1	// SPDX-License-Identifier: GPL-2.0-only
2	/*
3	* Copyright (C) 2001 Sistina Software (UK) Limited.
4	* Copyright (C) 2004-2008 Red Hat, Inc. All rights reserved.
5	*
6	* This file is released under the GPL.
7	*/
8
9	#include "dm-core.h"
10	#include "dm-rq.h"
11
12	#include <linux/module.h>
13	#include <linux/vmalloc.h>
14	#include <linux/blkdev.h>
15	#include <linux/blk-integrity.h>
16	#include <linux/namei.h>
17	#include <linux/ctype.h>
18	#include <linux/string.h>
19	#include <linux/slab.h>
20	#include <linux/interrupt.h>
21	#include <linux/mutex.h>
22	#include <linux/delay.h>
23	#include <linux/atomic.h>
24	#include <linux/blk-mq.h>
25	#include <linux/mount.h>
26	#include <linux/dax.h>
27
28	#define DM_MSG_PREFIX "table"
29
30	#define NODE_SIZE L1_CACHE_BYTES
31	#define KEYS_PER_NODE (NODE_SIZE / sizeof(sector_t))
32	#define CHILDREN_PER_NODE (KEYS_PER_NODE + 1)
33
34	/*
35	* Similar to ceiling(log_size(n))
36	*/
37	static unsigned int int_log(unsigned int n, unsigned int base)
38	{
39	int result = 0;
40
41	while (n > 1) {
42	n = dm_div_up(n, base);
43	result++;
44	}
45
46	return result;
47	}
48
49	/*
50	* Calculate the index of the child node of the n'th node k'th key.
51	*/
52	static inline unsigned int get_child(unsigned int n, unsigned int k)
53	{
54	return (n * CHILDREN_PER_NODE) + k;
55	}
56
57	/*
58	* Return the n'th node of level l from table t.
59	*/
60	static inline sector_t *get_node(struct dm_table *t,
61	unsigned int l, unsigned int n)
62	{
63	return t->index[l] + (n * KEYS_PER_NODE);
64	}
65
66	/*
67	* Return the highest key that you could lookup from the n'th
68	* node on level l of the btree.
69	*/
70	static sector_t high(struct dm_table *t, unsigned int l, unsigned int n)
71	{
72	for (; l < t->depth - 1; l++)
73	n = get_child(n, CHILDREN_PER_NODE - 1);
74
75	if (n >= t->counts[l])
76	return (sector_t) -1;
77
78	return get_node(t, l, n)[KEYS_PER_NODE - 1];
79	}
80
81	/*
82	* Fills in a level of the btree based on the highs of the level
83	* below it.
84	*/
85	static int setup_btree_index(unsigned int l, struct dm_table *t)
86	{
87	unsigned int n, k;
88	sector_t *node;
89
90	for (n = 0U; n < t->counts[l]; n++) {
91	node = get_node(t, l, n);
92
93	for (k = 0U; k < KEYS_PER_NODE; k++)
94	node[k] = high(t, l: l + 1, n: get_child(n, k));
95	}
96
97	return 0;
98	}
99
100	/*
101	* highs, and targets are managed as dynamic arrays during a
102	* table load.
103	*/
104	static int alloc_targets(struct dm_table *t, unsigned int num)
105	{
106	sector_t *n_highs;
107	struct dm_target *n_targets;
108
109	/*
110	* Allocate both the target array and offset array at once.
111	*/
112	n_highs = kvcalloc(n: num, size: sizeof(struct dm_target) + sizeof(sector_t),
113	GFP_KERNEL);
114	if (!n_highs)
115	return -ENOMEM;
116
117	n_targets = (struct dm_target *) (n_highs + num);
118
119	memset(n_highs, -1, sizeof(*n_highs) * num);
120	kvfree(addr: t->highs);
121
122	t->num_allocated = num;
123	t->highs = n_highs;
124	t->targets = n_targets;
125
126	return 0;
127	}
128
129	int dm_table_create(struct dm_table **result, blk_mode_t mode,
130	unsigned int num_targets, struct mapped_device *md)
131	{
132	struct dm_table *t;
133
134	if (num_targets > DM_MAX_TARGETS)
135	return -EOVERFLOW;
136
137	t = kzalloc(size: sizeof(*t), GFP_KERNEL);
138
139	if (!t)
140	return -ENOMEM;
141
142	INIT_LIST_HEAD(list: &t->devices);
143	init_rwsem(&t->devices_lock);
144
145	if (!num_targets)
146	num_targets = KEYS_PER_NODE;
147
148	num_targets = dm_round_up(num_targets, KEYS_PER_NODE);
149
150	if (!num_targets) {
151	kfree(objp: t);
152	return -EOVERFLOW;
153	}
154
155	if (alloc_targets(t, num: num_targets)) {
156	kfree(objp: t);
157	return -ENOMEM;
158	}
159
160	t->type = DM_TYPE_NONE;
161	t->mode = mode;
162	t->md = md;
163	*result = t;
164	return 0;
165	}
166
167	static void free_devices(struct list_head *devices, struct mapped_device *md)
168	{
169	struct list_head *tmp, *next;
170
171	list_for_each_safe(tmp, next, devices) {
172	struct dm_dev_internal *dd =
173	list_entry(tmp, struct dm_dev_internal, list);
174	DMWARN("%s: dm_table_destroy: dm_put_device call missing for %s",
175	dm_device_name(md), dd->dm_dev->name);
176	dm_put_table_device(md, d: dd->dm_dev);
177	kfree(objp: dd);
178	}
179	}
180
181	static void dm_table_destroy_crypto_profile(struct dm_table *t);
182
183	void dm_table_destroy(struct dm_table *t)
184	{
185	if (!t)
186	return;
187
188	/* free the indexes */
189	if (t->depth >= 2)
190	kvfree(addr: t->index[t->depth - 2]);
191
192	/* free the targets */
193	for (unsigned int i = 0; i < t->num_targets; i++) {
194	struct dm_target *ti = dm_table_get_target(t, index: i);
195
196	if (ti->type->dtr)
197	ti->type->dtr(ti);
198
199	dm_put_target_type(tt: ti->type);
200	}
201
202	kvfree(addr: t->highs);
203
204	/* free the device list */
205	free_devices(devices: &t->devices, md: t->md);
206
207	dm_free_md_mempools(pools: t->mempools);
208
209	dm_table_destroy_crypto_profile(t);
210
211	kfree(objp: t);
212	}
213
214	/*
215	* See if we've already got a device in the list.
216	*/
217	static struct dm_dev_internal *find_device(struct list_head *l, dev_t dev)
218	{
219	struct dm_dev_internal *dd;
220
221	list_for_each_entry(dd, l, list)
222	if (dd->dm_dev->bdev->bd_dev == dev)
223	return dd;
224
225	return NULL;
226	}
227
228	/*
229	* If possible, this checks an area of a destination device is invalid.
230	*/
231	static int device_area_is_invalid(struct dm_target *ti, struct dm_dev *dev,
232	sector_t start, sector_t len, void *data)
233	{
234	struct queue_limits *limits = data;
235	struct block_device *bdev = dev->bdev;
236	sector_t dev_size = bdev_nr_sectors(bdev);
237	unsigned short logical_block_size_sectors =
238	limits->logical_block_size >> SECTOR_SHIFT;
239
240	if (!dev_size)
241	return 0;
242
243	if ((start >= dev_size) || (start + len > dev_size)) {
244	DMERR("%s: %pg too small for target: start=%llu, len=%llu, dev_size=%llu",
245	dm_device_name(ti->table->md), bdev,
246	(unsigned long long)start,
247	(unsigned long long)len,
248	(unsigned long long)dev_size);
249	return 1;
250	}
251
252	/*
253	* If the target is mapped to zoned block device(s), check
254	* that the zones are not partially mapped.
255	*/
256	if (bdev_is_zoned(bdev)) {
257	unsigned int zone_sectors = bdev_zone_sectors(bdev);
258
259	if (start & (zone_sectors - 1)) {
260	DMERR("%s: start=%llu not aligned to h/w zone size %u of %pg",
261	dm_device_name(ti->table->md),
262	(unsigned long long)start,
263	zone_sectors, bdev);
264	return 1;
265	}
266
267	/*
268	* Note: The last zone of a zoned block device may be smaller
269	* than other zones. So for a target mapping the end of a
270	* zoned block device with such a zone, len would not be zone
271	* aligned. We do not allow such last smaller zone to be part
272	* of the mapping here to ensure that mappings with multiple
273	* devices do not end up with a smaller zone in the middle of
274	* the sector range.
275	*/
276	if (len & (zone_sectors - 1)) {
277	DMERR("%s: len=%llu not aligned to h/w zone size %u of %pg",
278	dm_device_name(ti->table->md),
279	(unsigned long long)len,
280	zone_sectors, bdev);
281	return 1;
282	}
283	}
284
285	if (logical_block_size_sectors <= 1)
286	return 0;
287
288	if (start & (logical_block_size_sectors - 1)) {
289	DMERR("%s: start=%llu not aligned to h/w logical block size %u of %pg",
290	dm_device_name(ti->table->md),
291	(unsigned long long)start,
292	limits->logical_block_size, bdev);
293	return 1;
294	}
295
296	if (len & (logical_block_size_sectors - 1)) {
297	DMERR("%s: len=%llu not aligned to h/w logical block size %u of %pg",
298	dm_device_name(ti->table->md),
299	(unsigned long long)len,
300	limits->logical_block_size, bdev);
301	return 1;
302	}
303
304	return 0;
305	}
306
307	/*
308	* This upgrades the mode on an already open dm_dev, being
309	* careful to leave things as they were if we fail to reopen the
310	* device and not to touch the existing bdev field in case
311	* it is accessed concurrently.
312	*/
313	static int upgrade_mode(struct dm_dev_internal *dd, blk_mode_t new_mode,
314	struct mapped_device *md)
315	{
316	int r;
317	struct dm_dev *old_dev, *new_dev;
318
319	old_dev = dd->dm_dev;
320
321	r = dm_get_table_device(md, dev: dd->dm_dev->bdev->bd_dev,
322	mode: dd->dm_dev->mode | new_mode, result: &new_dev);
323	if (r)
324	return r;
325
326	dd->dm_dev = new_dev;
327	dm_put_table_device(md, d: old_dev);
328
329	return 0;
330	}
331
332	/*
333	* Add a device to the list, or just increment the usage count if
334	* it's already present.
335	*
336	* Note: the __ref annotation is because this function can call the __init
337	* marked early_lookup_bdev when called during early boot code from dm-init.c.
338	*/
339	int __ref dm_get_device(struct dm_target *ti, const char *path, blk_mode_t mode,
340	struct dm_dev **result)
341	{
342	int r;
343	dev_t dev;
344	unsigned int major, minor;
345	char dummy;
346	struct dm_dev_internal *dd;
347	struct dm_table *t = ti->table;
348
349	BUG_ON(!t);
350
351	if (sscanf(path, "%u:%u%c", &major, &minor, &dummy) == 2) {
352	/* Extract the major/minor numbers */
353	dev = MKDEV(major, minor);
354	if (MAJOR(dev) != major || MINOR(dev) != minor)
355	return -EOVERFLOW;
356	} else {
357	r = lookup_bdev(pathname: path, dev: &dev);
358	#ifndef MODULE
359	if (r && system_state < SYSTEM_RUNNING)
360	r = early_lookup_bdev(pathname: path, dev: &dev);
361	#endif
362	if (r)
363	return r;
364	}
365	if (dev == disk_devt(disk: t->md->disk))
366	return -EINVAL;
367
368	down_write(sem: &t->devices_lock);
369
370	dd = find_device(l: &t->devices, dev);
371	if (!dd) {
372	dd = kmalloc(size: sizeof(*dd), GFP_KERNEL);
373	if (!dd) {
374	r = -ENOMEM;
375	goto unlock_ret_r;
376	}
377
378	r = dm_get_table_device(md: t->md, dev, mode, result: &dd->dm_dev);
379	if (r) {
380	kfree(objp: dd);
381	goto unlock_ret_r;
382	}
383
384	refcount_set(r: &dd->count, n: 1);
385	list_add(new: &dd->list, head: &t->devices);
386	goto out;
387
388	} else if (dd->dm_dev->mode != (mode | dd->dm_dev->mode)) {
389	r = upgrade_mode(dd, new_mode: mode, md: t->md);
390	if (r)
391	goto unlock_ret_r;
392	}
393	refcount_inc(r: &dd->count);
394	out:
395	up_write(sem: &t->devices_lock);
396	*result = dd->dm_dev;
397	return 0;
398
399	unlock_ret_r:
400	up_write(sem: &t->devices_lock);
401	return r;
402	}
403	EXPORT_SYMBOL(dm_get_device);
404
405	static int dm_set_device_limits(struct dm_target *ti, struct dm_dev *dev,
406	sector_t start, sector_t len, void *data)
407	{
408	struct queue_limits *limits = data;
409	struct block_device *bdev = dev->bdev;
410	struct request_queue *q = bdev_get_queue(bdev);
411
412	if (unlikely(!q)) {
413	DMWARN("%s: Cannot set limits for nonexistent device %pg",
414	dm_device_name(ti->table->md), bdev);
415	return 0;
416	}
417
418	if (blk_stack_limits(t: limits, b: &q->limits,
419	offset: get_start_sect(bdev) + start) < 0)
420	DMWARN("%s: adding target device %pg caused an alignment inconsistency: "
421	"physical_block_size=%u, logical_block_size=%u, "
422	"alignment_offset=%u, start=%llu",
423	dm_device_name(ti->table->md), bdev,
424	q->limits.physical_block_size,
425	q->limits.logical_block_size,
426	q->limits.alignment_offset,
427	(unsigned long long) start << SECTOR_SHIFT);
428	return 0;
429	}
430
431	/*
432	* Decrement a device's use count and remove it if necessary.
433	*/
434	void dm_put_device(struct dm_target *ti, struct dm_dev *d)
435	{
436	int found = 0;
437	struct dm_table *t = ti->table;
438	struct list_head *devices = &t->devices;
439	struct dm_dev_internal *dd;
440
441	down_write(sem: &t->devices_lock);
442
443	list_for_each_entry(dd, devices, list) {
444	if (dd->dm_dev == d) {
445	found = 1;
446	break;
447	}
448	}
449	if (!found) {
450	DMERR("%s: device %s not in table devices list",
451	dm_device_name(t->md), d->name);
452	goto unlock_ret;
453	}
454	if (refcount_dec_and_test(r: &dd->count)) {
455	dm_put_table_device(md: t->md, d);
456	list_del(entry: &dd->list);
457	kfree(objp: dd);
458	}
459
460	unlock_ret:
461	up_write(sem: &t->devices_lock);
462	}
463	EXPORT_SYMBOL(dm_put_device);
464
465	/*
466	* Checks to see if the target joins onto the end of the table.
467	*/
468	static int adjoin(struct dm_table *t, struct dm_target *ti)
469	{
470	struct dm_target *prev;
471
472	if (!t->num_targets)
473	return !ti->begin;
474
475	prev = &t->targets[t->num_targets - 1];
476	return (ti->begin == (prev->begin + prev->len));
477	}
478
479	/*
480	* Used to dynamically allocate the arg array.
481	*
482	* We do first allocation with GFP_NOIO because dm-mpath and dm-thin must
483	* process messages even if some device is suspended. These messages have a
484	* small fixed number of arguments.
485	*
486	* On the other hand, dm-switch needs to process bulk data using messages and
487	* excessive use of GFP_NOIO could cause trouble.
488	*/
489	static char **realloc_argv(unsigned int *size, char **old_argv)
490	{
491	char **argv;
492	unsigned int new_size;
493	gfp_t gfp;
494
495	if (*size) {
496	new_size = *size * 2;
497	gfp = GFP_KERNEL;
498	} else {
499	new_size = 8;
500	gfp = GFP_NOIO;
501	}
502	argv = kmalloc_array(n: new_size, size: sizeof(*argv), flags: gfp);
503	if (argv && old_argv) {
504	memcpy(argv, old_argv, *size * sizeof(*argv));
505	*size = new_size;
506	}
507
508	kfree(objp: old_argv);
509	return argv;
510	}
511
512	/*
513	* Destructively splits up the argument list to pass to ctr.
514	*/
515	int dm_split_args(int *argc, char ***argvp, char *input)
516	{
517	char *start, *end = input, *out, **argv = NULL;
518	unsigned int array_size = 0;
519
520	*argc = 0;
521
522	if (!input) {
523	*argvp = NULL;
524	return 0;
525	}
526
527	argv = realloc_argv(size: &array_size, old_argv: argv);
528	if (!argv)
529	return -ENOMEM;
530
531	while (1) {
532	/* Skip whitespace */
533	start = skip_spaces(end);
534
535	if (!*start)
536	break; /* success, we hit the end */
537
538	/* 'out' is used to remove any back-quotes */
539	end = out = start;
540	while (*end) {
541	/* Everything apart from '\0' can be quoted */
542	if (*end == '\\' && *(end + 1)) {
543	*out++ = *(end + 1);
544	end += 2;
545	continue;
546	}
547
548	if (isspace(*end))
549	break; /* end of token */
550
551	*out++ = *end++;
552	}
553
554	/* have we already filled the array ? */
555	if ((*argc + 1) > array_size) {
556	argv = realloc_argv(size: &array_size, old_argv: argv);
557	if (!argv)
558	return -ENOMEM;
559	}
560
561	/* we know this is whitespace */
562	if (*end)
563	end++;
564
565	/* terminate the string and put it in the array */
566	*out = '\0';
567	argv[*argc] = start;
568	(*argc)++;
569	}
570
571	*argvp = argv;
572	return 0;
573	}
574
575	/*
576	* Impose necessary and sufficient conditions on a devices's table such
577	* that any incoming bio which respects its logical_block_size can be
578	* processed successfully. If it falls across the boundary between
579	* two or more targets, the size of each piece it gets split into must
580	* be compatible with the logical_block_size of the target processing it.
581	*/
582	static int validate_hardware_logical_block_alignment(struct dm_table *t,
583	struct queue_limits *limits)
584	{
585	/*
586	* This function uses arithmetic modulo the logical_block_size
587	* (in units of 512-byte sectors).
588	*/
589	unsigned short device_logical_block_size_sects =
590	limits->logical_block_size >> SECTOR_SHIFT;
591
592	/*
593	* Offset of the start of the next table entry, mod logical_block_size.
594	*/
595	unsigned short next_target_start = 0;
596
597	/*
598	* Given an aligned bio that extends beyond the end of a
599	* target, how many sectors must the next target handle?
600	*/
601	unsigned short remaining = 0;
602
603	struct dm_target *ti;
604	struct queue_limits ti_limits;
605	unsigned int i;
606
607	/*
608	* Check each entry in the table in turn.
609	*/
610	for (i = 0; i < t->num_targets; i++) {
611	ti = dm_table_get_target(t, index: i);
612
613	blk_set_stacking_limits(lim: &ti_limits);
614
615	/* combine all target devices' limits */
616	if (ti->type->iterate_devices)
617	ti->type->iterate_devices(ti, dm_set_device_limits,
618	&ti_limits);
619
620	/*
621	* If the remaining sectors fall entirely within this
622	* table entry are they compatible with its logical_block_size?
623	*/
624	if (remaining < ti->len &&
625	remaining & ((ti_limits.logical_block_size >>
626	SECTOR_SHIFT) - 1))
627	break; /* Error */
628
629	next_target_start =
630	(unsigned short) ((next_target_start + ti->len) &
631	(device_logical_block_size_sects - 1));
632	remaining = next_target_start ?
633	device_logical_block_size_sects - next_target_start : 0;
634	}
635
636	if (remaining) {
637	DMERR("%s: table line %u (start sect %llu len %llu) "
638	"not aligned to h/w logical block size %u",
639	dm_device_name(t->md), i,
640	(unsigned long long) ti->begin,
641	(unsigned long long) ti->len,
642	limits->logical_block_size);
643	return -EINVAL;
644	}
645
646	return 0;
647	}
648
649	int dm_table_add_target(struct dm_table *t, const char *type,
650	sector_t start, sector_t len, char *params)
651	{
652	int r = -EINVAL, argc;
653	char **argv;
654	struct dm_target *ti;
655
656	if (t->singleton) {
657	DMERR("%s: target type %s must appear alone in table",
658	dm_device_name(t->md), t->targets->type->name);
659	return -EINVAL;
660	}
661
662	BUG_ON(t->num_targets >= t->num_allocated);
663
664	ti = t->targets + t->num_targets;
665	memset(ti, 0, sizeof(*ti));
666
667	if (!len) {
668	DMERR("%s: zero-length target", dm_device_name(t->md));
669	return -EINVAL;
670	}
671
672	ti->type = dm_get_target_type(name: type);
673	if (!ti->type) {
674	DMERR("%s: %s: unknown target type", dm_device_name(t->md), type);
675	return -EINVAL;
676	}
677
678	if (dm_target_needs_singleton(ti->type)) {
679	if (t->num_targets) {
680	ti->error = "singleton target type must appear alone in table";
681	goto bad;
682	}
683	t->singleton = true;
684	}
685
686	if (dm_target_always_writeable(ti->type) &&
687	!(t->mode & BLK_OPEN_WRITE)) {
688	ti->error = "target type may not be included in a read-only table";
689	goto bad;
690	}
691
692	if (t->immutable_target_type) {
693	if (t->immutable_target_type != ti->type) {
694	ti->error = "immutable target type cannot be mixed with other target types";
695	goto bad;
696	}
697	} else if (dm_target_is_immutable(ti->type)) {
698	if (t->num_targets) {
699	ti->error = "immutable target type cannot be mixed with other target types";
700	goto bad;
701	}
702	t->immutable_target_type = ti->type;
703	}
704
705	if (dm_target_has_integrity(ti->type))
706	t->integrity_added = 1;
707
708	ti->table = t;
709	ti->begin = start;
710	ti->len = len;
711	ti->error = "Unknown error";
712
713	/*
714	* Does this target adjoin the previous one ?
715	*/
716	if (!adjoin(t, ti)) {
717	ti->error = "Gap in table";
718	goto bad;
719	}
720
721	r = dm_split_args(argc: &argc, argvp: &argv, input: params);
722	if (r) {
723	ti->error = "couldn't split parameters";
724	goto bad;
725	}
726
727	r = ti->type->ctr(ti, argc, argv);
728	kfree(objp: argv);
729	if (r)
730	goto bad;
731
732	t->highs[t->num_targets++] = ti->begin + ti->len - 1;
733
734	if (!ti->num_discard_bios && ti->discards_supported)
735	DMWARN("%s: %s: ignoring discards_supported because num_discard_bios is zero.",
736	dm_device_name(t->md), type);
737
738	if (ti->limit_swap_bios && !static_key_enabled(&swap_bios_enabled.key))
739	static_branch_enable(&swap_bios_enabled);
740
741	return 0;
742
743	bad:
744	DMERR("%s: %s: %s (%pe)", dm_device_name(t->md), type, ti->error, ERR_PTR(r));
745	dm_put_target_type(tt: ti->type);
746	return r;
747	}
748
749	/*
750	* Target argument parsing helpers.
751	*/
752	static int validate_next_arg(const struct dm_arg *arg, struct dm_arg_set *arg_set,
753	unsigned int *value, char **error, unsigned int grouped)
754	{
755	const char *arg_str = dm_shift_arg(as: arg_set);
756	char dummy;
757
758	if (!arg_str ||
759	(sscanf(arg_str, "%u%c", value, &dummy) != 1) ||
760	(*value < arg->min) ||
761	(*value > arg->max) ||
762	(grouped && arg_set->argc < *value)) {
763	*error = arg->error;
764	return -EINVAL;
765	}
766
767	return 0;
768	}
769
770	int dm_read_arg(const struct dm_arg *arg, struct dm_arg_set *arg_set,
771	unsigned int *value, char **error)
772	{
773	return validate_next_arg(arg, arg_set, value, error, grouped: 0);
774	}
775	EXPORT_SYMBOL(dm_read_arg);
776
777	int dm_read_arg_group(const struct dm_arg *arg, struct dm_arg_set *arg_set,
778	unsigned int *value, char **error)
779	{
780	return validate_next_arg(arg, arg_set, value, error, grouped: 1);
781	}
782	EXPORT_SYMBOL(dm_read_arg_group);
783
784	const char *dm_shift_arg(struct dm_arg_set *as)
785	{
786	char *r;
787
788	if (as->argc) {
789	as->argc--;
790	r = *as->argv;
791	as->argv++;
792	return r;
793	}
794
795	return NULL;
796	}
797	EXPORT_SYMBOL(dm_shift_arg);
798
799	void dm_consume_args(struct dm_arg_set *as, unsigned int num_args)
800	{
801	BUG_ON(as->argc < num_args);
802	as->argc -= num_args;
803	as->argv += num_args;
804	}
805	EXPORT_SYMBOL(dm_consume_args);
806
807	static bool __table_type_bio_based(enum dm_queue_mode table_type)
808	{
809	return (table_type == DM_TYPE_BIO_BASED ||
810	table_type == DM_TYPE_DAX_BIO_BASED);
811	}
812
813	static bool __table_type_request_based(enum dm_queue_mode table_type)
814	{
815	return table_type == DM_TYPE_REQUEST_BASED;
816	}
817
818	void dm_table_set_type(struct dm_table *t, enum dm_queue_mode type)
819	{
820	t->type = type;
821	}
822	EXPORT_SYMBOL_GPL(dm_table_set_type);
823
824	/* validate the dax capability of the target device span */
825	static int device_not_dax_capable(struct dm_target *ti, struct dm_dev *dev,
826	sector_t start, sector_t len, void *data)
827	{
828	if (dev->dax_dev)
829	return false;
830
831	DMDEBUG("%pg: error: dax unsupported by block device", dev->bdev);
832	return true;
833	}
834
835	/* Check devices support synchronous DAX */
836	static int device_not_dax_synchronous_capable(struct dm_target *ti, struct dm_dev *dev,
837	sector_t start, sector_t len, void *data)
838	{
839	return !dev->dax_dev || !dax_synchronous(dax_dev: dev->dax_dev);
840	}
841
842	static bool dm_table_supports_dax(struct dm_table *t,
843	iterate_devices_callout_fn iterate_fn)
844	{
845	/* Ensure that all targets support DAX. */
846	for (unsigned int i = 0; i < t->num_targets; i++) {
847	struct dm_target *ti = dm_table_get_target(t, index: i);
848
849	if (!ti->type->direct_access)
850	return false;
851
852	if (dm_target_is_wildcard(ti->type) ||
853	!ti->type->iterate_devices ||
854	ti->type->iterate_devices(ti, iterate_fn, NULL))
855	return false;
856	}
857
858	return true;
859	}
860
861	static int device_is_rq_stackable(struct dm_target *ti, struct dm_dev *dev,
862	sector_t start, sector_t len, void *data)
863	{
864	struct block_device *bdev = dev->bdev;
865	struct request_queue *q = bdev_get_queue(bdev);
866
867	/* request-based cannot stack on partitions! */
868	if (bdev_is_partition(bdev))
869	return false;
870
871	return queue_is_mq(q);
872	}
873
874	static int dm_table_determine_type(struct dm_table *t)
875	{
876	unsigned int bio_based = 0, request_based = 0, hybrid = 0;
877	struct dm_target *ti;
878	struct list_head *devices = dm_table_get_devices(t);
879	enum dm_queue_mode live_md_type = dm_get_md_type(md: t->md);
880
881	if (t->type != DM_TYPE_NONE) {
882	/* target already set the table's type */
883	if (t->type == DM_TYPE_BIO_BASED) {
884	/* possibly upgrade to a variant of bio-based */
885	goto verify_bio_based;
886	}
887	BUG_ON(t->type == DM_TYPE_DAX_BIO_BASED);
888	goto verify_rq_based;
889	}
890
891	for (unsigned int i = 0; i < t->num_targets; i++) {
892	ti = dm_table_get_target(t, index: i);
893	if (dm_target_hybrid(ti))
894	hybrid = 1;
895	else if (dm_target_request_based(ti))
896	request_based = 1;
897	else
898	bio_based = 1;
899
900	if (bio_based && request_based) {
901	DMERR("Inconsistent table: different target types can't be mixed up");
902	return -EINVAL;
903	}
904	}
905
906	if (hybrid && !bio_based && !request_based) {
907	/*
908	* The targets can work either way.
909	* Determine the type from the live device.
910	* Default to bio-based if device is new.
911	*/
912	if (__table_type_request_based(table_type: live_md_type))
913	request_based = 1;
914	else
915	bio_based = 1;
916	}
917
918	if (bio_based) {
919	verify_bio_based:
920	/* We must use this table as bio-based */
921	t->type = DM_TYPE_BIO_BASED;
922	if (dm_table_supports_dax(t, iterate_fn: device_not_dax_capable) ||
923	(list_empty(head: devices) && live_md_type == DM_TYPE_DAX_BIO_BASED)) {
924	t->type = DM_TYPE_DAX_BIO_BASED;
925	}
926	return 0;
927	}
928
929	BUG_ON(!request_based); /* No targets in this table */
930
931	t->type = DM_TYPE_REQUEST_BASED;
932
933	verify_rq_based:
934	/*
935	* Request-based dm supports only tables that have a single target now.
936	* To support multiple targets, request splitting support is needed,
937	* and that needs lots of changes in the block-layer.
938	* (e.g. request completion process for partial completion.)
939	*/
940	if (t->num_targets > 1) {
941	DMERR("request-based DM doesn't support multiple targets");
942	return -EINVAL;
943	}
944
945	if (list_empty(head: devices)) {
946	int srcu_idx;
947	struct dm_table *live_table = dm_get_live_table(md: t->md, srcu_idx: &srcu_idx);
948
949	/* inherit live table's type */
950	if (live_table)
951	t->type = live_table->type;
952	dm_put_live_table(md: t->md, srcu_idx);
953	return 0;
954	}
955
956	ti = dm_table_get_immutable_target(t);
957	if (!ti) {
958	DMERR("table load rejected: immutable target is required");
959	return -EINVAL;
960	} else if (ti->max_io_len) {
961	DMERR("table load rejected: immutable target that splits IO is not supported");
962	return -EINVAL;
963	}
964
965	/* Non-request-stackable devices can't be used for request-based dm */
966	if (!ti->type->iterate_devices ||
967	!ti->type->iterate_devices(ti, device_is_rq_stackable, NULL)) {
968	DMERR("table load rejected: including non-request-stackable devices");
969	return -EINVAL;
970	}
971
972	return 0;
973	}
974
975	enum dm_queue_mode dm_table_get_type(struct dm_table *t)
976	{
977	return t->type;
978	}
979
980	struct target_type *dm_table_get_immutable_target_type(struct dm_table *t)
981	{
982	return t->immutable_target_type;
983	}
984
985	struct dm_target *dm_table_get_immutable_target(struct dm_table *t)
986	{
987	/* Immutable target is implicitly a singleton */
988	if (t->num_targets > 1 ||
989	!dm_target_is_immutable(t->targets[0].type))
990	return NULL;
991
992	return t->targets;
993	}
994
995	struct dm_target *dm_table_get_wildcard_target(struct dm_table *t)
996	{
997	for (unsigned int i = 0; i < t->num_targets; i++) {
998	struct dm_target *ti = dm_table_get_target(t, index: i);
999
1000	if (dm_target_is_wildcard(ti->type))
1001	return ti;
1002	}
1003
1004	return NULL;
1005	}
1006
1007	bool dm_table_bio_based(struct dm_table *t)
1008	{
1009	return __table_type_bio_based(table_type: dm_table_get_type(t));
1010	}
1011
1012	bool dm_table_request_based(struct dm_table *t)
1013	{
1014	return __table_type_request_based(table_type: dm_table_get_type(t));
1015	}
1016
1017	static bool dm_table_supports_poll(struct dm_table *t);
1018
1019	static int dm_table_alloc_md_mempools(struct dm_table *t, struct mapped_device *md)
1020	{
1021	enum dm_queue_mode type = dm_table_get_type(t);
1022	unsigned int per_io_data_size = 0, front_pad, io_front_pad;
1023	unsigned int min_pool_size = 0, pool_size;
1024	struct dm_md_mempools *pools;
1025
1026	if (unlikely(type == DM_TYPE_NONE)) {
1027	DMERR("no table type is set, can't allocate mempools");
1028	return -EINVAL;
1029	}
1030
1031	pools = kzalloc_node(size: sizeof(*pools), GFP_KERNEL, node: md->numa_node_id);
1032	if (!pools)
1033	return -ENOMEM;
1034
1035	if (type == DM_TYPE_REQUEST_BASED) {
1036	pool_size = dm_get_reserved_rq_based_ios();
1037	front_pad = offsetof(struct dm_rq_clone_bio_info, clone);
1038	goto init_bs;
1039	}
1040
1041	for (unsigned int i = 0; i < t->num_targets; i++) {
1042	struct dm_target *ti = dm_table_get_target(t, index: i);
1043
1044	per_io_data_size = max(per_io_data_size, ti->per_io_data_size);
1045	min_pool_size = max(min_pool_size, ti->num_flush_bios);
1046	}
1047	pool_size = max(dm_get_reserved_bio_based_ios(), min_pool_size);
1048	front_pad = roundup(per_io_data_size,
1049	__alignof__(struct dm_target_io)) + DM_TARGET_IO_BIO_OFFSET;
1050
1051	io_front_pad = roundup(per_io_data_size,
1052	__alignof__(struct dm_io)) + DM_IO_BIO_OFFSET;
1053	if (bioset_init(&pools->io_bs, pool_size, io_front_pad,
1054	flags: dm_table_supports_poll(t) ? BIOSET_PERCPU_CACHE : 0))
1055	goto out_free_pools;
1056	if (t->integrity_supported &&
1057	bioset_integrity_create(&pools->io_bs, pool_size))
1058	goto out_free_pools;
1059	init_bs:
1060	if (bioset_init(&pools->bs, pool_size, front_pad, flags: 0))
1061	goto out_free_pools;
1062	if (t->integrity_supported &&
1063	bioset_integrity_create(&pools->bs, pool_size))
1064	goto out_free_pools;
1065
1066	t->mempools = pools;
1067	return 0;
1068
1069	out_free_pools:
1070	dm_free_md_mempools(pools);
1071	return -ENOMEM;
1072	}
1073
1074	static int setup_indexes(struct dm_table *t)
1075	{
1076	int i;
1077	unsigned int total = 0;
1078	sector_t *indexes;
1079
1080	/* allocate the space for *all* the indexes */
1081	for (i = t->depth - 2; i >= 0; i--) {
1082	t->counts[i] = dm_div_up(t->counts[i + 1], CHILDREN_PER_NODE);
1083	total += t->counts[i];
1084	}
1085
1086	indexes = kvcalloc(n: total, NODE_SIZE, GFP_KERNEL);
1087	if (!indexes)
1088	return -ENOMEM;
1089
1090	/* set up internal nodes, bottom-up */
1091	for (i = t->depth - 2; i >= 0; i--) {
1092	t->index[i] = indexes;
1093	indexes += (KEYS_PER_NODE * t->counts[i]);
1094	setup_btree_index(l: i, t);
1095	}
1096
1097	return 0;
1098	}
1099
1100	/*
1101	* Builds the btree to index the map.
1102	*/
1103	static int dm_table_build_index(struct dm_table *t)
1104	{
1105	int r = 0;
1106	unsigned int leaf_nodes;
1107
1108	/* how many indexes will the btree have ? */
1109	leaf_nodes = dm_div_up(t->num_targets, KEYS_PER_NODE);
1110	t->depth = 1 + int_log(n: leaf_nodes, CHILDREN_PER_NODE);
1111
1112	/* leaf layer has already been set up */
1113	t->counts[t->depth - 1] = leaf_nodes;
1114	t->index[t->depth - 1] = t->highs;
1115
1116	if (t->depth >= 2)
1117	r = setup_indexes(t);
1118
1119	return r;
1120	}
1121
1122	static bool integrity_profile_exists(struct gendisk *disk)
1123	{
1124	return !!blk_get_integrity(disk);
1125	}
1126
1127	/*
1128	* Get a disk whose integrity profile reflects the table's profile.
1129	* Returns NULL if integrity support was inconsistent or unavailable.
1130	*/
1131	static struct gendisk *dm_table_get_integrity_disk(struct dm_table *t)
1132	{
1133	struct list_head *devices = dm_table_get_devices(t);
1134	struct dm_dev_internal *dd = NULL;
1135	struct gendisk *prev_disk = NULL, *template_disk = NULL;
1136
1137	for (unsigned int i = 0; i < t->num_targets; i++) {
1138	struct dm_target *ti = dm_table_get_target(t, index: i);
1139
1140	if (!dm_target_passes_integrity(ti->type))
1141	goto no_integrity;
1142	}
1143
1144	list_for_each_entry(dd, devices, list) {
1145	template_disk = dd->dm_dev->bdev->bd_disk;
1146	if (!integrity_profile_exists(disk: template_disk))
1147	goto no_integrity;
1148	else if (prev_disk &&
1149	blk_integrity_compare(prev_disk, template_disk) < 0)
1150	goto no_integrity;
1151	prev_disk = template_disk;
1152	}
1153
1154	return template_disk;
1155
1156	no_integrity:
1157	if (prev_disk)
1158	DMWARN("%s: integrity not set: %s and %s profile mismatch",
1159	dm_device_name(t->md),
1160	prev_disk->disk_name,
1161	template_disk->disk_name);
1162	return NULL;
1163	}
1164
1165	/*
1166	* Register the mapped device for blk_integrity support if the
1167	* underlying devices have an integrity profile. But all devices may
1168	* not have matching profiles (checking all devices isn't reliable
1169	* during table load because this table may use other DM device(s) which
1170	* must be resumed before they will have an initialized integity
1171	* profile). Consequently, stacked DM devices force a 2 stage integrity
1172	* profile validation: First pass during table load, final pass during
1173	* resume.
1174	*/
1175	static int dm_table_register_integrity(struct dm_table *t)
1176	{
1177	struct mapped_device *md = t->md;
1178	struct gendisk *template_disk = NULL;
1179
1180	/* If target handles integrity itself do not register it here. */
1181	if (t->integrity_added)
1182	return 0;
1183
1184	template_disk = dm_table_get_integrity_disk(t);
1185	if (!template_disk)
1186	return 0;
1187
1188	if (!integrity_profile_exists(disk: dm_disk(md))) {
1189	t->integrity_supported = true;
1190	/*
1191	* Register integrity profile during table load; we can do
1192	* this because the final profile must match during resume.
1193	*/
1194	blk_integrity_register(dm_disk(md),
1195	blk_get_integrity(disk: template_disk));
1196	return 0;
1197	}
1198
1199	/*
1200	* If DM device already has an initialized integrity
1201	* profile the new profile should not conflict.
1202	*/
1203	if (blk_integrity_compare(dm_disk(md), template_disk) < 0) {
1204	DMERR("%s: conflict with existing integrity profile: %s profile mismatch",
1205	dm_device_name(t->md),
1206	template_disk->disk_name);
1207	return 1;
1208	}
1209
1210	/* Preserve existing integrity profile */
1211	t->integrity_supported = true;
1212	return 0;
1213	}
1214
1215	#ifdef CONFIG_BLK_INLINE_ENCRYPTION
1216
1217	struct dm_crypto_profile {
1218	struct blk_crypto_profile profile;
1219	struct mapped_device *md;
1220	};
1221
1222	static int dm_keyslot_evict_callback(struct dm_target *ti, struct dm_dev *dev,
1223	sector_t start, sector_t len, void *data)
1224	{
1225	const struct blk_crypto_key *key = data;
1226
1227	blk_crypto_evict_key(bdev: dev->bdev, key);
1228	return 0;
1229	}
1230
1231	/*
1232	* When an inline encryption key is evicted from a device-mapper device, evict
1233	* it from all the underlying devices.
1234	*/
1235	static int dm_keyslot_evict(struct blk_crypto_profile *profile,
1236	const struct blk_crypto_key *key, unsigned int slot)
1237	{
1238	struct mapped_device *md =
1239	container_of(profile, struct dm_crypto_profile, profile)->md;
1240	struct dm_table *t;
1241	int srcu_idx;
1242
1243	t = dm_get_live_table(md, srcu_idx: &srcu_idx);
1244	if (!t)
1245	return 0;
1246
1247	for (unsigned int i = 0; i < t->num_targets; i++) {
1248	struct dm_target *ti = dm_table_get_target(t, index: i);
1249
1250	if (!ti->type->iterate_devices)
1251	continue;
1252	ti->type->iterate_devices(ti, dm_keyslot_evict_callback,
1253	(void *)key);
1254	}
1255
1256	dm_put_live_table(md, srcu_idx);
1257	return 0;
1258	}
1259
1260	static int
1261	device_intersect_crypto_capabilities(struct dm_target *ti, struct dm_dev *dev,
1262	sector_t start, sector_t len, void *data)
1263	{
1264	struct blk_crypto_profile *parent = data;
1265	struct blk_crypto_profile *child =
1266	bdev_get_queue(bdev: dev->bdev)->crypto_profile;
1267
1268	blk_crypto_intersect_capabilities(parent, child);
1269	return 0;
1270	}
1271
1272	void dm_destroy_crypto_profile(struct blk_crypto_profile *profile)
1273	{
1274	struct dm_crypto_profile *dmcp = container_of(profile,
1275	struct dm_crypto_profile,
1276	profile);
1277
1278	if (!profile)
1279	return;
1280
1281	blk_crypto_profile_destroy(profile);
1282	kfree(objp: dmcp);
1283	}
1284
1285	static void dm_table_destroy_crypto_profile(struct dm_table *t)
1286	{
1287	dm_destroy_crypto_profile(profile: t->crypto_profile);
1288	t->crypto_profile = NULL;
1289	}
1290
1291	/*
1292	* Constructs and initializes t->crypto_profile with a crypto profile that
1293	* represents the common set of crypto capabilities of the devices described by
1294	* the dm_table. However, if the constructed crypto profile doesn't support all
1295	* crypto capabilities that are supported by the current mapped_device, it
1296	* returns an error instead, since we don't support removing crypto capabilities
1297	* on table changes. Finally, if the constructed crypto profile is "empty" (has
1298	* no crypto capabilities at all), it just sets t->crypto_profile to NULL.
1299	*/
1300	static int dm_table_construct_crypto_profile(struct dm_table *t)
1301	{
1302	struct dm_crypto_profile *dmcp;
1303	struct blk_crypto_profile *profile;
1304	unsigned int i;
1305	bool empty_profile = true;
1306
1307	dmcp = kmalloc(size: sizeof(*dmcp), GFP_KERNEL);
1308	if (!dmcp)
1309	return -ENOMEM;
1310	dmcp->md = t->md;
1311
1312	profile = &dmcp->profile;
1313	blk_crypto_profile_init(profile, num_slots: 0);
1314	profile->ll_ops.keyslot_evict = dm_keyslot_evict;
1315	profile->max_dun_bytes_supported = UINT_MAX;
1316	memset(profile->modes_supported, 0xFF,
1317	sizeof(profile->modes_supported));
1318
1319	for (i = 0; i < t->num_targets; i++) {
1320	struct dm_target *ti = dm_table_get_target(t, index: i);
1321
1322	if (!dm_target_passes_crypto(ti->type)) {
1323	blk_crypto_intersect_capabilities(parent: profile, NULL);
1324	break;
1325	}
1326	if (!ti->type->iterate_devices)
1327	continue;
1328	ti->type->iterate_devices(ti,
1329	device_intersect_crypto_capabilities,
1330	profile);
1331	}
1332
1333	if (t->md->queue &&
1334	!blk_crypto_has_capabilities(target: profile,
1335	reference: t->md->queue->crypto_profile)) {
1336	DMERR("Inline encryption capabilities of new DM table were more restrictive than the old table's. This is not supported!");
1337	dm_destroy_crypto_profile(profile);
1338	return -EINVAL;
1339	}
1340
1341	/*
1342	* If the new profile doesn't actually support any crypto capabilities,
1343	* we may as well represent it with a NULL profile.
1344	*/
1345	for (i = 0; i < ARRAY_SIZE(profile->modes_supported); i++) {
1346	if (profile->modes_supported[i]) {
1347	empty_profile = false;
1348	break;
1349	}
1350	}
1351
1352	if (empty_profile) {
1353	dm_destroy_crypto_profile(profile);
1354	profile = NULL;
1355	}
1356
1357	/*
1358	* t->crypto_profile is only set temporarily while the table is being
1359	* set up, and it gets set to NULL after the profile has been
1360	* transferred to the request_queue.
1361	*/
1362	t->crypto_profile = profile;
1363
1364	return 0;
1365	}
1366
1367	static void dm_update_crypto_profile(struct request_queue *q,
1368	struct dm_table *t)
1369	{
1370	if (!t->crypto_profile)
1371	return;
1372
1373	/* Make the crypto profile less restrictive. */
1374	if (!q->crypto_profile) {
1375	blk_crypto_register(profile: t->crypto_profile, q);
1376	} else {
1377	blk_crypto_update_capabilities(dst: q->crypto_profile,
1378	src: t->crypto_profile);
1379	dm_destroy_crypto_profile(profile: t->crypto_profile);
1380	}
1381	t->crypto_profile = NULL;
1382	}
1383
1384	#else /* CONFIG_BLK_INLINE_ENCRYPTION */
1385
1386	static int dm_table_construct_crypto_profile(struct dm_table *t)
1387	{
1388	return 0;
1389	}
1390
1391	void dm_destroy_crypto_profile(struct blk_crypto_profile *profile)
1392	{
1393	}
1394
1395	static void dm_table_destroy_crypto_profile(struct dm_table *t)
1396	{
1397	}
1398
1399	static void dm_update_crypto_profile(struct request_queue *q,
1400	struct dm_table *t)
1401	{
1402	}
1403
1404	#endif /* !CONFIG_BLK_INLINE_ENCRYPTION */
1405
1406	/*
1407	* Prepares the table for use by building the indices,
1408	* setting the type, and allocating mempools.
1409	*/
1410	int dm_table_complete(struct dm_table *t)
1411	{
1412	int r;
1413
1414	r = dm_table_determine_type(t);
1415	if (r) {
1416	DMERR("unable to determine table type");
1417	return r;
1418	}
1419
1420	r = dm_table_build_index(t);
1421	if (r) {
1422	DMERR("unable to build btrees");
1423	return r;
1424	}
1425
1426	r = dm_table_register_integrity(t);
1427	if (r) {
1428	DMERR("could not register integrity profile.");
1429	return r;
1430	}
1431
1432	r = dm_table_construct_crypto_profile(t);
1433	if (r) {
1434	DMERR("could not construct crypto profile.");
1435	return r;
1436	}
1437
1438	r = dm_table_alloc_md_mempools(t, md: t->md);
1439	if (r)
1440	DMERR("unable to allocate mempools");
1441
1442	return r;
1443	}
1444
1445	static DEFINE_MUTEX(_event_lock);
1446	void dm_table_event_callback(struct dm_table *t,
1447	void (*fn)(void *), void *context)
1448	{
1449	mutex_lock(&_event_lock);
1450	t->event_fn = fn;
1451	t->event_context = context;
1452	mutex_unlock(lock: &_event_lock);
1453	}
1454
1455	void dm_table_event(struct dm_table *t)
1456	{
1457	mutex_lock(&_event_lock);
1458	if (t->event_fn)
1459	t->event_fn(t->event_context);
1460	mutex_unlock(lock: &_event_lock);
1461	}
1462	EXPORT_SYMBOL(dm_table_event);
1463
1464	inline sector_t dm_table_get_size(struct dm_table *t)
1465	{
1466	return t->num_targets ? (t->highs[t->num_targets - 1] + 1) : 0;
1467	}
1468	EXPORT_SYMBOL(dm_table_get_size);
1469
1470	/*
1471	* Search the btree for the correct target.
1472	*
1473	* Caller should check returned pointer for NULL
1474	* to trap I/O beyond end of device.
1475	*/
1476	struct dm_target *dm_table_find_target(struct dm_table *t, sector_t sector)
1477	{
1478	unsigned int l, n = 0, k = 0;
1479	sector_t *node;
1480
1481	if (unlikely(sector >= dm_table_get_size(t)))
1482	return NULL;
1483
1484	for (l = 0; l < t->depth; l++) {
1485	n = get_child(n, k);
1486	node = get_node(t, l, n);
1487
1488	for (k = 0; k < KEYS_PER_NODE; k++)
1489	if (node[k] >= sector)
1490	break;
1491	}
1492
1493	return &t->targets[(KEYS_PER_NODE * n) + k];
1494	}
1495
1496	static int device_not_poll_capable(struct dm_target *ti, struct dm_dev *dev,
1497	sector_t start, sector_t len, void *data)
1498	{
1499	struct request_queue *q = bdev_get_queue(bdev: dev->bdev);
1500
1501	return !test_bit(QUEUE_FLAG_POLL, &q->queue_flags);
1502	}
1503
1504	/*
1505	* type->iterate_devices() should be called when the sanity check needs to
1506	* iterate and check all underlying data devices. iterate_devices() will
1507	* iterate all underlying data devices until it encounters a non-zero return
1508	* code, returned by whether the input iterate_devices_callout_fn, or
1509	* iterate_devices() itself internally.
1510	*
1511	* For some target type (e.g. dm-stripe), one call of iterate_devices() may
1512	* iterate multiple underlying devices internally, in which case a non-zero
1513	* return code returned by iterate_devices_callout_fn will stop the iteration
1514	* in advance.
1515	*
1516	* Cases requiring _any_ underlying device supporting some kind of attribute,
1517	* should use the iteration structure like dm_table_any_dev_attr(), or call
1518	* it directly. @func should handle semantics of positive examples, e.g.
1519	* capable of something.
1520	*
1521	* Cases requiring _all_ underlying devices supporting some kind of attribute,
1522	* should use the iteration structure like dm_table_supports_nowait() or
1523	* dm_table_supports_discards(). Or introduce dm_table_all_devs_attr() that
1524	* uses an @anti_func that handle semantics of counter examples, e.g. not
1525	* capable of something. So: return !dm_table_any_dev_attr(t, anti_func, data);
1526	*/
1527	static bool dm_table_any_dev_attr(struct dm_table *t,
1528	iterate_devices_callout_fn func, void *data)
1529	{
1530	for (unsigned int i = 0; i < t->num_targets; i++) {
1531	struct dm_target *ti = dm_table_get_target(t, index: i);
1532
1533	if (ti->type->iterate_devices &&
1534	ti->type->iterate_devices(ti, func, data))
1535	return true;
1536	}
1537
1538	return false;
1539	}
1540
1541	static int count_device(struct dm_target *ti, struct dm_dev *dev,
1542	sector_t start, sector_t len, void *data)
1543	{
1544	unsigned int *num_devices = data;
1545
1546	(*num_devices)++;
1547
1548	return 0;
1549	}
1550
1551	static bool dm_table_supports_poll(struct dm_table *t)
1552	{
1553	for (unsigned int i = 0; i < t->num_targets; i++) {
1554	struct dm_target *ti = dm_table_get_target(t, index: i);
1555
1556	if (!ti->type->iterate_devices ||
1557	ti->type->iterate_devices(ti, device_not_poll_capable, NULL))
1558	return false;
1559	}
1560
1561	return true;
1562	}
1563
1564	/*
1565	* Check whether a table has no data devices attached using each
1566	* target's iterate_devices method.
1567	* Returns false if the result is unknown because a target doesn't
1568	* support iterate_devices.
1569	*/
1570	bool dm_table_has_no_data_devices(struct dm_table *t)
1571	{
1572	for (unsigned int i = 0; i < t->num_targets; i++) {
1573	struct dm_target *ti = dm_table_get_target(t, index: i);
1574	unsigned int num_devices = 0;
1575
1576	if (!ti->type->iterate_devices)
1577	return false;
1578
1579	ti->type->iterate_devices(ti, count_device, &num_devices);
1580	if (num_devices)
1581	return false;
1582	}
1583
1584	return true;
1585	}
1586
1587	static int device_not_zoned(struct dm_target *ti, struct dm_dev *dev,
1588	sector_t start, sector_t len, void *data)
1589	{
1590	bool *zoned = data;
1591
1592	return bdev_is_zoned(bdev: dev->bdev) != *zoned;
1593	}
1594
1595	static int device_is_zoned_model(struct dm_target *ti, struct dm_dev *dev,
1596	sector_t start, sector_t len, void *data)
1597	{
1598	return bdev_is_zoned(bdev: dev->bdev);
1599	}
1600
1601	/*
1602	* Check the device zoned model based on the target feature flag. If the target
1603	* has the DM_TARGET_ZONED_HM feature flag set, host-managed zoned devices are
1604	* also accepted but all devices must have the same zoned model. If the target
1605	* has the DM_TARGET_MIXED_ZONED_MODEL feature set, the devices can have any
1606	* zoned model with all zoned devices having the same zone size.
1607	*/
1608	static bool dm_table_supports_zoned(struct dm_table *t, bool zoned)
1609	{
1610	for (unsigned int i = 0; i < t->num_targets; i++) {
1611	struct dm_target *ti = dm_table_get_target(t, index: i);
1612
1613	/*
1614	* For the wildcard target (dm-error), if we do not have a
1615	* backing device, we must always return false. If we have a
1616	* backing device, the result must depend on checking zoned
1617	* model, like for any other target. So for this, check directly
1618	* if the target backing device is zoned as we get "false" when
1619	* dm-error was set without a backing device.
1620	*/
1621	if (dm_target_is_wildcard(ti->type) &&
1622	!ti->type->iterate_devices(ti, device_is_zoned_model, NULL))
1623	return false;
1624
1625	if (dm_target_supports_zoned_hm(ti->type)) {
1626	if (!ti->type->iterate_devices ||
1627	ti->type->iterate_devices(ti, device_not_zoned,
1628	&zoned))
1629	return false;
1630	} else if (!dm_target_supports_mixed_zoned_model(ti->type)) {
1631	if (zoned)
1632	return false;
1633	}
1634	}
1635
1636	return true;
1637	}
1638
1639	static int device_not_matches_zone_sectors(struct dm_target *ti, struct dm_dev *dev,
1640	sector_t start, sector_t len, void *data)
1641	{
1642	unsigned int *zone_sectors = data;
1643
1644	if (!bdev_is_zoned(bdev: dev->bdev))
1645	return 0;
1646	return bdev_zone_sectors(bdev: dev->bdev) != *zone_sectors;
1647	}
1648
1649	/*
1650	* Check consistency of zoned model and zone sectors across all targets. For
1651	* zone sectors, if the destination device is a zoned block device, it shall
1652	* have the specified zone_sectors.
1653	*/
1654	static int validate_hardware_zoned(struct dm_table *t, bool zoned,
1655	unsigned int zone_sectors)
1656	{
1657	if (!zoned)
1658	return 0;
1659
1660	if (!dm_table_supports_zoned(t, zoned)) {
1661	DMERR("%s: zoned model is not consistent across all devices",
1662	dm_device_name(t->md));
1663	return -EINVAL;
1664	}
1665
1666	/* Check zone size validity and compatibility */
1667	if (!zone_sectors || !is_power_of_2(n: zone_sectors))
1668	return -EINVAL;
1669
1670	if (dm_table_any_dev_attr(t, func: device_not_matches_zone_sectors, data: &zone_sectors)) {
1671	DMERR("%s: zone sectors is not consistent across all zoned devices",
1672	dm_device_name(t->md));
1673	return -EINVAL;
1674	}
1675
1676	return 0;
1677	}
1678
1679	/*
1680	* Establish the new table's queue_limits and validate them.
1681	*/
1682	int dm_calculate_queue_limits(struct dm_table *t,
1683	struct queue_limits *limits)
1684	{
1685	struct queue_limits ti_limits;
1686	unsigned int zone_sectors = 0;
1687	bool zoned = false;
1688
1689	blk_set_stacking_limits(lim: limits);
1690
1691	for (unsigned int i = 0; i < t->num_targets; i++) {
1692	struct dm_target *ti = dm_table_get_target(t, index: i);
1693
1694	blk_set_stacking_limits(lim: &ti_limits);
1695
1696	if (!ti->type->iterate_devices) {
1697	/* Set I/O hints portion of queue limits */
1698	if (ti->type->io_hints)
1699	ti->type->io_hints(ti, &ti_limits);
1700	goto combine_limits;
1701	}
1702
1703	/*
1704	* Combine queue limits of all the devices this target uses.
1705	*/
1706	ti->type->iterate_devices(ti, dm_set_device_limits,
1707	&ti_limits);
1708
1709	if (!zoned && ti_limits.zoned) {
1710	/*
1711	* After stacking all limits, validate all devices
1712	* in table support this zoned model and zone sectors.
1713	*/
1714	zoned = ti_limits.zoned;
1715	zone_sectors = ti_limits.chunk_sectors;
1716	}
1717
1718	/* Set I/O hints portion of queue limits */
1719	if (ti->type->io_hints)
1720	ti->type->io_hints(ti, &ti_limits);
1721
1722	/*
1723	* Check each device area is consistent with the target's
1724	* overall queue limits.
1725	*/
1726	if (ti->type->iterate_devices(ti, device_area_is_invalid,
1727	&ti_limits))
1728	return -EINVAL;
1729
1730	combine_limits:
1731	/*
1732	* Merge this target's queue limits into the overall limits
1733	* for the table.
1734	*/
1735	if (blk_stack_limits(t: limits, b: &ti_limits, offset: 0) < 0)
1736	DMWARN("%s: adding target device (start sect %llu len %llu) "
1737	"caused an alignment inconsistency",
1738	dm_device_name(t->md),
1739	(unsigned long long) ti->begin,
1740	(unsigned long long) ti->len);
1741	}
1742
1743	/*
1744	* Verify that the zoned model and zone sectors, as determined before
1745	* any .io_hints override, are the same across all devices in the table.
1746	* - this is especially relevant if .io_hints is emulating a disk-managed
1747	* zoned model on host-managed zoned block devices.
1748	* BUT...
1749	*/
1750	if (limits->zoned) {
1751	/*
1752	* ...IF the above limits stacking determined a zoned model
1753	* validate that all of the table's devices conform to it.
1754	*/
1755	zoned = limits->zoned;
1756	zone_sectors = limits->chunk_sectors;
1757	}
1758	if (validate_hardware_zoned(t, zoned, zone_sectors))
1759	return -EINVAL;
1760
1761	return validate_hardware_logical_block_alignment(t, limits);
1762	}
1763
1764	/*
1765	* Verify that all devices have an integrity profile that matches the
1766	* DM device's registered integrity profile. If the profiles don't
1767	* match then unregister the DM device's integrity profile.
1768	*/
1769	static void dm_table_verify_integrity(struct dm_table *t)
1770	{
1771	struct gendisk *template_disk = NULL;
1772
1773	if (t->integrity_added)
1774	return;
1775
1776	if (t->integrity_supported) {
1777	/*
1778	* Verify that the original integrity profile
1779	* matches all the devices in this table.
1780	*/
1781	template_disk = dm_table_get_integrity_disk(t);
1782	if (template_disk &&
1783	blk_integrity_compare(dm_disk(md: t->md), template_disk) >= 0)
1784	return;
1785	}
1786
1787	if (integrity_profile_exists(disk: dm_disk(md: t->md))) {
1788	DMWARN("%s: unable to establish an integrity profile",
1789	dm_device_name(t->md));
1790	blk_integrity_unregister(dm_disk(md: t->md));
1791	}
1792	}
1793
1794	static int device_flush_capable(struct dm_target *ti, struct dm_dev *dev,
1795	sector_t start, sector_t len, void *data)
1796	{
1797	unsigned long flush = (unsigned long) data;
1798	struct request_queue *q = bdev_get_queue(bdev: dev->bdev);
1799
1800	return (q->queue_flags & flush);
1801	}
1802
1803	static bool dm_table_supports_flush(struct dm_table *t, unsigned long flush)
1804	{
1805	/*
1806	* Require at least one underlying device to support flushes.
1807	* t->devices includes internal dm devices such as mirror logs
1808	* so we need to use iterate_devices here, which targets
1809	* supporting flushes must provide.
1810	*/
1811	for (unsigned int i = 0; i < t->num_targets; i++) {
1812	struct dm_target *ti = dm_table_get_target(t, index: i);
1813
1814	if (!ti->num_flush_bios)
1815	continue;
1816
1817	if (ti->flush_supported)
1818	return true;
1819
1820	if (ti->type->iterate_devices &&
1821	ti->type->iterate_devices(ti, device_flush_capable, (void *) flush))
1822	return true;
1823	}
1824
1825	return false;
1826	}
1827
1828	static int device_dax_write_cache_enabled(struct dm_target *ti,
1829	struct dm_dev *dev, sector_t start,
1830	sector_t len, void *data)
1831	{
1832	struct dax_device *dax_dev = dev->dax_dev;
1833
1834	if (!dax_dev)
1835	return false;
1836
1837	if (dax_write_cache_enabled(dax_dev))
1838	return true;
1839	return false;
1840	}
1841
1842	static int device_is_rotational(struct dm_target *ti, struct dm_dev *dev,
1843	sector_t start, sector_t len, void *data)
1844	{
1845	return !bdev_nonrot(bdev: dev->bdev);
1846	}
1847
1848	static int device_is_not_random(struct dm_target *ti, struct dm_dev *dev,
1849	sector_t start, sector_t len, void *data)
1850	{
1851	struct request_queue *q = bdev_get_queue(bdev: dev->bdev);
1852
1853	return !blk_queue_add_random(q);
1854	}
1855
1856	static int device_not_write_zeroes_capable(struct dm_target *ti, struct dm_dev *dev,
1857	sector_t start, sector_t len, void *data)
1858	{
1859	struct request_queue *q = bdev_get_queue(bdev: dev->bdev);
1860
1861	return !q->limits.max_write_zeroes_sectors;
1862	}
1863
1864	static bool dm_table_supports_write_zeroes(struct dm_table *t)
1865	{
1866	for (unsigned int i = 0; i < t->num_targets; i++) {
1867	struct dm_target *ti = dm_table_get_target(t, index: i);
1868
1869	if (!ti->num_write_zeroes_bios)
1870	return false;
1871
1872	if (!ti->type->iterate_devices ||
1873	ti->type->iterate_devices(ti, device_not_write_zeroes_capable, NULL))
1874	return false;
1875	}
1876
1877	return true;
1878	}
1879
1880	static int device_not_nowait_capable(struct dm_target *ti, struct dm_dev *dev,
1881	sector_t start, sector_t len, void *data)
1882	{
1883	return !bdev_nowait(bdev: dev->bdev);
1884	}
1885
1886	static bool dm_table_supports_nowait(struct dm_table *t)
1887	{
1888	for (unsigned int i = 0; i < t->num_targets; i++) {
1889	struct dm_target *ti = dm_table_get_target(t, index: i);
1890
1891	if (!dm_target_supports_nowait(ti->type))
1892	return false;
1893
1894	if (!ti->type->iterate_devices ||
1895	ti->type->iterate_devices(ti, device_not_nowait_capable, NULL))
1896	return false;
1897	}
1898
1899	return true;
1900	}
1901
1902	static int device_not_discard_capable(struct dm_target *ti, struct dm_dev *dev,
1903	sector_t start, sector_t len, void *data)
1904	{
1905	return !bdev_max_discard_sectors(bdev: dev->bdev);
1906	}
1907
1908	static bool dm_table_supports_discards(struct dm_table *t)
1909	{
1910	for (unsigned int i = 0; i < t->num_targets; i++) {
1911	struct dm_target *ti = dm_table_get_target(t, index: i);
1912
1913	if (!ti->num_discard_bios)
1914	return false;
1915
1916	/*
1917	* Either the target provides discard support (as implied by setting
1918	* 'discards_supported') or it relies on _all_ data devices having
1919	* discard support.
1920	*/
1921	if (!ti->discards_supported &&
1922	(!ti->type->iterate_devices ||
1923	ti->type->iterate_devices(ti, device_not_discard_capable, NULL)))
1924	return false;
1925	}
1926
1927	return true;
1928	}
1929
1930	static int device_not_secure_erase_capable(struct dm_target *ti,
1931	struct dm_dev *dev, sector_t start,
1932	sector_t len, void *data)
1933	{
1934	return !bdev_max_secure_erase_sectors(bdev: dev->bdev);
1935	}
1936
1937	static bool dm_table_supports_secure_erase(struct dm_table *t)
1938	{
1939	for (unsigned int i = 0; i < t->num_targets; i++) {
1940	struct dm_target *ti = dm_table_get_target(t, index: i);
1941
1942	if (!ti->num_secure_erase_bios)
1943	return false;
1944
1945	if (!ti->type->iterate_devices ||
1946	ti->type->iterate_devices(ti, device_not_secure_erase_capable, NULL))
1947	return false;
1948	}
1949
1950	return true;
1951	}
1952
1953	static int device_requires_stable_pages(struct dm_target *ti,
1954	struct dm_dev *dev, sector_t start,
1955	sector_t len, void *data)
1956	{
1957	return bdev_stable_writes(bdev: dev->bdev);
1958	}
1959
1960	int dm_table_set_restrictions(struct dm_table *t, struct request_queue *q,
1961	struct queue_limits *limits)
1962	{
1963	bool wc = false, fua = false;
1964	int r;
1965
1966	/*
1967	* Copy table's limits to the DM device's request_queue
1968	*/
1969	q->limits = *limits;
1970
1971	if (dm_table_supports_nowait(t))
1972	blk_queue_flag_set(QUEUE_FLAG_NOWAIT, q);
1973	else
1974	blk_queue_flag_clear(QUEUE_FLAG_NOWAIT, q);
1975
1976	if (!dm_table_supports_discards(t)) {
1977	q->limits.max_discard_sectors = 0;
1978	q->limits.max_hw_discard_sectors = 0;
1979	q->limits.discard_granularity = 0;
1980	q->limits.discard_alignment = 0;
1981	q->limits.discard_misaligned = 0;
1982	}
1983
1984	if (!dm_table_supports_secure_erase(t))
1985	q->limits.max_secure_erase_sectors = 0;
1986
1987	if (dm_table_supports_flush(t, flush: (1UL << QUEUE_FLAG_WC))) {
1988	wc = true;
1989	if (dm_table_supports_flush(t, flush: (1UL << QUEUE_FLAG_FUA)))
1990	fua = true;
1991	}
1992	blk_queue_write_cache(q, enabled: wc, fua);
1993
1994	if (dm_table_supports_dax(t, iterate_fn: device_not_dax_capable)) {
1995	blk_queue_flag_set(QUEUE_FLAG_DAX, q);
1996	if (dm_table_supports_dax(t, iterate_fn: device_not_dax_synchronous_capable))
1997	set_dax_synchronous(t->md->dax_dev);
1998	} else
1999	blk_queue_flag_clear(QUEUE_FLAG_DAX, q);
2000
2001	if (dm_table_any_dev_attr(t, func: device_dax_write_cache_enabled, NULL))
2002	dax_write_cache(dax_dev: t->md->dax_dev, wc: true);
2003
2004	/* Ensure that all underlying devices are non-rotational. */
2005	if (dm_table_any_dev_attr(t, func: device_is_rotational, NULL))
2006	blk_queue_flag_clear(QUEUE_FLAG_NONROT, q);
2007	else
2008	blk_queue_flag_set(QUEUE_FLAG_NONROT, q);
2009
2010	if (!dm_table_supports_write_zeroes(t))
2011	q->limits.max_write_zeroes_sectors = 0;
2012
2013	dm_table_verify_integrity(t);
2014
2015	/*
2016	* Some devices don't use blk_integrity but still want stable pages
2017	* because they do their own checksumming.
2018	* If any underlying device requires stable pages, a table must require
2019	* them as well. Only targets that support iterate_devices are considered:
2020	* don't want error, zero, etc to require stable pages.
2021	*/
2022	if (dm_table_any_dev_attr(t, func: device_requires_stable_pages, NULL))
2023	blk_queue_flag_set(QUEUE_FLAG_STABLE_WRITES, q);
2024	else
2025	blk_queue_flag_clear(QUEUE_FLAG_STABLE_WRITES, q);
2026
2027	/*
2028	* Determine whether or not this queue's I/O timings contribute
2029	* to the entropy pool, Only request-based targets use this.
2030	* Clear QUEUE_FLAG_ADD_RANDOM if any underlying device does not
2031	* have it set.
2032	*/
2033	if (blk_queue_add_random(q) &&
2034	dm_table_any_dev_attr(t, func: device_is_not_random, NULL))
2035	blk_queue_flag_clear(QUEUE_FLAG_ADD_RANDOM, q);
2036
2037	/*
2038	* For a zoned target, setup the zones related queue attributes
2039	* and resources necessary for zone append emulation if necessary.
2040	*/
2041	if (blk_queue_is_zoned(q)) {
2042	r = dm_set_zones_restrictions(t, q);
2043	if (r)
2044	return r;
2045	if (!static_key_enabled(&zoned_enabled.key))
2046	static_branch_enable(&zoned_enabled);
2047	}
2048
2049	dm_update_crypto_profile(q, t);
2050	disk_update_readahead(disk: t->md->disk);
2051
2052	/*
2053	* Check for request-based device is left to
2054	* dm_mq_init_request_queue()->blk_mq_init_allocated_queue().
2055	*
2056	* For bio-based device, only set QUEUE_FLAG_POLL when all
2057	* underlying devices supporting polling.
2058	*/
2059	if (__table_type_bio_based(table_type: t->type)) {
2060	if (dm_table_supports_poll(t))
2061	blk_queue_flag_set(QUEUE_FLAG_POLL, q);
2062	else
2063	blk_queue_flag_clear(QUEUE_FLAG_POLL, q);
2064	}
2065
2066	return 0;
2067	}
2068
2069	struct list_head *dm_table_get_devices(struct dm_table *t)
2070	{
2071	return &t->devices;
2072	}
2073
2074	blk_mode_t dm_table_get_mode(struct dm_table *t)
2075	{
2076	return t->mode;
2077	}
2078	EXPORT_SYMBOL(dm_table_get_mode);
2079
2080	enum suspend_mode {
2081	PRESUSPEND,
2082	PRESUSPEND_UNDO,
2083	POSTSUSPEND,
2084	};
2085
2086	static void suspend_targets(struct dm_table *t, enum suspend_mode mode)
2087	{
2088	lockdep_assert_held(&t->md->suspend_lock);
2089
2090	for (unsigned int i = 0; i < t->num_targets; i++) {
2091	struct dm_target *ti = dm_table_get_target(t, index: i);
2092
2093	switch (mode) {
2094	case PRESUSPEND:
2095	if (ti->type->presuspend)
2096	ti->type->presuspend(ti);
2097	break;
2098	case PRESUSPEND_UNDO:
2099	if (ti->type->presuspend_undo)
2100	ti->type->presuspend_undo(ti);
2101	break;
2102	case POSTSUSPEND:
2103	if (ti->type->postsuspend)
2104	ti->type->postsuspend(ti);
2105	break;
2106	}
2107	}
2108	}
2109
2110	void dm_table_presuspend_targets(struct dm_table *t)
2111	{
2112	if (!t)
2113	return;
2114
2115	suspend_targets(t, mode: PRESUSPEND);
2116	}
2117
2118	void dm_table_presuspend_undo_targets(struct dm_table *t)
2119	{
2120	if (!t)
2121	return;
2122
2123	suspend_targets(t, mode: PRESUSPEND_UNDO);
2124	}
2125
2126	void dm_table_postsuspend_targets(struct dm_table *t)
2127	{
2128	if (!t)
2129	return;
2130
2131	suspend_targets(t, mode: POSTSUSPEND);
2132	}
2133
2134	int dm_table_resume_targets(struct dm_table *t)
2135	{
2136	unsigned int i;
2137	int r = 0;
2138
2139	lockdep_assert_held(&t->md->suspend_lock);
2140
2141	for (i = 0; i < t->num_targets; i++) {
2142	struct dm_target *ti = dm_table_get_target(t, index: i);
2143
2144	if (!ti->type->preresume)
2145	continue;
2146
2147	r = ti->type->preresume(ti);
2148	if (r) {
2149	DMERR("%s: %s: preresume failed, error = %d",
2150	dm_device_name(t->md), ti->type->name, r);
2151	return r;
2152	}
2153	}
2154
2155	for (i = 0; i < t->num_targets; i++) {
2156	struct dm_target *ti = dm_table_get_target(t, index: i);
2157
2158	if (ti->type->resume)
2159	ti->type->resume(ti);
2160	}
2161
2162	return 0;
2163	}
2164
2165	struct mapped_device *dm_table_get_md(struct dm_table *t)
2166	{
2167	return t->md;
2168	}
2169	EXPORT_SYMBOL(dm_table_get_md);
2170
2171	const char *dm_table_device_name(struct dm_table *t)
2172	{
2173	return dm_device_name(md: t->md);
2174	}
2175	EXPORT_SYMBOL_GPL(dm_table_device_name);
2176
2177	void dm_table_run_md_queue_async(struct dm_table *t)
2178	{
2179	if (!dm_table_request_based(t))
2180	return;
2181
2182	if (t->md->queue)
2183	blk_mq_run_hw_queues(q: t->md->queue, async: true);
2184	}
2185	EXPORT_SYMBOL(dm_table_run_md_queue_async);
2186
2187