdkp->max_retries,
733  			       &exec_args);
734  	return ret <= 0 ? ret : -EIO;
735  }
736  #endif /* CONFIG_BLK_SED_OPAL */
737  
738  /*
739   * Look up the DIX operation based on whether the command is read or
740   * write and whether dix and dif are enabled.
741   */
sd_prot_op(bool write,bool dix,bool dif)742  static unsigned int sd_prot_op(bool write, bool dix, bool dif)
743  {
744  	/* Lookup table: bit 2 (write), bit 1 (dix), bit 0 (dif) */
745  	static const unsigned int ops[] = {	/* wrt dix dif */
746  		SCSI_PROT_NORMAL,		/*  0	0   0  */
747  		SCSI_PROT_READ_STRIP,		/*  0	0   1  */
748  		SCSI_PROT_READ_INSERT,		/*  0	1   0  */
749  		SCSI_PROT_READ_PASS,		/*  0	1   1  */
750  		SCSI_PROT_NORMAL,		/*  1	0   0  */
751  		SCSI_PROT_WRITE_INSERT,		/*  1	0   1  */
752  		SCSI_PROT_WRITE_STRIP,		/*  1	1   0  */
753  		SCSI_PROT_WRITE_PASS,		/*  1	1   1  */
754  	};
755  
756  	return ops[write << 2 | dix << 1 | dif];
757  }
758  
759  /*
760   * Returns a mask of the protection flags that are valid for a given DIX
761   * operation.
762   */
sd_prot_flag_mask(unsigned int prot_op)763  static unsigned int sd_prot_flag_mask(unsigned int prot_op)
764  {
765  	static const unsigned int flag_mask[] = {
766  		[SCSI_PROT_NORMAL]		= 0,
767  
768  		[SCSI_PROT_READ_STRIP]		= SCSI_PROT_TRANSFER_PI |
769  						  SCSI_PROT_GUARD_CHECK |
770  						  SCSI_PROT_REF_CHECK |
771  						  SCSI_PROT_REF_INCREMENT,
772  
773  		[SCSI_PROT_READ_INSERT]		= SCSI_PROT_REF_INCREMENT |
774  						  SCSI_PROT_IP_CHECKSUM,
775  
776  		[SCSI_PROT_READ_PASS]		= SCSI_PROT_TRANSFER_PI |
777  						  SCSI_PROT_GUARD_CHECK |
778  						  SCSI_PROT_REF_CHECK |
779  						  SCSI_PROT_REF_INCREMENT |
780  						  SCSI_PROT_IP_CHECKSUM,
781  
782  		[SCSI_PROT_WRITE_INSERT]	= SCSI_PROT_TRANSFER_PI |
783  						  SCSI_PROT_REF_INCREMENT,
784  
785  		[SCSI_PROT_WRITE_STRIP]		= SCSI_PROT_GUARD_CHECK |
786  						  SCSI_PROT_REF_CHECK |
787  						  SCSI_PROT_REF_INCREMENT |
788  						  SCSI_PROT_IP_CHECKSUM,
789  
790  		[SCSI_PROT_WRITE_PASS]		= SCSI_PROT_TRANSFER_PI |
791  						  SCSI_PROT_GUARD_CHECK |
792  						  SCSI_PROT_REF_CHECK |
793  						  SCSI_PROT_REF_INCREMENT |
794  						  SCSI_PROT_IP_CHECKSUM,
795  	};
796  
797  	return flag_mask[prot_op];
798  }
799  
sd_setup_protect_cmnd(struct scsi_cmnd * scmd,unsigned int dix,unsigned int dif)800  static unsigned char sd_setup_protect_cmnd(struct scsi_cmnd *scmd,
801  					   unsigned int dix, unsigned int dif)
802  {
803  	struct request *rq = scsi_cmd_to_rq(scmd);
804  	struct bio *bio = rq->bio;
805  	unsigned int prot_op = sd_prot_op(rq_data_dir(rq), dix, dif);
806  	unsigned int protect = 0;
807  
808  	if (dix) {				/* DIX Type 0, 1, 2, 3 */
809  		if (bio_integrity_flagged(bio, BIP_IP_CHECKSUM))
810  			scmd->prot_flags |= SCSI_PROT_IP_CHECKSUM;
811  
812  		if (bio_integrity_flagged(bio, BIP_CHECK_GUARD))
813  			scmd->prot_flags |= SCSI_PROT_GUARD_CHECK;
814  	}
815  
816  	if (dif != T10_PI_TYPE3_PROTECTION) {	/* DIX/DIF Type 0, 1, 2 */
817  		scmd->prot_flags |= SCSI_PROT_REF_INCREMENT;
818  
819  		if (bio_integrity_flagged(bio, BIP_CHECK_REFTAG))
820  			scmd->prot_flags |= SCSI_PROT_REF_CHECK;
821  	}
822  
823  	if (dif) {				/* DIX/DIF Type 1, 2, 3 */
824  		scmd->prot_flags |= SCSI_PROT_TRANSFER_PI;
825  
826  		if (bio_integrity_flagged(bio, BIP_DISK_NOCHECK))
827  			protect = 3 << 5;	/* Disable target PI checking */
828  		else
829  			protect = 1 << 5;	/* Enable target PI checking */
830  	}
831  
832  	scsi_set_prot_op(scmd, prot_op);
833  	scsi_set_prot_type(scmd, dif);
834  	scmd->prot_flags &= sd_prot_flag_mask(prot_op);
835  
836  	return protect;
837  }
838  
sd_disable_discard(struct scsi_disk * sdkp)839  static void sd_disable_discard(struct scsi_disk *sdkp)
840  {
841  	sdkp->provisioning_mode = SD_LBP_DISABLE;
842  	blk_queue_disable_discard(sdkp->disk->queue);
843  }
844  
sd_config_discard(struct scsi_disk * sdkp,struct queue_limits * lim,unsigned int mode)845  static void sd_config_discard(struct scsi_disk *sdkp, struct queue_limits *lim,
846  		unsigned int mode)
847  {
848  	unsigned int logical_block_size = sdkp->device->sector_size;
849  	unsigned int max_blocks = 0;
850  
851  	lim->discard_alignment = sdkp->unmap_alignment * logical_block_size;
852  	lim->discard_granularity = max(sdkp->physical_block_size,
853  			sdkp->unmap_granularity * logical_block_size);
854  	sdkp->provisioning_mode = mode;
855  
856  	switch (mode) {
857  
858  	case SD_LBP_FULL:
859  	case SD_LBP_DISABLE:
860  		break;
861  
862  	case SD_LBP_UNMAP:
863  		max_blocks = min_not_zero(sdkp->max_unmap_blocks,
864  					  (u32)SD_MAX_WS16_BLOCKS);
865  		break;
866  
867  	case SD_LBP_WS16:
868  		if (sdkp->device->unmap_limit_for_ws)
869  			max_blocks = sdkp->max_unmap_blocks;
870  		else
871  			max_blocks = sdkp->max_ws_blocks;
872  
873  		max_blocks = min_not_zero(max_blocks, (u32)SD_MAX_WS16_BLOCKS);
874  		break;
875  
876  	case SD_LBP_WS10:
877  		if (sdkp->device->unmap_limit_for_ws)
878  			max_blocks = sdkp->max_unmap_blocks;
879  		else
880  			max_blocks = sdkp->max_ws_blocks;
881  
882  		max_blocks = min_not_zero(max_blocks, (u32)SD_MAX_WS10_BLOCKS);
883  		break;
884  
885  	case SD_LBP_ZERO:
886  		max_blocks = min_not_zero(sdkp->max_ws_blocks,
887  					  (u32)SD_MAX_WS10_BLOCKS);
888  		break;
889  	}
890  
891  	lim->max_hw_discard_sectors = max_blocks *
892  		(logical_block_size >> SECTOR_SHIFT);
893  }
894  
sd_set_special_bvec(struct request * rq,unsigned int data_len)895  static void *sd_set_special_bvec(struct request *rq, unsigned int data_len)
896  {
897  	struct page *page;
898  
899  	page = mempool_alloc(sd_page_pool, GFP_ATOMIC);
900  	if (!page)
901  		return NULL;
902  	clear_highpage(page);
903  	bvec_set_page(&rq->special_vec, page, data_len, 0);
904  	rq->rq_flags |= RQF_SPECIAL_PAYLOAD;
905  	return bvec_virt(&rq->special_vec);
906  }
907  
sd_setup_unmap_cmnd(struct scsi_cmnd * cmd)908  static blk_status_t sd_setup_unmap_cmnd(struct scsi_cmnd *cmd)
909  {
910  	struct scsi_device *sdp = cmd->device;
911  	struct request *rq = scsi_cmd_to_rq(cmd);
912  	struct scsi_disk *sdkp = scsi_disk(rq->q->disk);
913  	u64 lba = sectors_to_logical(sdp, blk_rq_pos(rq));
914  	u32 nr_blocks = sectors_to_logical(sdp, blk_rq_sectors(rq));
915  	unsigned int data_len = 24;
916  	char *buf;
917  
918  	buf = sd_set_special_bvec(rq, data_len);
919  	if (!buf)
920  		return BLK_STS_RESOURCE;
921  
922  	cmd->cmd_len = 10;
923  	cmd->cmnd[0] = UNMAP;
924  	cmd->cmnd[8] = 24;
925  
926  	put_unaligned_be16(6 + 16, &buf[0]);
927  	put_unaligned_be16(16, &buf[2]);
928  	put_unaligned_be64(lba, &buf[8]);
929  	put_unaligned_be32(nr_blocks, &buf[16]);
930  
931  	cmd->allowed = sdkp->max_retries;
932  	cmd->transfersize = data_len;
933  	rq->timeout = SD_TIMEOUT;
934  
935  	return scsi_alloc_sgtables(cmd);
936  }
937  
sd_config_atomic(struct scsi_disk * sdkp,struct queue_limits * lim)938  static void sd_config_atomic(struct scsi_disk *sdkp, struct queue_limits *lim)
939  {
940  	unsigned int logical_block_size = sdkp->device->sector_size,
941  		physical_block_size_sectors, max_atomic, unit_min, unit_max;
942  
943  	if ((!sdkp->max_atomic && !sdkp->max_atomic_with_boundary) ||
944  	    sdkp->protection_type == T10_PI_TYPE2_PROTECTION)
945  		return;
946  
947  	physical_block_size_sectors = sdkp->physical_block_size /
948  					sdkp->device->sector_size;
949  
950  	unit_min = rounddown_pow_of_two(sdkp->atomic_granularity ?
951  					sdkp->atomic_granularity :
952  					physical_block_size_sectors);
953  
954  	/*
955  	 * Only use atomic boundary when we have the odd scenario of
956  	 * sdkp->max_atomic == 0, which the spec does permit.
957  	 */
958  	if (sdkp->max_atomic) {
959  		max_atomic = sdkp->max_atomic;
960  		unit_max = rounddown_pow_of_two(sdkp->max_atomic);
961  		sdkp->use_atomic_write_boundary = 0;
962  	} else {
963  		max_atomic = sdkp->max_atomic_with_boundary;
964  		unit_max = rounddown_pow_of_two(sdkp->max_atomic_boundary);
965  		sdkp->use_atomic_write_boundary = 1;
966  	}
967  
968  	/*
969  	 * Ensure compliance with granularity and alignment. For now, keep it
970  	 * simple and just don't support atomic writes for values mismatched
971  	 * with max_{boundary}atomic, physical block size, and
972  	 * atomic_granularity itself.
973  	 *
974  	 * We're really being distrustful by checking unit_max also...
975  	 */
976  	if (sdkp->atomic_granularity > 1) {
977  		if (unit_min > 1 && unit_min % sdkp->atomic_granularity)
978  			return;
979  		if (unit_max > 1 && unit_max % sdkp->atomic_granularity)
980  			return;
981  	}
982  
983  	if (sdkp->atomic_alignment > 1) {
984  		if (unit_min > 1 && unit_min % sdkp->atomic_alignment)
985  			return;
986  		if (unit_max > 1 && unit_max % sdkp->atomic_alignment)
987  			return;
988  	}
989  
990  	lim->atomic_write_hw_max = max_atomic * logical_block_size;
991  	lim->atomic_write_hw_boundary = 0;
992  	lim->atomic_write_hw_unit_min = unit_min * logical_block_size;
993  	lim->atomic_write_hw_unit_max = unit_max * logical_block_size;
994  	lim->features |= BLK_FEAT_ATOMIC_WRITES;
995  }
996  
sd_setup_write_same16_cmnd(struct scsi_cmnd * cmd,bool unmap)997  static blk_status_t sd_setup_write_same16_cmnd(struct scsi_cmnd *cmd,
998  		bool unmap)
999  {
1000  	struct scsi_device *sdp = cmd->device;
1001  	struct request *rq = scsi_cmd_to_rq(cmd);
1002  	struct scsi_disk *sdkp = scsi_disk(rq->q->disk);
1003  	u64 lba = sectors_to_logical(sdp, blk_rq_pos(rq));
1004  	u32 nr_blocks = sectors_to_logical(sdp, blk_rq_sectors(rq));
1005  	u32 data_len = sdp->sector_size;
1006  
1007  	if (!sd_set_special_bvec(rq, data_len))
1008  		return BLK_STS_RESOURCE;
1009  
1010  	cmd->cmd_len = 16;
1011  	cmd->cmnd[0] = WRITE_SAME_16;
1012  	if (unmap)
1013  		cmd->cmnd[1] = 0x8; /* UNMAP */
1014  	put_unaligned_be64(lba, &cmd->cmnd[2]);
1015  	put_unaligned_be32(nr_blocks, &cmd->cmnd[10]);
1016  
1017  	cmd->allowed = sdkp->max_retries;
1018  	cmd->transfersize = data_len;
1019  	rq->timeout = unmap ? SD_TIMEOUT : SD_WRITE_SAME_TIMEOUT;
1020  
1021  	return scsi_alloc_sgtables(cmd);
1022  }
1023  
sd_setup_write_same10_cmnd(struct scsi_cmnd * cmd,bool unmap)1024  static blk_status_t sd_setup_write_same10_cmnd(struct scsi_cmnd *cmd,
1025  		bool unmap)
1026  {
1027  	struct scsi_device *sdp = cmd->device;
1028  	struct request *rq = scsi_cmd_to_rq(cmd);
1029  	struct scsi_disk *sdkp = scsi_disk(rq->q->disk);
1030  	u64 lba = sectors_to_logical(sdp, blk_rq_pos(rq));
1031  	u32 nr_blocks = sectors_to_logical(sdp, blk_rq_sectors(rq));
1032  	u32 data_len = sdp->sector_size;
1033  
1034  	if (!sd_set_special_bvec(rq, data_len))
1035  		return BLK_STS_RESOURCE;
1036  
1037  	cmd->cmd_len = 10;
1038  	cmd->cmnd[0] = WRITE_SAME;
1039  	if (unmap)
1040  		cmd->cmnd[1] = 0x8; /* UNMAP */
1041  	put_unaligned_be32(lba, &cmd->cmnd[2]);
1042  	put_unaligned_be16(nr_blocks, &cmd->cmnd[7]);
1043  
1044  	cmd->allowed = sdkp->max_retries;
1045  	cmd->transfersize = data_len;
1046  	rq->timeout = unmap ? SD_TIMEOUT : SD_WRITE_SAME_TIMEOUT;
1047  
1048  	return scsi_alloc_sgtables(cmd);
1049  }
1050  
sd_setup_write_zeroes_cmnd(struct scsi_cmnd * cmd)1051  static blk_status_t sd_setup_write_zeroes_cmnd(struct scsi_cmnd *cmd)
1052  {
1053  	struct request *rq = scsi_cmd_to_rq(cmd);
1054  	struct scsi_device *sdp = cmd->device;
1055  	struct scsi_disk *sdkp = scsi_disk(rq->q->disk);
1056  	u64 lba = sectors_to_logical(sdp, blk_rq_pos(rq));
1057  	u32 nr_blocks = sectors_to_logical(sdp, blk_rq_sectors(rq));
1058  
1059  	if (!(rq->cmd_flags & REQ_NOUNMAP)) {
1060  		switch (sdkp->zeroing_mode) {
1061  		case SD_ZERO_WS16_UNMAP:
1062  			return sd_setup_write_same16_cmnd(cmd, true);
1063  		case SD_ZERO_WS10_UNMAP:
1064  			return sd_setup_write_same10_cmnd(cmd, true);
1065  		}
1066  	}
1067  
1068  	if (sdp->no_write_same) {
1069  		rq->rq_flags |= RQF_QUIET;
1070  		return BLK_STS_TARGET;
1071  	}
1072  
1073  	if (sdkp->ws16 || lba > 0xffffffff || nr_blocks > 0xffff)
1074  		return sd_setup_write_same16_cmnd(cmd, false);
1075  
1076  	return sd_setup_write_same10_cmnd(cmd, false);
1077  }
1078  
sd_disable_write_same(struct scsi_disk * sdkp)1079  static void sd_disable_write_same(struct scsi_disk *sdkp)
1080  {
1081  	sdkp->device->no_write_same = 1;
1082  	sdkp->max_ws_blocks = 0;
1083  	blk_queue_disable_write_zeroes(sdkp->disk->queue);
1084  }
1085  
sd_config_write_same(struct scsi_disk * sdkp,struct queue_limits * lim)1086  static void sd_config_write_same(struct scsi_disk *sdkp,
1087  		struct queue_limits *lim)
1088  {
1089  	unsigned int logical_block_size = sdkp->device->sector_size;
1090  
1091  	if (sdkp->device->no_write_same) {
1092  		sdkp->max_ws_blocks = 0;
1093  		goto out;
1094  	}
1095  
1096  	/* Some devices can not handle block counts above 0xffff despite
1097  	 * supporting WRITE SAME(16). Consequently we default to 64k
1098  	 * blocks per I/O unless the device explicitly advertises a
1099  	 * bigger limit.
1100  	 */
1101  	if (sdkp->max_ws_blocks > SD_MAX_WS10_BLOCKS)
1102  		sdkp->max_ws_blocks = min_not_zero(sdkp->max_ws_blocks,
1103  						   (u32)SD_MAX_WS16_BLOCKS);
1104  	else if (sdkp->ws16 || sdkp->ws10 || sdkp->device->no_report_opcodes)
1105  		sdkp->max_ws_blocks = min_not_zero(sdkp->max_ws_blocks,
1106  						   (u32)SD_MAX_WS10_BLOCKS);
1107  	else {
1108  		sdkp->device->no_write_same = 1;
1109  		sdkp->max_ws_blocks = 0;
1110  	}
1111  
1112  	if (sdkp->lbprz && sdkp->lbpws)
1113  		sdkp->zeroing_mode = SD_ZERO_WS16_UNMAP;
1114  	else if (sdkp->lbprz && sdkp->lbpws10)
1115  		sdkp->zeroing_mode = SD_ZERO_WS10_UNMAP;
1116  	else if (sdkp->max_ws_blocks)
1117  		sdkp->zeroing_mode = SD_ZERO_WS;
1118  	else
1119  		sdkp->zeroing_mode = SD_ZERO_WRITE;
1120  
1121  	if (sdkp->max_ws_blocks &&
1122  	    sdkp->physical_block_size > logical_block_size) {
1123  		/*
1124  		 * Reporting a maximum number of blocks that is not aligned
1125  		 * on the device physical size would cause a large write same
1126  		 * request to be split into physically unaligned chunks by
1127  		 * __blkdev_issue_write_zeroes() even if the caller of this
1128  		 * functions took care to align the large request. So make sure
1129  		 * the maximum reported is aligned to the device physical block
1130  		 * size. This is only an optional optimization for regular
1131  		 * disks, but this is mandatory to avoid failure of large write
1132  		 * same requests directed at sequential write required zones of
1133  		 * host-managed ZBC disks.
1134  		 */
1135  		sdkp->max_ws_blocks =
1136  			round_down(sdkp->max_ws_blocks,
1137  				   bytes_to_logical(sdkp->device,
1138  						    sdkp->physical_block_size));
1139  	}
1140  
1141  out:
1142  	lim->max_write_zeroes_sectors =
1143  		sdkp->max_ws_blocks * (logical_block_size >> SECTOR_SHIFT);
1144  
1145  	if (sdkp->zeroing_mode == SD_ZERO_WS16_UNMAP ||
1146  	    sdkp->zeroing_mode == SD_ZERO_WS10_UNMAP)
1147  		lim->max_hw_wzeroes_unmap_sectors =
1148  				lim->max_write_zeroes_sectors;
1149  }
1150  
sd_setup_flush_cmnd(struct scsi_cmnd * cmd)1151  static blk_status_t sd_setup_flush_cmnd(struct scsi_cmnd *cmd)
1152  {
1153  	struct request *rq = scsi_cmd_to_rq(cmd);
1154  	struct scsi_disk *sdkp = scsi_disk(rq->q->disk);
1155  
1156  	/* flush requests don't perform I/O, zero the S/G table */
1157  	memset(&cmd->sdb, 0, sizeof(cmd->sdb));
1158  
1159  	if (cmd->device->use_16_for_sync) {
1160  		cmd->cmnd[0] = SYNCHRONIZE_CACHE_16;
1161  		cmd->cmd_len = 16;
1162  	} else {
1163  		cmd->cmnd[0] = SYNCHRONIZE_CACHE;
1164  		cmd->cmd_len = 10;
1165  	}
1166  	cmd->transfersize = 0;
1167  	cmd->allowed = sdkp->max_retries;
1168  
1169  	rq->timeout = rq->q->rq_timeout * SD_FLUSH_TIMEOUT_MULTIPLIER;
1170  	return BLK_STS_OK;
1171  }
1172  
1173  /**
1174   * sd_group_number() - Compute the GROUP NUMBER field
1175   * @cmd: SCSI command for which to compute the value of the six-bit GROUP NUMBER
1176   *	field.
1177   *
1178   * From SBC-5 r05 (https://www.t10.org/cgi-bin/ac.pl?t=f&f=sbc5r05.pdf):
1179   * 0: no relative lifetime.
1180   * 1: shortest relative lifetime.
1181   * 2: second shortest relative lifetime.
1182   * 3 - 0x3d: intermediate relative lifetimes.
1183   * 0x3e: second longest relative lifetime.
1184   * 0x3f: longest relative lifetime.
1185   */
sd_group_number(struct scsi_cmnd * cmd)1186  static u8 sd_group_number(struct scsi_cmnd *cmd)
1187  {
1188  	const struct request *rq = scsi_cmd_to_rq(cmd);
1189  	struct scsi_disk *sdkp = scsi_disk(rq->q->disk);
1190  
1191  	if (!sdkp->rscs)
1192  		return 0;
1193  
1194  	return min3((u32)rq->bio->bi_write_hint,
1195  		    (u32)sdkp->permanent_stream_count, 0x3fu);
1196  }
1197  
sd_setup_rw32_cmnd(struct scsi_cmnd * cmd,bool write,sector_t lba,unsigned int nr_blocks,unsigned char flags,unsigned int dld)1198  static blk_status_t sd_setup_rw32_cmnd(struct scsi_cmnd *cmd, bool write,
1199  				       sector_t lba, unsigned int nr_blocks,
1200  				       unsigned char flags, unsigned int dld)
1201  {
1202  	cmd->cmd_len = SD_EXT_CDB_SIZE;
1203  	cmd->cmnd[0]  = VARIABLE_LENGTH_CMD;
1204  	cmd->cmnd[6]  = sd_group_number(cmd);
1205  	cmd->cmnd[7]  = 0x18; /* Additional CDB len */
1206  	cmd->cmnd[9]  = write ? WRITE_32 : READ_32;
1207  	cmd->cmnd[10] = flags;
1208  	cmd->cmnd[11] = dld & 0x07;
1209  	put_unaligned_be64(lba, &cmd->cmnd[12]);
1210  	put_unaligned_be32(lba, &cmd->cmnd[20]); /* Expected Indirect LBA */
1211  	put_unaligned_be32(nr_blocks, &cmd->cmnd[28]);
1212  
1213  	return BLK_STS_OK;
1214  }
1215  
sd_setup_rw16_cmnd(struct scsi_cmnd * cmd,bool write,sector_t lba,unsigned int nr_blocks,unsigned char flags,unsigned int dld)1216  static blk_status_t sd_setup_rw16_cmnd(struct scsi_cmnd *cmd, bool write,
1217  				       sector_t lba, unsigned int nr_blocks,
1218  				       unsigned char flags, unsigned int dld)
1219  {
1220  	cmd->cmd_len  = 16;
1221  	cmd->cmnd[0]  = write ? WRITE_16 : READ_16;
1222  	cmd->cmnd[1]  = flags | ((dld >> 2) & 0x01);
1223  	cmd->cmnd[14] = ((dld & 0x03) << 6) | sd_group_number(cmd);
1224  	cmd->cmnd[15] = 0;
1225  	put_unaligned_be64(lba, &cmd->cmnd[2]);
1226  	put_unaligned_be32(nr_blocks, &cmd->cmnd[10]);
1227  
1228  	return BLK_STS_OK;
1229  }
1230  
sd_setup_rw10_cmnd(struct scsi_cmnd * cmd,bool write,sector_t lba,unsigned int nr_blocks,unsigned char flags)1231  static blk_status_t sd_setup_rw10_cmnd(struct scsi_cmnd *cmd, bool write,
1232  				       sector_t lba, unsigned int nr_blocks,
1233  				       unsigned char flags)
1234  {
1235  	cmd->cmd_len = 10;
1236  	cmd->cmnd[0] = write ? WRITE_10 : READ_10;
1237  	cmd->cmnd[1] = flags;
1238  	cmd->cmnd[6] = sd_group_number(cmd);
1239  	cmd->cmnd[9] = 0;
1240  	put_unaligned_be32(lba, &cmd->cmnd[2]);
1241  	put_unaligned_be16(nr_blocks, &cmd->cmnd[7]);
1242  
1243  	return BLK_STS_OK;
1244  }
1245  
sd_setup_rw6_cmnd(struct scsi_cmnd * cmd,bool write,sector_t lba,unsigned int nr_blocks,unsigned char flags)1246  static blk_status_t sd_setup_rw6_cmnd(struct scsi_cmnd *cmd, bool write,
1247  				      sector_t lba, unsigned int nr_blocks,
1248  				      unsigned char flags)
1249  {
1250  	/* Avoid that 0 blocks gets translated into 256 blocks. */
1251  	if (WARN_ON_ONCE(nr_blocks == 0))
1252  		return BLK_STS_IOERR;
1253  
1254  	if (unlikely(flags & 0x8)) {
1255  		/*
1256  		 * This happens only if this drive failed 10byte rw
1257  		 * command with ILLEGAL_REQUEST during operation and
1258  		 * thus turned off use_10_for_rw.
1259  		 */
1260  		scmd_printk(KERN_ERR, cmd, "FUA write on READ/WRITE(6) drive\n");
1261  		return BLK_STS_IOERR;
1262  	}
1263  
1264  	cmd->cmd_len = 6;
1265  	cmd->cmnd[0] = write ? WRITE_6 : READ_6;
1266  	cmd->cmnd[1] = (lba >> 16) & 0x1f;
1267  	cmd->cmnd[2] = (lba >> 8) & 0xff;
1268  	cmd->cmnd[3] = lba & 0xff;
1269  	cmd->cmnd[4] = nr_blocks;
1270  	cmd->cmnd[5] = 0;
1271  
1272  	return BLK_STS_OK;
1273  }
1274  
1275  /*
1276   * Check if a command has a duration limit set. If it does, and the target
1277   * device supports CDL and the feature is enabled, return the limit
1278   * descriptor index to use. Return 0 (no limit) otherwise.
1279   */
sd_cdl_dld(struct scsi_disk * sdkp,struct scsi_cmnd * scmd)1280  static int sd_cdl_dld(struct scsi_disk *sdkp, struct scsi_cmnd *scmd)
1281  {
1282  	struct scsi_device *sdp = sdkp->device;
1283  	int hint;
1284  
1285  	if (!sdp->cdl_supported || !sdp->cdl_enable)
1286  		return 0;
1287  
1288  	/*
1289  	 * Use "no limit" if the request ioprio does not specify a duration
1290  	 * limit hint.
1291  	 */
1292  	hint = IOPRIO_PRIO_HINT(req_get_ioprio(scsi_cmd_to_rq(scmd)));
1293  	if (hint < IOPRIO_HINT_DEV_DURATION_LIMIT_1 ||
1294  	    hint > IOPRIO_HINT_DEV_DURATION_LIMIT_7)
1295  		return 0;
1296  
1297  	return (hint - IOPRIO_HINT_DEV_DURATION_LIMIT_1) + 1;
1298  }
1299  
sd_setup_atomic_cmnd(struct scsi_cmnd * cmd,sector_t lba,unsigned int nr_blocks,bool boundary,unsigned char flags)1300  static blk_status_t sd_setup_atomic_cmnd(struct scsi_cmnd *cmd,
1301  					sector_t lba, unsigned int nr_blocks,
1302  					bool boundary, unsigned char flags)
1303  {
1304  	cmd->cmd_len  = 16;
1305  	cmd->cmnd[0]  = WRITE_ATOMIC_16;
1306  	cmd->cmnd[1]  = flags;
1307  	put_unaligned_be64(lba, &cmd->cmnd[2]);
1308  	put_unaligned_be16(nr_blocks, &cmd->cmnd[12]);
1309  	if (boundary)
1310  		put_unaligned_be16(nr_blocks, &cmd->cmnd[10]);
1311  	else
1312  		put_unaligned_be16(0, &cmd->cmnd[10]);
1313  	put_unaligned_be16(nr_blocks, &cmd->cmnd[12]);
1314  	cmd->cmnd[14] = 0;
1315  	cmd->cmnd[15] = 0;
1316  
1317  	return BLK_STS_OK;
1318  }
1319  
sd_setup_read_write_cmnd(struct scsi_cmnd * cmd)1320  static blk_status_t sd_setup_read_write_cmnd(struct scsi_cmnd *cmd)
1321  {
1322  	struct request *rq = scsi_cmd_to_rq(cmd);
1323  	struct scsi_device *sdp = cmd->device;
1324  	struct scsi_disk *sdkp = scsi_disk(rq->q->disk);
1325  	sector_t lba = sectors_to_logical(sdp, blk_rq_pos(rq));
1326  	sector_t threshold;
1327  	unsigned int nr_blocks = sectors_to_logical(sdp, blk_rq_sectors(rq));
1328  	unsigned int mask = logical_to_sectors(sdp, 1) - 1;
1329  	bool write = rq_data_dir(rq) == WRITE;
1330  	unsigned char protect, fua;
1331  	unsigned int dld;
1332  	blk_status_t ret;
1333  	unsigned int dif;
1334  	bool dix;
1335  
1336  	ret = scsi_alloc_sgtables(cmd);
1337  	if (ret != BLK_STS_OK)
1338  		return ret;
1339  
1340  	ret = BLK_STS_IOERR;
1341  	if (!scsi_device_online(sdp) || sdp->changed) {
1342  		scmd_printk(KERN_ERR, cmd, "device offline or changed\n");
1343  		goto fail;
1344  	}
1345  
1346  	if (blk_rq_pos(rq) + blk_rq_sectors(rq) > get_capacity(rq->q->disk)) {
1347  		scmd_printk(KERN_ERR, cmd, "access beyond end of device\n");
1348  		goto fail;
1349  	}
1350  
1351  	if ((blk_rq_pos(rq) & mask) || (blk_rq_sectors(rq) & mask)) {
1352  		scmd_printk(KERN_ERR, cmd, "request not aligned to the logical block size\n");
1353  		goto fail;
1354  	}
1355  
1356  	/*
1357  	 * Some SD card readers can't handle accesses which touch the
1358  	 * last one or two logical blocks. Split accesses as needed.
1359  	 */
1360  	threshold = sdkp->capacity - SD_LAST_BUGGY_SECTORS;
1361  
1362  	if (unlikely(sdp->last_sector_bug && lba + nr_blocks > threshold)) {
1363  		if (lba < threshold) {
1364  			/* Access up to the threshold but not beyond */
1365  			nr_blocks = threshold - lba;
1366  		} else {
1367  			/* Access only a single logical block */
1368  			nr_blocks = 1;
1369  		}
1370  	}
1371  
1372  	fua = rq->cmd_flags & REQ_FUA ? 0x8 : 0;
1373  	dix = scsi_prot_sg_count(cmd);
1374  	dif = scsi_host_dif_capable(cmd->device->host, sdkp->protection_type);
1375  	dld = sd_cdl_dld(sdkp, cmd);
1376  
1377  	if (dif || dix)
1378  		protect = sd_setup_protect_cmnd(cmd, dix, dif);
1379  	else
1380  		protect = 0;
1381  
1382  	if (protect && sdkp->protection_type == T10_PI_TYPE2_PROTECTION) {
1383  		ret = sd_setup_rw32_cmnd(cmd, write, lba, nr_blocks,
1384  					 protect | fua, dld);
1385  	} else if (rq->cmd_flags & REQ_ATOMIC) {
1386  		ret = sd_setup_atomic_cmnd(cmd, lba, nr_blocks,
1387  				sdkp->use_atomic_write_boundary,
1388  				protect | fua);
1389  	} else if (sdp->use_16_for_rw || (nr_blocks > 0xffff)) {
1390  		ret = sd_setup_rw16_cmnd(cmd, write, lba, nr_blocks,
1391  					 protect | fua, dld);
1392  	} else if ((nr_blocks > 0xff) || (lba > 0x1fffff) ||
1393  		   sdp->use_10_for_rw || protect || rq->bio->bi_write_hint) {
1394  		ret = sd_setup_rw10_cmnd(cmd, write, lba, nr_blocks,
1395  					 protect | fua);
1396  	} else {
1397  		ret = sd_setup_rw6_cmnd(cmd, write, lba, nr_blocks,
1398  					protect | fua);
1399  	}
1400  
1401  	if (unlikely(ret != BLK_STS_OK))
1402  		goto fail;
1403  
1404  	/*
1405  	 * We shouldn't disconnect in the middle of a sector, so with a dumb
1406  	 * host adapter, it's safe to assume that we can at least transfer
1407  	 * this many bytes between each connect / disconnect.
1408  	 */
1409  	cmd->transfersize = sdp->sector_size;
1410  	cmd->underflow = nr_blocks << 9;
1411  	cmd->allowed = sdkp->max_retries;
1412  	cmd->sdb.length = nr_blocks * sdp->sector_size;
1413  
1414  	SCSI_LOG_HLQUEUE(1,
1415  			 scmd_printk(KERN_INFO, cmd,
1416  				     "%s: block=%llu, count=%d\n", __func__,
1417  				     (unsigned long long)blk_rq_pos(rq),
1418  				     blk_rq_sectors(rq)));
1419  	SCSI_LOG_HLQUEUE(2,
1420  			 scmd_printk(KERN_INFO, cmd,
1421  				     "%s %d/%u 512 byte blocks.\n",
1422  				     write ? "writing" : "reading", nr_blocks,
1423  				     blk_rq_sectors(rq)));
1424  
1425  	/*
1426  	 * This indicates that the command is ready from our end to be queued.
1427  	 */
1428  	return BLK_STS_OK;
1429  fail:
1430  	scsi_free_sgtables(cmd);
1431  	return ret;
1432  }
1433  
sd_init_command(struct scsi_cmnd * cmd)1434  static blk_status_t sd_init_command(struct scsi_cmnd *cmd)
1435  {
1436  	struct request *rq = scsi_cmd_to_rq(cmd);
1437  
1438  	switch (req_op(rq)) {
1439  	case REQ_OP_DISCARD:
1440  		switch (scsi_disk(rq->q->disk)->provisioning_mode) {
1441  		case SD_LBP_UNMAP:
1442  			return sd_setup_unmap_cmnd(cmd);
1443  		case SD_LBP_WS16:
1444  			return sd_setup_write_same16_cmnd(cmd, true);
1445  		case SD_LBP_WS10:
1446  			return sd_setup_write_same10_cmnd(cmd, true);
1447  		case SD_LBP_ZERO:
1448  			return sd_setup_write_same10_cmnd(cmd, false);
1449  		default:
1450  			return BLK_STS_TARGET;
1451  		}
1452  	case REQ_OP_WRITE_ZEROES:
1453  		return sd_setup_write_zeroes_cmnd(cmd);
1454  	case REQ_OP_FLUSH:
1455  		return sd_setup_flush_cmnd(cmd);
1456  	case REQ_OP_READ:
1457  	case REQ_OP_WRITE:
1458  		return sd_setup_read_write_cmnd(cmd);
1459  	case REQ_OP_ZONE_RESET:
1460  		return sd_zbc_setup_zone_mgmt_cmnd(cmd, ZO_RESET_WRITE_POINTER,
1461  						   false);
1462  	case REQ_OP_ZONE_RESET_ALL:
1463  		return sd_zbc_setup_zone_mgmt_cmnd(cmd, ZO_RESET_WRITE_POINTER,
1464  						   true);
1465  	case REQ_OP_ZONE_OPEN:
1466  		return sd_zbc_setup_zone_mgmt_cmnd(cmd, ZO_OPEN_ZONE, false);
1467  	case REQ_OP_ZONE_CLOSE:
1468  		return sd_zbc_setup_zone_mgmt_cmnd(cmd, ZO_CLOSE_ZONE, false);
1469  	case REQ_OP_ZONE_FINISH:
1470  		return sd_zbc_setup_zone_mgmt_cmnd(cmd, ZO_FINISH_ZONE, false);
1471  	default:
1472  		WARN_ON_ONCE(1);
1473  		return BLK_STS_NOTSUPP;
1474  	}
1475  }
1476  
sd_uninit_command(struct scsi_cmnd * SCpnt)1477  static void sd_uninit_command(struct scsi_cmnd *SCpnt)
1478  {
1479  	struct request *rq = scsi_cmd_to_rq(SCpnt);
1480  
1481  	if (rq->rq_flags & RQF_SPECIAL_PAYLOAD)
1482  		mempool_free(rq->special_vec.bv_page, sd_page_pool);
1483  }
1484  
sd_need_revalidate(struct gendisk * disk,struct scsi_disk * sdkp)1485  static bool sd_need_revalidate(struct gendisk *disk, struct scsi_disk *sdkp)
1486  {
1487  	if (sdkp->device->removable || sdkp->write_prot) {
1488  		if (disk_check_media_change(disk))
1489  			return true;
1490  	}
1491  
1492  	/*
1493  	 * Force a full rescan after ioctl(BLKRRPART).  While the disk state has
1494  	 * nothing to do with partitions, BLKRRPART is used to force a full
1495  	 * revalidate after things like a format for historical reasons.
1496  	 */
1497  	return test_bit(GD_NEED_PART_SCAN, &disk->state);
1498  }
1499  
1500  /**
1501   *	sd_open - open a scsi disk device
1502   *	@disk: disk to open
1503   *	@mode: open mode
1504   *
1505   *	Returns 0 if successful. Returns a negated errno value in case
1506   *	of error.
1507   *
1508   *	Note: This can be called from a user context (e.g. fsck(1) )
1509   *	or from within the kernel (e.g. as a result of a mount(1) ).
1510   *	In the latter case @inode and @filp carry an abridged amount
1511   *	of information as noted above.
1512   *
1513   *	Locking: called with disk->open_mutex held.
1514   **/
sd_open(struct gendisk * disk,blk_mode_t mode)1515  static int sd_open(struct gendisk *disk, blk_mode_t mode)
1516  {
1517  	struct scsi_disk *sdkp = scsi_disk(disk);
1518  	struct scsi_device *sdev = sdkp->device;
1519  	int retval;
1520  
1521  	if (scsi_device_get(sdev))
1522  		return -ENXIO;
1523  
1524  	SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp, "sd_open\n"));
1525  
1526  	/*
1527  	 * If the device is in error recovery, wait until it is done.
1528  	 * If the device is offline, then disallow any access to it.
1529  	 */
1530  	retval = -ENXIO;
1531  	if (!scsi_block_when_processing_errors(sdev))
1532  		goto error_out;
1533  
1534  	if (sd_need_revalidate(disk, sdkp))
1535  		sd_revalidate_disk(disk);
1536  
1537  	/*
1538  	 * If the drive is empty, just let the open fail.
1539  	 */
1540  	retval = -ENOMEDIUM;
1541  	if (sdev->removable && !sdkp->media_present &&
1542  	    !(mode & BLK_OPEN_NDELAY))
1543  		goto error_out;
1544  
1545  	/*
1546  	 * If the device has the write protect tab set, have the open fail
1547  	 * if the user expects to be able to write to the thing.
1548  	 */
1549  	retval = -EROFS;
1550  	if (sdkp->write_prot && (mode & BLK_OPEN_WRITE))
1551  		goto error_out;
1552  
1553  	/*
1554  	 * It is possible that the disk changing stuff resulted in
1555  	 * the device being taken offline.  If this is the case,
1556  	 * report this to the user, and don't pretend that the
1557  	 * open actually succeeded.
1558  	 */
1559  	retval = -ENXIO;
1560  	if (!scsi_device_online(sdev))
1561  		goto error_out;
1562  
1563  	if ((atomic_inc_return(&sdkp->openers) == 1) && sdev->removable) {
1564  		if (scsi_block_when_processing_errors(sdev))
1565  			scsi_set_medium_removal(sdev, SCSI_REMOVAL_PREVENT);
1566  	}
1567  
1568  	return 0;
1569  
1570  error_out:
1571  	scsi_device_put(sdev);
1572  	return retval;
1573  }
1574  
1575  /**
1576   *	sd_release - invoked when the (last) close(2) is called on this
1577   *	scsi disk.
1578   *	@disk: disk to release
1579   *
1580   *	Returns 0.
1581   *
1582   *	Note: may block (uninterruptible) if error recovery is underway
1583   *	on this disk.
1584   *
1585   *	Locking: called with disk->open_mutex held.
1586   **/
sd_release(struct gendisk * disk)1587  static void sd_release(struct gendisk *disk)
1588  {
1589  	struct scsi_disk *sdkp = scsi_disk(disk);
1590  	struct scsi_device *sdev = sdkp->device;
1591  
1592  	SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp, "sd_release\n"));
1593  
1594  	if (atomic_dec_return(&sdkp->openers) == 0 && sdev->removable) {
1595  		if (scsi_block_when_processing_errors(sdev))
1596  			scsi_set_medium_removal(sdev, SCSI_REMOVAL_ALLOW);
1597  	}
1598  
1599  	scsi_device_put(sdev);
1600  }
1601  
sd_getgeo(struct block_device * bdev,struct hd_geometry * geo)1602  static int sd_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1603  {
1604  	struct scsi_disk *sdkp = scsi_disk(bdev->bd_disk);
1605  	struct scsi_device *sdp = sdkp->device;
1606  	struct Scsi_Host *host = sdp->host;
1607  	sector_t capacity = logical_to_sectors(sdp, sdkp->capacity);
1608  	int diskinfo[4];
1609  
1610  	/* default to most commonly used values */
1611  	diskinfo[0] = 0x40;	/* 1 << 6 */
1612  	diskinfo[1] = 0x20;	/* 1 << 5 */
1613  	diskinfo[2] = capacity >> 11;
1614  
1615  	/* override with calculated, extended default, or driver values */
1616  	if (host->hostt->bios_param)
1617  		host->hostt->bios_param(sdp, bdev, capacity, diskinfo);
1618  	else
1619  		scsicam_bios_param(bdev, capacity, diskinfo);
1620  
1621  	geo->heads = diskinfo[0];
1622  	geo->sectors = diskinfo[1];
1623  	geo->cylinders = diskinfo[2];
1624  	return 0;
1625  }
1626  
1627  /**
1628   *	sd_ioctl - process an ioctl
1629   *	@bdev: target block device
1630   *	@mode: open mode
1631   *	@cmd: ioctl command number
1632   *	@arg: this is third argument given to ioctl(2) system call.
1633   *	Often contains a pointer.
1634   *
1635   *	Returns 0 if successful (some ioctls return positive numbers on
1636   *	success as well). Returns a negated errno value in case of error.
1637   *
1638   *	Note: most ioctls are forward onto the block subsystem or further
1639   *	down in the scsi subsystem.
1640   **/
sd_ioctl(struct block_device * bdev,blk_mode_t mode,unsigned int cmd,unsigned long arg)1641  static int sd_ioctl(struct block_device *bdev, blk_mode_t mode,
1642  		    unsigned int cmd, unsigned long arg)
1643  {
1644  	struct gendisk *disk = bdev->bd_disk;
1645  	struct scsi_disk *sdkp = scsi_disk(disk);
1646  	struct scsi_device *sdp = sdkp->device;
1647  	void __user *p = (void __user *)arg;
1648  	int error;
1649  
1650  	SCSI_LOG_IOCTL(1, sd_printk(KERN_INFO, sdkp, "sd_ioctl: disk=%s, "
1651  				    "cmd=0x%x\n", disk->disk_name, cmd));
1652  
1653  	if (bdev_is_partition(bdev) && !capable(CAP_SYS_RAWIO))
1654  		return -ENOIOCTLCMD;
1655  
1656  	/*
1657  	 * If we are in the middle of error recovery, don't let anyone
1658  	 * else try and use this device.  Also, if error recovery fails, it
1659  	 * may try and take the device offline, in which case all further
1660  	 * access to the device is prohibited.
1661  	 */
1662  	error = scsi_ioctl_block_when_processing_errors(sdp, cmd,
1663  			(mode & BLK_OPEN_NDELAY));
1664  	if (error)
1665  		return error;
1666  
1667  	if (is_sed_ioctl(cmd))
1668  		return sed_ioctl(sdkp->opal_dev, cmd, p);
1669  	return scsi_ioctl(sdp, mode & BLK_OPEN_WRITE, cmd, p);
1670  }
1671  
set_media_not_present(struct scsi_disk * sdkp)1672  static void set_media_not_present(struct scsi_disk *sdkp)
1673  {
1674  	if (sdkp->media_present)
1675  		sdkp->device->changed = 1;
1676  
1677  	if (sdkp->device->removable) {
1678  		sdkp->media_present = 0;
1679  		sdkp->capacity = 0;
1680  	}
1681  }
1682  
media_not_present(struct scsi_disk * sdkp,struct scsi_sense_hdr * sshdr)1683  static int media_not_present(struct scsi_disk *sdkp,
1684  			     struct scsi_sense_hdr *sshdr)
1685  {
1686  	if (!scsi_sense_valid(sshdr))
1687  		return 0;
1688  
1689  	/* not invoked for commands that could return deferred errors */
1690  	switch (sshdr->sense_key) {
1691  	case UNIT_ATTENTION:
1692  	case NOT_READY:
1693  		/* medium not present */
1694  		if (sshdr->asc == 0x3A) {
1695  			set_media_not_present(sdkp);
1696  			return 1;
1697  		}
1698  	}
1699  	return 0;
1700  }
1701  
1702  /**
1703   *	sd_check_events - check media events
1704   *	@disk: kernel device descriptor
1705   *	@clearing: disk events currently being cleared
1706   *
1707   *	Returns mask of DISK_EVENT_*.
1708   *
1709   *	Note: this function is invoked from the block subsystem.
1710   **/
sd_check_events(struct gendisk * disk,unsigned int clearing)1711  static unsigned int sd_check_events(struct gendisk *disk, unsigned int clearing)
1712  {
1713  	struct scsi_disk *sdkp = disk->private_data;
1714  	struct scsi_device *sdp;
1715  	int retval;
1716  	bool disk_changed;
1717  
1718  	if (!sdkp)
1719  		return 0;
1720  
1721  	sdp = sdkp->device;
1722  	SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp, "sd_check_events\n"));
1723  
1724  	/*
1725  	 * If the device is offline, don't send any commands - just pretend as
1726  	 * if the command failed.  If the device ever comes back online, we
1727  	 * can deal with it then.  It is only because of unrecoverable errors
1728  	 * that we would ever take a device offline in the first place.
1729  	 */
1730  	if (!scsi_device_online(sdp)) {
1731  		set_media_not_present(sdkp);
1732  		goto out;
1733  	}
1734  
1735  	/*
1736  	 * Using TEST_UNIT_READY enables differentiation between drive with
1737  	 * no cartridge loaded - NOT READY, drive with changed cartridge -
1738  	 * UNIT ATTENTION, or with same cartridge - GOOD STATUS.
1739  	 *
1740  	 * Drives that auto spin down. eg iomega jaz 1G, will be started
1741  	 * by sd_spinup_disk() from sd_revalidate_disk(), which happens whenever
1742  	 * sd_revalidate() is called.
1743  	 */
1744  	if (scsi_block_when_processing_errors(sdp)) {
1745  		struct scsi_sense_hdr sshdr = { 0, };
1746  
1747  		retval = scsi_test_unit_ready(sdp, SD_TIMEOUT, sdkp->max_retries,
1748  					      &sshdr);
1749  
1750  		/* failed to execute TUR, assume media not present */
1751  		if (retval < 0 || host_byte(retval)) {
1752  			set_media_not_present(sdkp);
1753  			goto out;
1754  		}
1755  
1756  		if (media_not_present(sdkp, &sshdr))
1757  			goto out;
1758  	}
1759  
1760  	/*
1761  	 * For removable scsi disk we have to recognise the presence
1762  	 * of a disk in the drive.
1763  	 */
1764  	if (!sdkp->media_present)
1765  		sdp->changed = 1;
1766  	sdkp->media_present = 1;
1767  out:
1768  	/*
1769  	 * sdp->changed is set under the following conditions:
1770  	 *
1771  	 *	Medium present state has changed in either direction.
1772  	 *	Device has indicated UNIT_ATTENTION.
1773  	 */
1774  	disk_changed = sdp->changed;
1775  	sdp->changed = 0;
1776  	return disk_changed ? DISK_EVENT_MEDIA_CHANGE : 0;
1777  }
1778  
sd_sync_cache(struct scsi_disk * sdkp)1779  static int sd_sync_cache(struct scsi_disk *sdkp)
1780  {
1781  	int res;
1782  	struct scsi_device *sdp = sdkp->device;
1783  	const int timeout = sdp->request_queue->rq_timeout
1784  		* SD_FLUSH_TIMEOUT_MULTIPLIER;
1785  	/* Leave the rest of the command zero to indicate flush everything. */
1786  	const unsigned char cmd[16] = { sdp->use_16_for_sync ?
1787  				SYNCHRONIZE_CACHE_16 : SYNCHRONIZE_CACHE };
1788  	struct scsi_sense_hdr sshdr;
1789  	struct scsi_failure failure_defs[] = {
1790  		{
1791  			.allowed = 3,
1792  			.result = SCMD_FAILURE_RESULT_ANY,
1793  		},
1794  		{}
1795  	};
1796  	struct scsi_failures failures = {
1797  		.failure_definitions = failure_defs,
1798  	};
1799  	const struct scsi_exec_args exec_args = {
1800  		.req_flags = BLK_MQ_REQ_PM,
1801  		.sshdr = &sshdr,
1802  		.failures = &failures,
1803  	};
1804  
1805  	if (!scsi_device_online(sdp))
1806  		return -ENODEV;
1807  
1808  	res = scsi_execute_cmd(sdp, cmd, REQ_OP_DRV_IN, NULL, 0, timeout,
1809  			       sdkp->max_retries, &exec_args);
1810  	if (res) {
1811  		sd_print_result(sdkp, "Synchronize Cache(10) failed", res);
1812  
1813  		if (res < 0)
1814  			return res;
1815  
1816  		if (scsi_status_is_check_condition(res) &&
1817  		    scsi_sense_valid(&sshdr)) {
1818  			sd_print_sense_hdr(sdkp, &sshdr);
1819  
1820  			/* we need to evaluate the error return  */
1821  			if (sshdr.asc == 0x3a ||	/* medium not present */
1822  			    sshdr.asc == 0x20 ||	/* invalid command */
1823  			    (sshdr.asc == 0x74 && sshdr.ascq == 0x71))	/* drive is password locked */
1824  				/* this is no error here */
1825  				return 0;
1826  
1827  			/*
1828  			 * If a format is in progress or if the drive does not
1829  			 * support sync, there is not much we can do because
1830  			 * this is called during shutdown or suspend so just
1831  			 * return success so those operations can proceed.
1832  			 */
1833  			if ((sshdr.asc == 0x04 && sshdr.ascq == 0x04) ||
1834  			    sshdr.sense_key == ILLEGAL_REQUEST)
1835  				return 0;
1836  		}
1837  
1838  		switch (host_byte(res)) {
1839  		/* ignore errors due to racing a disconnection */
1840  		case DID_BAD_TARGET:
1841  		case DID_NO_CONNECT:
1842  			return 0;
1843  		/* signal the upper layer it might try again */
1844  		case DID_BUS_BUSY:
1845  		case DID_IMM_RETRY:
1846  		case DID_REQUEUE:
1847  		case DID_SOFT_ERROR:
1848  			return -EBUSY;
1849  		default:
1850  			return -EIO;
1851  		}
1852  	}
1853  	return 0;
1854  }
1855  
sd_rescan(struct device * dev)1856  static void sd_rescan(struct device *dev)
1857  {
1858  	struct scsi_disk *sdkp = dev_get_drvdata(dev);
1859  
1860  	sd_revalidate_disk(sdkp->disk);
1861  }
1862  
sd_get_unique_id(struct gendisk * disk,u8 id[16],enum blk_unique_id type)1863  static int sd_get_unique_id(struct gendisk *disk, u8 id[16],
1864  		enum blk_unique_id type)
1865  {
1866  	struct scsi_device *sdev = scsi_disk(disk)->device;
1867  	const struct scsi_vpd *vpd;
1868  	const unsigned char *d;
1869  	int ret = -ENXIO, len;
1870  
1871  	rcu_read_lock();
1872  	vpd = rcu_dereference(sdev->vpd_pg83);
1873  	if (!vpd)
1874  		goto out_unlock;
1875  
1876  	ret = -EINVAL;
1877  	for (d = vpd->data + 4; d < vpd->data + vpd->len; d += d[3] + 4) {
1878  		/* we only care about designators with LU association */
1879  		if (((d[1] >> 4) & 0x3) != 0x00)
1880  			continue;
1881  		if ((d[1] & 0xf) != type)
1882  			continue;
1883  
1884  		/*
1885  		 * Only exit early if a 16-byte descriptor was found.  Otherwise
1886  		 * keep looking as one with more entropy might still show up.
1887  		 */
1888  		len = d[3];
1889  		if (len != 8 && len != 12 && len != 16)
1890  			continue;
1891  		ret = len;
1892  		memcpy(id, d + 4, len);
1893  		if (len == 16)
1894  			break;
1895  	}
1896  out_unlock:
1897  	rcu_read_unlock();
1898  	return ret;
1899  }
1900  
sd_scsi_to_pr_err(struct scsi_sense_hdr * sshdr,int result)1901  static int sd_scsi_to_pr_err(struct scsi_sense_hdr *sshdr, int result)
1902  {
1903  	switch (host_byte(result)) {
1904  	case DID_TRANSPORT_MARGINAL:
1905  	case DID_TRANSPORT_DISRUPTED:
1906  	case DID_BUS_BUSY:
1907  		return PR_STS_RETRY_PATH_FAILURE;
1908  	case DID_NO_CONNECT:
1909  		return PR_STS_PATH_FAILED;
1910  	case DID_TRANSPORT_FAILFAST:
1911  		return PR_STS_PATH_FAST_FAILED;
1912  	}
1913  
1914  	switch (status_byte(result)) {
1915  	case SAM_STAT_RESERVATION_CONFLICT:
1916  		return PR_STS_RESERVATION_CONFLICT;
1917  	case SAM_STAT_CHECK_CONDITION:
1918  		if (!scsi_sense_valid(sshdr))
1919  			return PR_STS_IOERR;
1920  
1921  		if (sshdr->sense_key == ILLEGAL_REQUEST &&
1922  		    (sshdr->asc == 0x26 || sshdr->asc == 0x24))
1923  			return -EINVAL;
1924  
1925  		fallthrough;
1926  	default:
1927  		return PR_STS_IOERR;
1928  	}
1929  }
1930  
sd_pr_in_command(struct block_device * bdev,u8 sa,unsigned char * data,int data_len)1931  static int sd_pr_in_command(struct block_device *bdev, u8 sa,
1932  			    unsigned char *data, int data_len)
1933  {
1934  	struct scsi_disk *sdkp = scsi_disk(bdev->bd_disk);
1935  	struct scsi_device *sdev = sdkp->device;
1936  	struct scsi_sense_hdr sshdr;
1937  	u8 cmd[10] = { PERSISTENT_RESERVE_IN, sa };
1938  	struct scsi_failure failure_defs[] = {
1939  		{
1940  			.sense = UNIT_ATTENTION,
1941  			.asc = SCMD_FAILURE_ASC_ANY,
1942  			.ascq = SCMD_FAILURE_ASCQ_ANY,
1943  			.allowed = 5,
1944  			.result = SAM_STAT_CHECK_CONDITION,
1945  		},
1946  		{}
1947  	};
1948  	struct scsi_failures failures = {
1949  		.failure_definitions = failure_defs,
1950  	};
1951  	const struct scsi_exec_args exec_args = {
1952  		.sshdr = &sshdr,
1953  		.failures = &failures,
1954  	};
1955  	int result;
1956  
1957  	put_unaligned_be16(data_len, &cmd[7]);
1958  
1959  	result = scsi_execute_cmd(sdev, cmd, REQ_OP_DRV_IN, data, data_len,
1960  				  SD_TIMEOUT, sdkp->max_retries, &exec_args);
1961  	if (scsi_status_is_check_condition(result) &&
1962  	    scsi_sense_valid(&sshdr)) {
1963  		sdev_printk(KERN_INFO, sdev, "PR command failed: %d\n", result);
1964  		scsi_print_sense_hdr(sdev, NULL, &sshdr);
1965  	}
1966  
1967  	if (result <= 0)
1968  		return result;
1969  
1970  	return sd_scsi_to_pr_err(&sshdr, result);
1971  }
1972  
sd_pr_read_keys(struct block_device * bdev,struct pr_keys * keys_info)1973  static int sd_pr_read_keys(struct block_device *bdev, struct pr_keys *keys_info)
1974  {
1975  	int result, i, data_offset, num_copy_keys;
1976  	u32 num_keys = keys_info->num_keys;
1977  	int data_len = num_keys * 8 + 8;
1978  	u8 *data;
1979  
1980  	data = kzalloc(data_len, GFP_KERNEL);
1981  	if (!data)
1982  		return -ENOMEM;
1983  
1984  	result = sd_pr_in_command(bdev, READ_KEYS, data, data_len);
1985  	if (result)
1986  		goto free_data;
1987  
1988  	keys_info->generation = get_unaligned_be32(&data[0]);
1989  	keys_info->num_keys = get_unaligned_be32(&data[4]) / 8;
1990  
1991  	data_offset = 8;
1992  	num_copy_keys = min(num_keys, keys_info->num_keys);
1993  
1994  	for (i = 0; i < num_copy_keys; i++) {
1995  		keys_info->keys[i] = get_unaligned_be64(&data[data_offset]);
1996  		data_offset += 8;
1997  	}
1998  
1999  free_data:
2000  	kfree(data);
2001  	return result;
2002  }
2003  
sd_pr_read_reservation(struct block_device * bdev,struct pr_held_reservation * rsv)2004  static int sd_pr_read_reservation(struct block_device *bdev,
2005  				  struct pr_held_reservation *rsv)
2006  {
2007  	struct scsi_disk *sdkp = scsi_disk(bdev->bd_disk);
2008  	struct scsi_device *sdev = sdkp->device;
2009  	u8 data[24] = { };
2010  	int result, len;
2011  
2012  	result = sd_pr_in_command(bdev, READ_RESERVATION, data, sizeof(data));
2013  	if (result)
2014  		return result;
2015  
2016  	len = get_unaligned_be32(&data[4]);
2017  	if (!len)
2018  		return 0;
2019  
2020  	/* Make sure we have at least the key and type */
2021  	if (len < 14) {
2022  		sdev_printk(KERN_INFO, sdev,
2023  			    "READ RESERVATION failed due to short return buffer of %d bytes\n",
2024  			    len);
2025  		return -EINVAL;
2026  	}
2027  
2028  	rsv->generation = get_unaligned_be32(&data[0]);
2029  	rsv->key = get_unaligned_be64(&data[8]);
2030  	rsv->type = scsi_pr_type_to_block(data[21] & 0x0f);
2031  	return 0;
2032  }
2033  
sd_pr_out_command(struct block_device * bdev,u8 sa,u64 key,u64 sa_key,enum scsi_pr_type type,u8 flags)2034  static int sd_pr_out_command(struct block_device *bdev, u8 sa, u64 key,
2035  			     u64 sa_key, enum scsi_pr_type type, u8 flags)
2036  {
2037  	struct scsi_disk *sdkp = scsi_disk(bdev->bd_disk);
2038  	struct scsi_device *sdev = sdkp->device;
2039  	struct scsi_sense_hdr sshdr;
2040  	struct scsi_failure failure_defs[] = {
2041  		{
2042  			.sense = UNIT_ATTENTION,
2043  			.asc = SCMD_FAILURE_ASC_ANY,
2044  			.ascq = SCMD_FAILURE_ASCQ_ANY,
2045  			.allowed = 5,
2046  			.result = SAM_STAT_CHECK_CONDITION,
2047  		},
2048  		{}
2049  	};
2050  	struct scsi_failures failures = {
2051  		.failure_definitions = failure_defs,
2052  	};
2053  	const struct scsi_exec_args exec_args = {
2054  		.sshdr = &sshdr,
2055  		.failures = &failures,
2056  	};
2057  	int result;
2058  	u8 cmd[16] = { 0, };
2059  	u8 data[24] = { 0, };
2060  
2061  	cmd[0] = PERSISTENT_RESERVE_OUT;
2062  	cmd[1] = sa;
2063  	cmd[2] = type;
2064  	put_unaligned_be32(sizeof(data), &cmd[5]);
2065  
2066  	put_unaligned_be64(key, &data[0]);
2067  	put_unaligned_be64(sa_key, &data[8]);
2068  	data[20] = flags;
2069  
2070  	result = scsi_execute_cmd(sdev, cmd, REQ_OP_DRV_OUT, &data,
2071  				  sizeof(data), SD_TIMEOUT, sdkp->max_retries,
2072  				  &exec_args);
2073  
2074  	if (scsi_status_is_check_condition(result) &&
2075  	    scsi_sense_valid(&sshdr)) {
2076  		sdev_printk(KERN_INFO, sdev, "PR command failed: %d\n", result);
2077  		scsi_print_sense_hdr(sdev, NULL, &sshdr);
2078  	}
2079  
2080  	if (result <= 0)
2081  		return result;
2082  
2083  	return sd_scsi_to_pr_err(&sshdr, result);
2084  }
2085  
sd_pr_register(struct block_device * bdev,u64 old_key,u64 new_key,u32 flags)2086  static int sd_pr_register(struct block_device *bdev, u64 old_key, u64 new_key,
2087  		u32 flags)
2088  {
2089  	if (flags & ~PR_FL_IGNORE_KEY)
2090  		return -EOPNOTSUPP;
2091  	return sd_pr_out_command(bdev, (flags & PR_FL_IGNORE_KEY) ? 0x06 : 0x00,
2092  			old_key, new_key, 0,
2093  			(1 << 0) /* APTPL */);
2094  }
2095  
sd_pr_reserve(struct block_device * bdev,u64 key,enum pr_type type,u32 flags)2096  static int sd_pr_reserve(struct block_device *bdev, u64 key, enum pr_type type,
2097  		u32 flags)
2098  {
2099  	if (flags)
2100  		return -EOPNOTSUPP;
2101  	return sd_pr_out_command(bdev, 0x01, key, 0,
2102  				 block_pr_type_to_scsi(type), 0);
2103  }
2104  
sd_pr_release(struct block_device * bdev,u64 key,enum pr_type type)2105  static int sd_pr_release(struct block_device *bdev, u64 key, enum pr_type type)
2106  {
2107  	return sd_pr_out_command(bdev, 0x02, key, 0,
2108  				 block_pr_type_to_scsi(type), 0);
2109  }
2110  
sd_pr_preempt(struct block_device * bdev,u64 old_key,u64 new_key,enum pr_type type,bool abort)2111  static int sd_pr_preempt(struct block_device *bdev, u64 old_key, u64 new_key,
2112  		enum pr_type type, bool abort)
2113  {
2114  	return sd_pr_out_command(bdev, abort ? 0x05 : 0x04, old_key, new_key,
2115  				 block_pr_type_to_scsi(type), 0);
2116  }
2117  
sd_pr_clear(struct block_device * bdev,u64 key)2118  static int sd_pr_clear(struct block_device *bdev, u64 key)
2119  {
2120  	return sd_pr_out_command(bdev, 0x03, key, 0, 0, 0);
2121  }
2122  
2123  static const struct pr_ops sd_pr_ops = {
2124  	.pr_register	= sd_pr_register,
2125  	.pr_reserve	= sd_pr_reserve,
2126  	.pr_release	= sd_pr_release,
2127  	.pr_preempt	= sd_pr_preempt,
2128  	.pr_clear	= sd_pr_clear,
2129  	.pr_read_keys	= sd_pr_read_keys,
2130  	.pr_read_reservation = sd_pr_read_reservation,
2131  };
2132  
scsi_disk_free_disk(struct gendisk * disk)2133  static void scsi_disk_free_disk(struct gendisk *disk)
2134  {
2135  	struct scsi_disk *sdkp = scsi_disk(disk);
2136  
2137  	put_device(&sdkp->disk_dev);
2138  }
2139  
2140  static const struct block_device_operations sd_fops = {
2141  	.owner			= THIS_MODULE,
2142  	.open			= sd_open,
2143  	.release		= sd_release,
2144  	.ioctl			= sd_ioctl,
2145  	.getgeo			= sd_getgeo,
2146  	.compat_ioctl		= blkdev_compat_ptr_ioctl,
2147  	.check_events		= sd_check_events,
2148  	.unlock_native_capacity	= sd_unlock_native_capacity,
2149  	.report_zones		= sd_zbc_report_zones,
2150  	.get_unique_id		= sd_get_unique_id,
2151  	.free_disk		= scsi_disk_free_disk,
2152  	.pr_ops			= &sd_pr_ops,
2153  };
2154  
2155  /**
2156   *	sd_eh_reset - reset error handling callback
2157   *	@scmd:		sd-issued command that has failed
2158   *
2159   *	This function is called by the SCSI midlayer before starting
2160   *	SCSI EH. When counting medium access failures we have to be
2161   *	careful to register it only only once per device and SCSI EH run;
2162   *	there might be several timed out commands which will cause the
2163   *	'max_medium_access_timeouts' counter to trigger after the first
2164   *	SCSI EH run already and set the device to offline.
2165   *	So this function resets the internal counter before starting SCSI EH.
2166   **/
sd_eh_reset(struct scsi_cmnd * scmd)2167  static void sd_eh_reset(struct scsi_cmnd *scmd)
2168  {
2169  	struct scsi_disk *sdkp = scsi_disk(scsi_cmd_to_rq(scmd)->q->disk);
2170  
2171  	/* New SCSI EH run, reset gate variable */
2172  	sdkp->ignore_medium_access_errors = false;
2173  }
2174  
2175  /**
2176   *	sd_eh_action - error handling callback
2177   *	@scmd:		sd-issued command that has failed
2178   *	@eh_disp:	The recovery disposition suggested by the midlayer
2179   *
2180   *	This function is called by the SCSI midlayer upon completion of an
2181   *	error test command (currently TEST UNIT READY). The result of sending
2182   *	the eh command is passed in eh_disp.  We're looking for devices that
2183   *	fail medium access commands but are OK with non access commands like
2184   *	test unit ready (so wrongly see the device as having a successful
2185   *	recovery)
2186   **/
sd_eh_action(struct scsi_cmnd * scmd,int eh_disp)2187  static int sd_eh_action(struct scsi_cmnd *scmd, int eh_disp)
2188  {
2189  	struct scsi_disk *sdkp = scsi_disk(scsi_cmd_to_rq(scmd)->q->disk);
2190  	struct scsi_device *sdev = scmd->device;
2191  
2192  	if (!scsi_device_online(sdev) ||
2193  	    !scsi_medium_access_command(scmd) ||
2194  	    host_byte(scmd->result) != DID_TIME_OUT ||
2195  	    eh_disp != SUCCESS)
2196  		return eh_disp;
2197  
2198  	/*
2199  	 * The device has timed out executing a medium access command.
2200  	 * However, the TEST UNIT READY command sent during error
2201  	 * handling completed successfully. Either the device is in the
2202  	 * process of recovering or has it suffered an internal failure
2203  	 * that prevents access to the storage medium.
2204  	 */
2205  	if (!sdkp->ignore_medium_access_errors) {
2206  		sdkp->medium_access_timed_out++;
2207  		sdkp->ignore_medium_access_errors = true;
2208  	}
2209  
2210  	/*
2211  	 * If the device keeps failing read/write commands but TEST UNIT
2212  	 * READY always completes successfully we assume that medium
2213  	 * access is no longer possible and take the device offline.
2214  	 */
2215  	if (sdkp->medium_access_timed_out >= sdkp->max_medium_access_timeouts) {
2216  		scmd_printk(KERN_ERR, scmd,
2217  			    "Medium access timeout failure. Offlining disk!\n");
2218  		mutex_lock(&sdev->state_mutex);
2219  		scsi_device_set_state(sdev, SDEV_OFFLINE);
2220  		mutex_unlock(&sdev->state_mutex);
2221  
2222  		return SUCCESS;
2223  	}
2224  
2225  	return eh_disp;
2226  }
2227  
sd_completed_bytes(struct scsi_cmnd * scmd)2228  static unsigned int sd_completed_bytes(struct scsi_cmnd *scmd)
2229  {
2230  	struct request *req = scsi_cmd_to_rq(scmd);
2231  	struct scsi_device *sdev = scmd->device;
2232  	unsigned int transferred, good_bytes;
2233  	u64 start_lba, end_lba, bad_lba;
2234  
2235  	/*
2236  	 * Some commands have a payload smaller than the device logical
2237  	 * block size (e.g. INQUIRY on a 4K disk).
2238  	 */
2239  	if (scsi_bufflen(scmd) <= sdev->sector_size)
2240  		return 0;
2241  
2242  	/* Check if we have a 'bad_lba' information */
2243  	if (!scsi_get_sense_info_fld(scmd->sense_buffer,
2244  				     SCSI_SENSE_BUFFERSIZE,
2245  				     &bad_lba))
2246  		return 0;
2247  
2248  	/*
2249  	 * If the bad lba was reported incorrectly, we have no idea where
2250  	 * the error is.
2251  	 */
2252  	start_lba = sectors_to_logical(sdev, blk_rq_pos(req));
2253  	end_lba = start_lba + bytes_to_logical(sdev, scsi_bufflen(scmd));
2254  	if (bad_lba < start_lba || bad_lba >= end_lba)
2255  		return 0;
2256  
2257  	/*
2258  	 * resid is optional but mostly filled in.  When it's unused,
2259  	 * its value is zero, so we assume the whole buffer transferred
2260  	 */
2261  	transferred = scsi_bufflen(scmd) - scsi_get_resid(scmd);
2262  
2263  	/* This computation should always be done in terms of the
2264  	 * resolution of the device's medium.
2265  	 */
2266  	good_bytes = logical_to_bytes(sdev, bad_lba - start_lba);
2267  
2268  	return min(good_bytes, transferred);
2269  }
2270  
2271  /**
2272   *	sd_done - bottom half handler: called when the lower level
2273   *	driver has completed (successfully or otherwise) a scsi command.
2274   *	@SCpnt: mid-level's per command structure.
2275   *
2276   *	Note: potentially run from within an ISR. Must not block.
2277   **/
sd_done(struct scsi_cmnd * SCpnt)2278  static int sd_done(struct scsi_cmnd *SCpnt)
2279  {
2280  	int result = SCpnt->result;
2281  	unsigned int good_bytes = result ? 0 : scsi_bufflen(SCpnt);
2282  	unsigned int sector_size = SCpnt->device->sector_size;
2283  	unsigned int resid;
2284  	struct scsi_sense_hdr sshdr;
2285  	struct request *req = scsi_cmd_to_rq(SCpnt);
2286  	struct scsi_disk *sdkp = scsi_disk(req->q->disk);
2287  	int sense_valid = 0;
2288  	int sense_deferred = 0;
2289  
2290  	switch (req_op(req)) {
2291  	case REQ_OP_DISCARD:
2292  	case REQ_OP_WRITE_ZEROES:
2293  	case REQ_OP_ZONE_RESET:
2294  	case REQ_OP_ZONE_RESET_ALL:
2295  	case REQ_OP_ZONE_OPEN:
2296  	case REQ_OP_ZONE_CLOSE:
2297  	case REQ_OP_ZONE_FINISH:
2298  		if (!result) {
2299  			good_bytes = blk_rq_bytes(req);
2300  			scsi_set_resid(SCpnt, 0);
2301  		} else {
2302  			good_bytes = 0;
2303  			scsi_set_resid(SCpnt, blk_rq_bytes(req));
2304  		}
2305  		break;
2306  	default:
2307  		/*
2308  		 * In case of bogus fw or device, we could end up having
2309  		 * an unaligned partial completion. Check this here and force
2310  		 * alignment.
2311  		 */
2312  		resid = scsi_get_resid(SCpnt);
2313  		if (resid & (sector_size - 1)) {
2314  			sd_printk(KERN_INFO, sdkp,
2315  				"Unaligned partial completion (resid=%u, sector_sz=%u)\n",
2316  				resid, sector_size);
2317  			scsi_print_command(SCpnt);
2318  			resid = min(scsi_bufflen(SCpnt),
2319  				    round_up(resid, sector_size));
2320  			scsi_set_resid(SCpnt, resid);
2321  		}
2322  	}
2323  
2324  	if (result) {
2325  		sense_valid = scsi_command_normalize_sense(SCpnt, &sshdr);
2326  		if (sense_valid)
2327  			sense_deferred = scsi_sense_is_deferred(&sshdr);
2328  	}
2329  	sdkp->medium_access_timed_out = 0;
2330  
2331  	if (!scsi_status_is_check_condition(result) &&
2332  	    (!sense_valid || sense_deferred))
2333  		goto out;
2334  
2335  	switch (sshdr.sense_key) {
2336  	case HARDWARE_ERROR:
2337  	case MEDIUM_ERROR:
2338  		good_bytes = sd_completed_bytes(SCpnt);
2339  		break;
2340  	case RECOVERED_ERROR:
2341  		good_bytes = scsi_bufflen(SCpnt);
2342  		break;
2343  	case NO_SENSE:
2344  		/* This indicates a false check condition, so ignore it.  An
2345  		 * unknown amount of data was transferred so treat it as an
2346  		 * error.
2347  		 */
2348  		SCpnt->result = 0;
2349  		memset(SCpnt->sense_buffer, 0, SCSI_SENSE_BUFFERSIZE);
2350  		break;
2351  	case ABORTED_COMMAND:
2352  		if (sshdr.asc == 0x10)  /* DIF: Target detected corruption */
2353  			good_bytes = sd_completed_bytes(SCpnt);
2354  		break;
2355  	case ILLEGAL_REQUEST:
2356  		switch (sshdr.asc) {
2357  		case 0x10:	/* DIX: Host detected corruption */
2358  			good_bytes = sd_completed_bytes(SCpnt);
2359  			break;
2360  		case 0x20:	/* INVALID COMMAND OPCODE */
2361  		case 0x24:	/* INVALID FIELD IN CDB */
2362  			switch (SCpnt->cmnd[0]) {
2363  			case UNMAP:
2364  				sd_disable_discard(sdkp);
2365  				break;
2366  			case WRITE_SAME_16:
2367  			case WRITE_SAME:
2368  				if (SCpnt->cmnd[1] & 8) { /* UNMAP */
2369  					sd_disable_discard(sdkp);
2370  				} else {
2371  					sd_disable_write_same(sdkp);
2372  					req->rq_flags |= RQF_QUIET;
2373  				}
2374  				break;
2375  			}
2376  		}
2377  		break;
2378  	default:
2379  		break;
2380  	}
2381  
2382   out:
2383  	if (sdkp->device->type == TYPE_ZBC)
2384  		good_bytes = sd_zbc_complete(SCpnt, good_bytes, &sshdr);
2385  
2386  	SCSI_LOG_HLCOMPLETE(1, scmd_printk(KERN_INFO, SCpnt,
2387  					   "sd_done: completed %d of %d bytes\n",
2388  					   good_bytes, scsi_bufflen(SCpnt)));
2389  
2390  	return good_bytes;
2391  }
2392  
2393  /*
2394   * spinup disk - called only in sd_revalidate_disk()
2395   */
2396  static void
sd_spinup_disk(struct scsi_disk * sdkp)2397  sd_spinup_disk(struct scsi_disk *sdkp)
2398  {
2399  	static const u8 cmd[10] = { TEST_UNIT_READY };
2400  	unsigned long spintime_expire = 0;
2401  	int spintime, sense_valid = 0;
2402  	unsigned int the_result;
2403  	struct scsi_sense_hdr sshdr;
2404  	struct scsi_failure failure_defs[] = {
2405  		/* Do not retry Medium Not Present */
2406  		{
2407  			.sense = UNIT_ATTENTION,
2408  			.asc = 0x3A,
2409  			.ascq = SCMD_FAILURE_ASCQ_ANY,
2410  			.result = SAM_STAT_CHECK_CONDITION,
2411  		},
2412  		{
2413  			.sense = NOT_READY,
2414  			.asc = 0x3A,
2415  			.ascq = SCMD_FAILURE_ASCQ_ANY,
2416  			.result = SAM_STAT_CHECK_CONDITION,
2417  		},
2418  		/* Retry when scsi_status_is_good would return false 3 times */
2419  		{
2420  			.result = SCMD_FAILURE_STAT_ANY,
2421  			.allowed = 3,
2422  		},
2423  		{}
2424  	};
2425  	struct scsi_failures failures = {
2426  		.failure_definitions = failure_defs,
2427  	};
2428  	const struct scsi_exec_args exec_args = {
2429  		.sshdr = &sshdr,
2430  		.failures = &failures,
2431  	};
2432  
2433  	spintime = 0;
2434  
2435  	/* Spin up drives, as required.  Only do this at boot time */
2436  	/* Spinup needs to be done for module loads too. */
2437  	do {
2438  		bool media_was_present = sdkp->media_present;
2439  
2440  		scsi_failures_reset_retries(&failures);
2441  
2442  		the_result = scsi_execute_cmd(sdkp->device, cmd, REQ_OP_DRV_IN,
2443  					      NULL, 0, SD_TIMEOUT,
2444  					      sdkp->max_retries, &exec_args);
2445  
2446  
2447  		if (the_result > 0) {
2448  			/*
2449  			 * If the drive has indicated to us that it doesn't
2450  			 * have any media in it, don't bother with any more
2451  			 * polling.
2452  			 */
2453  			if (media_not_present(sdkp, &sshdr)) {
2454  				if (media_was_present)
2455  					sd_printk(KERN_NOTICE, sdkp,
2456  						  "Media removed, stopped polling\n");
2457  				return;
2458  			}
2459  			sense_valid = scsi_sense_valid(&sshdr);
2460  		}
2461  
2462  		if (!scsi_status_is_check_condition(the_result)) {
2463  			/* no sense, TUR either succeeded or failed
2464  			 * with a status error */
2465  			if(!spintime && !scsi_status_is_good(the_result)) {
2466  				sd_print_result(sdkp, "Test Unit Ready failed",
2467  						the_result);
2468  			}
2469  			break;
2470  		}
2471  
2472  		/*
2473  		 * The device does not want the automatic start to be issued.
2474  		 */
2475  		if (sdkp->device->no_start_on_add)
2476  			break;
2477  
2478  		if (sense_valid && sshdr.sense_key == NOT_READY) {
2479  			if (sshdr.asc == 4 && sshdr.ascq == 3)
2480  				break;	/* manual intervention required */
2481  			if (sshdr.asc == 4 && sshdr.ascq == 0xb)
2482  				break;	/* standby */
2483  			if (sshdr.asc == 4 && sshdr.ascq == 0xc)
2484  				break;	/* unavailable */
2485  			if (sshdr.asc == 4 && sshdr.ascq == 0x1b)
2486  				break;	/* sanitize in progress */
2487  			if (sshdr.asc == 4 && sshdr.ascq == 0x24)
2488  				break;	/* depopulation in progress */
2489  			if (sshdr.asc == 4 && sshdr.ascq == 0x25)
2490  				break;	/* depopulation restoration in progress */
2491  			/*
2492  			 * Issue command to spin up drive when not ready
2493  			 */
2494  			if (!spintime) {
2495  				/* Return immediately and start spin cycle */
2496  				const u8 start_cmd[10] = {
2497  					[0] = START_STOP,
2498  					[1] = 1,
2499  					[4] = sdkp->device->start_stop_pwr_cond ?
2500  						0x11 : 1,
2501  				};
2502  
2503  				sd_printk(KERN_NOTICE, sdkp, "Spinning up disk...");
2504  				scsi_execute_cmd(sdkp->device, start_cmd,
2505  						 REQ_OP_DRV_IN, NULL, 0,
2506  						 SD_TIMEOUT, sdkp->max_retries,
2507  						 &exec_args);
2508  				spintime_expire = jiffies + 100 * HZ;
2509  				spintime = 1;
2510  			}
2511  			/* Wait 1 second for next try */
2512  			msleep(1000);
2513  			printk(KERN_CONT ".");
2514  
2515  		/*
2516  		 * Wait for USB flash devices with slow firmware.
2517  		 * Yes, this sense key/ASC combination shouldn't
2518  		 * occur here.  It's characteristic of these devices.
2519  		 */
2520  		} else if (sense_valid &&
2521  				sshdr.sense_key == UNIT_ATTENTION &&
2522  				sshdr.asc == 0x28) {
2523  			if (!spintime) {
2524  				spintime_expire = jiffies + 5 * HZ;
2525  				spintime = 1;
2526  			}
2527  			/* Wait 1 second for next try */
2528  			msleep(1000);
2529  		} else {
2530  			/* we don't understand the sense code, so it's
2531  			 * probably pointless to loop */
2532  			if(!spintime) {
2533  				sd_printk(KERN_NOTICE, sdkp, "Unit Not Ready\n");
2534  				sd_print_sense_hdr(sdkp, &sshdr);
2535  			}
2536  			break;
2537  		}
2538  
2539  	} while (spintime && time_before_eq(jiffies, spintime_expire));
2540  
2541  	if (spintime) {
2542  		if (scsi_status_is_good(the_result))
2543  			printk(KERN_CONT "ready\n");
2544  		else
2545  			printk(KERN_CONT "not responding...\n");
2546  	}
2547  }
2548  
2549  /*
2550   * Determine whether disk supports Data Integrity Field.
2551   */
sd_read_protection_type(struct scsi_disk * sdkp,unsigned char * buffer)2552  static int sd_read_protection_type(struct scsi_disk *sdkp, unsigned char *buffer)
2553  {
2554  	struct scsi_device *sdp = sdkp->device;
2555  	u8 type;
2556  
2557  	if (scsi_device_protection(sdp) == 0 || (buffer[12] & 1) == 0) {
2558  		sdkp->protection_type = 0;
2559  		return 0;
2560  	}
2561  
2562  	type = ((buffer[12] >> 1) & 7) + 1; /* P_TYPE 0 = Type 1 */
2563  
2564  	if (type > T10_PI_TYPE3_PROTECTION) {
2565  		sd_printk(KERN_ERR, sdkp, "formatted with unsupported"	\
2566  			  " protection type %u. Disabling disk!\n",
2567  			  type);
2568  		sdkp->protection_type = 0;
2569  		return -ENODEV;
2570  	}
2571  
2572  	sdkp->protection_type = type;
2573  
2574  	return 0;
2575  }
2576  
sd_config_protection(struct scsi_disk * sdkp,struct queue_limits * lim)2577  static void sd_config_protection(struct scsi_disk *sdkp,
2578  		struct queue_limits *lim)
2579  {
2580  	struct scsi_device *sdp = sdkp->device;
2581  
2582  	if (IS_ENABLED(CONFIG_BLK_DEV_INTEGRITY))
2583  		sd_dif_config_host(sdkp, lim);
2584  
2585  	if (!sdkp->protection_type)
2586  		return;
2587  
2588  	if (!scsi_host_dif_capable(sdp->host, sdkp->protection_type)) {
2589  		sd_first_printk(KERN_NOTICE, sdkp,
2590  				"Disabling DIF Type %u protection\n",
2591  				sdkp->protection_type);
2592  		sdkp->protection_type = 0;
2593  	}
2594  
2595  	sd_first_printk(KERN_NOTICE, sdkp, "Enabling DIF Type %u protection\n",
2596  			sdkp->protection_type);
2597  }
2598  
read_capacity_error(struct scsi_disk * sdkp,struct scsi_device * sdp,struct scsi_sense_hdr * sshdr,int sense_valid,int the_result)2599  static void read_capacity_error(struct scsi_disk *sdkp, struct scsi_device *sdp,
2600  			struct scsi_sense_hdr *sshdr, int sense_valid,
2601  			int the_result)
2602  {
2603  	if (sense_valid)
2604  		sd_print_sense_hdr(sdkp, sshdr);
2605  	else
2606  		sd_printk(KERN_NOTICE, sdkp, "Sense not available.\n");
2607  
2608  	/*
2609  	 * Set dirty bit for removable devices if not ready -
2610  	 * sometimes drives will not report this properly.
2611  	 */
2612  	if (sdp->removable &&
2613  	    sense_valid && sshdr->sense_key == NOT_READY)
2614  		set_media_not_present(sdkp);
2615  
2616  	/*
2617  	 * We used to set media_present to 0 here to indicate no media
2618  	 * in the drive, but some drives fail read capacity even with
2619  	 * media present, so we can't do that.
2620  	 */
2621  	sdkp->capacity = 0; /* unknown mapped to zero - as usual */
2622  }
2623  
2624  #define RC16_LEN 32
2625  #if RC16_LEN > SD_BUF_SIZE
2626  #error RC16_LEN must not be more than SD_BUF_SIZE
2627  #endif
2628  
2629  #define READ_CAPACITY_RETRIES_ON_RESET	10
2630  
read_capacity_16(struct scsi_disk * sdkp,struct scsi_device * sdp,struct queue_limits * lim,unsigned char * buffer)2631  static int read_capacity_16(struct scsi_disk *sdkp, struct scsi_device *sdp,
2632  		struct queue_limits *lim, unsigned char *buffer)
2633  {
2634  	unsigned char cmd[16];
2635  	struct scsi_sense_hdr sshdr;
2636  	const struct scsi_exec_args exec_args = {
2637  		.sshdr = &sshdr,
2638  	};
2639  	int sense_valid = 0;
2640  	int the_result;
2641  	int retries = 3, reset_retries = READ_CAPACITY_RETRIES_ON_RESET;
2642  	unsigned int alignment;
2643  	unsigned long long lba;
2644  	unsigned sector_size;
2645  
2646  	if (sdp->no_read_capacity_16)
2647  		return -EINVAL;
2648  
2649  	do {
2650  		memset(cmd, 0, 16);
2651  		cmd[0] = SERVICE_ACTION_IN_16;
2652  		cmd[1] = SAI_READ_CAPACITY_16;
2653  		cmd[13] = RC16_LEN;
2654  		memset(buffer, 0, RC16_LEN);
2655  
2656  		the_result = scsi_execute_cmd(sdp, cmd, REQ_OP_DRV_IN,
2657  					      buffer, RC16_LEN, SD_TIMEOUT,
2658  					      sdkp->max_retries, &exec_args);
2659  		if (the_result > 0) {
2660  			if (media_not_present(sdkp, &sshdr))
2661  				return -ENODEV;
2662  
2663  			sense_valid = scsi_sense_valid(&sshdr);
2664  			if (sense_valid &&
2665  			    sshdr.sense_key == ILLEGAL_REQUEST &&
2666  			    (sshdr.asc == 0x20 || sshdr.asc == 0x24) &&
2667  			    sshdr.ascq == 0x00)
2668  				/* Invalid Command Operation Code or
2669  				 * Invalid Field in CDB, just retry
2670  				 * silently with RC10 */
2671  				return -EINVAL;
2672  			if (sense_valid &&
2673  			    sshdr.sense_key == UNIT_ATTENTION &&
2674  			    sshdr.asc == 0x29 && sshdr.ascq == 0x00)
2675  				/* Device reset might occur several times,
2676  				 * give it one more chance */
2677  				if (--reset_retries > 0)
2678  					continue;
2679  		}
2680  		retries--;
2681  
2682  	} while (the_result && retries);
2683  
2684  	if (the_result) {
2685  		sd_print_result(sdkp, "Read Capacity(16) failed", the_result);
2686  		read_capacity_error(sdkp, sdp, &sshdr, sense_valid, the_result);
2687  		return -EINVAL;
2688  	}
2689  
2690  	sector_size = get_unaligned_be32(&buffer[8]);
2691  	lba = get_unaligned_be64(&buffer[0]);
2692  
2693  	if (sd_read_protection_type(sdkp, buffer) < 0) {
2694  		sdkp->capacity = 0;
2695  		return -ENODEV;
2696  	}
2697  
2698  	/* Logical blocks per physical block exponent */
2699  	sdkp->physical_block_size = (1 << (buffer[13] & 0xf)) * sector_size;
2700  
2701  	/* RC basis */
2702  	sdkp->rc_basis = (buffer[12] >> 4) & 0x3;
2703  
2704  	/* Lowest aligned logical block */
2705  	alignment = ((buffer[14] & 0x3f) << 8 | buffer[15]) * sector_size;
2706  	lim->alignment_offset = alignment;
2707  	if (alignment && sdkp->first_scan)
2708  		sd_printk(KERN_NOTICE, sdkp,
2709  			  "physical block alignment offset: %u\n", alignment);
2710  
2711  	if (buffer[14] & 0x80) { /* LBPME */
2712  		sdkp->lbpme = 1;
2713  
2714  		if (buffer[14] & 0x40) /* LBPRZ */
2715  			sdkp->lbprz = 1;
2716  	}
2717  
2718  	sdkp->capacity = lba + 1;
2719  	return sector_size;
2720  }
2721  
read_capacity_10(struct scsi_disk * sdkp,struct scsi_device * sdp,unsigned char * buffer)2722  static int read_capacity_10(struct scsi_disk *sdkp, struct scsi_device *sdp,
2723  						unsigned char *buffer)
2724  {
2725  	static const u8 cmd[10] = { READ_CAPACITY };
2726  	struct scsi_sense_hdr sshdr;
2727  	struct scsi_failure failure_defs[] = {
2728  		/* Do not retry Medium Not Present */
2729  		{
2730  			.sense = UNIT_ATTENTION,
2731  			.asc = 0x3A,
2732  			.result = SAM_STAT_CHECK_CONDITION,
2733  		},
2734  		{
2735  			.sense = NOT_READY,
2736  			.asc = 0x3A,
2737  			.result = SAM_STAT_CHECK_CONDITION,
2738  		},
2739  		 /* Device reset might occur several times so retry a lot */
2740  		{
2741  			.sense = UNIT_ATTENTION,
2742  			.asc = 0x29,
2743  			.allowed = READ_CAPACITY_RETRIES_ON_RESET,
2744  			.result = SAM_STAT_CHECK_CONDITION,
2745  		},
2746  		/* Any other error not listed above retry 3 times */
2747  		{
2748  			.result = SCMD_FAILURE_RESULT_ANY,
2749  			.allowed = 3,
2750  		},
2751  		{}
2752  	};
2753  	struct scsi_failures failures = {
2754  		.failure_definitions = failure_defs,
2755  	};
2756  	const struct scsi_exec_args exec_args = {
2757  		.sshdr = &sshdr,
2758  		.failures = &failures,
2759  	};
2760  	int sense_valid = 0;
2761  	int the_result;
2762  	sector_t lba;
2763  	unsigned sector_size;
2764  
2765  	memset(buffer, 0, 8);
2766  
2767  	the_result = scsi_execute_cmd(sdp, cmd, REQ_OP_DRV_IN, buffer,
2768  				      8, SD_TIMEOUT, sdkp->max_retries,
2769  				      &exec_args);
2770  
2771  	if (the_result > 0) {
2772  		sense_valid = scsi_sense_valid(&sshdr);
2773  
2774  		if (media_not_present(sdkp, &sshdr))
2775  			return -ENODEV;
2776  	}
2777  
2778  	if (the_result) {
2779  		sd_print_result(sdkp, "Read Capacity(10) failed", the_result);
2780  		read_capacity_error(sdkp, sdp, &sshdr, sense_valid, the_result);
2781  		return -EINVAL;
2782  	}
2783  
2784  	sector_size = get_unaligned_be32(&buffer[4]);
2785  	lba = get_unaligned_be32(&buffer[0]);
2786  
2787  	if (sdp->no_read_capacity_16 && (lba == 0xffffffff)) {
2788  		/* Some buggy (usb cardreader) devices return an lba of
2789  		   0xffffffff when the want to report a size of 0 (with
2790  		   which they really mean no media is present) */
2791  		sdkp->capacity = 0;
2792  		sdkp->physical_block_size = sector_size;
2793  		return sector_size;
2794  	}
2795  
2796  	sdkp->capacity = lba + 1;
2797  	sdkp->physical_block_size = sector_size;
2798  	return sector_size;
2799  }
2800  
sd_try_rc16_first(struct scsi_device * sdp)2801  static int sd_try_rc16_first(struct scsi_device *sdp)
2802  {
2803  	if (sdp->host->max_cmd_len < 16)
2804  		return 0;
2805  	if (sdp->try_rc_10_first)
2806  		return 0;
2807  	if (sdp->scsi_level > SCSI_SPC_2)
2808  		return 1;
2809  	if (scsi_device_protection(sdp))
2810  		return 1;
2811  	return 0;
2812  }
2813  
2814  /*
2815   * read disk capacity
2816   */
2817  static void
sd_read_capacity(struct scsi_disk * sdkp,struct queue_limits * lim,unsigned char * buffer)2818  sd_read_capacity(struct scsi_disk *sdkp, struct queue_limits *lim,
2819  		unsigned char *buffer)
2820  {
2821  	int sector_size;
2822  	struct scsi_device *sdp = sdkp->device;
2823  
2824  	if (sd_try_rc16_first(sdp)) {
2825  		sector_size = read_capacity_16(sdkp, sdp, lim, buffer);
2826  		if (sector_size == -EOVERFLOW)
2827  			goto got_data;
2828  		if (sector_size == -ENODEV)
2829  			return;
2830  		if (sector_size < 0)
2831  			sector_size = read_capacity_10(sdkp, sdp, buffer);
2832  		if (sector_size < 0)
2833  			return;
2834  	} else {
2835  		sector_size = read_capacity_10(sdkp, sdp, buffer);
2836  		if (sector_size == -EOVERFLOW)
2837  			goto got_data;
2838  		if (sector_size < 0)
2839  			return;
2840  		if ((sizeof(sdkp->capacity) > 4) &&
2841  		    (sdkp->capacity > 0xffffffffULL)) {
2842  			int old_sector_size = sector_size;
2843  			sd_printk(KERN_NOTICE, sdkp, "Very big device. "
2844  					"Trying to use READ CAPACITY(16).\n");
2845  			sector_size = read_capacity_16(sdkp, sdp, lim, buffer);
2846  			if (sector_size < 0) {
2847  				sd_printk(KERN_NOTICE, sdkp,
2848  					"Using 0xffffffff as device size\n");
2849  				sdkp->capacity = 1 + (sector_t) 0xffffffff;
2850  				sector_size = old_sector_size;
2851  				goto got_data;
2852  			}
2853  			/* Remember that READ CAPACITY(16) succeeded */
2854  			sdp->try_rc_10_first = 0;
2855  		}
2856  	}
2857  
2858  	/* Some devices are known to return the total number of blocks,
2859  	 * not the highest block number.  Some devices have versions
2860  	 * which do this and others which do not.  Some devices we might
2861  	 * suspect of doing this but we don't know for certain.
2862  	 *
2863  	 * If we know the reported capacity is wrong, decrement it.  If
2864  	 * we can only guess, then assume the number of blocks is even
2865  	 * (usually true but not always) and err on the side of lowering
2866  	 * the capacity.
2867  	 */
2868  	if (sdp->fix_capacity ||
2869  	    (sdp->guess_capacity && (sdkp->capacity & 0x01))) {
2870  		sd_printk(KERN_INFO, sdkp, "Adjusting the sector count "
2871  				"from its reported value: %llu\n",
2872  				(unsigned long long) sdkp->capacity);
2873  		--sdkp->capacity;
2874  	}
2875  
2876  got_data:
2877  	if (sector_size == 0) {
2878  		sector_size = 512;
2879  		sd_printk(KERN_NOTICE, sdkp, "Sector size 0 reported, "
2880  			  "assuming 512.\n");
2881  	}
2882  
2883  	if (sector_size != 512 &&
2884  	    sector_size != 1024 &&
2885  	    sector_size != 2048 &&
2886  	    sector_size != 4096) {
2887  		sd_printk(KERN_NOTICE, sdkp, "Unsupported sector size %d.\n",
2888  			  sector_size);
2889  		/*
2890  		 * The user might want to re-format the drive with
2891  		 * a supported sectorsize.  Once this happens, it
2892  		 * would be relatively trivial to set the thing up.
2893  		 * For this reason, we leave the thing in the table.
2894  		 */
2895  		sdkp->capacity = 0;
2896  		/*
2897  		 * set a bogus sector size so the normal read/write
2898  		 * logic in the block layer will eventually refuse any
2899  		 * request on this device without tripping over power
2900  		 * of two sector size assumptions
2901  		 */
2902  		sector_size = 512;
2903  	}
2904  	lim->logical_block_size = sector_size;
2905  	lim->physical_block_size = sdkp->physical_block_size;
2906  	sdkp->device->sector_size = sector_size;
2907  
2908  	if (sdkp->capacity > 0xffffffff)
2909  		sdp->use_16_for_rw = 1;
2910  
2911  }
2912  
2913  /*
2914   * Print disk capacity
2915   */
2916  static void
sd_print_capacity(struct scsi_disk * sdkp,sector_t old_capacity)2917  sd_print_capacity(struct scsi_disk *sdkp,
2918  		  sector_t old_capacity)
2919  {
2920  	int sector_size = sdkp->device->sector_size;
2921  	char cap_str_2[10], cap_str_10[10];
2922  
2923  	if (!sdkp->first_scan && old_capacity == sdkp->capacity)
2924  		return;
2925  
2926  	string_get_size(sdkp->capacity, sector_size,
2927  			STRING_UNITS_2, cap_str_2, sizeof(cap_str_2));
2928  	string_get_size(sdkp->capacity, sector_size,
2929  			STRING_UNITS_10, cap_str_10, sizeof(cap_str_10));
2930  
2931  	sd_printk(KERN_NOTICE, sdkp,
2932  		  "%llu %d-byte logical blocks: (%s/%s)\n",
2933  		  (unsigned long long)sdkp->capacity,
2934  		  sector_size, cap_str_10, cap_str_2);
2935  
2936  	if (sdkp->physical_block_size != sector_size)
2937  		sd_printk(KERN_NOTICE, sdkp,
2938  			  "%u-byte physical blocks\n",
2939  			  sdkp->physical_block_size);
2940  }
2941  
2942  /* called with buffer of length 512 */
2943  static inline int
sd_do_mode_sense(struct scsi_disk * sdkp,int dbd,int modepage,unsigned char * buffer,int len,struct scsi_mode_data * data,struct scsi_sense_hdr * sshdr)2944  sd_do_mode_sense(struct scsi_disk *sdkp, int dbd, int modepage,
2945  		 unsigned char *buffer, int len, struct scsi_mode_data *data,
2946  		 struct scsi_sense_hdr *sshdr)
2947  {
2948  	/*
2949  	 * If we must use MODE SENSE(10), make sure that the buffer length
2950  	 * is at least 8 bytes so that the mode sense header fits.
2951  	 */
2952  	if (sdkp->device->use_10_for_ms && len < 8)
2953  		len = 8;
2954  
2955  	return scsi_mode_sense(sdkp->device, dbd, modepage, 0, buffer, len,
2956  			       SD_TIMEOUT, sdkp->max_retries, data, sshdr);
2957  }
2958  
2959  /*
2960   * read write protect setting, if possible - called only in sd_revalidate_disk()
2961   * called with buffer of length SD_BUF_SIZE
2962   */
2963  static void
sd_read_write_protect_flag(struct scsi_disk * sdkp,unsigned char * buffer)2964  sd_read_write_protect_flag(struct scsi_disk *sdkp, unsigned char *buffer)
2965  {
2966  	int res;
2967  	struct scsi_device *sdp = sdkp->device;
2968  	struct scsi_mode_data data;
2969  	int old_wp = sdkp->write_prot;
2970  
2971  	set_disk_ro(sdkp->disk, 0);
2972  	if (sdp->skip_ms_page_3f) {
2973  		sd_first_printk(KERN_NOTICE, sdkp, "Assuming Write Enabled\n");
2974  		return;
2975  	}
2976  
2977  	if (sdp->use_192_bytes_for_3f) {
2978  		res = sd_do_mode_sense(sdkp, 0, 0x3F, buffer, 192, &data, NULL);
2979  	} else {
2980  		/*
2981  		 * First attempt: ask for all pages (0x3F), but only 4 bytes.
2982  		 * We have to start carefully: some devices hang if we ask
2983  		 * for more than is available.
2984  		 */
2985  		res = sd_do_mode_sense(sdkp, 0, 0x3F, buffer, 4, &data, NULL);
2986  
2987  		/*
2988  		 * Second attempt: ask for page 0 When only page 0 is
2989  		 * implemented, a request for page 3F may return Sense Key
2990  		 * 5: Illegal Request, Sense Code 24: Invalid field in
2991  		 * CDB.
2992  		 */
2993  		if (res < 0)
2994  			res = sd_do_mode_sense(sdkp, 0, 0, buffer, 4, &data, NULL);
2995  
2996  		/*
2997  		 * Third attempt: ask 255 bytes, as we did earlier.
2998  		 */
2999  		if (res < 0)
3000  			res = sd_do_mode_sense(sdkp, 0, 0x3F, buffer, 255,
3001  					       &data, NULL);
3002  	}
3003  
3004  	if (res < 0) {
3005  		sd_first_printk(KERN_WARNING, sdkp,
3006  			  "Test WP failed, assume Write Enabled\n");
3007  	} else {
3008  		sdkp->write_prot = ((data.device_specific & 0x80) != 0);
3009  		set_disk_ro(sdkp->disk, sdkp->write_prot);
3010  		if (sdkp->first_scan || old_wp != sdkp->write_prot) {
3011  			sd_printk(KERN_NOTICE, sdkp, "Write Protect is %s\n",
3012  				  sdkp->write_prot ? "on" : "off");
3013  			sd_printk(KERN_DEBUG, sdkp, "Mode Sense: %4ph\n", buffer);
3014  		}
3015  	}
3016  }
3017  
3018  /*
3019   * sd_read_cache_type - called only from sd_revalidate_disk()
3020   * called with buffer of length SD_BUF_SIZE
3021   */
3022  static void
sd_read_cache_type(struct scsi_disk * sdkp,unsigned char * buffer)3023  sd_read_cache_type(struct scsi_disk *sdkp, unsigned char *buffer)
3024  {
3025  	int len = 0, res;
3026  	struct scsi_device *sdp = sdkp->device;
3027  
3028  	int dbd;
3029  	int modepage;
3030  	int first_len;
3031  	struct scsi_mode_data data;
3032  	struct scsi_sense_hdr sshdr;
3033  	int old_wce = sdkp->WCE;
3034  	int old_rcd = sdkp->RCD;
3035  	int old_dpofua = sdkp->DPOFUA;
3036  
3037  
3038  	if (sdkp->cache_override)
3039  		return;
3040  
3041  	first_len = 4;
3042  	if (sdp->skip_ms_page_8) {
3043  		if (sdp->type == TYPE_RBC)
3044  			goto defaults;
3045  		else {
3046  			if (sdp->skip_ms_page_3f)
3047  				goto defaults;
3048  			modepage = 0x3F;
3049  			if (sdp->use_192_bytes_for_3f)
3050  				first_len = 192;
3051  			dbd = 0;
3052  		}
3053  	} else if (sdp->type == TYPE_RBC) {
3054  		modepage = 6;
3055  		dbd = 8;
3056  	} else {
3057  		modepage = 8;
3058  		dbd = 0;
3059  	}
3060  
3061  	/* cautiously ask */
3062  	res = sd_do_mode_sense(sdkp, dbd, modepage, buffer, first_len,
3063  			&data, &sshdr);
3064  
3065  	if (res < 0)
3066  		goto bad_sense;
3067  
3068  	if (!data.header_length) {
3069  		modepage = 6;
3070  		first_len = 0;
3071  		sd_first_printk(KERN_ERR, sdkp,
3072  				"Missing header in MODE_SENSE response\n");
3073  	}
3074  
3075  	/* that went OK, now ask for the proper length */
3076  	len = data.length;
3077  
3078  	/*
3079  	 * We're only interested in the first three bytes, actually.
3080  	 * But the data cache page is defined for the first 20.
3081  	 */
3082  	if (len < 3)
3083  		goto bad_sense;
3084  	else if (len > SD_BUF_SIZE) {
3085  		sd_first_printk(KERN_NOTICE, sdkp, "Truncating mode parameter "
3086  			  "data from %d to %d bytes\n", len, SD_BUF_SIZE);
3087  		len = SD_BUF_SIZE;
3088  	}
3089  	if (modepage == 0x3F && sdp->use_192_bytes_for_3f)
3090  		len = 192;
3091  
3092  	/* Get the data */
3093  	if (len > first_len)
3094  		res = sd_do_mode_sense(sdkp, dbd, modepage, buffer, len,
3095  				&data, &sshdr);
3096  
3097  	if (!res) {
3098  		int offset = data.header_length + data.block_descriptor_length;
3099  
3100  		while (offset < len) {
3101  			u8 page_code = buffer[offset] & 0x3F;
3102  			u8 spf       = buffer[offset] & 0x40;
3103  
3104  			if (page_code == 8 || page_code == 6) {
3105  				/* We're interested only in the first 3 bytes.
3106  				 */
3107  				if (len - offset <= 2) {
3108  					sd_first_printk(KERN_ERR, sdkp,
3109  						"Incomplete mode parameter "
3110  							"data\n");
3111  					goto defaults;
3112  				} else {
3113  					modepage = page_code;
3114  					goto Page_found;
3115  				}
3116  			} else {
3117  				/* Go to the next page */
3118  				if (spf && len - offset > 3)
3119  					offset += 4 + (buffer[offset+2] << 8) +
3120  						buffer[offset+3];
3121  				else if (!spf && len - offset > 1)
3122  					offset += 2 + buffer[offset+1];
3123  				else {
3124  					sd_first_printk(KERN_ERR, sdkp,
3125  							"Incomplete mode "
3126  							"parameter data\n");
3127  					goto defaults;
3128  				}
3129  			}
3130  		}
3131  
3132  		sd_first_printk(KERN_WARNING, sdkp,
3133  				"No Caching mode page found\n");
3134  		goto defaults;
3135  
3136  	Page_found:
3137  		if (modepage == 8) {
3138  			sdkp->WCE = ((buffer[offset + 2] & 0x04) != 0);
3139  			sdkp->RCD = ((buffer[offset + 2] & 0x01) != 0);
3140  		} else {
3141  			sdkp->WCE = ((buffer[offset + 2] & 0x01) == 0);
3142  			sdkp->RCD = 0;
3143  		}
3144  
3145  		sdkp->DPOFUA = (data.device_specific & 0x10) != 0;
3146  		if (sdp->broken_fua) {
3147  			sd_first_printk(KERN_NOTICE, sdkp, "Disabling FUA\n");
3148  			sdkp->DPOFUA = 0;
3149  		} else if (sdkp->DPOFUA && !sdkp->device->use_10_for_rw &&
3150  			   !sdkp->device->use_16_for_rw) {
3151  			sd_first_printk(KERN_NOTICE, sdkp,
3152  				  "Uses READ/WRITE(6), disabling FUA\n");
3153  			sdkp->DPOFUA = 0;
3154  		}
3155  
3156  		/* No cache flush allowed for write protected devices */
3157  		if (sdkp->WCE && sdkp->write_prot)
3158  			sdkp->WCE = 0;
3159  
3160  		if (sdkp->first_scan || old_wce != sdkp->WCE ||
3161  		    old_rcd != sdkp->RCD || old_dpofua != sdkp->DPOFUA)
3162  			sd_printk(KERN_NOTICE, sdkp,
3163  				  "Write cache: %s, read cache: %s, %s\n",
3164  				  sdkp->WCE ? "enabled" : "disabled",
3165  				  sdkp->RCD ? "disabled" : "enabled",
3166  				  sdkp->DPOFUA ? "supports DPO and FUA"
3167  				  : "doesn't support DPO or FUA");
3168  
3169  		return;
3170  	}
3171  
3172  bad_sense:
3173  	if (res == -EIO && scsi_sense_valid(&sshdr) &&
3174  	    sshdr.sense_key == ILLEGAL_REQUEST &&
3175  	    sshdr.asc == 0x24 && sshdr.ascq == 0x0)
3176  		/* Invalid field in CDB */
3177  		sd_first_printk(KERN_NOTICE, sdkp, "Cache data unavailable\n");
3178  	else
3179  		sd_first_printk(KERN_ERR, sdkp,
3180  				"Asking for cache data failed\n");
3181  
3182  defaults:
3183  	if (sdp->wce_default_on) {
3184  		sd_first_printk(KERN_NOTICE, sdkp,
3185  				"Assuming drive cache: write back\n");
3186  		sdkp->WCE = 1;
3187  	} else {
3188  		sd_first_printk(KERN_WARNING, sdkp,
3189  				"Assuming drive cache: write through\n");
3190  		sdkp->WCE = 0;
3191  	}
3192  	sdkp->RCD = 0;
3193  	sdkp->DPOFUA = 0;
3194  }
3195  
sd_is_perm_stream(struct scsi_disk * sdkp,unsigned int stream_id)3196  static bool sd_is_perm_stream(struct scsi_disk *sdkp, unsigned int stream_id)
3197  {
3198  	u8 cdb[16] = { SERVICE_ACTION_IN_16, SAI_GET_STREAM_STATUS };
3199  	struct {
3200  		struct scsi_stream_status_header h;
3201  		struct scsi_stream_status s;
3202  	} buf;
3203  	struct scsi_device *sdev = sdkp->device;
3204  	struct scsi_sense_hdr sshdr;
3205  	const struct scsi_exec_args exec_args = {
3206  		.sshdr = &sshdr,
3207  	};
3208  	int res;
3209  
3210  	put_unaligned_be16(stream_id, &cdb[4]);
3211  	put_unaligned_be32(sizeof(buf), &cdb[10]);
3212  
3213  	res = scsi_execute_cmd(sdev, cdb, REQ_OP_DRV_IN, &buf, sizeof(buf),
3214  			       SD_TIMEOUT, sdkp->max_retries, &exec_args);
3215  	if (res < 0)
3216  		return false;
3217  	if (scsi_status_is_check_condition(res) && scsi_sense_valid(&sshdr))
3218  		sd_print_sense_hdr(sdkp, &sshdr);
3219  	if (res)
3220  		return false;
3221  	if (get_unaligned_be32(&buf.h.len) < sizeof(struct scsi_stream_status))
3222  		return false;
3223  	return buf.s.perm;
3224  }
3225  
sd_read_io_hints(struct scsi_disk * sdkp,unsigned char * buffer)3226  static void sd_read_io_hints(struct scsi_disk *sdkp, unsigned char *buffer)
3227  {
3228  	struct scsi_device *sdp = sdkp->device;
3229  	const struct scsi_io_group_descriptor *desc, *start, *end;
3230  	u16 permanent_stream_count_old;
3231  	struct scsi_sense_hdr sshdr;
3232  	struct scsi_mode_data data;
3233  	int res;
3234  
3235  	if (sdp->sdev_bflags & BLIST_SKIP_IO_HINTS)
3236  		return;
3237  
3238  	res = scsi_mode_sense(sdp, /*dbd=*/0x8, /*modepage=*/0x0a,
3239  			      /*subpage=*/0x05, buffer, SD_BUF_SIZE, SD_TIMEOUT,
3240  			      sdkp->max_retries, &data, &sshdr);
3241  	if (res < 0)
3242  		return;
3243  	start = (void *)buffer + data.header_length + 16;
3244  	end = (void *)buffer + ALIGN_DOWN(data.header_length + data.length,
3245  					  sizeof(*end));
3246  	/*
3247  	 * From "SBC-5 Constrained Streams with Data Lifetimes": Device severs
3248  	 * should assign the lowest numbered stream identifiers to permanent
3249  	 * streams.
3250  	 */
3251  	for (desc = start; desc < end; desc++)
3252  		if (!desc->st_enble || !sd_is_perm_stream(sdkp, desc - start))
3253  			break;
3254  	permanent_stream_count_old = sdkp->permanent_stream_count;
3255  	sdkp->permanent_stream_count = desc - start;
3256  	if (sdkp->rscs && sdkp->permanent_stream_count < 2)
3257  		sd_printk(KERN_INFO, sdkp,
3258  			  "Unexpected: RSCS has been set and the permanent stream count is %u\n",
3259  			  sdkp->permanent_stream_count);
3260  	else if (sdkp->permanent_stream_count != permanent_stream_count_old)
3261  		sd_printk(KERN_INFO, sdkp, "permanent stream count = %d\n",
3262  			  sdkp->permanent_stream_count);
3263  }
3264  
3265  /*
3266   * The ATO bit indicates whether the DIF application tag is available
3267   * for use by the operating system.
3268   */
sd_read_app_tag_own(struct scsi_disk * sdkp,unsigned char * buffer)3269  static void sd_read_app_tag_own(struct scsi_disk *sdkp, unsigned char *buffer)
3270  {
3271  	int res, offset;
3272  	struct scsi_device *sdp = sdkp->device;
3273  	struct scsi_mode_data data;
3274  	struct scsi_sense_hdr sshdr;
3275  
3276  	if (sdp->type != TYPE_DISK && sdp->type != TYPE_ZBC)
3277  		return;
3278  
3279  	if (sdkp->protection_type == 0)
3280  		return;
3281  
3282  	res = scsi_mode_sense(sdp, 1, 0x0a, 0, buffer, 36, SD_TIMEOUT,
3283  			      sdkp->max_retries, &data, &sshdr);
3284  
3285  	if (res < 0 || !data.header_length ||
3286  	    data.length < 6) {
3287  		sd_first_printk(KERN_WARNING, sdkp,
3288  			  "getting Control mode page failed, assume no ATO\n");
3289  
3290  		if (res == -EIO && scsi_sense_valid(&sshdr))
3291  			sd_print_sense_hdr(sdkp, &sshdr);
3292  
3293  		return;
3294  	}
3295  
3296  	offset = data.header_length + data.block_descriptor_length;
3297  
3298  	if ((buffer[offset] & 0x3f) != 0x0a) {
3299  		sd_first_printk(KERN_ERR, sdkp, "ATO Got wrong page\n");
3300  		return;
3301  	}
3302  
3303  	if ((buffer[offset + 5] & 0x80) == 0)
3304  		return;
3305  
3306  	sdkp->ATO = 1;
3307  
3308  	return;
3309  }
3310  
sd_discard_mode(struct scsi_disk * sdkp)3311  static unsigned int sd_discard_mode(struct scsi_disk *sdkp)
3312  {
3313  	if (!sdkp->lbpme)
3314  		return SD_LBP_FULL;
3315  
3316  	if (!sdkp->lbpvpd) {
3317  		/* LBP VPD page not provided */
3318  		if (sdkp->max_unmap_blocks)
3319  			return SD_LBP_UNMAP;
3320  		return SD_LBP_WS16;
3321  	}
3322  
3323  	/* LBP VPD page tells us what to use */
3324  	if (sdkp->lbpu && sdkp->max_unmap_blocks)
3325  		return SD_LBP_UNMAP;
3326  	if (sdkp->lbpws)
3327  		return SD_LBP_WS16;
3328  	if (sdkp->lbpws10)
3329  		return SD_LBP_WS10;
3330  	return SD_LBP_DISABLE;
3331  }
3332  
3333  /*
3334   * Query disk device for preferred I/O sizes.
3335   */
sd_read_block_limits(struct scsi_disk * sdkp,struct queue_limits * lim)3336  static void sd_read_block_limits(struct scsi_disk *sdkp,
3337  		struct queue_limits *lim)
3338  {
3339  	struct scsi_vpd *vpd;
3340  
3341  	rcu_read_lock();
3342  
3343  	vpd = rcu_dereference(sdkp->device->vpd_pgb0);
3344  	if (!vpd || vpd->len < 16)
3345  		goto out;
3346  
3347  	sdkp->min_xfer_blocks = get_unaligned_be16(&vpd->data[6]);
3348  	sdkp->max_xfer_blocks = get_unaligned_be32(&vpd->data[8]);
3349  	sdkp->opt_xfer_blocks = get_unaligned_be32(&vpd->data[12]);
3350  
3351  	if (vpd->len >= 64) {
3352  		unsigned int lba_count, desc_count;
3353  
3354  		sdkp->max_ws_blocks = (u32)get_unaligned_be64(&vpd->data[36]);
3355  
3356  		if (!sdkp->lbpme)
3357  			goto config_atomic;
3358  
3359  		lba_count = get_unaligned_be32(&vpd->data[20]);
3360  		desc_count = get_unaligned_be32(&vpd->data[24]);
3361  
3362  		if (lba_count && desc_count)
3363  			sdkp->max_unmap_blocks = lba_count;
3364  
3365  		sdkp->unmap_granularity = get_unaligned_be32(&vpd->data[28]);
3366  
3367  		if (vpd->data[32] & 0x80)
3368  			sdkp->unmap_alignment =
3369  				get_unaligned_be32(&vpd->data[32]) & ~(1 << 31);
3370  
3371  config_atomic:
3372  		sdkp->max_atomic = get_unaligned_be32(&vpd->data[44]);
3373  		sdkp->atomic_alignment = get_unaligned_be32(&vpd->data[48]);
3374  		sdkp->atomic_granularity = get_unaligned_be32(&vpd->data[52]);
3375  		sdkp->max_atomic_with_boundary = get_unaligned_be32(&vpd->data[56]);
3376  		sdkp->max_atomic_boundary = get_unaligned_be32(&vpd->data[60]);
3377  
3378  		sd_config_atomic(sdkp, lim);
3379  	}
3380  
3381   out:
3382  	rcu_read_unlock();
3383  }
3384  
3385  /* Parse the Block Limits Extension VPD page (0xb7) */
sd_read_block_limits_ext(struct scsi_disk * sdkp)3386  static void sd_read_block_limits_ext(struct scsi_disk *sdkp)
3387  {
3388  	struct scsi_vpd *vpd;
3389  
3390  	rcu_read_lock();
3391  	vpd = rcu_dereference(sdkp->device->vpd_pgb7);
3392  	if (vpd && vpd->len >= 6)
3393  		sdkp->rscs = vpd->data[5] & 1;
3394  	rcu_read_unlock();
3395  }
3396  
3397  /* Query block device characteristics */
sd_read_block_characteristics(struct scsi_disk * sdkp,struct queue_limits * lim)3398  static void sd_read_block_characteristics(struct scsi_disk *sdkp,
3399  		struct queue_limits *lim)
3400  {
3401  	struct scsi_vpd *vpd;
3402  	u16 rot;
3403  
3404  	rcu_read_lock();
3405  	vpd = rcu_dereference(sdkp->device->vpd_pgb1);
3406  
3407  	if (!vpd || vpd->len <= 8) {
3408  		rcu_read_unlock();
3409  	        return;
3410  	}
3411  
3412  	rot = get_unaligned_be16(&vpd->data[4]);
3413  	sdkp->zoned = (vpd->data[8] >> 4) & 3;
3414  	rcu_read_unlock();
3415  
3416  	if (rot == 1)
3417  		lim->features &= ~(BLK_FEAT_ROTATIONAL | BLK_FEAT_ADD_RANDOM);
3418  
3419  	if (!sdkp->first_scan)
3420  		return;
3421  
3422  	if (sdkp->device->type == TYPE_ZBC)
3423  		sd_printk(KERN_NOTICE, sdkp, "Host-managed zoned block device\n");
3424  	else if (sdkp->zoned == 1)
3425  		sd_printk(KERN_NOTICE, sdkp, "Host-aware SMR disk used as regular disk\n");
3426  	else if (sdkp->zoned == 2)
3427  		sd_printk(KERN_NOTICE, sdkp, "Drive-managed SMR disk\n");
3428  }
3429  
3430  /**
3431   * sd_read_block_provisioning - Query provisioning VPD page
3432   * @sdkp: disk to query
3433   */
sd_read_block_provisioning(struct scsi_disk * sdkp)3434  static void sd_read_block_provisioning(struct scsi_disk *sdkp)
3435  {
3436  	struct scsi_vpd *vpd;
3437  
3438  	if (sdkp->lbpme == 0)
3439  		return;
3440  
3441  	rcu_read_lock();
3442  	vpd = rcu_dereference(sdkp->device->vpd_pgb2);
3443  
3444  	if (!vpd || vpd->len < 8) {
3445  		rcu_read_unlock();
3446  		return;
3447  	}
3448  
3449  	sdkp->lbpvpd	= 1;
3450  	sdkp->lbpu	= (vpd->data[5] >> 7) & 1; /* UNMAP */
3451  	sdkp->lbpws	= (vpd->data[5] >> 6) & 1; /* WRITE SAME(16) w/ UNMAP */
3452  	sdkp->lbpws10	= (vpd->data[5] >> 5) & 1; /* WRITE SAME(10) w/ UNMAP */
3453  	rcu_read_unlock();
3454  }
3455  
sd_read_write_same(struct scsi_disk * sdkp,unsigned char * buffer)3456  static void sd_read_write_same(struct scsi_disk *sdkp, unsigned char *buffer)
3457  {
3458  	struct scsi_device *sdev = sdkp->device;
3459  
3460  	if (sdev->host->no_write_same) {
3461  		sdev->no_write_same = 1;
3462  
3463  		return;
3464  	}
3465  
3466  	if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE, INQUIRY, 0) < 0) {
3467  		sdev->no_report_opcodes = 1;
3468  
3469  		/*
3470  		 * Disable WRITE SAME if REPORT SUPPORTED OPERATION CODES is
3471  		 * unsupported and this is an ATA device.
3472  		 */
3473  		if (sdev->is_ata)
3474  			sdev->no_write_same = 1;
3475  	}
3476  
3477  	if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE, WRITE_SAME_16, 0) == 1)
3478  		sdkp->ws16 = 1;
3479  
3480  	if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE, WRITE_SAME, 0) == 1)
3481  		sdkp->ws10 = 1;
3482  }
3483  
sd_read_security(struct scsi_disk * sdkp,unsigned char * buffer)3484  static void sd_read_security(struct scsi_disk *sdkp, unsigned char *buffer)
3485  {
3486  	struct scsi_device *sdev = sdkp->device;
3487  
3488  	if (!sdev->security_supported)
3489  		return;
3490  
3491  	if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE,
3492  			SECURITY_PROTOCOL_IN, 0) == 1 &&
3493  	    scsi_report_opcode(sdev, buffer, SD_BUF_SIZE,
3494  			SECURITY_PROTOCOL_OUT, 0) == 1)
3495  		sdkp->security = 1;
3496  }
3497  
sd64_to_sectors(struct scsi_disk * sdkp,u8 * buf)3498  static inline sector_t sd64_to_sectors(struct scsi_disk *sdkp, u8 *buf)
3499  {
3500  	return logical_to_sectors(sdkp->device, get_unaligned_be64(buf));
3501  }
3502  
3503  /**
3504   * sd_read_cpr - Query concurrent positioning ranges
3505   * @sdkp:	disk to query
3506   */
sd_read_cpr(struct scsi_disk * sdkp)3507  static void sd_read_cpr(struct scsi_disk *sdkp)
3508  {
3509  	struct blk_independent_access_ranges *iars = NULL;
3510  	unsigned char *buffer = NULL;
3511  	unsigned int nr_cpr = 0;
3512  	int i, vpd_len, buf_len = SD_BUF_SIZE;
3513  	u8 *desc;
3514  
3515  	/*
3516  	 * We need to have the capacity set first for the block layer to be
3517  	 * able to check the ranges.
3518  	 */
3519  	if (sdkp->first_scan)
3520  		return;
3521  
3522  	if (!sdkp->capacity)
3523  		goto out;
3524  
3525  	/*
3526  	 * Concurrent Positioning Ranges VPD: there can be at most 256 ranges,
3527  	 * leading to a maximum page size of 64 + 256*32 bytes.
3528  	 */
3529  	buf_len = 64 + 256*32;
3530  	buffer = kmalloc(buf_len, GFP_KERNEL);
3531  	if (!buffer || scsi_get_vpd_page(sdkp->device, 0xb9, buffer, buf_len))
3532  		goto out;
3533  
3534  	/* We must have at least a 64B header and one 32B range descriptor */
3535  	vpd_len = get_unaligned_be16(&buffer[2]) + 4;
3536  	if (vpd_len > buf_len || vpd_len < 64 + 32 || (vpd_len & 31)) {
3537  		sd_printk(KERN_ERR, sdkp,
3538  			  "Invalid Concurrent Positioning Ranges VPD page\n");
3539  		goto out;
3540  	}
3541  
3542  	nr_cpr = (vpd_len - 64) / 32;
3543  	if (nr_cpr == 1) {
3544  		nr_cpr = 0;
3545  		goto out;
3546  	}
3547  
3548  	iars = disk_alloc_independent_access_ranges(sdkp->disk, nr_cpr);
3549  	if (!iars) {
3550  		nr_cpr = 0;
3551  		goto out;
3552  	}
3553  
3554  	desc = &buffer[64];
3555  	for (i = 0; i < nr_cpr; i++, desc += 32) {
3556  		if (desc[0] != i) {
3557  			sd_printk(KERN_ERR, sdkp,
3558  				"Invalid Concurrent Positioning Range number\n");
3559  			nr_cpr = 0;
3560  			break;
3561  		}
3562  
3563  		iars->ia_range[i].sector = sd64_to_sectors(sdkp, desc + 8);
3564  		iars->ia_range[i].nr_sectors = sd64_to_sectors(sdkp, desc + 16);
3565  	}
3566  
3567  out:
3568  	disk_set_independent_access_ranges(sdkp->disk, iars);
3569  	if (nr_cpr && sdkp->nr_actuators != nr_cpr) {
3570  		sd_printk(KERN_NOTICE, sdkp,
3571  			  "%u concurrent positioning ranges\n", nr_cpr);
3572  		sdkp->nr_actuators = nr_cpr;
3573  	}
3574  
3575  	kfree(buffer);
3576  }
3577  
sd_validate_min_xfer_size(struct scsi_disk * sdkp)3578  static bool sd_validate_min_xfer_size(struct scsi_disk *sdkp)
3579  {
3580  	struct scsi_device *sdp = sdkp->device;
3581  	unsigned int min_xfer_bytes =
3582  		logical_to_bytes(sdp, sdkp->min_xfer_blocks);
3583  
3584  	if (sdkp->min_xfer_blocks == 0)
3585  		return false;
3586  
3587  	if (min_xfer_bytes & (sdkp->physical_block_size - 1)) {
3588  		sd_first_printk(KERN_WARNING, sdkp,
3589  				"Preferred minimum I/O size %u bytes not a " \
3590  				"multiple of physical block size (%u bytes)\n",
3591  				min_xfer_bytes, sdkp->physical_block_size);
3592  		sdkp->min_xfer_blocks = 0;
3593  		return false;
3594  	}
3595  
3596  	sd_first_printk(KERN_INFO, sdkp, "Preferred minimum I/O size %u bytes\n",
3597  			min_xfer_bytes);
3598  	return true;
3599  }
3600  
3601  /*
3602   * Determine the device's preferred I/O size for reads and writes
3603   * unless the reported value is unreasonably small, large, not a
3604   * multiple of the physical block size, or simply garbage.
3605   */
sd_validate_opt_xfer_size(struct scsi_disk * sdkp,unsigned int dev_max)3606  static bool sd_validate_opt_xfer_size(struct scsi_disk *sdkp,
3607  				      unsigned int dev_max)
3608  {
3609  	struct scsi_device *sdp = sdkp->device;
3610  	unsigned int opt_xfer_bytes =
3611  		logical_to_bytes(sdp, sdkp->opt_xfer_blocks);
3612  	unsigned int min_xfer_bytes =
3613  		logical_to_bytes(sdp, sdkp->min_xfer_blocks);
3614  
3615  	if (sdkp->opt_xfer_blocks == 0)
3616  		return false;
3617  
3618  	if (sdkp->opt_xfer_blocks > dev_max) {
3619  		sd_first_printk(KERN_WARNING, sdkp,
3620  				"Optimal transfer size %u logical blocks " \
3621  				"> dev_max (%u logical blocks)\n",
3622  				sdkp->opt_xfer_blocks, dev_max);
3623  		return false;
3624  	}
3625  
3626  	if (sdkp->opt_xfer_blocks > SD_DEF_XFER_BLOCKS) {
3627  		sd_first_printk(KERN_WARNING, sdkp,
3628  				"Optimal transfer size %u logical blocks " \
3629  				"> sd driver limit (%u logical blocks)\n",
3630  				sdkp->opt_xfer_blocks, SD_DEF_XFER_BLOCKS);
3631  		return false;
3632  	}
3633  
3634  	if (opt_xfer_bytes < PAGE_SIZE) {
3635  		sd_first_printk(KERN_WARNING, sdkp,
3636  				"Optimal transfer size %u bytes < " \
3637  				"PAGE_SIZE (%u bytes)\n",
3638  				opt_xfer_bytes, (unsigned int)PAGE_SIZE);
3639  		return false;
3640  	}
3641  
3642  	if (min_xfer_bytes && opt_xfer_bytes % min_xfer_bytes) {
3643  		sd_first_printk(KERN_WARNING, sdkp,
3644  				"Optimal transfer size %u bytes not a " \
3645  				"multiple of preferred minimum block " \
3646  				"size (%u bytes)\n",
3647  				opt_xfer_bytes, min_xfer_bytes);
3648  		return false;
3649  	}
3650  
3651  	if (opt_xfer_bytes & (sdkp->physical_block_size - 1)) {
3652  		sd_first_printk(KERN_WARNING, sdkp,
3653  				"Optimal transfer size %u bytes not a " \
3654  				"multiple of physical block size (%u bytes)\n",
3655  				opt_xfer_bytes, sdkp->physical_block_size);
3656  		return false;
3657  	}
3658  
3659  	sd_first_printk(KERN_INFO, sdkp, "Optimal transfer size %u bytes\n",
3660  			opt_xfer_bytes);
3661  	return true;
3662  }
3663  
sd_read_block_zero(struct scsi_disk * sdkp)3664  static void sd_read_block_zero(struct scsi_disk *sdkp)
3665  {
3666  	struct scsi_device *sdev = sdkp->device;
3667  	unsigned int buf_len = sdev->sector_size;
3668  	u8 *buffer, cmd[16] = { };
3669  
3670  	buffer = kmalloc(buf_len, GFP_KERNEL);
3671  	if (!buffer)
3672  		return;
3673  
3674  	if (sdev->use_16_for_rw) {
3675  		cmd[0] = READ_16;
3676  		put_unaligned_be64(0, &cmd[2]); /* Logical block address 0 */
3677  		put_unaligned_be32(1, &cmd[10]);/* Transfer 1 logical block */
3678  	} else {
3679  		cmd[0] = READ_10;
3680  		put_unaligned_be32(0, &cmd[2]); /* Logical block address 0 */
3681  		put_unaligned_be16(1, &cmd[7]);	/* Transfer 1 logical block */
3682  	}
3683  
3684  	scsi_execute_cmd(sdkp->device, cmd, REQ_OP_DRV_IN, buffer, buf_len,
3685  			 SD_TIMEOUT, sdkp->max_retries, NULL);
3686  	kfree(buffer);
3687  }
3688  
3689  /**
3690   *	sd_revalidate_disk - called the first time a new disk is seen,
3691   *	performs disk spin up, read_capacity, etc.
3692   *	@disk: struct gendisk we care about
3693   **/
sd_revalidate_disk(struct gendisk * disk)3694  static int sd_revalidate_disk(struct gendisk *disk)
3695  {
3696  	struct scsi_disk *sdkp = scsi_disk(disk);
3697  	struct scsi_device *sdp = sdkp->device;
3698  	sector_t old_capacity = sdkp->capacity;
3699  	struct queue_limits lim;
3700  	unsigned char *buffer;
3701  	unsigned int dev_max;
3702  	int err;
3703  
3704  	SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp,
3705  				      "sd_revalidate_disk\n"));
3706  
3707  	/*
3708  	 * If the device is offline, don't try and read capacity or any
3709  	 * of the other niceties.
3710  	 */
3711  	if (!scsi_device_online(sdp))
3712  		goto out;
3713  
3714  	buffer = kmalloc(SD_BUF_SIZE, GFP_KERNEL);
3715  	if (!buffer) {
3716  		sd_printk(KERN_WARNING, sdkp, "sd_revalidate_disk: Memory "
3717  			  "allocation failure.\n");
3718  		goto out;
3719  	}
3720  
3721  	sd_spinup_disk(sdkp);
3722  
3723  	lim = queue_limits_start_update(sdkp->disk->queue);
3724  
3725  	/*
3726  	 * Without media there is no reason to ask; moreover, some devices
3727  	 * react badly if we do.
3728  	 */
3729  	if (sdkp->media_present) {
3730  		sd_read_capacity(sdkp, &lim, buffer);
3731  		/*
3732  		 * Some USB/UAS devices return generic values for mode pages
3733  		 * until the media has been accessed. Trigger a READ operation
3734  		 * to force the device to populate mode pages.
3735  		 */
3736  		if (sdp->read_before_ms)
3737  			sd_read_block_zero(sdkp);
3738  		/*
3739  		 * set the default to rotational.  All non-rotational devices
3740  		 * support the block characteristics VPD page, which will
3741  		 * cause this to be updated correctly and any device which
3742  		 * doesn't support it should be treated as rotational.
3743  		 */
3744  		lim.features |= (BLK_FEAT_ROTATIONAL | BLK_FEAT_ADD_RANDOM);
3745  
3746  		if (scsi_device_supports_vpd(sdp)) {
3747  			sd_read_block_provisioning(sdkp);
3748  			sd_read_block_limits(sdkp, &lim);
3749  			sd_read_block_limits_ext(sdkp);
3750  			sd_read_block_characteristics(sdkp, &lim);
3751  			sd_zbc_read_zones(sdkp, &lim, buffer);
3752  		}
3753  
3754  		sd_config_discard(sdkp, &lim, sd_discard_mode(sdkp));
3755  
3756  		sd_print_capacity(sdkp, old_capacity);
3757  
3758  		sd_read_write_protect_flag(sdkp, buffer);
3759  		sd_read_cache_type(sdkp, buffer);
3760  		sd_read_io_hints(sdkp, buffer);
3761  		sd_read_app_tag_own(sdkp, buffer);
3762  		sd_read_write_same(sdkp, buffer);
3763  		sd_read_security(sdkp, buffer);
3764  		sd_config_protection(sdkp, &lim);
3765  	}
3766  
3767  	/*
3768  	 * We now have all cache related info, determine how we deal
3769  	 * with flush requests.
3770  	 */
3771  	sd_set_flush_flag(sdkp, &lim);
3772  
3773  	/* Initial block count limit based on CDB TRANSFER LENGTH field size. */
3774  	dev_max = sdp->use_16_for_rw ? SD_MAX_XFER_BLOCKS : SD_DEF_XFER_BLOCKS;
3775  
3776  	/* Some devices report a maximum block count for READ/WRITE requests. */
3777  	dev_max = min_not_zero(dev_max, sdkp->max_xfer_blocks);
3778  	lim.max_dev_sectors = logical_to_sectors(sdp, dev_max);
3779  
3780  	if (sd_validate_min_xfer_size(sdkp))
3781  		lim.io_min = logical_to_bytes(sdp, sdkp->min_xfer_blocks);
3782  	else
3783  		lim.io_min = 0;
3784  
3785  	/*
3786  	 * Limit default to SCSI host optimal sector limit if set. There may be
3787  	 * an impact on performance for when the size of a request exceeds this
3788  	 * host limit.
3789  	 */
3790  	lim.io_opt = sdp->host->opt_sectors << SECTOR_SHIFT;
3791  	if (sd_validate_opt_xfer_size(sdkp, dev_max)) {
3792  		lim.io_opt = min_not_zero(lim.io_opt,
3793  				logical_to_bytes(sdp, sdkp->opt_xfer_blocks));
3794  	}
3795  
3796  	sdkp->first_scan = 0;
3797  
3798  	set_capacity_and_notify(disk, logical_to_sectors(sdp, sdkp->capacity));
3799  	sd_config_write_same(sdkp, &lim);
3800  	kfree(buffer);
3801  
3802  	err = queue_limits_commit_update_frozen(sdkp->disk->queue, &lim);
3803  	if (err)
3804  		return err;
3805  
3806  	/*
3807  	 * Query concurrent positioning ranges after
3808  	 * queue_limits_commit_update() unlocked q->limits_lock to avoid
3809  	 * deadlock with q->sysfs_dir_lock and q->sysfs_lock.
3810  	 */
3811  	if (sdkp->media_present && scsi_device_supports_vpd(sdp))
3812  		sd_read_cpr(sdkp);
3813  
3814  	/*
3815  	 * For a zoned drive, revalidating the zones can be done only once
3816  	 * the gendisk capacity is set. So if this fails, set back the gendisk
3817  	 * capacity to 0.
3818  	 */
3819  	if (sd_zbc_revalidate_zones(sdkp))
3820  		set_capacity_and_notify(disk, 0);
3821  
3822   out:
3823  	return 0;
3824  }
3825  
3826  /**
3827   *	sd_unlock_native_capacity - unlock native capacity
3828   *	@disk: struct gendisk to set capacity for
3829   *
3830   *	Block layer calls this function if it detects that partitions
3831   *	on @disk reach beyond the end of the device.  If the SCSI host
3832   *	implements ->unlock_native_capacity() method, it's invoked to
3833   *	give it a chance to adjust the device capacity.
3834   *
3835   *	CONTEXT:
3836   *	Defined by block layer.  Might sleep.
3837   */
sd_unlock_native_capacity(struct gendisk * disk)3838  static void sd_unlock_native_capacity(struct gendisk *disk)
3839  {
3840  	struct scsi_device *sdev = scsi_disk(disk)->device;
3841  
3842  	if (sdev->host->hostt->unlock_native_capacity)
3843  		sdev->host->hostt->unlock_native_capacity(sdev);
3844  }
3845  
3846  /**
3847   *	sd_format_disk_name - format disk name
3848   *	@prefix: name prefix - ie. "sd" for SCSI disks
3849   *	@index: index of the disk to format name for
3850   *	@buf: output buffer
3851   *	@buflen: length of the output buffer
3852   *
3853   *	SCSI disk names starts at sda.  The 26th device is sdz and the
3854   *	27th is sdaa.  The last one for two lettered suffix is sdzz
3855   *	which is followed by sdaaa.
3856   *
3857   *	This is basically 26 base counting with one extra 'nil' entry
3858   *	at the beginning from the second digit on and can be
3859   *	determined using similar method as 26 base conversion with the
3860   *	index shifted -1 after each digit is computed.
3861   *
3862   *	CONTEXT:
3863   *	Don't care.
3864   *
3865   *	RETURNS:
3866   *	0 on success, -errno on failure.
3867   */
sd_format_disk_name(char * prefix,int index,char * buf,int buflen)3868  static int sd_format_disk_name(char *prefix, int index, char *buf, int buflen)
3869  {
3870  	const int base = 'z' - 'a' + 1;
3871  	char *begin = buf + strlen(prefix);
3872  	char *end = buf + buflen;
3873  	char *p;
3874  	int unit;
3875  
3876  	p = end - 1;
3877  	*p = '\0';
3878  	unit = base;
3879  	do {
3880  		if (p == begin)
3881  			return -EINVAL;
3882  		*--p = 'a' + (index % unit);
3883  		index = (index / unit) - 1;
3884  	} while (index >= 0);
3885  
3886  	memmove(begin, p, end - p);
3887  	memcpy(buf, prefix, strlen(prefix));
3888  
3889  	return 0;
3890  }
3891  
3892  /**
3893   *	sd_probe - called during driver initialization and whenever a
3894   *	new scsi device is attached to the system. It is called once
3895   *	for each scsi device (not just disks) present.
3896   *	@dev: pointer to device object
3897   *
3898   *	Returns 0 if successful (or not interested in this scsi device
3899   *	(e.g. scanner)); 1 when there is an error.
3900   *
3901   *	Note: this function is invoked from the scsi mid-level.
3902   *	This function sets up the mapping between a given
3903   *	<host,channel,id,lun> (found in sdp) and new device name
3904   *	(e.g. /dev/sda). More precisely it is the block device major
3905   *	and minor number that is chosen here.
3906   *
3907   *	Assume sd_probe is not re-entrant (for time being)
3908   *	Also think about sd_probe() and sd_remove() running coincidentally.
3909   **/
sd_probe(struct device * dev)3910  static int sd_probe(struct device *dev)
3911  {
3912  	struct scsi_device *sdp = to_scsi_device(dev);
3913  	struct scsi_disk *sdkp;
3914  	struct gendisk *gd;
3915  	int index;
3916  	int error;
3917  
3918  	scsi_autopm_get_device(sdp);
3919  	error = -ENODEV;
3920  	if (sdp->type != TYPE_DISK &&
3921  	    sdp->type != TYPE_ZBC &&
3922  	    sdp->type != TYPE_MOD &&
3923  	    sdp->type != TYPE_RBC)
3924  		goto out;
3925  
3926  	if (!IS_ENABLED(CONFIG_BLK_DEV_ZONED) && sdp->type == TYPE_ZBC) {
3927  		sdev_printk(KERN_WARNING, sdp,
3928  			    "Unsupported ZBC host-managed device.\n");
3929  		goto out;
3930  	}
3931  
3932  	SCSI_LOG_HLQUEUE(3, sdev_printk(KERN_INFO, sdp,
3933  					"sd_probe\n"));
3934  
3935  	error = -ENOMEM;
3936  	sdkp = kzalloc(sizeof(*sdkp), GFP_KERNEL);
3937  	if (!sdkp)
3938  		goto out;
3939  
3940  	gd = blk_mq_alloc_disk_for_queue(sdp->request_queue,
3941  					 &sd_bio_compl_lkclass);
3942  	if (!gd)
3943  		goto out_free;
3944  
3945  	index = ida_alloc(&sd_index_ida, GFP_KERNEL);
3946  	if (index < 0) {
3947  		sdev_printk(KERN_WARNING, sdp, "sd_probe: memory exhausted.\n");
3948  		goto out_put;
3949  	}
3950  
3951  	error = sd_format_disk_name("sd", index, gd->disk_name, DISK_NAME_LEN);
3952  	if (error) {
3953  		sdev_printk(KERN_WARNING, sdp, "SCSI disk (sd) name length exceeded.\n");
3954  		goto out_free_index;
3955  	}
3956  
3957  	sdkp->device = sdp;
3958  	sdkp->disk = gd;
3959  	sdkp->index = index;
3960  	sdkp->max_retries = SD_MAX_RETRIES;
3961  	atomic_set(&sdkp->openers, 0);
3962  	atomic_set(&sdkp->device->ioerr_cnt, 0);
3963  
3964  	if (!sdp->request_queue->rq_timeout) {
3965  		if (sdp->type != TYPE_MOD)
3966  			blk_queue_rq_timeout(sdp->request_queue, SD_TIMEOUT);
3967  		else
3968  			blk_queue_rq_timeout(sdp->request_queue,
3969  					     SD_MOD_TIMEOUT);
3970  	}
3971  
3972  	device_initialize(&sdkp->disk_dev);
3973  	sdkp->disk_dev.parent = get_device(dev);
3974  	sdkp->disk_dev.class = &sd_disk_class;
3975  	dev_set_name(&sdkp->disk_dev, "%s", dev_name(dev));
3976  
3977  	error = device_add(&sdkp->disk_dev);
3978  	if (error) {
3979  		put_device(&sdkp->disk_dev);
3980  		goto out;
3981  	}
3982  
3983  	dev_set_drvdata(dev, sdkp);
3984  
3985  	gd->major = sd_major((index & 0xf0) >> 4);
3986  	gd->first_minor = ((index & 0xf) << 4) | (index & 0xfff00);
3987  	gd->minors = SD_MINORS;
3988  
3989  	gd->fops = &sd_fops;
3990  	gd->private_data = sdkp;
3991  
3992  	/* defaults, until the device tells us otherwise */
3993  	sdp->sector_size = 512;
3994  	sdkp->capacity = 0;
3995  	sdkp->media_present = 1;
3996  	sdkp->write_prot = 0;
3997  	sdkp->cache_override = 0;
3998  	sdkp->WCE = 0;
3999  	sdkp->RCD = 0;
4000  	sdkp->ATO = 0;
4001  	sdkp->first_scan = 1;
4002  	sdkp->max_medium_access_timeouts = SD_MAX_MEDIUM_TIMEOUTS;
4003  
4004  	sd_revalidate_disk(gd);
4005  
4006  	if (sdp->removable) {
4007  		gd->flags |= GENHD_FL_REMOVABLE;
4008  		gd->events |= DISK_EVENT_MEDIA_CHANGE;
4009  		gd->event_flags = DISK_EVENT_FLAG_POLL | DISK_EVENT_FLAG_UEVENT;
4010  	}
4011  
4012  	blk_pm_runtime_init(sdp->request_queue, dev);
4013  	if (sdp->rpm_autosuspend) {
4014  		pm_runtime_set_autosuspend_delay(dev,
4015  			sdp->host->rpm_autosuspend_delay);
4016  	}
4017  
4018  	error = device_add_disk(dev, gd, NULL);
4019  	if (error) {
4020  		device_unregister(&sdkp->disk_dev);
4021  		put_disk(gd);
4022  		goto out;
4023  	}
4024  
4025  	if (sdkp->security) {
4026  		sdkp->opal_dev = init_opal_dev(sdkp, &sd_sec_submit);
4027  		if (sdkp->opal_dev)
4028  			sd_printk(KERN_NOTICE, sdkp, "supports TCG Opal\n");
4029  	}
4030  
4031  	sd_printk(KERN_NOTICE, sdkp, "Attached SCSI %sdisk\n",
4032  		  sdp->removable ? "removable " : "");
4033  	scsi_autopm_put_device(sdp);
4034  
4035  	return 0;
4036  
4037   out_free_index:
4038  	ida_free(&sd_index_ida, index);
4039   out_put:
4040  	put_disk(gd);
4041   out_free:
4042  	kfree(sdkp);
4043   out:
4044  	scsi_autopm_put_device(sdp);
4045  	return error;
4046  }
4047  
4048  /**
4049   *	sd_remove - called whenever a scsi disk (previously recognized by
4050   *	sd_probe) is detached from the system. It is called (potentially
4051   *	multiple times) during sd module unload.
4052   *	@dev: pointer to device object
4053   *
4054   *	Note: this function is invoked from the scsi mid-level.
4055   *	This function potentially frees up a device name (e.g. /dev/sdc)
4056   *	that could be re-used by a subsequent sd_probe().
4057   *	This function is not called when the built-in sd driver is "exit-ed".
4058   **/
sd_remove(struct device * dev)4059  static int sd_remove(struct device *dev)
4060  {
4061  	struct scsi_disk *sdkp = dev_get_drvdata(dev);
4062  
4063  	scsi_autopm_get_device(sdkp->device);
4064  
4065  	device_del(&sdkp->disk_dev);
4066  	del_gendisk(sdkp->disk);
4067  	if (!sdkp->suspended)
4068  		sd_shutdown(dev);
4069  
4070  	put_disk(sdkp->disk);
4071  	return 0;
4072  }
4073  
scsi_disk_release(struct device * dev)4074  static void scsi_disk_release(struct device *dev)
4075  {
4076  	struct scsi_disk *sdkp = to_scsi_disk(dev);
4077  
4078  	ida_free(&sd_index_ida, sdkp->index);
4079  	put_device(&sdkp->device->sdev_gendev);
4080  	free_opal_dev(sdkp->opal_dev);
4081  
4082  	kfree(sdkp);
4083  }
4084  
sd_start_stop_device(struct scsi_disk * sdkp,int start)4085  static int sd_start_stop_device(struct scsi_disk *sdkp, int start)
4086  {
4087  	unsigned char cmd[6] = { START_STOP };	/* START_VALID */
4088  	struct scsi_sense_hdr sshdr;
4089  	struct scsi_failure failure_defs[] = {
4090  		{
4091  			/* Power on, reset, or bus device reset occurred */
4092  			.sense = UNIT_ATTENTION,
4093  			.asc = 0x29,
4094  			.ascq = 0,
4095  			.result = SAM_STAT_CHECK_CONDITION,
4096  		},
4097  		{
4098  			/* Power on occurred */
4099  			.sense = UNIT_ATTENTION,
4100  			.asc = 0x29,
4101  			.ascq = 1,
4102  			.result = SAM_STAT_CHECK_CONDITION,
4103  		},
4104  		{
4105  			/* SCSI bus reset */
4106  			.sense = UNIT_ATTENTION,
4107  			.asc = 0x29,
4108  			.ascq = 2,
4109  			.result = SAM_STAT_CHECK_CONDITION,
4110  		},
4111  		{}
4112  	};
4113  	struct scsi_failures failures = {
4114  		.total_allowed = 3,
4115  		.failure_definitions = failure_defs,
4116  	};
4117  	const struct scsi_exec_args exec_args = {
4118  		.sshdr = &sshdr,
4119  		.req_flags = BLK_MQ_REQ_PM,
4120  		.failures = &failures,
4121  	};
4122  	struct scsi_device *sdp = sdkp->device;
4123  	int res;
4124  
4125  	if (start)
4126  		cmd[4] |= 1;	/* START */
4127  
4128  	if (sdp->start_stop_pwr_cond)
4129  		cmd[4] |= start ? 1 << 4 : 3 << 4;	/* Active or Standby */
4130  
4131  	if (!scsi_device_online(sdp))
4132  		return -ENODEV;
4133  
4134  	res = scsi_execute_cmd(sdp, cmd, REQ_OP_DRV_IN, NULL, 0, SD_TIMEOUT,
4135  			       sdkp->max_retries, &exec_args);
4136  	if (res) {
4137  		sd_print_result(sdkp, "Start/Stop Unit failed", res);
4138  		if (res > 0 && scsi_sense_valid(&sshdr)) {
4139  			sd_print_sense_hdr(sdkp, &sshdr);
4140  			/* 0x3a is medium not present */
4141  			if (sshdr.asc == 0x3a)
4142  				res = 0;
4143  		}
4144  	}
4145  
4146  	/* SCSI error codes must not go to the generic layer */
4147  	if (res)
4148  		return -EIO;
4149  
4150  	return 0;
4151  }
4152  
4153  /*
4154   * Send a SYNCHRONIZE CACHE instruction down to the device through
4155   * the normal SCSI command structure.  Wait for the command to
4156   * complete.
4157   */
sd_shutdown(struct device * dev)4158  static void sd_shutdown(struct device *dev)
4159  {
4160  	struct scsi_disk *sdkp = dev_get_drvdata(dev);
4161  
4162  	if (!sdkp)
4163  		return;         /* this can happen */
4164  
4165  	if (pm_runtime_suspended(dev))
4166  		return;
4167  
4168  	if (sdkp->WCE && sdkp->media_present) {
4169  		sd_printk(KERN_NOTICE, sdkp, "Synchronizing SCSI cache\n");
4170  		sd_sync_cache(sdkp);
4171  	}
4172  
4173  	if ((system_state != SYSTEM_RESTART &&
4174  	     sdkp->device->manage_system_start_stop) ||
4175  	    (system_state == SYSTEM_POWER_OFF &&
4176  	     sdkp->device->manage_shutdown) ||
4177  	    (system_state == SYSTEM_RUNNING &&
4178  	     sdkp->device->manage_runtime_start_stop)) {
4179  		sd_printk(KERN_NOTICE, sdkp, "Stopping disk\n");
4180  		sd_start_stop_device(sdkp, 0);
4181  	}
4182  }
4183  
sd_do_start_stop(struct scsi_device * sdev,bool runtime)4184  static inline bool sd_do_start_stop(struct scsi_device *sdev, bool runtime)
4185  {
4186  	return (sdev->manage_system_start_stop && !runtime) ||
4187  		(sdev->manage_runtime_start_stop && runtime);
4188  }
4189  
sd_suspend_common(struct device * dev,bool runtime)4190  static int sd_suspend_common(struct device *dev, bool runtime)
4191  {
4192  	struct scsi_disk *sdkp = dev_get_drvdata(dev);
4193  	int ret = 0;
4194  
4195  	if (!sdkp)	/* E.g.: runtime suspend following sd_remove() */
4196  		return 0;
4197  
4198  	if (sdkp->WCE && sdkp->media_present) {
4199  		if (!sdkp->device->silence_suspend)
4200  			sd_printk(KERN_NOTICE, sdkp, "Synchronizing SCSI cache\n");
4201  		ret = sd_sync_cache(sdkp);
4202  		/* ignore OFFLINE device */
4203  		if (ret == -ENODEV)
4204  			return 0;
4205  
4206  		if (ret)
4207  			return ret;
4208  	}
4209  
4210  	if (sd_do_start_stop(sdkp->device, runtime)) {
4211  		if (!sdkp->device->silence_suspend)
4212  			sd_printk(KERN_NOTICE, sdkp, "Stopping disk\n");
4213  		/* an error is not worth aborting a system sleep */
4214  		ret = sd_start_stop_device(sdkp, 0);
4215  		if (!runtime)
4216  			ret = 0;
4217  	}
4218  
4219  	if (!ret)
4220  		sdkp->suspended = true;
4221  
4222  	return ret;
4223  }
4224  
sd_suspend_system(struct device * dev)4225  static int sd_suspend_system(struct device *dev)
4226  {
4227  	if (pm_runtime_suspended(dev))
4228  		return 0;
4229  
4230  	return sd_suspend_common(dev, false);
4231  }
4232  
sd_suspend_runtime(struct device * dev)4233  static int sd_suspend_runtime(struct device *dev)
4234  {
4235  	return sd_suspend_common(dev, true);
4236  }
4237  
sd_resume(struct device * dev)4238  static int sd_resume(struct device *dev)
4239  {
4240  	struct scsi_disk *sdkp = dev_get_drvdata(dev);
4241  
4242  	sd_printk(KERN_NOTICE, sdkp, "Starting disk\n");
4243  
4244  	if (opal_unlock_from_suspend(sdkp->opal_dev)) {
4245  		sd_printk(KERN_NOTICE, sdkp, "OPAL unlock failed\n");
4246  		return -EIO;
4247  	}
4248  
4249  	return 0;
4250  }
4251  
sd_resume_common(struct device * dev,bool runtime)4252  static int sd_resume_common(struct device *dev, bool runtime)
4253  {
4254  	struct scsi_disk *sdkp = dev_get_drvdata(dev);
4255  	int ret;
4256  
4257  	if (!sdkp)	/* E.g.: runtime resume at the start of sd_probe() */
4258  		return 0;
4259  
4260  	if (!sd_do_start_stop(sdkp->device, runtime)) {
4261  		sdkp->suspended = false;
4262  		return 0;
4263  	}
4264  
4265  	sd_printk(KERN_NOTICE, sdkp, "Starting disk\n");
4266  	ret = sd_start_stop_device(sdkp, 1);
4267  	if (!ret) {
4268  		sd_resume(dev);
4269  		sdkp->suspended = false;
4270  	}
4271  
4272  	return ret;
4273  }
4274  
sd_resume_system(struct device * dev)4275  static int sd_resume_system(struct device *dev)
4276  {
4277  	if (pm_runtime_suspended(dev)) {
4278  		struct scsi_disk *sdkp = dev_get_drvdata(dev);
4279  		struct scsi_device *sdp = sdkp ? sdkp->device : NULL;
4280  
4281  		if (sdp && sdp->force_runtime_start_on_system_start)
4282  			pm_request_resume(dev);
4283  
4284  		return 0;
4285  	}
4286  
4287  	return sd_resume_common(dev, false);
4288  }
4289  
sd_resume_runtime(struct device * dev)4290  static int sd_resume_runtime(struct device *dev)
4291  {
4292  	struct scsi_disk *sdkp = dev_get_drvdata(dev);
4293  	struct scsi_device *sdp;
4294  
4295  	if (!sdkp)	/* E.g.: runtime resume at the start of sd_probe() */
4296  		return 0;
4297  
4298  	sdp = sdkp->device;
4299  
4300  	if (sdp->ignore_media_change) {
4301  		/* clear the device's sense data */
4302  		static const u8 cmd[10] = { REQUEST_SENSE };
4303  		const struct scsi_exec_args exec_args = {
4304  			.req_flags = BLK_MQ_REQ_PM,
4305  		};
4306  
4307  		if (scsi_execute_cmd(sdp, cmd, REQ_OP_DRV_IN, NULL, 0,
4308  				     sdp->request_queue->rq_timeout, 1,
4309  				     &exec_args))
4310  			sd_printk(KERN_NOTICE, sdkp,
4311  				  "Failed to clear sense data\n");
4312  	}
4313  
4314  	return sd_resume_common(dev, true);
4315  }
4316  
4317  static const struct dev_pm_ops sd_pm_ops = {
4318  	.suspend		= sd_suspend_system,
4319  	.resume			= sd_resume_system,
4320  	.poweroff		= sd_suspend_system,
4321  	.restore		= sd_resume_system,
4322  	.runtime_suspend	= sd_suspend_runtime,
4323  	.runtime_resume		= sd_resume_runtime,
4324  };
4325  
4326  static struct scsi_driver sd_template = {
4327  	.gendrv = {
4328  		.name		= "sd",
4329  		.probe		= sd_probe,
4330  		.probe_type	= PROBE_PREFER_ASYNCHRONOUS,
4331  		.remove		= sd_remove,
4332  		.shutdown	= sd_shutdown,
4333  		.pm		= &sd_pm_ops,
4334  	},
4335  	.rescan			= sd_rescan,
4336  	.resume			= sd_resume,
4337  	.init_command		= sd_init_command,
4338  	.uninit_command		= sd_uninit_command,
4339  	.done			= sd_done,
4340  	.eh_action		= sd_eh_action,
4341  	.eh_reset		= sd_eh_reset,
4342  };
4343  
4344  /**
4345   *	init_sd - entry point for this driver (both when built in or when
4346   *	a module).
4347   *
4348   *	Note: this function registers this driver with the scsi mid-level.
4349   **/
init_sd(void)4350  static int __init init_sd(void)
4351  {
4352  	int majors = 0, i, err;
4353  
4354  	SCSI_LOG_HLQUEUE(3, printk("init_sd: sd driver entry point\n"));
4355  
4356  	for (i = 0; i < SD_MAJORS; i++) {
4357  		if (__register_blkdev(sd_major(i), "sd", sd_default_probe))
4358  			continue;
4359  		majors++;
4360  	}
4361  
4362  	if (!majors)
4363  		return -ENODEV;
4364  
4365  	err = class_register(&sd_disk_class);
4366  	if (err)
4367  		goto err_out;
4368  
4369  	sd_page_pool = mempool_create_page_pool(SD_MEMPOOL_SIZE, 0);
4370  	if (!sd_page_pool) {
4371  		printk(KERN_ERR "sd: can't init discard page pool\n");
4372  		err = -ENOMEM;
4373  		goto err_out_class;
4374  	}
4375  
4376  	err = scsi_register_driver(&sd_template.gendrv);
4377  	if (err)
4378  		goto err_out_driver;
4379  
4380  	return 0;
4381  
4382  err_out_driver:
4383  	mempool_destroy(sd_page_pool);
4384  err_out_class:
4385  	class_unregister(&sd_disk_class);
4386  err_out:
4387  	for (i = 0; i < SD_MAJORS; i++)
4388  		unregister_blkdev(sd_major(i), "sd");
4389  	return err;
4390  }
4391  
4392  /**
4393   *	exit_sd - exit point for this driver (when it is a module).
4394   *
4395   *	Note: this function unregisters this driver from the scsi mid-level.
4396   **/
exit_sd(void)4397  static void __exit exit_sd(void)
4398  {
4399  	int i;
4400  
4401  	SCSI_LOG_HLQUEUE(3, printk("exit_sd: exiting sd driver\n"));
4402  
4403  	scsi_unregister_driver(&sd_template.gendrv);
4404  	mempool_destroy(sd_page_pool);
4405  
4406  	class_unregister(&sd_disk_class);
4407  
4408  	for (i = 0; i < SD_MAJORS; i++)
4409  		unregister_blkdev(sd_major(i), "sd");
4410  }
4411  
4412  module_init(init_sd);
4413  module_exit(exit_sd);
4414  
sd_print_sense_hdr(struct scsi_disk * sdkp,struct scsi_sense_hdr * sshdr)4415  void sd_print_sense_hdr(struct scsi_disk *sdkp, struct scsi_sense_hdr *sshdr)
4416  {
4417  	scsi_print_sense_hdr(sdkp->device,
4418  			     sdkp->disk ? sdkp->disk->disk_name : NULL, sshdr);
4419  }
4420  
sd_print_result(const struct scsi_disk * sdkp,const char * msg,int result)4421  void sd_print_result(const struct scsi_disk *sdkp, const char *msg, int result)
4422  {
4423  	const char *hb_string = scsi_hostbyte_string(result);
4424  
4425  	if (hb_string)
4426  		sd_printk(KERN_INFO, sdkp,
4427  			  "%s: Result: hostbyte=%s driverbyte=%s\n", msg,
4428  			  hb_string ? hb_string : "invalid",
4429  			  "DRIVER_OK");
4430  	else
4431  		sd_printk(KERN_INFO, sdkp,
4432  			  "%s: Result: hostbyte=0x%02x driverbyte=%s\n",
4433  			  msg, host_byte(result), "DRIVER_OK");
4434  }
4435