Resguardo inicial

linux-7.2.3/drivers/md/persistent-data/Kconfig

@@ -0,0 +1,10 @@
+# SPDX-License-Identifier: GPL-2.0-only
+config DM_PERSISTENT_DATA
+       tristate
+       depends on BLK_DEV_DM
+       select CRC32
+       select DM_BUFIO
+	help
+	 Library providing immutable on-disk data structure support for
+	 device-mapper targets such as the thin provisioning target.
+

linux-7.2.3/drivers/md/persistent-data/Makefile

@@ -0,0 +1,13 @@
+# SPDX-License-Identifier: GPL-2.0
+obj-$(CONFIG_DM_PERSISTENT_DATA) += dm-persistent-data.o
+dm-persistent-data-objs := \
+	dm-array.o \
+	dm-bitset.o \
+	dm-block-manager.o \
+	dm-space-map-common.o \
+	dm-space-map-disk.o \
+	dm-space-map-metadata.o \
+	dm-transaction-manager.o \
+	dm-btree.o \
+	dm-btree-remove.o \
+	dm-btree-spine.o

linux-7.2.3/drivers/md/persistent-data/dm-array.h

@@ -0,0 +1,220 @@
+/* SPDX-License-Identifier: GPL-2.0-only */
+/*
+ * Copyright (C) 2012 Red Hat, Inc.
+ *
+ * This file is released under the GPL.
+ */
+#ifndef _LINUX_DM_ARRAY_H
+#define _LINUX_DM_ARRAY_H
+
+#include "dm-btree.h"
+
+/*----------------------------------------------------------------*/
+
+/*
+ * The dm-array is a persistent version of an array.  It packs the data
+ * more efficiently than a btree which will result in less disk space use,
+ * and a performance boost.  The element get and set operations are still
+ * O(ln(n)), but with a much smaller constant.
+ *
+ * The value type structure is reused from the btree type to support proper
+ * reference counting of values.
+ *
+ * The arrays implicitly know their length, and bounds are checked for
+ * lookups and updated.  It doesn't store this in an accessible place
+ * because it would waste a whole metadata block.  Make sure you store the
+ * size along with the array root in your encompassing data.
+ *
+ * Array entries are indexed via an unsigned integer starting from zero.
+ * Arrays are not sparse; if you resize an array to have 'n' entries then
+ * 'n - 1' will be the last valid index.
+ *
+ * Typical use:
+ *
+ * a) initialise a dm_array_info structure.  This describes the array
+ *    values and ties it into a specific transaction manager.  It holds no
+ *    instance data; the same info can be used for many similar arrays if
+ *    you wish.
+ *
+ * b) Get yourself a root.  The root is the index of a block of data on the
+ *    disk that holds a particular instance of an array.  You may have a
+ *    pre existing root in your metadata that you wish to use, or you may
+ *    want to create a brand new, empty array with dm_array_empty().
+ *
+ * Like the other data structures in this library, dm_array objects are
+ * immutable between transactions.  Update functions will return you the
+ * root for a _new_ array.  If you've incremented the old root, via
+ * dm_tm_inc(), before calling the update function you may continue to use
+ * it in parallel with the new root.
+ *
+ * c) resize an array with dm_array_resize().
+ *
+ * d) Get a value from the array with dm_array_get_value().
+ *
+ * e) Set a value in the array with dm_array_set_value().
+ *
+ * f) Walk an array of values in index order with dm_array_walk().  More
+ *    efficient than making many calls to dm_array_get_value().
+ *
+ * g) Destroy the array with dm_array_del().  This tells the transaction
+ *    manager that you're no longer using this data structure so it can
+ *    recycle it's blocks.  (dm_array_dec() would be a better name for it,
+ *    but del is in keeping with dm_btree_del()).
+ */
+
+/*
+ * Describes an array.  Don't initialise this structure yourself, use the
+ * init function below.
+ */
+struct dm_array_info {
+	struct dm_transaction_manager *tm;
+	struct dm_btree_value_type value_type;
+	struct dm_btree_info btree_info;
+};
+
+/*
+ * Sets up a dm_array_info structure.  You don't need to do anything with
+ * this structure when you finish using it.
+ *
+ * info - the structure being filled in.
+ * tm   - the transaction manager that should supervise this structure.
+ * vt   - describes the leaf values.
+ */
+void dm_array_info_init(struct dm_array_info *info,
+			struct dm_transaction_manager *tm,
+			struct dm_btree_value_type *vt);
+
+/*
+ * Create an empty, zero length array.
+ *
+ * info - describes the array
+ * root - on success this will be filled out with the root block
+ */
+int dm_array_empty(struct dm_array_info *info, dm_block_t *root);
+
+/*
+ * Resizes the array.
+ *
+ * info - describes the array
+ * root - the root block of the array on disk
+ * old_size - the caller is responsible for remembering the size of
+ *            the array
+ * new_size - can be bigger or smaller than old_size
+ * value - if we're growing the array the new entries will have this value
+ * new_root - on success, points to the new root block
+ *
+ * If growing the inc function for 'value' will be called the appropriate
+ * number of times.  So if the caller is holding a reference they may want
+ * to drop it.
+ */
+int dm_array_resize(struct dm_array_info *info, dm_block_t root,
+		    uint32_t old_size, uint32_t new_size,
+		    const void *value, dm_block_t *new_root)
+	__dm_written_to_disk(value);
+
+/*
+ * Creates a new array populated with values provided by a callback
+ * function.  This is more efficient than creating an empty array,
+ * resizing, and then setting values since that process incurs a lot of
+ * copying.
+ *
+ * Assumes 32bit values for now since it's only used by the cache hint
+ * array.
+ *
+ * info - describes the array
+ * root - the root block of the array on disk
+ * size - the number of entries in the array
+ * fn - the callback
+ * context - passed to the callback
+ */
+typedef int (*value_fn)(uint32_t index, void *value_le, void *context);
+int dm_array_new(struct dm_array_info *info, dm_block_t *root,
+		 uint32_t size, value_fn fn, void *context);
+
+/*
+ * Frees a whole array.  The value_type's decrement operation will be called
+ * for all values in the array
+ */
+int dm_array_del(struct dm_array_info *info, dm_block_t root);
+
+/*
+ * Lookup a value in the array
+ *
+ * info - describes the array
+ * root - root block of the array
+ * index - array index
+ * value - the value to be read.  Will be in on-disk format of course.
+ *
+ * -ENODATA will be returned if the index is out of bounds.
+ */
+int dm_array_get_value(struct dm_array_info *info, dm_block_t root,
+		       uint32_t index, void *value);
+
+/*
+ * Set an entry in the array.
+ *
+ * info - describes the array
+ * root - root block of the array
+ * index - array index
+ * value - value to be written to disk.  Make sure you confirm the value is
+ *         in on-disk format with__dm_bless_for_disk() before calling.
+ * new_root - the new root block
+ *
+ * The old value being overwritten will be decremented, the new value
+ * incremented.
+ *
+ * -ENODATA will be returned if the index is out of bounds.
+ */
+int dm_array_set_value(struct dm_array_info *info, dm_block_t root,
+		       uint32_t index, const void *value, dm_block_t *new_root)
+	__dm_written_to_disk(value);
+
+/*
+ * Walk through all the entries in an array.
+ *
+ * info - describes the array
+ * root - root block of the array
+ * fn - called back for every element
+ * context - passed to the callback
+ */
+int dm_array_walk(struct dm_array_info *info, dm_block_t root,
+		  int (*fn)(void *context, uint64_t key, void *leaf),
+		  void *context);
+
+/*----------------------------------------------------------------*/
+
+/*
+ * Cursor api.
+ *
+ * This lets you iterate through all the entries in an array efficiently
+ * (it will preload metadata).
+ *
+ * I'm using a cursor, rather than a walk function with a callback because
+ * the cache target needs to iterate both the mapping and hint arrays in
+ * unison.
+ */
+struct dm_array_cursor {
+	struct dm_array_info *info;
+	struct dm_btree_cursor cursor;
+
+	struct dm_block *block;
+	struct array_block *ab;
+	unsigned int index;
+};
+
+int dm_array_cursor_begin(struct dm_array_info *info,
+			  dm_block_t root, struct dm_array_cursor *c);
+void dm_array_cursor_end(struct dm_array_cursor *c);
+
+uint32_t dm_array_cursor_index(struct dm_array_cursor *c);
+int dm_array_cursor_next(struct dm_array_cursor *c);
+int dm_array_cursor_skip(struct dm_array_cursor *c, uint32_t count);
+
+/*
+ * value_le is only valid while the cursor points at the current value.
+ */
+void dm_array_cursor_get_value(struct dm_array_cursor *c, void **value_le);
+
+/*----------------------------------------------------------------*/
+
+#endif	/* _LINUX_DM_ARRAY_H */

linux-7.2.3/drivers/md/persistent-data/dm-bitset.c

@@ -0,0 +1,319 @@
+// SPDX-License-Identifier: GPL-2.0-only
+/*
+ * Copyright (C) 2012 Red Hat, Inc.
+ *
+ * This file is released under the GPL.
+ */
+
+#include "dm-bitset.h"
+#include "dm-transaction-manager.h"
+
+#include <linux/export.h>
+#include <linux/device-mapper.h>
+
+#define DM_MSG_PREFIX "bitset"
+#define BITS_PER_ARRAY_ENTRY 64
+
+/*----------------------------------------------------------------*/
+
+static struct dm_btree_value_type bitset_bvt = {
+	.context = NULL,
+	.size = sizeof(__le64),
+	.inc = NULL,
+	.dec = NULL,
+	.equal = NULL,
+};
+
+/*----------------------------------------------------------------*/
+
+void dm_disk_bitset_init(struct dm_transaction_manager *tm,
+			 struct dm_disk_bitset *info)
+{
+	dm_array_info_init(&info->array_info, tm, &bitset_bvt);
+	info->current_index_set = false;
+}
+EXPORT_SYMBOL_GPL(dm_disk_bitset_init);
+
+int dm_bitset_empty(struct dm_disk_bitset *info, dm_block_t *root)
+{
+	return dm_array_empty(&info->array_info, root);
+}
+EXPORT_SYMBOL_GPL(dm_bitset_empty);
+
+struct packer_context {
+	bit_value_fn fn;
+	unsigned int nr_bits;
+	void *context;
+};
+
+static int pack_bits(uint32_t index, void *value, void *context)
+{
+	int r;
+	struct packer_context *p = context;
+	unsigned int bit, nr = min(64u, p->nr_bits - (index * 64));
+	uint64_t word = 0;
+	bool bv;
+
+	for (bit = 0; bit < nr; bit++) {
+		r = p->fn(index * 64 + bit, &bv, p->context);
+		if (r)
+			return r;
+
+		if (bv)
+			set_bit(bit, (unsigned long *) &word);
+		else
+			clear_bit(bit, (unsigned long *) &word);
+	}
+
+	*((__le64 *) value) = cpu_to_le64(word);
+
+	return 0;
+}
+
+int dm_bitset_new(struct dm_disk_bitset *info, dm_block_t *root,
+		  uint32_t size, bit_value_fn fn, void *context)
+{
+	struct packer_context p;
+
+	p.fn = fn;
+	p.nr_bits = size;
+	p.context = context;
+
+	return dm_array_new(&info->array_info, root, dm_div_up(size, 64), pack_bits, &p);
+}
+EXPORT_SYMBOL_GPL(dm_bitset_new);
+
+int dm_bitset_resize(struct dm_disk_bitset *info, dm_block_t root,
+		     uint32_t old_nr_entries, uint32_t new_nr_entries,
+		     bool default_value, dm_block_t *new_root)
+{
+	uint32_t old_blocks = dm_div_up(old_nr_entries, BITS_PER_ARRAY_ENTRY);
+	uint32_t new_blocks = dm_div_up(new_nr_entries, BITS_PER_ARRAY_ENTRY);
+	__le64 value = default_value ? cpu_to_le64(~0) : cpu_to_le64(0);
+
+	__dm_bless_for_disk(&value);
+	return dm_array_resize(&info->array_info, root, old_blocks, new_blocks,
+			       &value, new_root);
+}
+EXPORT_SYMBOL_GPL(dm_bitset_resize);
+
+int dm_bitset_del(struct dm_disk_bitset *info, dm_block_t root)
+{
+	return dm_array_del(&info->array_info, root);
+}
+EXPORT_SYMBOL_GPL(dm_bitset_del);
+
+int dm_bitset_flush(struct dm_disk_bitset *info, dm_block_t root,
+		    dm_block_t *new_root)
+{
+	int r;
+	__le64 value;
+
+	if (!info->current_index_set || !info->dirty)
+		return 0;
+
+	value = cpu_to_le64(info->current_bits);
+
+	__dm_bless_for_disk(&value);
+	r = dm_array_set_value(&info->array_info, root, info->current_index,
+			       &value, new_root);
+	if (r)
+		return r;
+
+	info->current_index_set = false;
+	info->dirty = false;
+
+	return 0;
+}
+EXPORT_SYMBOL_GPL(dm_bitset_flush);
+
+static int read_bits(struct dm_disk_bitset *info, dm_block_t root,
+		     uint32_t array_index)
+{
+	int r;
+	__le64 value;
+
+	r = dm_array_get_value(&info->array_info, root, array_index, &value);
+	if (r)
+		return r;
+
+	info->current_bits = le64_to_cpu(value);
+	info->current_index_set = true;
+	info->current_index = array_index;
+	info->dirty = false;
+
+	return 0;
+}
+
+static int get_array_entry(struct dm_disk_bitset *info, dm_block_t root,
+			   uint32_t index, dm_block_t *new_root)
+{
+	int r;
+	unsigned int array_index = index / BITS_PER_ARRAY_ENTRY;
+
+	if (info->current_index_set) {
+		if (info->current_index == array_index)
+			return 0;
+
+		r = dm_bitset_flush(info, root, new_root);
+		if (r)
+			return r;
+	}
+
+	return read_bits(info, root, array_index);
+}
+
+int dm_bitset_set_bit(struct dm_disk_bitset *info, dm_block_t root,
+		      uint32_t index, dm_block_t *new_root)
+{
+	int r;
+	unsigned int b = index % BITS_PER_ARRAY_ENTRY;
+
+	r = get_array_entry(info, root, index, new_root);
+	if (r)
+		return r;
+
+	set_bit(b, (unsigned long *) &info->current_bits);
+	info->dirty = true;
+
+	return 0;
+}
+EXPORT_SYMBOL_GPL(dm_bitset_set_bit);
+
+int dm_bitset_clear_bit(struct dm_disk_bitset *info, dm_block_t root,
+			uint32_t index, dm_block_t *new_root)
+{
+	int r;
+	unsigned int b = index % BITS_PER_ARRAY_ENTRY;
+
+	r = get_array_entry(info, root, index, new_root);
+	if (r)
+		return r;
+
+	clear_bit(b, (unsigned long *) &info->current_bits);
+	info->dirty = true;
+
+	return 0;
+}
+EXPORT_SYMBOL_GPL(dm_bitset_clear_bit);
+
+int dm_bitset_test_bit(struct dm_disk_bitset *info, dm_block_t root,
+		       uint32_t index, dm_block_t *new_root, bool *result)
+{
+	int r;
+	unsigned int b = index % BITS_PER_ARRAY_ENTRY;
+
+	r = get_array_entry(info, root, index, new_root);
+	if (r)
+		return r;
+
+	*result = test_bit(b, (unsigned long *) &info->current_bits);
+	return 0;
+}
+EXPORT_SYMBOL_GPL(dm_bitset_test_bit);
+
+static int cursor_next_array_entry(struct dm_bitset_cursor *c)
+{
+	int r;
+	__le64 *value;
+
+	r = dm_array_cursor_next(&c->cursor);
+	if (r)
+		return r;
+
+	dm_array_cursor_get_value(&c->cursor, (void **) &value);
+	c->array_index++;
+	c->bit_index = 0;
+	c->current_bits = le64_to_cpu(*value);
+	return 0;
+}
+
+int dm_bitset_cursor_begin(struct dm_disk_bitset *info,
+			   dm_block_t root, uint32_t nr_entries,
+			   struct dm_bitset_cursor *c)
+{
+	int r;
+	__le64 *value;
+
+	if (!nr_entries)
+		return -ENODATA;
+
+	c->info = info;
+	c->entries_remaining = nr_entries;
+
+	r = dm_array_cursor_begin(&info->array_info, root, &c->cursor);
+	if (r)
+		return r;
+
+	dm_array_cursor_get_value(&c->cursor, (void **) &value);
+	c->array_index = 0;
+	c->bit_index = 0;
+	c->current_bits = le64_to_cpu(*value);
+
+	return r;
+}
+EXPORT_SYMBOL_GPL(dm_bitset_cursor_begin);
+
+void dm_bitset_cursor_end(struct dm_bitset_cursor *c)
+{
+	return dm_array_cursor_end(&c->cursor);
+}
+EXPORT_SYMBOL_GPL(dm_bitset_cursor_end);
+
+int dm_bitset_cursor_next(struct dm_bitset_cursor *c)
+{
+	int r = 0;
+
+	if (!c->entries_remaining)
+		return -ENODATA;
+
+	c->entries_remaining--;
+	if (++c->bit_index > 63)
+		r = cursor_next_array_entry(c);
+
+	return r;
+}
+EXPORT_SYMBOL_GPL(dm_bitset_cursor_next);
+
+int dm_bitset_cursor_skip(struct dm_bitset_cursor *c, uint32_t count)
+{
+	int r;
+	__le64 *value;
+	uint32_t nr_array_skip;
+	uint32_t remaining_in_word = 64 - c->bit_index;
+
+	if (c->entries_remaining < count)
+		return -ENODATA;
+
+	if (count < remaining_in_word) {
+		c->bit_index += count;
+		c->entries_remaining -= count;
+		return 0;
+
+	} else {
+		c->entries_remaining -= remaining_in_word;
+		count -= remaining_in_word;
+	}
+
+	nr_array_skip = (count / 64) + 1;
+	r = dm_array_cursor_skip(&c->cursor, nr_array_skip);
+	if (r)
+		return r;
+
+	dm_array_cursor_get_value(&c->cursor, (void **) &value);
+	c->entries_remaining -= count;
+	c->array_index += nr_array_skip;
+	c->bit_index = count & 63;
+	c->current_bits = le64_to_cpu(*value);
+
+	return 0;
+}
+EXPORT_SYMBOL_GPL(dm_bitset_cursor_skip);
+
+bool dm_bitset_cursor_get_value(struct dm_bitset_cursor *c)
+{
+	return test_bit(c->bit_index, (unsigned long *) &c->current_bits);
+}
+EXPORT_SYMBOL_GPL(dm_bitset_cursor_get_value);
+
+/*----------------------------------------------------------------*/

linux-7.2.3/drivers/md/persistent-data/dm-bitset.h

@@ -0,0 +1,206 @@
+/* SPDX-License-Identifier: GPL-2.0-only */
+/*
+ * Copyright (C) 2012 Red Hat, Inc.
+ *
+ * This file is released under the GPL.
+ */
+#ifndef _LINUX_DM_BITSET_H
+#define _LINUX_DM_BITSET_H
+
+#include "dm-array.h"
+
+/*----------------------------------------------------------------*/
+
+/*
+ * This bitset type is a thin wrapper round a dm_array of 64bit words.  It
+ * uses a tiny, one word cache to reduce the number of array lookups and so
+ * increase performance.
+ *
+ * Like the dm-array that it's based on, the caller needs to keep track of
+ * the size of the bitset separately.  The underlying dm-array implicitly
+ * knows how many words it's storing and will return -ENODATA if you try
+ * and access an out of bounds word.  However, an out of bounds bit in the
+ * final word will _not_ be detected, you have been warned.
+ *
+ * Bits are indexed from zero.
+
+ * Typical use:
+ *
+ * a) Initialise a dm_disk_bitset structure with dm_disk_bitset_init().
+ *    This describes the bitset and includes the cache.  It's not called it
+ *    dm_bitset_info in line with other data structures because it does
+ *    include instance data.
+ *
+ * b) Get yourself a root.  The root is the index of a block of data on the
+ *    disk that holds a particular instance of an bitset.  You may have a
+ *    pre existing root in your metadata that you wish to use, or you may
+ *    want to create a brand new, empty bitset with dm_bitset_empty().
+ *
+ * Like the other data structures in this library, dm_bitset objects are
+ * immutable between transactions.  Update functions will return you the
+ * root for a _new_ array.  If you've incremented the old root, via
+ * dm_tm_inc(), before calling the update function you may continue to use
+ * it in parallel with the new root.
+ *
+ * Even read operations may trigger the cache to be flushed and as such
+ * return a root for a new, updated bitset.
+ *
+ * c) resize a bitset with dm_bitset_resize().
+ *
+ * d) Set a bit with dm_bitset_set_bit().
+ *
+ * e) Clear a bit with dm_bitset_clear_bit().
+ *
+ * f) Test a bit with dm_bitset_test_bit().
+ *
+ * g) Flush all updates from the cache with dm_bitset_flush().
+ *
+ * h) Destroy the bitset with dm_bitset_del().  This tells the transaction
+ *    manager that you're no longer using this data structure so it can
+ *    recycle it's blocks.  (dm_bitset_dec() would be a better name for it,
+ *    but del is in keeping with dm_btree_del()).
+ */
+
+/*
+ * Opaque object.  Unlike dm_array_info, you should have one of these per
+ * bitset.  Initialise with dm_disk_bitset_init().
+ */
+struct dm_disk_bitset {
+	struct dm_array_info array_info;
+
+	uint32_t current_index;
+	uint64_t current_bits;
+
+	bool current_index_set:1;
+	bool dirty:1;
+};
+
+/*
+ * Sets up a dm_disk_bitset structure.  You don't need to do anything with
+ * this structure when you finish using it.
+ *
+ * tm - the transaction manager that should supervise this structure
+ * info - the structure being initialised
+ */
+void dm_disk_bitset_init(struct dm_transaction_manager *tm,
+			 struct dm_disk_bitset *info);
+
+/*
+ * Create an empty, zero length bitset.
+ *
+ * info - describes the bitset
+ * new_root - on success, points to the new root block
+ */
+int dm_bitset_empty(struct dm_disk_bitset *info, dm_block_t *new_root);
+
+/*
+ * Creates a new bitset populated with values provided by a callback
+ * function.  This is more efficient than creating an empty bitset,
+ * resizing, and then setting values since that process incurs a lot of
+ * copying.
+ *
+ * info - describes the array
+ * root - the root block of the array on disk
+ * size - the number of entries in the array
+ * fn - the callback
+ * context - passed to the callback
+ */
+typedef int (*bit_value_fn)(uint32_t index, bool *value, void *context);
+int dm_bitset_new(struct dm_disk_bitset *info, dm_block_t *root,
+		  uint32_t size, bit_value_fn fn, void *context);
+
+/*
+ * Resize the bitset.
+ *
+ * info - describes the bitset
+ * old_root - the root block of the array on disk
+ * old_nr_entries - the number of bits in the old bitset
+ * new_nr_entries - the number of bits you want in the new bitset
+ * default_value - the value for any new bits
+ * new_root - on success, points to the new root block
+ */
+int dm_bitset_resize(struct dm_disk_bitset *info, dm_block_t old_root,
+		     uint32_t old_nr_entries, uint32_t new_nr_entries,
+		     bool default_value, dm_block_t *new_root);
+
+/*
+ * Frees the bitset.
+ */
+int dm_bitset_del(struct dm_disk_bitset *info, dm_block_t root);
+
+/*
+ * Set a bit.
+ *
+ * info - describes the bitset
+ * root - the root block of the bitset
+ * index - the bit index
+ * new_root - on success, points to the new root block
+ *
+ * -ENODATA will be returned if the index is out of bounds.
+ */
+int dm_bitset_set_bit(struct dm_disk_bitset *info, dm_block_t root,
+		      uint32_t index, dm_block_t *new_root);
+
+/*
+ * Clears a bit.
+ *
+ * info - describes the bitset
+ * root - the root block of the bitset
+ * index - the bit index
+ * new_root - on success, points to the new root block
+ *
+ * -ENODATA will be returned if the index is out of bounds.
+ */
+int dm_bitset_clear_bit(struct dm_disk_bitset *info, dm_block_t root,
+			uint32_t index, dm_block_t *new_root);
+
+/*
+ * Tests a bit.
+ *
+ * info - describes the bitset
+ * root - the root block of the bitset
+ * index - the bit index
+ * new_root - on success, points to the new root block (cached values may have been written)
+ * result - the bit value you're after
+ *
+ * -ENODATA will be returned if the index is out of bounds.
+ */
+int dm_bitset_test_bit(struct dm_disk_bitset *info, dm_block_t root,
+		       uint32_t index, dm_block_t *new_root, bool *result);
+
+/*
+ * Flush any cached changes to disk.
+ *
+ * info - describes the bitset
+ * root - the root block of the bitset
+ * new_root - on success, points to the new root block
+ */
+int dm_bitset_flush(struct dm_disk_bitset *info, dm_block_t root,
+		    dm_block_t *new_root);
+
+struct dm_bitset_cursor {
+	struct dm_disk_bitset *info;
+	struct dm_array_cursor cursor;
+
+	uint32_t entries_remaining;
+	uint32_t array_index;
+	uint32_t bit_index;
+	uint64_t current_bits;
+};
+
+/*
+ * Make sure you've flush any dm_disk_bitset and updated the root before
+ * using this.
+ */
+int dm_bitset_cursor_begin(struct dm_disk_bitset *info,
+			   dm_block_t root, uint32_t nr_entries,
+			   struct dm_bitset_cursor *c);
+void dm_bitset_cursor_end(struct dm_bitset_cursor *c);
+
+int dm_bitset_cursor_next(struct dm_bitset_cursor *c);
+int dm_bitset_cursor_skip(struct dm_bitset_cursor *c, uint32_t count);
+bool dm_bitset_cursor_get_value(struct dm_bitset_cursor *c);
+
+/*----------------------------------------------------------------*/
+
+#endif /* _LINUX_DM_BITSET_H */

linux-7.2.3/drivers/md/persistent-data/dm-block-manager.c

@@ -0,0 +1,662 @@
+// SPDX-License-Identifier: GPL-2.0-only
+/*
+ * Copyright (C) 2011 Red Hat, Inc.
+ *
+ * This file is released under the GPL.
+ */
+#include "dm-block-manager.h"
+#include "dm-persistent-data-internal.h"
+
+#include <linux/dm-bufio.h>
+#include <linux/crc32c.h>
+#include <linux/module.h>
+#include <linux/slab.h>
+#include <linux/rwsem.h>
+#include <linux/device-mapper.h>
+#include <linux/stacktrace.h>
+#include <linux/sched/task.h>
+
+#define DM_MSG_PREFIX "block manager"
+
+/*----------------------------------------------------------------*/
+
+#ifdef CONFIG_DM_DEBUG_BLOCK_MANAGER_LOCKING
+
+/*
+ * This is a read/write semaphore with a couple of differences.
+ *
+ * i) There is a restriction on the number of concurrent read locks that
+ * may be held at once.  This is just an implementation detail.
+ *
+ * ii) Recursive locking attempts are detected and return EINVAL.  A stack
+ * trace is also emitted for the previous lock acquisition.
+ *
+ * iii) Priority is given to write locks.
+ */
+#define MAX_HOLDERS 4
+#define MAX_STACK 10
+
+struct stack_store {
+	unsigned int	nr_entries;
+	unsigned long	entries[MAX_STACK];
+};
+
+struct block_lock {
+	spinlock_t lock;
+	__s32 count;
+	struct list_head waiters;
+	struct task_struct *holders[MAX_HOLDERS];
+
+#ifdef CONFIG_DM_DEBUG_BLOCK_STACK_TRACING
+	struct stack_store traces[MAX_HOLDERS];
+#endif
+};
+
+struct waiter {
+	struct list_head list;
+	struct task_struct *task;
+	int wants_write;
+};
+
+static unsigned int __find_holder(struct block_lock *lock,
+			      struct task_struct *task)
+{
+	unsigned int i;
+
+	for (i = 0; i < MAX_HOLDERS; i++)
+		if (lock->holders[i] == task)
+			break;
+
+	BUG_ON(i == MAX_HOLDERS);
+	return i;
+}
+
+/* call this *after* you increment lock->count */
+static void __add_holder(struct block_lock *lock, struct task_struct *task)
+{
+	unsigned int h = __find_holder(lock, NULL);
+#ifdef CONFIG_DM_DEBUG_BLOCK_STACK_TRACING
+	struct stack_store *t;
+#endif
+
+	get_task_struct(task);
+	lock->holders[h] = task;
+
+#ifdef CONFIG_DM_DEBUG_BLOCK_STACK_TRACING
+	t = lock->traces + h;
+	t->nr_entries = stack_trace_save(t->entries, MAX_STACK, 2);
+#endif
+}
+
+/* call this *before* you decrement lock->count */
+static void __del_holder(struct block_lock *lock, struct task_struct *task)
+{
+	unsigned int h = __find_holder(lock, task);
+
+	lock->holders[h] = NULL;
+	put_task_struct(task);
+}
+
+static int __check_holder(struct block_lock *lock)
+{
+	unsigned int i;
+
+	for (i = 0; i < MAX_HOLDERS; i++) {
+		if (lock->holders[i] == current) {
+			DMERR("recursive lock detected in metadata");
+#ifdef CONFIG_DM_DEBUG_BLOCK_STACK_TRACING
+			DMERR("previously held here:");
+			stack_trace_print(lock->traces[i].entries,
+					  lock->traces[i].nr_entries, 4);
+
+			DMERR("subsequent acquisition attempted here:");
+			dump_stack();
+#endif
+			return -EINVAL;
+		}
+	}
+
+	return 0;
+}
+
+static void __wait(struct waiter *w)
+{
+	for (;;) {
+		set_current_state(TASK_UNINTERRUPTIBLE);
+
+		if (!w->task)
+			break;
+
+		schedule();
+	}
+
+	set_current_state(TASK_RUNNING);
+}
+
+static void __wake_waiter(struct waiter *w)
+{
+	struct task_struct *task;
+
+	list_del(&w->list);
+	task = w->task;
+	smp_mb();
+	w->task = NULL;
+	wake_up_process(task);
+}
+
+/*
+ * We either wake a few readers or a single writer.
+ */
+static void __wake_many(struct block_lock *lock)
+{
+	struct waiter *w, *tmp;
+
+	BUG_ON(lock->count < 0);
+	list_for_each_entry_safe(w, tmp, &lock->waiters, list) {
+		if (lock->count >= MAX_HOLDERS)
+			return;
+
+		if (w->wants_write) {
+			if (lock->count > 0)
+				return; /* still read locked */
+
+			lock->count = -1;
+			__add_holder(lock, w->task);
+			__wake_waiter(w);
+			return;
+		}
+
+		lock->count++;
+		__add_holder(lock, w->task);
+		__wake_waiter(w);
+	}
+}
+
+static void bl_init(struct block_lock *lock)
+{
+	int i;
+
+	spin_lock_init(&lock->lock);
+	lock->count = 0;
+	INIT_LIST_HEAD(&lock->waiters);
+	for (i = 0; i < MAX_HOLDERS; i++)
+		lock->holders[i] = NULL;
+}
+
+static int __available_for_read(struct block_lock *lock)
+{
+	return lock->count >= 0 &&
+		lock->count < MAX_HOLDERS &&
+		list_empty(&lock->waiters);
+}
+
+static int bl_down_read(struct block_lock *lock)
+{
+	int r;
+	struct waiter w;
+
+	spin_lock(&lock->lock);
+	r = __check_holder(lock);
+	if (r) {
+		spin_unlock(&lock->lock);
+		return r;
+	}
+
+	if (__available_for_read(lock)) {
+		lock->count++;
+		__add_holder(lock, current);
+		spin_unlock(&lock->lock);
+		return 0;
+	}
+
+	get_task_struct(current);
+
+	w.task = current;
+	w.wants_write = 0;
+	list_add_tail(&w.list, &lock->waiters);
+	spin_unlock(&lock->lock);
+
+	__wait(&w);
+	put_task_struct(current);
+	return 0;
+}
+
+static int bl_down_read_nonblock(struct block_lock *lock)
+{
+	int r;
+
+	spin_lock(&lock->lock);
+	r = __check_holder(lock);
+	if (r)
+		goto out;
+
+	if (__available_for_read(lock)) {
+		lock->count++;
+		__add_holder(lock, current);
+		r = 0;
+	} else
+		r = -EWOULDBLOCK;
+
+out:
+	spin_unlock(&lock->lock);
+	return r;
+}
+
+static void bl_up_read(struct block_lock *lock)
+{
+	spin_lock(&lock->lock);
+	BUG_ON(lock->count <= 0);
+	__del_holder(lock, current);
+	--lock->count;
+	if (!list_empty(&lock->waiters))
+		__wake_many(lock);
+	spin_unlock(&lock->lock);
+}
+
+static int bl_down_write(struct block_lock *lock)
+{
+	int r;
+	struct waiter w;
+
+	spin_lock(&lock->lock);
+	r = __check_holder(lock);
+	if (r) {
+		spin_unlock(&lock->lock);
+		return r;
+	}
+
+	if (lock->count == 0 && list_empty(&lock->waiters)) {
+		lock->count = -1;
+		__add_holder(lock, current);
+		spin_unlock(&lock->lock);
+		return 0;
+	}
+
+	get_task_struct(current);
+	w.task = current;
+	w.wants_write = 1;
+
+	/*
+	 * Writers given priority. We know there's only one mutator in the
+	 * system, so ignoring the ordering reversal.
+	 */
+	list_add(&w.list, &lock->waiters);
+	spin_unlock(&lock->lock);
+
+	__wait(&w);
+	put_task_struct(current);
+
+	return 0;
+}
+
+static void bl_up_write(struct block_lock *lock)
+{
+	spin_lock(&lock->lock);
+	__del_holder(lock, current);
+	lock->count = 0;
+	if (!list_empty(&lock->waiters))
+		__wake_many(lock);
+	spin_unlock(&lock->lock);
+}
+
+static void report_recursive_bug(dm_block_t b, int r)
+{
+	if (r == -EINVAL)
+		DMERR("recursive acquisition of block %llu requested.",
+		      (unsigned long long) b);
+}
+
+#else  /* !CONFIG_DM_DEBUG_BLOCK_MANAGER_LOCKING */
+
+#define bl_init(x) do { } while (0)
+#define bl_down_read(x) 0
+#define bl_down_read_nonblock(x) 0
+#define bl_up_read(x) do { } while (0)
+#define bl_down_write(x) 0
+#define bl_up_write(x) do { } while (0)
+#define report_recursive_bug(x, y) do { } while (0)
+
+#endif /* CONFIG_DM_DEBUG_BLOCK_MANAGER_LOCKING */
+
+/*----------------------------------------------------------------*/
+
+/*
+ * Block manager is currently implemented using dm-bufio.  struct
+ * dm_block_manager and struct dm_block map directly onto a couple of
+ * structs in the bufio interface.  I want to retain the freedom to move
+ * away from bufio in the future.  So these structs are just cast within
+ * this .c file, rather than making it through to the public interface.
+ */
+static struct dm_buffer *to_buffer(struct dm_block *b)
+{
+	return (struct dm_buffer *) b;
+}
+
+dm_block_t dm_block_location(struct dm_block *b)
+{
+	return dm_bufio_get_block_number(to_buffer(b));
+}
+EXPORT_SYMBOL_GPL(dm_block_location);
+
+void *dm_block_data(struct dm_block *b)
+{
+	return dm_bufio_get_block_data(to_buffer(b));
+}
+EXPORT_SYMBOL_GPL(dm_block_data);
+
+struct buffer_aux {
+	const struct dm_block_validator *validator;
+	int write_locked;
+
+#ifdef CONFIG_DM_DEBUG_BLOCK_MANAGER_LOCKING
+	struct block_lock lock;
+#endif
+};
+
+static void dm_block_manager_alloc_callback(struct dm_buffer *buf)
+{
+	struct buffer_aux *aux = dm_bufio_get_aux_data(buf);
+
+	aux->validator = NULL;
+	bl_init(&aux->lock);
+}
+
+static void dm_block_manager_write_callback(struct dm_buffer *buf)
+{
+	struct buffer_aux *aux = dm_bufio_get_aux_data(buf);
+
+	if (aux->validator) {
+		aux->validator->prepare_for_write(aux->validator, (struct dm_block *) buf,
+			 dm_bufio_get_block_size(dm_bufio_get_client(buf)));
+	}
+}
+
+/*
+ * -------------------------------------------------------------
+ * Public interface
+ *--------------------------------------------------------------
+ */
+struct dm_block_manager {
+	struct dm_bufio_client *bufio;
+	bool read_only:1;
+};
+
+struct dm_block_manager *dm_block_manager_create(struct block_device *bdev,
+						 unsigned int block_size,
+						 unsigned int max_held_per_thread)
+{
+	int r;
+	struct dm_block_manager *bm;
+
+	bm = kmalloc_obj(*bm);
+	if (!bm) {
+		r = -ENOMEM;
+		goto bad;
+	}
+
+	bm->bufio = dm_bufio_client_create(bdev, block_size, max_held_per_thread,
+					   sizeof(struct buffer_aux),
+					   dm_block_manager_alloc_callback,
+					   dm_block_manager_write_callback,
+					   0);
+	if (IS_ERR(bm->bufio)) {
+		r = PTR_ERR(bm->bufio);
+		kfree(bm);
+		goto bad;
+	}
+
+	bm->read_only = false;
+
+	return bm;
+
+bad:
+	return ERR_PTR(r);
+}
+EXPORT_SYMBOL_GPL(dm_block_manager_create);
+
+void dm_block_manager_destroy(struct dm_block_manager *bm)
+{
+	dm_bufio_client_destroy(bm->bufio);
+	kfree(bm);
+}
+EXPORT_SYMBOL_GPL(dm_block_manager_destroy);
+
+void dm_block_manager_reset(struct dm_block_manager *bm)
+{
+	dm_bufio_client_reset(bm->bufio);
+}
+EXPORT_SYMBOL_GPL(dm_block_manager_reset);
+
+unsigned int dm_bm_block_size(struct dm_block_manager *bm)
+{
+	return dm_bufio_get_block_size(bm->bufio);
+}
+EXPORT_SYMBOL_GPL(dm_bm_block_size);
+
+dm_block_t dm_bm_nr_blocks(struct dm_block_manager *bm)
+{
+	return dm_bufio_get_device_size(bm->bufio);
+}
+
+static int dm_bm_validate_buffer(struct dm_block_manager *bm,
+				 struct dm_buffer *buf,
+				 struct buffer_aux *aux,
+				 const struct dm_block_validator *v)
+{
+	if (unlikely(!aux->validator)) {
+		int r;
+
+		if (!v)
+			return 0;
+		r = v->check(v, (struct dm_block *) buf, dm_bufio_get_block_size(bm->bufio));
+		if (unlikely(r)) {
+			DMERR_LIMIT("%s validator check failed for block %llu", v->name,
+				    (unsigned long long) dm_bufio_get_block_number(buf));
+			return r;
+		}
+		aux->validator = v;
+	} else {
+		if (unlikely(aux->validator != v)) {
+			DMERR_LIMIT("validator mismatch (old=%s vs new=%s) for block %llu",
+				    aux->validator->name, v ? v->name : "NULL",
+				    (unsigned long long) dm_bufio_get_block_number(buf));
+			return -EINVAL;
+		}
+	}
+
+	return 0;
+}
+int dm_bm_read_lock(struct dm_block_manager *bm, dm_block_t b,
+		    const struct dm_block_validator *v,
+		    struct dm_block **result)
+{
+	struct buffer_aux *aux;
+	void *p;
+	int r;
+
+	p = dm_bufio_read(bm->bufio, b, (struct dm_buffer **) result);
+	if (IS_ERR(p))
+		return PTR_ERR(p);
+
+	aux = dm_bufio_get_aux_data(to_buffer(*result));
+	r = bl_down_read(&aux->lock);
+	if (unlikely(r)) {
+		dm_bufio_release(to_buffer(*result));
+		report_recursive_bug(b, r);
+		return r;
+	}
+
+	aux->write_locked = 0;
+
+	r = dm_bm_validate_buffer(bm, to_buffer(*result), aux, v);
+	if (unlikely(r)) {
+		bl_up_read(&aux->lock);
+		dm_bufio_release(to_buffer(*result));
+		return r;
+	}
+
+	return 0;
+}
+EXPORT_SYMBOL_GPL(dm_bm_read_lock);
+
+int dm_bm_write_lock(struct dm_block_manager *bm,
+		     dm_block_t b, const struct dm_block_validator *v,
+		     struct dm_block **result)
+{
+	struct buffer_aux *aux;
+	void *p;
+	int r;
+
+	if (dm_bm_is_read_only(bm))
+		return -EPERM;
+
+	p = dm_bufio_read(bm->bufio, b, (struct dm_buffer **) result);
+	if (IS_ERR(p))
+		return PTR_ERR(p);
+
+	aux = dm_bufio_get_aux_data(to_buffer(*result));
+	r = bl_down_write(&aux->lock);
+	if (r) {
+		dm_bufio_release(to_buffer(*result));
+		report_recursive_bug(b, r);
+		return r;
+	}
+
+	aux->write_locked = 1;
+
+	r = dm_bm_validate_buffer(bm, to_buffer(*result), aux, v);
+	if (unlikely(r)) {
+		bl_up_write(&aux->lock);
+		dm_bufio_release(to_buffer(*result));
+		return r;
+	}
+
+	return 0;
+}
+EXPORT_SYMBOL_GPL(dm_bm_write_lock);
+
+int dm_bm_read_try_lock(struct dm_block_manager *bm,
+			dm_block_t b, const struct dm_block_validator *v,
+			struct dm_block **result)
+{
+	struct buffer_aux *aux;
+	void *p;
+	int r;
+
+	p = dm_bufio_get(bm->bufio, b, (struct dm_buffer **) result);
+	if (IS_ERR(p))
+		return PTR_ERR(p);
+	if (unlikely(!p))
+		return -EWOULDBLOCK;
+
+	aux = dm_bufio_get_aux_data(to_buffer(*result));
+	r = bl_down_read_nonblock(&aux->lock);
+	if (r < 0) {
+		dm_bufio_release(to_buffer(*result));
+		report_recursive_bug(b, r);
+		return r;
+	}
+	aux->write_locked = 0;
+
+	r = dm_bm_validate_buffer(bm, to_buffer(*result), aux, v);
+	if (unlikely(r)) {
+		bl_up_read(&aux->lock);
+		dm_bufio_release(to_buffer(*result));
+		return r;
+	}
+
+	return 0;
+}
+
+int dm_bm_write_lock_zero(struct dm_block_manager *bm,
+			  dm_block_t b, const struct dm_block_validator *v,
+			  struct dm_block **result)
+{
+	int r;
+	struct buffer_aux *aux;
+	void *p;
+
+	if (dm_bm_is_read_only(bm))
+		return -EPERM;
+
+	p = dm_bufio_new(bm->bufio, b, (struct dm_buffer **) result);
+	if (IS_ERR(p))
+		return PTR_ERR(p);
+
+	memset(p, 0, dm_bm_block_size(bm));
+
+	aux = dm_bufio_get_aux_data(to_buffer(*result));
+	r = bl_down_write(&aux->lock);
+	if (r) {
+		dm_bufio_release(to_buffer(*result));
+		return r;
+	}
+
+	aux->write_locked = 1;
+	aux->validator = v;
+
+	return 0;
+}
+EXPORT_SYMBOL_GPL(dm_bm_write_lock_zero);
+
+void dm_bm_unlock(struct dm_block *b)
+{
+	struct buffer_aux *aux = dm_bufio_get_aux_data(to_buffer(b));
+
+	if (aux->write_locked) {
+		dm_bufio_mark_buffer_dirty(to_buffer(b));
+		bl_up_write(&aux->lock);
+	} else
+		bl_up_read(&aux->lock);
+
+	dm_bufio_release(to_buffer(b));
+}
+EXPORT_SYMBOL_GPL(dm_bm_unlock);
+
+int dm_bm_flush(struct dm_block_manager *bm)
+{
+	if (dm_bm_is_read_only(bm))
+		return -EPERM;
+
+	return dm_bufio_write_dirty_buffers(bm->bufio);
+}
+EXPORT_SYMBOL_GPL(dm_bm_flush);
+
+void dm_bm_prefetch(struct dm_block_manager *bm, dm_block_t b)
+{
+	dm_bufio_prefetch(bm->bufio, b, 1);
+}
+
+bool dm_bm_is_read_only(struct dm_block_manager *bm)
+{
+	return bm ? bm->read_only : true;
+}
+EXPORT_SYMBOL_GPL(dm_bm_is_read_only);
+
+void dm_bm_set_read_only(struct dm_block_manager *bm)
+{
+	if (bm)
+		bm->read_only = true;
+}
+EXPORT_SYMBOL_GPL(dm_bm_set_read_only);
+
+void dm_bm_set_read_write(struct dm_block_manager *bm)
+{
+	if (bm)
+		bm->read_only = false;
+}
+EXPORT_SYMBOL_GPL(dm_bm_set_read_write);
+
+u32 dm_bm_checksum(const void *data, size_t len, u32 init_xor)
+{
+	return crc32c(~(u32) 0, data, len) ^ init_xor;
+}
+EXPORT_SYMBOL_GPL(dm_bm_checksum);
+
+/*----------------------------------------------------------------*/
+
+MODULE_LICENSE("GPL");
+MODULE_AUTHOR("Joe Thornber <dm-devel@lists.linux.dev>");
+MODULE_DESCRIPTION("Immutable metadata library for dm");
+
+/*----------------------------------------------------------------*/

linux-7.2.3/drivers/md/persistent-data/dm-block-manager.h

@@ -0,0 +1,138 @@
+/* SPDX-License-Identifier: GPL-2.0-only */
+/*
+ * Copyright (C) 2011 Red Hat, Inc.
+ *
+ * This file is released under the GPL.
+ */
+
+#ifndef _LINUX_DM_BLOCK_MANAGER_H
+#define _LINUX_DM_BLOCK_MANAGER_H
+
+#include <linux/types.h>
+#include <linux/blkdev.h>
+
+/*----------------------------------------------------------------*/
+
+/*
+ * Block number.
+ */
+typedef uint64_t dm_block_t;
+struct dm_block;
+
+dm_block_t dm_block_location(struct dm_block *b);
+void *dm_block_data(struct dm_block *b);
+
+/*----------------------------------------------------------------*/
+
+/*
+ * @name should be a unique identifier for the block manager, no longer
+ * than 32 chars.
+ *
+ * @max_held_per_thread should be the maximum number of locks, read or
+ * write, that an individual thread holds at any one time.
+ */
+struct dm_block_manager;
+struct dm_block_manager *dm_block_manager_create(
+	struct block_device *bdev, unsigned int block_size,
+	unsigned int max_held_per_thread);
+void dm_block_manager_destroy(struct dm_block_manager *bm);
+void dm_block_manager_reset(struct dm_block_manager *bm);
+
+unsigned int dm_bm_block_size(struct dm_block_manager *bm);
+dm_block_t dm_bm_nr_blocks(struct dm_block_manager *bm);
+
+/*----------------------------------------------------------------*/
+
+/*
+ * The validator allows the caller to verify newly-read data and modify
+ * the data just before writing, e.g. to calculate checksums.  It's
+ * important to be consistent with your use of validators.  The only time
+ * you can change validators is if you call dm_bm_write_lock_zero.
+ */
+struct dm_block_validator {
+	const char *name;
+	void (*prepare_for_write)(const struct dm_block_validator *v,
+				  struct dm_block *b, size_t block_size);
+
+	/*
+	 * Return 0 if the checksum is valid or < 0 on error.
+	 */
+	int (*check)(const struct dm_block_validator *v,
+		     struct dm_block *b, size_t block_size);
+};
+
+/*----------------------------------------------------------------*/
+
+/*
+ * You can have multiple concurrent readers or a single writer holding a
+ * block lock.
+ */
+
+/*
+ * dm_bm_lock() locks a block and returns through @result a pointer to
+ * memory that holds a copy of that block.  If you have write-locked the
+ * block then any changes you make to memory pointed to by @result will be
+ * written back to the disk sometime after dm_bm_unlock is called.
+ */
+int dm_bm_read_lock(struct dm_block_manager *bm, dm_block_t b,
+		    const struct dm_block_validator *v,
+		    struct dm_block **result);
+
+int dm_bm_write_lock(struct dm_block_manager *bm, dm_block_t b,
+		     const struct dm_block_validator *v,
+		     struct dm_block **result);
+
+/*
+ * The *_try_lock variants return -EWOULDBLOCK if the block isn't
+ * available immediately.
+ */
+int dm_bm_read_try_lock(struct dm_block_manager *bm, dm_block_t b,
+			const struct dm_block_validator *v,
+			struct dm_block **result);
+
+/*
+ * Use dm_bm_write_lock_zero() when you know you're going to
+ * overwrite the block completely.  It saves a disk read.
+ */
+int dm_bm_write_lock_zero(struct dm_block_manager *bm, dm_block_t b,
+			  const struct dm_block_validator *v,
+			  struct dm_block **result);
+
+void dm_bm_unlock(struct dm_block *b);
+
+/*
+ * It's a common idiom to have a superblock that should be committed last.
+ *
+ * @superblock should be write-locked on entry. It will be unlocked during
+ * this function.  All dirty blocks are guaranteed to be written and flushed
+ * before the superblock.
+ *
+ * This method always blocks.
+ */
+int dm_bm_flush(struct dm_block_manager *bm);
+
+/*
+ * Request data is prefetched into the cache.
+ */
+void dm_bm_prefetch(struct dm_block_manager *bm, dm_block_t b);
+
+/*
+ * Switches the bm to a read only mode.  Once read-only mode
+ * has been entered the following functions will return -EPERM.
+ *
+ *   dm_bm_write_lock
+ *   dm_bm_write_lock_zero
+ *   dm_bm_flush_and_unlock
+ *
+ * Additionally you should not use dm_bm_unlock_move, however no error will
+ * be returned if you do.
+ */
+bool dm_bm_is_read_only(struct dm_block_manager *bm);
+void dm_bm_set_read_only(struct dm_block_manager *bm);
+void dm_bm_set_read_write(struct dm_block_manager *bm);
+
+u32 dm_bm_checksum(const void *data, size_t len, u32 init_xor);
+
+/*----------------------------------------------------------------*/
+
+#endif	/* _LINUX_DM_BLOCK_MANAGER_H */

linux-7.2.3/drivers/md/persistent-data/dm-btree-internal.h

@@ -0,0 +1,162 @@
+/* SPDX-License-Identifier: GPL-2.0-only */
+/*
+ * Copyright (C) 2011 Red Hat, Inc.
+ *
+ * This file is released under the GPL.
+ */
+
+#ifndef DM_BTREE_INTERNAL_H
+#define DM_BTREE_INTERNAL_H
+
+#include "dm-btree.h"
+
+/*----------------------------------------------------------------*/
+
+/*
+ * We'll need 2 accessor functions for n->csum and n->blocknr
+ * to support dm-btree-spine.c in that case.
+ */
+
+enum node_flags {
+	INTERNAL_NODE = 1,
+	LEAF_NODE = 1 << 1
+};
+
+/*
+ * Every btree node begins with this structure.  Make sure it's a multiple
+ * of 8-bytes in size, otherwise the 64bit keys will be mis-aligned.
+ */
+struct node_header {
+	__le32 csum;
+	__le32 flags;
+	__le64 blocknr; /* Block this node is supposed to live in. */
+
+	__le32 nr_entries;
+	__le32 max_entries;
+	__le32 value_size;
+	__le32 padding;
+} __packed __aligned(8);
+
+struct btree_node {
+	struct node_header header;
+	__le64 keys[];
+} __packed __aligned(8);
+
+
+/*
+ * Locks a block using the btree node validator.
+ */
+int bn_read_lock(struct dm_btree_info *info, dm_block_t b,
+		 struct dm_block **result);
+
+void inc_children(struct dm_transaction_manager *tm, struct btree_node *n,
+		  struct dm_btree_value_type *vt);
+
+int new_block(struct dm_btree_info *info, struct dm_block **result);
+void unlock_block(struct dm_btree_info *info, struct dm_block *b);
+
+/*
+ * Spines keep track of the rolling locks.  There are 2 variants, read-only
+ * and one that uses shadowing.  These are separate structs to allow the
+ * type checker to spot misuse, for example accidentally calling read_lock
+ * on a shadow spine.
+ */
+struct ro_spine {
+	struct dm_btree_info *info;
+
+	int count;
+	struct dm_block *nodes[2];
+};
+
+void init_ro_spine(struct ro_spine *s, struct dm_btree_info *info);
+void exit_ro_spine(struct ro_spine *s);
+int ro_step(struct ro_spine *s, dm_block_t new_child);
+void ro_pop(struct ro_spine *s);
+struct btree_node *ro_node(struct ro_spine *s);
+
+struct shadow_spine {
+	struct dm_btree_info *info;
+
+	int count;
+	struct dm_block *nodes[2];
+
+	dm_block_t root;
+};
+
+void init_shadow_spine(struct shadow_spine *s, struct dm_btree_info *info);
+void exit_shadow_spine(struct shadow_spine *s);
+
+int shadow_step(struct shadow_spine *s, dm_block_t b,
+		struct dm_btree_value_type *vt);
+
+/*
+ * The spine must have at least one entry before calling this.
+ */
+struct dm_block *shadow_current(struct shadow_spine *s);
+
+/*
+ * The spine must have at least two entries before calling this.
+ */
+struct dm_block *shadow_parent(struct shadow_spine *s);
+
+int shadow_has_parent(struct shadow_spine *s);
+
+dm_block_t shadow_root(struct shadow_spine *s);
+
+/*
+ * Some inlines.
+ */
+static inline __le64 *key_ptr(struct btree_node *n, uint32_t index)
+{
+	return n->keys + index;
+}
+
+static inline void *value_base(struct btree_node *n)
+{
+	return &n->keys[le32_to_cpu(n->header.max_entries)];
+}
+
+static inline void *value_ptr(struct btree_node *n, uint32_t index)
+{
+	uint32_t value_size = le32_to_cpu(n->header.value_size);
+
+	return value_base(n) + (value_size * index);
+}
+
+/*
+ * Assumes the values are suitably-aligned and converts to core format.
+ */
+static inline uint64_t value64(struct btree_node *n, uint32_t index)
+{
+	__le64 *values_le = value_base(n);
+
+	return le64_to_cpu(values_le[index]);
+}
+
+/*
+ * Searching for a key within a single node.
+ */
+int lower_bound(struct btree_node *n, uint64_t key);
+
+extern const struct dm_block_validator btree_node_validator;
+
+/*
+ * Value type for upper levels of multi-level btrees.
+ */
+extern void init_le64_type(struct dm_transaction_manager *tm,
+			   struct dm_btree_value_type *vt);
+
+/*
+ * This returns a shadowed btree leaf that you may modify.  In practise
+ * this means overwrites only, since an insert could cause a node to
+ * be split.  Useful if you need access to the old value to calculate the
+ * new one.
+ *
+ * This only works with single level btrees.  The given key must be present in
+ * the tree, otherwise -EINVAL will be returned.
+ */
+int btree_get_overwrite_leaf(struct dm_btree_info *info, dm_block_t root,
+			     uint64_t key, int *index,
+			     dm_block_t *new_root, struct dm_block **leaf);
+
+#endif	/* DM_BTREE_INTERNAL_H */

linux-7.2.3/drivers/md/persistent-data/dm-btree-remove.c

@@ -0,0 +1,773 @@
+// SPDX-License-Identifier: GPL-2.0-only
+/*
+ * Copyright (C) 2011 Red Hat, Inc.
+ *
+ * This file is released under the GPL.
+ */
+
+#include "dm-btree.h"
+#include "dm-btree-internal.h"
+#include "dm-transaction-manager.h"
+
+#include <linux/export.h>
+#include <linux/device-mapper.h>
+
+#define DM_MSG_PREFIX "btree"
+
+/*
+ * Removing an entry from a btree
+ * ==============================
+ *
+ * A very important constraint for our btree is that no node, except the
+ * root, may have fewer than a certain number of entries.
+ * (MIN_ENTRIES <= nr_entries <= MAX_ENTRIES).
+ *
+ * Ensuring this is complicated by the way we want to only ever hold the
+ * locks on 2 nodes concurrently, and only change nodes in a top to bottom
+ * fashion.
+ *
+ * Each node may have a left or right sibling.  When decending the spine,
+ * if a node contains only MIN_ENTRIES then we try and increase this to at
+ * least MIN_ENTRIES + 1.  We do this in the following ways:
+ *
+ * [A] No siblings => this can only happen if the node is the root, in which
+ *     case we copy the childs contents over the root.
+ *
+ * [B] No left sibling
+ *     ==> rebalance(node, right sibling)
+ *
+ * [C] No right sibling
+ *     ==> rebalance(left sibling, node)
+ *
+ * [D] Both siblings, total_entries(left, node, right) <= DEL_THRESHOLD
+ *     ==> delete node adding it's contents to left and right
+ *
+ * [E] Both siblings, total_entries(left, node, right) > DEL_THRESHOLD
+ *     ==> rebalance(left, node, right)
+ *
+ * After these operations it's possible that the our original node no
+ * longer contains the desired sub tree.  For this reason this rebalancing
+ * is performed on the children of the current node.  This also avoids
+ * having a special case for the root.
+ *
+ * Once this rebalancing has occurred we can then step into the child node
+ * for internal nodes.  Or delete the entry for leaf nodes.
+ */
+
+/*
+ * Some little utilities for moving node data around.
+ */
+static void node_shift(struct btree_node *n, int shift)
+{
+	uint32_t nr_entries = le32_to_cpu(n->header.nr_entries);
+	uint32_t value_size = le32_to_cpu(n->header.value_size);
+
+	if (shift < 0) {
+		shift = -shift;
+		BUG_ON(shift > nr_entries);
+		BUG_ON((void *) key_ptr(n, shift) >= value_ptr(n, shift));
+		memmove(key_ptr(n, 0),
+			key_ptr(n, shift),
+			(nr_entries - shift) * sizeof(__le64));
+		memmove(value_ptr(n, 0),
+			value_ptr(n, shift),
+			(nr_entries - shift) * value_size);
+	} else {
+		BUG_ON(nr_entries + shift > le32_to_cpu(n->header.max_entries));
+		memmove(key_ptr(n, shift),
+			key_ptr(n, 0),
+			nr_entries * sizeof(__le64));
+		memmove(value_ptr(n, shift),
+			value_ptr(n, 0),
+			nr_entries * value_size);
+	}
+}
+
+static int node_copy(struct btree_node *left, struct btree_node *right, int shift)
+{
+	uint32_t nr_left = le32_to_cpu(left->header.nr_entries);
+	uint32_t value_size = le32_to_cpu(left->header.value_size);
+
+	if (value_size != le32_to_cpu(right->header.value_size)) {
+		DMERR("mismatched value size");
+		return -EILSEQ;
+	}
+
+	if (shift < 0) {
+		shift = -shift;
+
+		if (nr_left + shift > le32_to_cpu(left->header.max_entries)) {
+			DMERR("bad shift");
+			return -EINVAL;
+		}
+
+		memcpy(key_ptr(left, nr_left),
+		       key_ptr(right, 0),
+		       shift * sizeof(__le64));
+		memcpy(value_ptr(left, nr_left),
+		       value_ptr(right, 0),
+		       shift * value_size);
+	} else {
+		if (shift > le32_to_cpu(right->header.max_entries)) {
+			DMERR("bad shift");
+			return -EINVAL;
+		}
+
+		memcpy(key_ptr(right, 0),
+		       key_ptr(left, nr_left - shift),
+		       shift * sizeof(__le64));
+		memcpy(value_ptr(right, 0),
+		       value_ptr(left, nr_left - shift),
+		       shift * value_size);
+	}
+	return 0;
+}
+
+/*
+ * Delete a specific entry from a leaf node.
+ */
+static void delete_at(struct btree_node *n, unsigned int index)
+{
+	unsigned int nr_entries = le32_to_cpu(n->header.nr_entries);
+	unsigned int nr_to_copy = nr_entries - (index + 1);
+	uint32_t value_size = le32_to_cpu(n->header.value_size);
+
+	BUG_ON(index >= nr_entries);
+
+	if (nr_to_copy) {
+		memmove(key_ptr(n, index),
+			key_ptr(n, index + 1),
+			nr_to_copy * sizeof(__le64));
+
+		memmove(value_ptr(n, index),
+			value_ptr(n, index + 1),
+			nr_to_copy * value_size);
+	}
+
+	n->header.nr_entries = cpu_to_le32(nr_entries - 1);
+}
+
+static unsigned int merge_threshold(struct btree_node *n)
+{
+	return le32_to_cpu(n->header.max_entries) / 3;
+}
+
+struct child {
+	unsigned int index;
+	struct dm_block *block;
+	struct btree_node *n;
+};
+
+static int init_child(struct dm_btree_info *info, struct dm_btree_value_type *vt,
+		      struct btree_node *parent,
+		      unsigned int index, struct child *result)
+{
+	int r, inc;
+	dm_block_t root;
+
+	result->index = index;
+	root = value64(parent, index);
+
+	r = dm_tm_shadow_block(info->tm, root, &btree_node_validator,
+			       &result->block, &inc);
+	if (r)
+		return r;
+
+	result->n = dm_block_data(result->block);
+
+	if (inc)
+		inc_children(info->tm, result->n, vt);
+
+	*((__le64 *) value_ptr(parent, index)) =
+		cpu_to_le64(dm_block_location(result->block));
+
+	return 0;
+}
+
+static void exit_child(struct dm_btree_info *info, struct child *c)
+{
+	dm_tm_unlock(info->tm, c->block);
+}
+
+static int shift(struct btree_node *left, struct btree_node *right, int count)
+{
+	int r;
+	uint32_t nr_left = le32_to_cpu(left->header.nr_entries);
+	uint32_t nr_right = le32_to_cpu(right->header.nr_entries);
+	uint32_t max_entries = le32_to_cpu(left->header.max_entries);
+	uint32_t r_max_entries = le32_to_cpu(right->header.max_entries);
+
+	if (max_entries != r_max_entries) {
+		DMERR("node max_entries mismatch");
+		return -EILSEQ;
+	}
+
+	if (nr_left - count > max_entries) {
+		DMERR("node shift out of bounds");
+		return -EINVAL;
+	}
+
+	if (nr_right + count > max_entries) {
+		DMERR("node shift out of bounds");
+		return -EINVAL;
+	}
+
+	if (!count)
+		return 0;
+
+	if (count > 0) {
+		node_shift(right, count);
+		r = node_copy(left, right, count);
+		if (r)
+			return r;
+	} else {
+		r = node_copy(left, right, count);
+		if (r)
+			return r;
+		node_shift(right, count);
+	}
+
+	left->header.nr_entries = cpu_to_le32(nr_left - count);
+	right->header.nr_entries = cpu_to_le32(nr_right + count);
+
+	return 0;
+}
+
+static int __rebalance2(struct dm_btree_info *info, struct btree_node *parent,
+			struct child *l, struct child *r)
+{
+	int ret;
+	struct btree_node *left = l->n;
+	struct btree_node *right = r->n;
+	uint32_t nr_left = le32_to_cpu(left->header.nr_entries);
+	uint32_t nr_right = le32_to_cpu(right->header.nr_entries);
+	/*
+	 * Ensure the number of entries in each child will be greater
+	 * than or equal to (max_entries / 3 + 1), so no matter which
+	 * child is used for removal, the number will still be not
+	 * less than (max_entries / 3).
+	 */
+	unsigned int threshold = 2 * (merge_threshold(left) + 1);
+
+	if (nr_left + nr_right < threshold) {
+		/*
+		 * Merge
+		 */
+		node_copy(left, right, -nr_right);
+		left->header.nr_entries = cpu_to_le32(nr_left + nr_right);
+		delete_at(parent, r->index);
+
+		/*
+		 * We need to decrement the right block, but not it's
+		 * children, since they're still referenced by left.
+		 */
+		dm_tm_dec(info->tm, dm_block_location(r->block));
+	} else {
+		/*
+		 * Rebalance.
+		 */
+		unsigned int target_left = (nr_left + nr_right) / 2;
+
+		ret = shift(left, right, nr_left - target_left);
+		if (ret)
+			return ret;
+		*key_ptr(parent, r->index) = right->keys[0];
+	}
+	return 0;
+}
+
+static int rebalance2(struct shadow_spine *s, struct dm_btree_info *info,
+		      struct dm_btree_value_type *vt, unsigned int left_index)
+{
+	int r;
+	struct btree_node *parent;
+	struct child left, right;
+
+	parent = dm_block_data(shadow_current(s));
+
+	r = init_child(info, vt, parent, left_index, &left);
+	if (r)
+		return r;
+
+	r = init_child(info, vt, parent, left_index + 1, &right);
+	if (r) {
+		exit_child(info, &left);
+		return r;
+	}
+
+	r = __rebalance2(info, parent, &left, &right);
+
+	exit_child(info, &left);
+	exit_child(info, &right);
+
+	return r;
+}
+
+/*
+ * We dump as many entries from center as possible into left, then the rest
+ * in right, then rebalance2.  This wastes some cpu, but I want something
+ * simple atm.
+ */
+static int delete_center_node(struct dm_btree_info *info, struct btree_node *parent,
+			      struct child *l, struct child *c, struct child *r,
+			      struct btree_node *left, struct btree_node *center, struct btree_node *right,
+			      uint32_t nr_left, uint32_t nr_center, uint32_t nr_right)
+{
+	uint32_t max_entries = le32_to_cpu(left->header.max_entries);
+	unsigned int shift = min(max_entries - nr_left, nr_center);
+
+	if (nr_left + shift > max_entries) {
+		DMERR("node shift out of bounds");
+		return -EINVAL;
+	}
+
+	node_copy(left, center, -shift);
+	left->header.nr_entries = cpu_to_le32(nr_left + shift);
+
+	if (shift != nr_center) {
+		shift = nr_center - shift;
+
+		if ((nr_right + shift) > max_entries) {
+			DMERR("node shift out of bounds");
+			return -EINVAL;
+		}
+
+		node_shift(right, shift);
+		node_copy(center, right, shift);
+		right->header.nr_entries = cpu_to_le32(nr_right + shift);
+	}
+	*key_ptr(parent, r->index) = right->keys[0];
+
+	delete_at(parent, c->index);
+	r->index--;
+
+	dm_tm_dec(info->tm, dm_block_location(c->block));
+	return __rebalance2(info, parent, l, r);
+}
+
+/*
+ * Redistributes entries among 3 sibling nodes.
+ */
+static int redistribute3(struct dm_btree_info *info, struct btree_node *parent,
+			 struct child *l, struct child *c, struct child *r,
+			 struct btree_node *left, struct btree_node *center, struct btree_node *right,
+			 uint32_t nr_left, uint32_t nr_center, uint32_t nr_right)
+{
+	int s, ret;
+	uint32_t max_entries = le32_to_cpu(left->header.max_entries);
+	unsigned int total = nr_left + nr_center + nr_right;
+	unsigned int target_right = total / 3;
+	unsigned int remainder = (target_right * 3) != total;
+	unsigned int target_left = target_right + remainder;
+
+	BUG_ON(target_left > max_entries);
+	BUG_ON(target_right > max_entries);
+
+	if (nr_left < nr_right) {
+		s = nr_left - target_left;
+
+		if (s < 0 && nr_center < -s) {
+			/* not enough in central node */
+			ret = shift(left, center, -nr_center);
+			if (ret)
+				return ret;
+
+			s += nr_center;
+			ret = shift(left, right, s);
+			if (ret)
+				return ret;
+
+			nr_right += s;
+		} else {
+			ret = shift(left, center, s);
+			if (ret)
+				return ret;
+		}
+
+		ret = shift(center, right, target_right - nr_right);
+		if (ret)
+			return ret;
+	} else {
+		s = target_right - nr_right;
+		if (s > 0 && nr_center < s) {
+			/* not enough in central node */
+			ret = shift(center, right, nr_center);
+			if (ret)
+				return ret;
+			s -= nr_center;
+			ret = shift(left, right, s);
+			if (ret)
+				return ret;
+			nr_left -= s;
+		} else {
+			ret = shift(center, right, s);
+			if (ret)
+				return ret;
+		}
+
+		ret = shift(left, center, nr_left - target_left);
+		if (ret)
+			return ret;
+	}
+
+	*key_ptr(parent, c->index) = center->keys[0];
+	*key_ptr(parent, r->index) = right->keys[0];
+	return 0;
+}
+
+static int __rebalance3(struct dm_btree_info *info, struct btree_node *parent,
+			struct child *l, struct child *c, struct child *r)
+{
+	struct btree_node *left = l->n;
+	struct btree_node *center = c->n;
+	struct btree_node *right = r->n;
+
+	uint32_t nr_left = le32_to_cpu(left->header.nr_entries);
+	uint32_t nr_center = le32_to_cpu(center->header.nr_entries);
+	uint32_t nr_right = le32_to_cpu(right->header.nr_entries);
+
+	unsigned int threshold = merge_threshold(left) * 4 + 1;
+
+	if ((left->header.max_entries != center->header.max_entries) ||
+	    (center->header.max_entries != right->header.max_entries)) {
+		DMERR("bad btree metadata, max_entries differ");
+		return -EILSEQ;
+	}
+
+	if ((nr_left + nr_center + nr_right) < threshold) {
+		return delete_center_node(info, parent, l, c, r, left, center, right,
+					  nr_left, nr_center, nr_right);
+	}
+
+	return redistribute3(info, parent, l, c, r, left, center, right,
+			     nr_left, nr_center, nr_right);
+}
+
+static int rebalance3(struct shadow_spine *s, struct dm_btree_info *info,
+		      struct dm_btree_value_type *vt, unsigned int left_index)
+{
+	int r;
+	struct btree_node *parent = dm_block_data(shadow_current(s));
+	struct child left, center, right;
+
+	/*
+	 * FIXME: fill out an array?
+	 */
+	r = init_child(info, vt, parent, left_index, &left);
+	if (r)
+		return r;
+
+	r = init_child(info, vt, parent, left_index + 1, &center);
+	if (r) {
+		exit_child(info, &left);
+		return r;
+	}
+
+	r = init_child(info, vt, parent, left_index + 2, &right);
+	if (r) {
+		exit_child(info, &left);
+		exit_child(info, &center);
+		return r;
+	}
+
+	r = __rebalance3(info, parent, &left, &center, &right);
+
+	exit_child(info, &left);
+	exit_child(info, &center);
+	exit_child(info, &right);
+
+	return r;
+}
+
+static int rebalance_children(struct shadow_spine *s,
+			      struct dm_btree_info *info,
+			      struct dm_btree_value_type *vt, uint64_t key)
+{
+	int i, r, has_left_sibling, has_right_sibling;
+	struct btree_node *n;
+
+	n = dm_block_data(shadow_current(s));
+
+	if (le32_to_cpu(n->header.nr_entries) == 1) {
+		struct dm_block *child;
+		int is_shared;
+		dm_block_t b = value64(n, 0);
+
+		r = dm_tm_block_is_shared(info->tm, b, &is_shared);
+		if (r)
+			return r;
+
+		r = dm_tm_read_lock(info->tm, b, &btree_node_validator, &child);
+		if (r)
+			return r;
+
+		if (is_shared)
+			inc_children(info->tm, dm_block_data(child), vt);
+
+		memcpy(n, dm_block_data(child),
+		       dm_bm_block_size(dm_tm_get_bm(info->tm)));
+
+		dm_tm_dec(info->tm, dm_block_location(child));
+		dm_tm_unlock(info->tm, child);
+		return 0;
+	}
+
+	i = lower_bound(n, key);
+	if (i < 0)
+		return -ENODATA;
+
+	has_left_sibling = i > 0;
+	has_right_sibling = i < (le32_to_cpu(n->header.nr_entries) - 1);
+
+	if (!has_left_sibling)
+		r = rebalance2(s, info, vt, i);
+
+	else if (!has_right_sibling)
+		r = rebalance2(s, info, vt, i - 1);
+
+	else
+		r = rebalance3(s, info, vt, i - 1);
+
+	return r;
+}
+
+static int do_leaf(struct btree_node *n, uint64_t key, unsigned int *index)
+{
+	int i = lower_bound(n, key);
+
+	if ((i < 0) ||
+	    (i >= le32_to_cpu(n->header.nr_entries)) ||
+	    (le64_to_cpu(n->keys[i]) != key))
+		return -ENODATA;
+
+	*index = i;
+
+	return 0;
+}
+
+/*
+ * Prepares for removal from one level of the hierarchy.  The caller must
+ * call delete_at() to remove the entry at index.
+ */
+static int remove_raw(struct shadow_spine *s, struct dm_btree_info *info,
+		      struct dm_btree_value_type *vt, dm_block_t root,
+		      uint64_t key, unsigned int *index)
+{
+	int i = *index, r;
+	struct btree_node *n;
+
+	for (;;) {
+		r = shadow_step(s, root, vt);
+		if (r < 0)
+			break;
+
+		/*
+		 * We have to patch up the parent node, ugly, but I don't
+		 * see a way to do this automatically as part of the spine
+		 * op.
+		 */
+		if (shadow_has_parent(s)) {
+			__le64 location = cpu_to_le64(dm_block_location(shadow_current(s)));
+
+			memcpy(value_ptr(dm_block_data(shadow_parent(s)), i),
+			       &location, sizeof(__le64));
+		}
+
+		n = dm_block_data(shadow_current(s));
+
+		if (le32_to_cpu(n->header.flags) & LEAF_NODE)
+			return do_leaf(n, key, index);
+
+		r = rebalance_children(s, info, vt, key);
+		if (r)
+			break;
+
+		n = dm_block_data(shadow_current(s));
+		if (le32_to_cpu(n->header.flags) & LEAF_NODE)
+			return do_leaf(n, key, index);
+
+		i = lower_bound(n, key);
+
+		/*
+		 * We know the key is present, or else
+		 * rebalance_children would have returned
+		 * -ENODATA
+		 */
+		root = value64(n, i);
+	}
+
+	return r;
+}
+
+int dm_btree_remove(struct dm_btree_info *info, dm_block_t root,
+		    uint64_t *keys, dm_block_t *new_root)
+{
+	unsigned int level, last_level = info->levels - 1;
+	int index = 0, r = 0;
+	struct shadow_spine spine;
+	struct btree_node *n;
+	struct dm_btree_value_type le64_vt;
+
+	init_le64_type(info->tm, &le64_vt);
+	init_shadow_spine(&spine, info);
+	for (level = 0; level < info->levels; level++) {
+		r = remove_raw(&spine, info,
+			       (level == last_level ?
+				&info->value_type : &le64_vt),
+			       root, keys[level], (unsigned int *)&index);
+		if (r < 0)
+			break;
+
+		n = dm_block_data(shadow_current(&spine));
+		if (level != last_level) {
+			root = value64(n, index);
+			continue;
+		}
+
+		BUG_ON(index < 0 || index >= le32_to_cpu(n->header.nr_entries));
+
+		if (info->value_type.dec)
+			info->value_type.dec(info->value_type.context,
+					     value_ptr(n, index), 1);
+
+		delete_at(n, index);
+	}
+
+	if (!r)
+		*new_root = shadow_root(&spine);
+	exit_shadow_spine(&spine);
+
+	return r;
+}
+EXPORT_SYMBOL_GPL(dm_btree_remove);
+
+/*----------------------------------------------------------------*/
+
+static int remove_nearest(struct shadow_spine *s, struct dm_btree_info *info,
+			  struct dm_btree_value_type *vt, dm_block_t root,
+			  uint64_t key, int *index)
+{
+	int i = *index, r;
+	struct btree_node *n;
+
+	for (;;) {
+		r = shadow_step(s, root, vt);
+		if (r < 0)
+			break;
+
+		/*
+		 * We have to patch up the parent node, ugly, but I don't
+		 * see a way to do this automatically as part of the spine
+		 * op.
+		 */
+		if (shadow_has_parent(s)) {
+			__le64 location = cpu_to_le64(dm_block_location(shadow_current(s)));
+
+			memcpy(value_ptr(dm_block_data(shadow_parent(s)), i),
+			       &location, sizeof(__le64));
+		}
+
+		n = dm_block_data(shadow_current(s));
+
+		if (le32_to_cpu(n->header.flags) & LEAF_NODE) {
+			*index = lower_bound(n, key);
+			return 0;
+		}
+
+		r = rebalance_children(s, info, vt, key);
+		if (r)
+			break;
+
+		n = dm_block_data(shadow_current(s));
+		if (le32_to_cpu(n->header.flags) & LEAF_NODE) {
+			*index = lower_bound(n, key);
+			return 0;
+		}
+
+		i = lower_bound(n, key);
+
+		/*
+		 * We know the key is present, or else
+		 * rebalance_children would have returned
+		 * -ENODATA
+		 */
+		root = value64(n, i);
+	}
+
+	return r;
+}
+
+static int remove_one(struct dm_btree_info *info, dm_block_t root,
+		      uint64_t *keys, uint64_t end_key,
+		      dm_block_t *new_root, unsigned int *nr_removed)
+{
+	unsigned int level, last_level = info->levels - 1;
+	int index = 0, r = 0;
+	struct shadow_spine spine;
+	struct btree_node *n;
+	struct dm_btree_value_type le64_vt;
+	uint64_t k;
+
+	init_le64_type(info->tm, &le64_vt);
+	init_shadow_spine(&spine, info);
+	for (level = 0; level < last_level; level++) {
+		r = remove_raw(&spine, info, &le64_vt,
+			       root, keys[level], (unsigned int *) &index);
+		if (r < 0)
+			goto out;
+
+		n = dm_block_data(shadow_current(&spine));
+		root = value64(n, index);
+	}
+
+	r = remove_nearest(&spine, info, &info->value_type,
+			   root, keys[last_level], &index);
+	if (r < 0)
+		goto out;
+
+	n = dm_block_data(shadow_current(&spine));
+
+	if (index < 0)
+		index = 0;
+
+	if (index >= le32_to_cpu(n->header.nr_entries)) {
+		r = -ENODATA;
+		goto out;
+	}
+
+	k = le64_to_cpu(n->keys[index]);
+	if (k >= keys[last_level] && k < end_key) {
+		if (info->value_type.dec)
+			info->value_type.dec(info->value_type.context,
+					     value_ptr(n, index), 1);
+
+		delete_at(n, index);
+		keys[last_level] = k + 1ull;
+
+	} else
+		r = -ENODATA;
+
+out:
+	*new_root = shadow_root(&spine);
+	exit_shadow_spine(&spine);
+
+	return r;
+}
+
+int dm_btree_remove_leaves(struct dm_btree_info *info, dm_block_t root,
+			   uint64_t *first_key, uint64_t end_key,
+			   dm_block_t *new_root, unsigned int *nr_removed)
+{
+	int r;
+
+	*nr_removed = 0;
+	do {
+		r = remove_one(info, root, first_key, end_key, &root, nr_removed);
+		if (!r)
+			(*nr_removed)++;
+	} while (!r);
+
+	*new_root = root;
+	return r == -ENODATA ? 0 : r;
+}
+EXPORT_SYMBOL_GPL(dm_btree_remove_leaves);

linux-7.2.3/drivers/md/persistent-data/dm-btree-spine.c

@@ -0,0 +1,263 @@
+// SPDX-License-Identifier: GPL-2.0-only
+/*
+ * Copyright (C) 2011 Red Hat, Inc.
+ *
+ * This file is released under the GPL.
+ */
+
+#include "dm-btree-internal.h"
+#include "dm-transaction-manager.h"
+
+#include <linux/device-mapper.h>
+
+#define DM_MSG_PREFIX "btree spine"
+
+/*----------------------------------------------------------------*/
+
+#define BTREE_CSUM_XOR 121107
+
+static void node_prepare_for_write(const struct dm_block_validator *v,
+				   struct dm_block *b,
+				   size_t block_size)
+{
+	struct btree_node *n = dm_block_data(b);
+	struct node_header *h = &n->header;
+
+	h->blocknr = cpu_to_le64(dm_block_location(b));
+	h->csum = cpu_to_le32(dm_bm_checksum(&h->flags,
+					     block_size - sizeof(__le32),
+					     BTREE_CSUM_XOR));
+}
+
+static int node_check(const struct dm_block_validator *v,
+		      struct dm_block *b,
+		      size_t block_size)
+{
+	struct btree_node *n = dm_block_data(b);
+	struct node_header *h = &n->header;
+	size_t value_size;
+	__le32 csum_disk;
+	uint32_t flags, nr_entries, max_entries;
+
+	if (dm_block_location(b) != le64_to_cpu(h->blocknr)) {
+		DMERR_LIMIT("%s failed: blocknr %llu != wanted %llu", __func__,
+			    le64_to_cpu(h->blocknr), dm_block_location(b));
+		return -ENOTBLK;
+	}
+
+	csum_disk = cpu_to_le32(dm_bm_checksum(&h->flags,
+					       block_size - sizeof(__le32),
+					       BTREE_CSUM_XOR));
+	if (csum_disk != h->csum) {
+		DMERR_LIMIT("%s failed: csum %u != wanted %u", __func__,
+			    le32_to_cpu(csum_disk), le32_to_cpu(h->csum));
+		return -EILSEQ;
+	}
+
+	nr_entries = le32_to_cpu(h->nr_entries);
+	max_entries = le32_to_cpu(h->max_entries);
+	value_size = le32_to_cpu(h->value_size);
+
+	if (sizeof(struct node_header) +
+	    (sizeof(__le64) + value_size) * max_entries > block_size) {
+		DMERR_LIMIT("%s failed: max_entries too large", __func__);
+		return -EILSEQ;
+	}
+
+	if (nr_entries > max_entries) {
+		DMERR_LIMIT("%s failed: too many entries", __func__);
+		return -EILSEQ;
+	}
+
+	/*
+	 * The node must be either INTERNAL or LEAF.
+	 */
+	flags = le32_to_cpu(h->flags);
+	if (!(flags & INTERNAL_NODE) && !(flags & LEAF_NODE)) {
+		DMERR_LIMIT("%s failed: node is neither INTERNAL or LEAF", __func__);
+		return -EILSEQ;
+	}
+
+	return 0;
+}
+
+const struct dm_block_validator btree_node_validator = {
+	.name = "btree_node",
+	.prepare_for_write = node_prepare_for_write,
+	.check = node_check
+};
+
+/*----------------------------------------------------------------*/
+
+int bn_read_lock(struct dm_btree_info *info, dm_block_t b,
+		 struct dm_block **result)
+{
+	return dm_tm_read_lock(info->tm, b, &btree_node_validator, result);
+}
+
+static int bn_shadow(struct dm_btree_info *info, dm_block_t orig,
+	      struct dm_btree_value_type *vt,
+	      struct dm_block **result)
+{
+	int r, inc;
+
+	r = dm_tm_shadow_block(info->tm, orig, &btree_node_validator,
+			       result, &inc);
+	if (!r && inc)
+		inc_children(info->tm, dm_block_data(*result), vt);
+
+	return r;
+}
+
+int new_block(struct dm_btree_info *info, struct dm_block **result)
+{
+	return dm_tm_new_block(info->tm, &btree_node_validator, result);
+}
+
+void unlock_block(struct dm_btree_info *info, struct dm_block *b)
+{
+	dm_tm_unlock(info->tm, b);
+}
+
+/*----------------------------------------------------------------*/
+
+void init_ro_spine(struct ro_spine *s, struct dm_btree_info *info)
+{
+	s->info = info;
+	s->count = 0;
+	s->nodes[0] = NULL;
+	s->nodes[1] = NULL;
+}
+
+void exit_ro_spine(struct ro_spine *s)
+{
+	int i;
+
+	for (i = 0; i < s->count; i++)
+		unlock_block(s->info, s->nodes[i]);
+}
+
+int ro_step(struct ro_spine *s, dm_block_t new_child)
+{
+	int r;
+
+	if (s->count == 2) {
+		unlock_block(s->info, s->nodes[0]);
+		s->nodes[0] = s->nodes[1];
+		s->count--;
+	}
+
+	r = bn_read_lock(s->info, new_child, s->nodes + s->count);
+	if (!r)
+		s->count++;
+
+	return r;
+}
+
+void ro_pop(struct ro_spine *s)
+{
+	BUG_ON(!s->count);
+	--s->count;
+	unlock_block(s->info, s->nodes[s->count]);
+}
+
+struct btree_node *ro_node(struct ro_spine *s)
+{
+	struct dm_block *block;
+
+	BUG_ON(!s->count);
+	block = s->nodes[s->count - 1];
+
+	return dm_block_data(block);
+}
+
+/*----------------------------------------------------------------*/
+
+void init_shadow_spine(struct shadow_spine *s, struct dm_btree_info *info)
+{
+	s->info = info;
+	s->count = 0;
+}
+
+void exit_shadow_spine(struct shadow_spine *s)
+{
+	int i;
+
+	for (i = 0; i < s->count; i++)
+		unlock_block(s->info, s->nodes[i]);
+}
+
+int shadow_step(struct shadow_spine *s, dm_block_t b,
+		struct dm_btree_value_type *vt)
+{
+	int r;
+
+	if (s->count == 2) {
+		unlock_block(s->info, s->nodes[0]);
+		s->nodes[0] = s->nodes[1];
+		s->count--;
+	}
+
+	r = bn_shadow(s->info, b, vt, s->nodes + s->count);
+	if (!r) {
+		if (!s->count)
+			s->root = dm_block_location(s->nodes[0]);
+
+		s->count++;
+	}
+
+	return r;
+}
+
+struct dm_block *shadow_current(struct shadow_spine *s)
+{
+	BUG_ON(!s->count);
+
+	return s->nodes[s->count - 1];
+}
+
+struct dm_block *shadow_parent(struct shadow_spine *s)
+{
+	BUG_ON(s->count != 2);
+
+	return s->count == 2 ? s->nodes[0] : NULL;
+}
+
+int shadow_has_parent(struct shadow_spine *s)
+{
+	return s->count >= 2;
+}
+
+dm_block_t shadow_root(struct shadow_spine *s)
+{
+	return s->root;
+}
+
+static void le64_inc(void *context, const void *value_le, unsigned int count)
+{
+	dm_tm_with_runs(context, value_le, count, dm_tm_inc_range);
+}
+
+static void le64_dec(void *context, const void *value_le, unsigned int count)
+{
+	dm_tm_with_runs(context, value_le, count, dm_tm_dec_range);
+}
+
+static int le64_equal(void *context, const void *value1_le, const void *value2_le)
+{
+	__le64 v1_le, v2_le;
+
+	memcpy(&v1_le, value1_le, sizeof(v1_le));
+	memcpy(&v2_le, value2_le, sizeof(v2_le));
+	return v1_le == v2_le;
+}
+
+void init_le64_type(struct dm_transaction_manager *tm,
+		    struct dm_btree_value_type *vt)
+{
+	vt->context = tm;
+	vt->size = sizeof(__le64);
+	vt->inc = le64_inc;
+	vt->dec = le64_dec;
+	vt->equal = le64_equal;
+}

linux-7.2.3/drivers/md/persistent-data/dm-btree.h

@@ -0,0 +1,216 @@
+/* SPDX-License-Identifier: GPL-2.0-only */
+/*
+ * Copyright (C) 2011 Red Hat, Inc.
+ *
+ * This file is released under the GPL.
+ */
+#ifndef _LINUX_DM_BTREE_H
+#define _LINUX_DM_BTREE_H
+
+#include "dm-block-manager.h"
+
+struct dm_transaction_manager;
+
+/*----------------------------------------------------------------*/
+
+/*
+ * Annotations used to check on-disk metadata is handled as little-endian.
+ */
+#ifdef __CHECKER__
+#  define __dm_written_to_disk(x) __releases(x)
+#  define __dm_reads_from_disk(x) __acquires(x)
+#  define __dm_bless_for_disk(x) __acquire(x)
+#  define __dm_unbless_for_disk(x) __release(x)
+#else
+#  define __dm_written_to_disk(x)
+#  define __dm_reads_from_disk(x)
+#  define __dm_bless_for_disk(x)
+#  define __dm_unbless_for_disk(x)
+#endif
+
+/*----------------------------------------------------------------*/
+
+/*
+ * Manipulates hierarchical B+ trees with 64-bit keys and arbitrary-sized
+ * values.
+ */
+
+/*
+ * Information about the values stored within the btree.
+ */
+struct dm_btree_value_type {
+	void *context;
+
+	/*
+	 * The size in bytes of each value.
+	 */
+	uint32_t size;
+
+	/*
+	 * Any of these methods can be safely set to NULL if you do not
+	 * need the corresponding feature.
+	 */
+
+	/*
+	 * The btree is making a duplicate of a run of values, for instance
+	 * because previously-shared btree nodes have now diverged.
+	 * @value argument is the new copy that the copy function may modify.
+	 * (Probably it just wants to increment a reference count
+	 * somewhere.) This method is _not_ called for insertion of a new
+	 * value: It is assumed the ref count is already 1.
+	 */
+	void (*inc)(void *context, const void *value, unsigned int count);
+
+	/*
+	 * These values are being deleted.  The btree takes care of freeing
+	 * the memory pointed to by @value.  Often the del function just
+	 * needs to decrement a reference counts somewhere.
+	 */
+	void (*dec)(void *context, const void *value, unsigned int count);
+
+	/*
+	 * A test for equality between two values.  When a value is
+	 * overwritten with a new one, the old one has the dec method
+	 * called _unless_ the new and old value are deemed equal.
+	 */
+	int (*equal)(void *context, const void *value1, const void *value2);
+};
+
+/*
+ * The shape and contents of a btree.
+ */
+struct dm_btree_info {
+	struct dm_transaction_manager *tm;
+
+	/*
+	 * Number of nested btrees. (Not the depth of a single tree.)
+	 */
+	unsigned int levels;
+	struct dm_btree_value_type value_type;
+};
+
+/*
+ * Set up an empty tree.  O(1).
+ */
+int dm_btree_empty(struct dm_btree_info *info, dm_block_t *root);
+
+/*
+ * Delete a tree.  O(n) - this is the slow one!  It can also block, so
+ * please don't call it on an IO path.
+ */
+int dm_btree_del(struct dm_btree_info *info, dm_block_t root);
+
+/*
+ * All the lookup functions return -ENODATA if the key cannot be found.
+ */
+
+/*
+ * Tries to find a key that matches exactly.  O(ln(n))
+ */
+int dm_btree_lookup(struct dm_btree_info *info, dm_block_t root,
+		    uint64_t *keys, void *value_le);
+
+/*
+ * Tries to find the first key where the bottom level key is >= to that
+ * given.  Useful for skipping empty sections of the btree.
+ */
+int dm_btree_lookup_next(struct dm_btree_info *info, dm_block_t root,
+			 uint64_t *keys, uint64_t *rkey, void *value_le);
+
+/*
+ * Insertion (or overwrite an existing value).  O(ln(n))
+ */
+int dm_btree_insert(struct dm_btree_info *info, dm_block_t root,
+		    uint64_t *keys, void *value, dm_block_t *new_root)
+	__dm_written_to_disk(value);
+
+/*
+ * A variant of insert that indicates whether it actually inserted or just
+ * overwrote.  Useful if you're keeping track of the number of entries in a
+ * tree.
+ */
+int dm_btree_insert_notify(struct dm_btree_info *info, dm_block_t root,
+			   uint64_t *keys, void *value, dm_block_t *new_root,
+			   int *inserted)
+			   __dm_written_to_disk(value);
+
+/*
+ * Remove a key if present.  This doesn't remove empty sub trees.  Normally
+ * subtrees represent a separate entity, like a snapshot map, so this is
+ * correct behaviour.  O(ln(n)).
+ */
+int dm_btree_remove(struct dm_btree_info *info, dm_block_t root,
+		    uint64_t *keys, dm_block_t *new_root);
+
+/*
+ * Removes a _contiguous_ run of values starting from 'keys' and not
+ * reaching keys2 (where keys2 is keys with the final key replaced with
+ * 'end_key').  'end_key' is the one-past-the-end value.  'keys' may be
+ * altered.
+ */
+int dm_btree_remove_leaves(struct dm_btree_info *info, dm_block_t root,
+			   uint64_t *keys, uint64_t end_key,
+			   dm_block_t *new_root, unsigned int *nr_removed);
+
+/*
+ * Returns < 0 on failure.  Otherwise the number of key entries that have
+ * been filled out.  Remember trees can have zero entries, and as such have
+ * no lowest key.
+ */
+int dm_btree_find_lowest_key(struct dm_btree_info *info, dm_block_t root,
+			     uint64_t *result_keys);
+
+/*
+ * Returns < 0 on failure.  Otherwise the number of key entries that have
+ * been filled out.  Remember trees can have zero entries, and as such have
+ * no highest key.
+ */
+int dm_btree_find_highest_key(struct dm_btree_info *info, dm_block_t root,
+			      uint64_t *result_keys);
+
+/*
+ * Iterate through the a btree, calling fn() on each entry.
+ * It only works for single level trees and is internally recursive, so
+ * monitor stack usage carefully.
+ */
+int dm_btree_walk(struct dm_btree_info *info, dm_block_t root,
+		  int (*fn)(void *context, uint64_t *keys, void *leaf),
+		  void *context);
+
+
+/*----------------------------------------------------------------*/
+
+/*
+ * Cursor API.  This does not follow the rolling lock convention.  Since we
+ * know the order that values are required we can issue prefetches to speed
+ * up iteration.  Use on a single level btree only.
+ */
+#define DM_BTREE_CURSOR_MAX_DEPTH 16
+
+struct cursor_node {
+	struct dm_block *b;
+	unsigned int index;
+};
+
+struct dm_btree_cursor {
+	struct dm_btree_info *info;
+	dm_block_t root;
+
+	bool prefetch_leaves;
+	unsigned int depth;
+	struct cursor_node nodes[DM_BTREE_CURSOR_MAX_DEPTH];
+};
+
+/*
+ * Creates a fresh cursor.  If prefetch_leaves is set then it is assumed
+ * the btree contains block indexes that will be prefetched.  The cursor is
+ * quite large, so you probably don't want to put it on the stack.
+ */
+int dm_btree_cursor_begin(struct dm_btree_info *info, dm_block_t root,
+			  bool prefetch_leaves, struct dm_btree_cursor *c);
+void dm_btree_cursor_end(struct dm_btree_cursor *c);
+int dm_btree_cursor_next(struct dm_btree_cursor *c);
+int dm_btree_cursor_skip(struct dm_btree_cursor *c, uint32_t count);
+int dm_btree_cursor_get_value(struct dm_btree_cursor *c, uint64_t *key, void *value_le);
+
+#endif	/* _LINUX_DM_BTREE_H */

linux-7.2.3/drivers/md/persistent-data/dm-persistent-data-internal.h

@@ -0,0 +1,20 @@
+/* SPDX-License-Identifier: GPL-2.0-only */
+/*
+ * Copyright (C) 2011 Red Hat, Inc.
+ *
+ * This file is released under the GPL.
+ */
+
+#ifndef _DM_PERSISTENT_DATA_INTERNAL_H
+#define _DM_PERSISTENT_DATA_INTERNAL_H
+
+#include "dm-block-manager.h"
+
+static inline unsigned int dm_hash_block(dm_block_t b, unsigned int hash_mask)
+{
+	const unsigned int BIG_PRIME = 4294967291UL;
+
+	return (((unsigned int) b) * BIG_PRIME) & hash_mask;
+}
+
+#endif	/* _PERSISTENT_DATA_INTERNAL_H */

linux-7.2.3/drivers/md/persistent-data/dm-space-map-common.h

@@ -0,0 +1,146 @@
+/* SPDX-License-Identifier: GPL-2.0-only */
+/*
+ * Copyright (C) 2011 Red Hat, Inc.
+ *
+ * This file is released under the GPL.
+ */
+
+#ifndef DM_SPACE_MAP_COMMON_H
+#define DM_SPACE_MAP_COMMON_H
+
+#include "dm-btree.h"
+
+/*----------------------------------------------------------------*/
+
+/*
+ * Low level disk format
+ *
+ * Bitmap btree
+ * ------------
+ *
+ * Each value stored in the btree is an index_entry.  This points to a
+ * block that is used as a bitmap.  Within the bitmap hold 2 bits per
+ * entry, which represent UNUSED = 0, REF_COUNT = 1, REF_COUNT = 2 and
+ * REF_COUNT = many.
+ *
+ * Refcount btree
+ * --------------
+ *
+ * Any entry that has a ref count higher than 2 gets entered in the ref
+ * count tree.  The leaf values for this tree is the 32-bit ref count.
+ */
+
+struct disk_index_entry {
+	__le64 blocknr;
+	__le32 nr_free;
+	__le32 none_free_before;
+} __packed __aligned(8);
+
+
+#define MAX_METADATA_BITMAPS 255
+struct disk_metadata_index {
+	__le32 csum;
+	__le32 padding;
+	__le64 blocknr;
+
+	struct disk_index_entry index[MAX_METADATA_BITMAPS];
+} __packed __aligned(8);
+
+struct ll_disk;
+
+typedef int (*load_ie_fn)(struct ll_disk *ll, dm_block_t index, struct disk_index_entry *result);
+typedef int (*save_ie_fn)(struct ll_disk *ll, dm_block_t index, struct disk_index_entry *ie);
+typedef int (*init_index_fn)(struct ll_disk *ll);
+typedef int (*open_index_fn)(struct ll_disk *ll);
+typedef dm_block_t (*max_index_entries_fn)(struct ll_disk *ll);
+typedef int (*commit_fn)(struct ll_disk *ll);
+
+/*
+ * A lot of time can be wasted reading and writing the same
+ * index entry.  So we cache a few entries.
+ */
+#define IE_CACHE_SIZE 64
+#define IE_CACHE_MASK (IE_CACHE_SIZE - 1)
+
+struct ie_cache {
+	bool valid;
+	bool dirty;
+	dm_block_t index;
+	struct disk_index_entry ie;
+};
+
+struct ll_disk {
+	struct dm_transaction_manager *tm;
+	struct dm_btree_info bitmap_info;
+	struct dm_btree_info ref_count_info;
+
+	uint32_t block_size;
+	uint32_t entries_per_block;
+	dm_block_t nr_blocks;
+	dm_block_t nr_allocated;
+
+	/*
+	 * bitmap_root may be a btree root or a simple index.
+	 */
+	dm_block_t bitmap_root;
+
+	dm_block_t ref_count_root;
+
+	struct disk_metadata_index mi_le;
+	load_ie_fn load_ie;
+	save_ie_fn save_ie;
+	init_index_fn init_index;
+	open_index_fn open_index;
+	max_index_entries_fn max_entries;
+	commit_fn commit;
+	bool bitmap_index_changed:1;
+
+	struct ie_cache ie_cache[IE_CACHE_SIZE];
+};
+
+struct disk_sm_root {
+	__le64 nr_blocks;
+	__le64 nr_allocated;
+	__le64 bitmap_root;
+	__le64 ref_count_root;
+} __packed __aligned(8);
+
+#define ENTRIES_PER_BYTE 4
+
+struct disk_bitmap_header {
+	__le32 csum;
+	__le32 not_used;
+	__le64 blocknr;
+} __packed __aligned(8);
+
+/*----------------------------------------------------------------*/
+
+int sm_ll_extend(struct ll_disk *ll, dm_block_t extra_blocks);
+int sm_ll_lookup_bitmap(struct ll_disk *ll, dm_block_t b, uint32_t *result);
+int sm_ll_lookup(struct ll_disk *ll, dm_block_t b, uint32_t *result);
+int sm_ll_find_free_block(struct ll_disk *ll, dm_block_t begin,
+			  dm_block_t end, dm_block_t *result);
+int sm_ll_find_common_free_block(struct ll_disk *old_ll, struct ll_disk *new_ll,
+				 dm_block_t begin, dm_block_t end, dm_block_t *result);
+
+/*
+ * The next three functions return (via nr_allocations) the net number of
+ * allocations that were made.  This number may be negative if there were
+ * more frees than allocs.
+ */
+int sm_ll_insert(struct ll_disk *ll, dm_block_t b, uint32_t ref_count, int32_t *nr_allocations);
+int sm_ll_inc(struct ll_disk *ll, dm_block_t b, dm_block_t e, int32_t *nr_allocations);
+int sm_ll_dec(struct ll_disk *ll, dm_block_t b, dm_block_t e, int32_t *nr_allocations);
+int sm_ll_commit(struct ll_disk *ll);
+
+int sm_ll_new_metadata(struct ll_disk *ll, struct dm_transaction_manager *tm);
+int sm_ll_open_metadata(struct ll_disk *ll, struct dm_transaction_manager *tm,
+			void *root_le, size_t len);
+
+int sm_ll_new_disk(struct ll_disk *ll, struct dm_transaction_manager *tm);
+int sm_ll_open_disk(struct ll_disk *ll, struct dm_transaction_manager *tm,
+		    void *root_le, size_t len);
+
+/*----------------------------------------------------------------*/
+
+#endif	/* DM_SPACE_MAP_COMMON_H */

linux-7.2.3/drivers/md/persistent-data/dm-space-map-disk.c

@@ -0,0 +1,281 @@
+// SPDX-License-Identifier: GPL-2.0-only
+/*
+ * Copyright (C) 2011 Red Hat, Inc.
+ *
+ * This file is released under the GPL.
+ */
+
+#include "dm-space-map-common.h"
+#include "dm-space-map-disk.h"
+#include "dm-space-map.h"
+#include "dm-transaction-manager.h"
+
+#include <linux/list.h>
+#include <linux/slab.h>
+#include <linux/export.h>
+#include <linux/device-mapper.h>
+
+#define DM_MSG_PREFIX "space map disk"
+
+/*----------------------------------------------------------------*/
+
+/*
+ * Space map interface.
+ */
+struct sm_disk {
+	struct dm_space_map sm;
+
+	struct ll_disk ll;
+	struct ll_disk old_ll;
+
+	dm_block_t begin;
+	dm_block_t nr_allocated_this_transaction;
+};
+
+static void sm_disk_destroy(struct dm_space_map *sm)
+{
+	struct sm_disk *smd = container_of(sm, struct sm_disk, sm);
+
+	kfree(smd);
+}
+
+static int sm_disk_extend(struct dm_space_map *sm, dm_block_t extra_blocks)
+{
+	struct sm_disk *smd = container_of(sm, struct sm_disk, sm);
+
+	return sm_ll_extend(&smd->ll, extra_blocks);
+}
+
+static int sm_disk_get_nr_blocks(struct dm_space_map *sm, dm_block_t *count)
+{
+	struct sm_disk *smd = container_of(sm, struct sm_disk, sm);
+
+	*count = smd->old_ll.nr_blocks;
+
+	return 0;
+}
+
+static int sm_disk_get_nr_free(struct dm_space_map *sm, dm_block_t *count)
+{
+	struct sm_disk *smd = container_of(sm, struct sm_disk, sm);
+
+	*count = (smd->old_ll.nr_blocks - smd->old_ll.nr_allocated) - smd->nr_allocated_this_transaction;
+
+	return 0;
+}
+
+static int sm_disk_get_count(struct dm_space_map *sm, dm_block_t b,
+			     uint32_t *result)
+{
+	struct sm_disk *smd = container_of(sm, struct sm_disk, sm);
+
+	return sm_ll_lookup(&smd->ll, b, result);
+}
+
+static int sm_disk_count_is_more_than_one(struct dm_space_map *sm, dm_block_t b,
+					  int *result)
+{
+	int r;
+	uint32_t count;
+
+	r = sm_disk_get_count(sm, b, &count);
+	if (r)
+		return r;
+
+	*result = count > 1;
+
+	return 0;
+}
+
+static int sm_disk_set_count(struct dm_space_map *sm, dm_block_t b,
+			     uint32_t count)
+{
+	int r;
+	int32_t nr_allocations;
+	struct sm_disk *smd = container_of(sm, struct sm_disk, sm);
+
+	r = sm_ll_insert(&smd->ll, b, count, &nr_allocations);
+	if (!r)
+		smd->nr_allocated_this_transaction += nr_allocations;
+
+	return r;
+}
+
+static int sm_disk_inc_blocks(struct dm_space_map *sm, dm_block_t b, dm_block_t e)
+{
+	int r;
+	int32_t nr_allocations;
+	struct sm_disk *smd = container_of(sm, struct sm_disk, sm);
+
+	r = sm_ll_inc(&smd->ll, b, e, &nr_allocations);
+	if (!r)
+		smd->nr_allocated_this_transaction += nr_allocations;
+
+	return r;
+}
+
+static int sm_disk_dec_blocks(struct dm_space_map *sm, dm_block_t b, dm_block_t e)
+{
+	int r;
+	int32_t nr_allocations;
+	struct sm_disk *smd = container_of(sm, struct sm_disk, sm);
+
+	r = sm_ll_dec(&smd->ll, b, e, &nr_allocations);
+	if (!r)
+		smd->nr_allocated_this_transaction += nr_allocations;
+
+	return r;
+}
+
+static int sm_disk_new_block(struct dm_space_map *sm, dm_block_t *b)
+{
+	int r;
+	int32_t nr_allocations;
+	struct sm_disk *smd = container_of(sm, struct sm_disk, sm);
+
+	/*
+	 * Any block we allocate has to be free in both the old and current ll.
+	 */
+	r = sm_ll_find_common_free_block(&smd->old_ll, &smd->ll, smd->begin, smd->ll.nr_blocks, b);
+	if (r == -ENOSPC)
+		/*
+		 * There's no free block between smd->begin and the end of the metadata device.
+		 * We search before smd->begin in case something has been freed.
+		 */
+		r = sm_ll_find_common_free_block(&smd->old_ll, &smd->ll, 0, smd->begin, b);
+
+	if (r)
+		return r;
+
+	smd->begin = *b + 1;
+	r = sm_ll_inc(&smd->ll, *b, *b + 1, &nr_allocations);
+	if (!r)
+		smd->nr_allocated_this_transaction += nr_allocations;
+
+	return r;
+}
+
+static int sm_disk_commit(struct dm_space_map *sm)
+{
+	int r;
+	struct sm_disk *smd = container_of(sm, struct sm_disk, sm);
+
+	r = sm_ll_commit(&smd->ll);
+	if (r)
+		return r;
+
+	memcpy(&smd->old_ll, &smd->ll, sizeof(smd->old_ll));
+	smd->nr_allocated_this_transaction = 0;
+
+	return 0;
+}
+
+static int sm_disk_root_size(struct dm_space_map *sm, size_t *result)
+{
+	*result = sizeof(struct disk_sm_root);
+
+	return 0;
+}
+
+static int sm_disk_copy_root(struct dm_space_map *sm, void *where_le, size_t max)
+{
+	struct sm_disk *smd = container_of(sm, struct sm_disk, sm);
+	struct disk_sm_root root_le;
+
+	root_le.nr_blocks = cpu_to_le64(smd->ll.nr_blocks);
+	root_le.nr_allocated = cpu_to_le64(smd->ll.nr_allocated);
+	root_le.bitmap_root = cpu_to_le64(smd->ll.bitmap_root);
+	root_le.ref_count_root = cpu_to_le64(smd->ll.ref_count_root);
+
+	if (max < sizeof(root_le))
+		return -ENOSPC;
+
+	memcpy(where_le, &root_le, sizeof(root_le));
+
+	return 0;
+}
+
+/*----------------------------------------------------------------*/
+
+static struct dm_space_map ops = {
+	.destroy = sm_disk_destroy,
+	.extend = sm_disk_extend,
+	.get_nr_blocks = sm_disk_get_nr_blocks,
+	.get_nr_free = sm_disk_get_nr_free,
+	.get_count = sm_disk_get_count,
+	.count_is_more_than_one = sm_disk_count_is_more_than_one,
+	.set_count = sm_disk_set_count,
+	.inc_blocks = sm_disk_inc_blocks,
+	.dec_blocks = sm_disk_dec_blocks,
+	.new_block = sm_disk_new_block,
+	.commit = sm_disk_commit,
+	.root_size = sm_disk_root_size,
+	.copy_root = sm_disk_copy_root,
+	.register_threshold_callback = NULL
+};
+
+struct dm_space_map *dm_sm_disk_create(struct dm_transaction_manager *tm,
+				       dm_block_t nr_blocks)
+{
+	int r;
+	struct sm_disk *smd;
+
+	smd = kmalloc_obj(*smd);
+	if (!smd)
+		return ERR_PTR(-ENOMEM);
+
+	smd->begin = 0;
+	smd->nr_allocated_this_transaction = 0;
+	memcpy(&smd->sm, &ops, sizeof(smd->sm));
+
+	r = sm_ll_new_disk(&smd->ll, tm);
+	if (r)
+		goto bad;
+
+	r = sm_ll_extend(&smd->ll, nr_blocks);
+	if (r)
+		goto bad;
+
+	r = sm_disk_commit(&smd->sm);
+	if (r)
+		goto bad;
+
+	return &smd->sm;
+
+bad:
+	kfree(smd);
+	return ERR_PTR(r);
+}
+EXPORT_SYMBOL_GPL(dm_sm_disk_create);
+
+struct dm_space_map *dm_sm_disk_open(struct dm_transaction_manager *tm,
+				     void *root_le, size_t len)
+{
+	int r;
+	struct sm_disk *smd;
+
+	smd = kmalloc_obj(*smd);
+	if (!smd)
+		return ERR_PTR(-ENOMEM);
+
+	smd->begin = 0;
+	smd->nr_allocated_this_transaction = 0;
+	memcpy(&smd->sm, &ops, sizeof(smd->sm));
+
+	r = sm_ll_open_disk(&smd->ll, tm, root_le, len);
+	if (r)
+		goto bad;
+
+	r = sm_disk_commit(&smd->sm);
+	if (r)
+		goto bad;
+
+	return &smd->sm;
+
+bad:
+	kfree(smd);
+	return ERR_PTR(r);
+}
+EXPORT_SYMBOL_GPL(dm_sm_disk_open);
+
+/*----------------------------------------------------------------*/

linux-7.2.3/drivers/md/persistent-data/dm-space-map-disk.h

@@ -0,0 +1,26 @@
+/* SPDX-License-Identifier: GPL-2.0-only */
+/*
+ * Copyright (C) 2011 Red Hat, Inc.
+ *
+ * This file is released under the GPL.
+ */
+
+#ifndef _LINUX_DM_SPACE_MAP_DISK_H
+#define _LINUX_DM_SPACE_MAP_DISK_H
+
+#include "dm-block-manager.h"
+
+struct dm_space_map;
+struct dm_transaction_manager;
+
+/*
+ * Unfortunately we have to use two-phase construction due to the cycle
+ * between the tm and sm.
+ */
+struct dm_space_map *dm_sm_disk_create(struct dm_transaction_manager *tm,
+				       dm_block_t nr_blocks);
+
+struct dm_space_map *dm_sm_disk_open(struct dm_transaction_manager *tm,
+				     void *root, size_t len);
+
+#endif /* _LINUX_DM_SPACE_MAP_DISK_H */

linux-7.2.3/drivers/md/persistent-data/dm-space-map-metadata.c

@@ -0,0 +1,846 @@
+// SPDX-License-Identifier: GPL-2.0-only
+/*
+ * Copyright (C) 2011 Red Hat, Inc.
+ *
+ * This file is released under the GPL.
+ */
+
+#include "dm-space-map.h"
+#include "dm-space-map-common.h"
+#include "dm-space-map-metadata.h"
+
+#include <linux/list.h>
+#include <linux/slab.h>
+#include <linux/device-mapper.h>
+#include <linux/kernel.h>
+
+#define DM_MSG_PREFIX "space map metadata"
+
+/*----------------------------------------------------------------*/
+
+/*
+ * An edge triggered threshold.
+ */
+struct threshold {
+	bool threshold_set;
+	bool value_set;
+	dm_block_t threshold;
+	dm_block_t current_value;
+	dm_sm_threshold_fn fn;
+	void *context;
+};
+
+static void threshold_init(struct threshold *t)
+{
+	t->threshold_set = false;
+	t->value_set = false;
+}
+
+static void set_threshold(struct threshold *t, dm_block_t value,
+			  dm_sm_threshold_fn fn, void *context)
+{
+	t->threshold_set = true;
+	t->threshold = value;
+	t->fn = fn;
+	t->context = context;
+}
+
+static bool below_threshold(struct threshold *t, dm_block_t value)
+{
+	return t->threshold_set && value <= t->threshold;
+}
+
+static bool threshold_already_triggered(struct threshold *t)
+{
+	return t->value_set && below_threshold(t, t->current_value);
+}
+
+static void check_threshold(struct threshold *t, dm_block_t value)
+{
+	if (below_threshold(t, value) &&
+	    !threshold_already_triggered(t))
+		t->fn(t->context);
+
+	t->value_set = true;
+	t->current_value = value;
+}
+
+/*----------------------------------------------------------------*/
+
+/*
+ * Space map interface.
+ *
+ * The low level disk format is written using the standard btree and
+ * transaction manager.  This means that performing disk operations may
+ * cause us to recurse into the space map in order to allocate new blocks.
+ * For this reason we have a pool of pre-allocated blocks large enough to
+ * service any metadata_ll_disk operation.
+ */
+
+/*
+ * FIXME: we should calculate this based on the size of the device.
+ * Only the metadata space map needs this functionality.
+ */
+#define MAX_RECURSIVE_ALLOCATIONS 1024
+
+enum block_op_type {
+	BOP_INC,
+	BOP_DEC
+};
+
+struct block_op {
+	enum block_op_type type;
+	dm_block_t b;
+	dm_block_t e;
+};
+
+struct bop_ring_buffer {
+	unsigned int begin;
+	unsigned int end;
+	struct block_op bops[MAX_RECURSIVE_ALLOCATIONS + 1];
+};
+
+static void brb_init(struct bop_ring_buffer *brb)
+{
+	brb->begin = 0;
+	brb->end = 0;
+}
+
+static bool brb_empty(struct bop_ring_buffer *brb)
+{
+	return brb->begin == brb->end;
+}
+
+static unsigned int brb_next(struct bop_ring_buffer *brb, unsigned int old)
+{
+	unsigned int r = old + 1;
+
+	return r >= ARRAY_SIZE(brb->bops) ? 0 : r;
+}
+
+static int brb_push(struct bop_ring_buffer *brb,
+		    enum block_op_type type, dm_block_t b, dm_block_t e)
+{
+	struct block_op *bop;
+	unsigned int next = brb_next(brb, brb->end);
+
+	/*
+	 * We don't allow the last bop to be filled, this way we can
+	 * differentiate between full and empty.
+	 */
+	if (next == brb->begin)
+		return -ENOMEM;
+
+	bop = brb->bops + brb->end;
+	bop->type = type;
+	bop->b = b;
+	bop->e = e;
+
+	brb->end = next;
+
+	return 0;
+}
+
+static int brb_peek(struct bop_ring_buffer *brb, struct block_op *result)
+{
+	struct block_op *bop;
+
+	if (brb_empty(brb))
+		return -ENODATA;
+
+	bop = brb->bops + brb->begin;
+	memcpy(result, bop, sizeof(*result));
+	return 0;
+}
+
+static int brb_pop(struct bop_ring_buffer *brb)
+{
+	if (brb_empty(brb))
+		return -ENODATA;
+
+	brb->begin = brb_next(brb, brb->begin);
+
+	return 0;
+}
+
+/*----------------------------------------------------------------*/
+
+struct sm_metadata {
+	struct dm_space_map sm;
+
+	struct ll_disk ll;
+	struct ll_disk old_ll;
+
+	dm_block_t begin;
+
+	unsigned int recursion_count;
+	unsigned int allocated_this_transaction;
+	struct bop_ring_buffer uncommitted;
+
+	struct threshold threshold;
+};
+
+static int add_bop(struct sm_metadata *smm, enum block_op_type type, dm_block_t b, dm_block_t e)
+{
+	int r = brb_push(&smm->uncommitted, type, b, e);
+
+	if (r) {
+		DMERR("too many recursive allocations");
+		return -ENOMEM;
+	}
+
+	return 0;
+}
+
+static int commit_bop(struct sm_metadata *smm, struct block_op *op)
+{
+	int r = 0;
+	int32_t nr_allocations;
+
+	switch (op->type) {
+	case BOP_INC:
+		r = sm_ll_inc(&smm->ll, op->b, op->e, &nr_allocations);
+		break;
+
+	case BOP_DEC:
+		r = sm_ll_dec(&smm->ll, op->b, op->e, &nr_allocations);
+		break;
+	}
+
+	return r;
+}
+
+static void in(struct sm_metadata *smm)
+{
+	smm->recursion_count++;
+}
+
+static int apply_bops(struct sm_metadata *smm)
+{
+	int r = 0;
+
+	while (!brb_empty(&smm->uncommitted)) {
+		struct block_op bop;
+
+		r = brb_peek(&smm->uncommitted, &bop);
+		if (r) {
+			DMERR("bug in bop ring buffer");
+			break;
+		}
+
+		r = commit_bop(smm, &bop);
+		if (r)
+			break;
+
+		brb_pop(&smm->uncommitted);
+	}
+
+	return r;
+}
+
+static int out(struct sm_metadata *smm)
+{
+	int r = 0;
+
+	/*
+	 * If we're not recursing then very bad things are happening.
+	 */
+	if (!smm->recursion_count) {
+		DMERR("lost track of recursion depth");
+		return -ENOMEM;
+	}
+
+	if (smm->recursion_count == 1)
+		r = apply_bops(smm);
+
+	smm->recursion_count--;
+
+	return r;
+}
+
+/*
+ * When using the out() function above, we often want to combine an error
+ * code for the operation run in the recursive context with that from
+ * out().
+ */
+static int combine_errors(int r1, int r2)
+{
+	return r1 ? r1 : r2;
+}
+
+static int recursing(struct sm_metadata *smm)
+{
+	return smm->recursion_count;
+}
+
+static void sm_metadata_destroy(struct dm_space_map *sm)
+{
+	struct sm_metadata *smm = container_of(sm, struct sm_metadata, sm);
+
+	kvfree(smm);
+}
+
+static int sm_metadata_get_nr_blocks(struct dm_space_map *sm, dm_block_t *count)
+{
+	struct sm_metadata *smm = container_of(sm, struct sm_metadata, sm);
+
+	*count = smm->ll.nr_blocks;
+
+	return 0;
+}
+
+static int sm_metadata_get_nr_free(struct dm_space_map *sm, dm_block_t *count)
+{
+	struct sm_metadata *smm = container_of(sm, struct sm_metadata, sm);
+
+	*count = smm->old_ll.nr_blocks - smm->old_ll.nr_allocated -
+		 smm->allocated_this_transaction;
+
+	return 0;
+}
+
+static int sm_metadata_get_count(struct dm_space_map *sm, dm_block_t b,
+				 uint32_t *result)
+{
+	int r;
+	unsigned int i;
+	struct sm_metadata *smm = container_of(sm, struct sm_metadata, sm);
+	unsigned int adjustment = 0;
+
+	/*
+	 * We may have some uncommitted adjustments to add.  This list
+	 * should always be really short.
+	 */
+	for (i = smm->uncommitted.begin;
+	     i != smm->uncommitted.end;
+	     i = brb_next(&smm->uncommitted, i)) {
+		struct block_op *op = smm->uncommitted.bops + i;
+
+		if (b < op->b || b >= op->e)
+			continue;
+
+		switch (op->type) {
+		case BOP_INC:
+			adjustment++;
+			break;
+
+		case BOP_DEC:
+			adjustment--;
+			break;
+		}
+	}
+
+	r = sm_ll_lookup(&smm->ll, b, result);
+	if (r)
+		return r;
+
+	*result += adjustment;
+
+	return 0;
+}
+
+static int sm_metadata_count_is_more_than_one(struct dm_space_map *sm,
+					      dm_block_t b, int *result)
+{
+	int r, adjustment = 0;
+	unsigned int i;
+	struct sm_metadata *smm = container_of(sm, struct sm_metadata, sm);
+	uint32_t rc;
+
+	/*
+	 * We may have some uncommitted adjustments to add.  This list
+	 * should always be really short.
+	 */
+	for (i = smm->uncommitted.begin;
+	     i != smm->uncommitted.end;
+	     i = brb_next(&smm->uncommitted, i)) {
+
+		struct block_op *op = smm->uncommitted.bops + i;
+
+		if (b < op->b || b >= op->e)
+			continue;
+
+		switch (op->type) {
+		case BOP_INC:
+			adjustment++;
+			break;
+
+		case BOP_DEC:
+			adjustment--;
+			break;
+		}
+	}
+
+	if (adjustment > 1) {
+		*result = 1;
+		return 0;
+	}
+
+	r = sm_ll_lookup_bitmap(&smm->ll, b, &rc);
+	if (r)
+		return r;
+
+	if (rc == 3)
+		/*
+		 * We err on the side of caution, and always return true.
+		 */
+		*result = 1;
+	else
+		*result = rc + adjustment > 1;
+
+	return 0;
+}
+
+static int sm_metadata_set_count(struct dm_space_map *sm, dm_block_t b,
+				 uint32_t count)
+{
+	int r, r2;
+	int32_t nr_allocations;
+	struct sm_metadata *smm = container_of(sm, struct sm_metadata, sm);
+
+	if (smm->recursion_count) {
+		DMERR("cannot recurse set_count()");
+		return -EINVAL;
+	}
+
+	in(smm);
+	r = sm_ll_insert(&smm->ll, b, count, &nr_allocations);
+	r2 = out(smm);
+
+	return combine_errors(r, r2);
+}
+
+static int sm_metadata_inc_blocks(struct dm_space_map *sm, dm_block_t b, dm_block_t e)
+{
+	int r, r2 = 0;
+	int32_t nr_allocations;
+	struct sm_metadata *smm = container_of(sm, struct sm_metadata, sm);
+
+	if (recursing(smm)) {
+		r = add_bop(smm, BOP_INC, b, e);
+		if (r)
+			return r;
+	} else {
+		in(smm);
+		r = sm_ll_inc(&smm->ll, b, e, &nr_allocations);
+		r2 = out(smm);
+	}
+
+	return combine_errors(r, r2);
+}
+
+static int sm_metadata_dec_blocks(struct dm_space_map *sm, dm_block_t b, dm_block_t e)
+{
+	int r, r2 = 0;
+	int32_t nr_allocations;
+	struct sm_metadata *smm = container_of(sm, struct sm_metadata, sm);
+
+	if (recursing(smm))
+		r = add_bop(smm, BOP_DEC, b, e);
+	else {
+		in(smm);
+		r = sm_ll_dec(&smm->ll, b, e, &nr_allocations);
+		r2 = out(smm);
+	}
+
+	return combine_errors(r, r2);
+}
+
+static int sm_metadata_new_block_(struct dm_space_map *sm, dm_block_t *b)
+{
+	int r, r2 = 0;
+	int32_t nr_allocations;
+	struct sm_metadata *smm = container_of(sm, struct sm_metadata, sm);
+
+	/*
+	 * Any block we allocate has to be free in both the old and current ll.
+	 */
+	r = sm_ll_find_common_free_block(&smm->old_ll, &smm->ll, smm->begin, smm->ll.nr_blocks, b);
+	if (r == -ENOSPC) {
+		/*
+		 * There's no free block between smm->begin and the end of the metadata device.
+		 * We search before smm->begin in case something has been freed.
+		 */
+		r = sm_ll_find_common_free_block(&smm->old_ll, &smm->ll, 0, smm->begin, b);
+	}
+
+	if (r)
+		return r;
+
+	smm->begin = *b + 1;
+
+	if (recursing(smm))
+		r = add_bop(smm, BOP_INC, *b, *b + 1);
+	else {
+		in(smm);
+		r = sm_ll_inc(&smm->ll, *b, *b + 1, &nr_allocations);
+		r2 = out(smm);
+	}
+
+	if (!r)
+		smm->allocated_this_transaction++;
+
+	return combine_errors(r, r2);
+}
+
+static int sm_metadata_new_block(struct dm_space_map *sm, dm_block_t *b)
+{
+	dm_block_t count;
+	struct sm_metadata *smm = container_of(sm, struct sm_metadata, sm);
+
+	int r = sm_metadata_new_block_(sm, b);
+
+	if (r) {
+		DMERR_LIMIT("unable to allocate new metadata block");
+		return r;
+	}
+
+	r = sm_metadata_get_nr_free(sm, &count);
+	if (r) {
+		DMERR_LIMIT("couldn't get free block count");
+		return r;
+	}
+
+	check_threshold(&smm->threshold, count);
+
+	return r;
+}
+
+static int sm_metadata_commit(struct dm_space_map *sm)
+{
+	int r;
+	struct sm_metadata *smm = container_of(sm, struct sm_metadata, sm);
+
+	r = sm_ll_commit(&smm->ll);
+	if (r)
+		return r;
+
+	memcpy(&smm->old_ll, &smm->ll, sizeof(smm->old_ll));
+	smm->allocated_this_transaction = 0;
+
+	return 0;
+}
+
+static int sm_metadata_register_threshold_callback(struct dm_space_map *sm,
+						   dm_block_t threshold,
+						   dm_sm_threshold_fn fn,
+						   void *context)
+{
+	struct sm_metadata *smm = container_of(sm, struct sm_metadata, sm);
+
+	set_threshold(&smm->threshold, threshold, fn, context);
+
+	return 0;
+}
+
+static int sm_metadata_root_size(struct dm_space_map *sm, size_t *result)
+{
+	*result = sizeof(struct disk_sm_root);
+
+	return 0;
+}
+
+static int sm_metadata_copy_root(struct dm_space_map *sm, void *where_le, size_t max)
+{
+	struct sm_metadata *smm = container_of(sm, struct sm_metadata, sm);
+	struct disk_sm_root root_le;
+
+	root_le.nr_blocks = cpu_to_le64(smm->ll.nr_blocks);
+	root_le.nr_allocated = cpu_to_le64(smm->ll.nr_allocated);
+	root_le.bitmap_root = cpu_to_le64(smm->ll.bitmap_root);
+	root_le.ref_count_root = cpu_to_le64(smm->ll.ref_count_root);
+
+	if (max < sizeof(root_le))
+		return -ENOSPC;
+
+	memcpy(where_le, &root_le, sizeof(root_le));
+
+	return 0;
+}
+
+static int sm_metadata_extend(struct dm_space_map *sm, dm_block_t extra_blocks);
+
+static const struct dm_space_map ops = {
+	.destroy = sm_metadata_destroy,
+	.extend = sm_metadata_extend,
+	.get_nr_blocks = sm_metadata_get_nr_blocks,
+	.get_nr_free = sm_metadata_get_nr_free,
+	.get_count = sm_metadata_get_count,
+	.count_is_more_than_one = sm_metadata_count_is_more_than_one,
+	.set_count = sm_metadata_set_count,
+	.inc_blocks = sm_metadata_inc_blocks,
+	.dec_blocks = sm_metadata_dec_blocks,
+	.new_block = sm_metadata_new_block,
+	.commit = sm_metadata_commit,
+	.root_size = sm_metadata_root_size,
+	.copy_root = sm_metadata_copy_root,
+	.register_threshold_callback = sm_metadata_register_threshold_callback
+};
+
+/*----------------------------------------------------------------*/
+
+/*
+ * When a new space map is created that manages its own space.  We use
+ * this tiny bootstrap allocator.
+ */
+static void sm_bootstrap_destroy(struct dm_space_map *sm)
+{
+}
+
+static int sm_bootstrap_extend(struct dm_space_map *sm, dm_block_t extra_blocks)
+{
+	DMERR("bootstrap doesn't support extend");
+
+	return -EINVAL;
+}
+
+static int sm_bootstrap_get_nr_blocks(struct dm_space_map *sm, dm_block_t *count)
+{
+	struct sm_metadata *smm = container_of(sm, struct sm_metadata, sm);
+
+	*count = smm->ll.nr_blocks;
+
+	return 0;
+}
+
+static int sm_bootstrap_get_nr_free(struct dm_space_map *sm, dm_block_t *count)
+{
+	struct sm_metadata *smm = container_of(sm, struct sm_metadata, sm);
+
+	*count = smm->ll.nr_blocks - smm->begin;
+
+	return 0;
+}
+
+static int sm_bootstrap_get_count(struct dm_space_map *sm, dm_block_t b,
+				  uint32_t *result)
+{
+	struct sm_metadata *smm = container_of(sm, struct sm_metadata, sm);
+
+	*result = (b < smm->begin) ? 1 : 0;
+
+	return 0;
+}
+
+static int sm_bootstrap_count_is_more_than_one(struct dm_space_map *sm,
+					       dm_block_t b, int *result)
+{
+	*result = 0;
+
+	return 0;
+}
+
+static int sm_bootstrap_set_count(struct dm_space_map *sm, dm_block_t b,
+				  uint32_t count)
+{
+	DMERR("bootstrap doesn't support set_count");
+
+	return -EINVAL;
+}
+
+static int sm_bootstrap_new_block(struct dm_space_map *sm, dm_block_t *b)
+{
+	struct sm_metadata *smm = container_of(sm, struct sm_metadata, sm);
+
+	/*
+	 * We know the entire device is unused.
+	 */
+	if (smm->begin == smm->ll.nr_blocks)
+		return -ENOSPC;
+
+	*b = smm->begin++;
+
+	return 0;
+}
+
+static int sm_bootstrap_inc_blocks(struct dm_space_map *sm, dm_block_t b, dm_block_t e)
+{
+	int r;
+	struct sm_metadata *smm = container_of(sm, struct sm_metadata, sm);
+
+	r = add_bop(smm, BOP_INC, b, e);
+	if (r)
+		return r;
+
+	return 0;
+}
+
+static int sm_bootstrap_dec_blocks(struct dm_space_map *sm, dm_block_t b, dm_block_t e)
+{
+	int r;
+	struct sm_metadata *smm = container_of(sm, struct sm_metadata, sm);
+
+	r = add_bop(smm, BOP_DEC, b, e);
+	if (r)
+		return r;
+
+	return 0;
+}
+
+static int sm_bootstrap_commit(struct dm_space_map *sm)
+{
+	return 0;
+}
+
+static int sm_bootstrap_root_size(struct dm_space_map *sm, size_t *result)
+{
+	DMERR("bootstrap doesn't support root_size");
+
+	return -EINVAL;
+}
+
+static int sm_bootstrap_copy_root(struct dm_space_map *sm, void *where,
+				  size_t max)
+{
+	DMERR("bootstrap doesn't support copy_root");
+
+	return -EINVAL;
+}
+
+static const struct dm_space_map bootstrap_ops = {
+	.destroy = sm_bootstrap_destroy,
+	.extend = sm_bootstrap_extend,
+	.get_nr_blocks = sm_bootstrap_get_nr_blocks,
+	.get_nr_free = sm_bootstrap_get_nr_free,
+	.get_count = sm_bootstrap_get_count,
+	.count_is_more_than_one = sm_bootstrap_count_is_more_than_one,
+	.set_count = sm_bootstrap_set_count,
+	.inc_blocks = sm_bootstrap_inc_blocks,
+	.dec_blocks = sm_bootstrap_dec_blocks,
+	.new_block = sm_bootstrap_new_block,
+	.commit = sm_bootstrap_commit,
+	.root_size = sm_bootstrap_root_size,
+	.copy_root = sm_bootstrap_copy_root,
+	.register_threshold_callback = NULL
+};
+
+/*----------------------------------------------------------------*/
+
+static int sm_metadata_extend(struct dm_space_map *sm, dm_block_t extra_blocks)
+{
+	int r;
+	struct sm_metadata *smm = container_of(sm, struct sm_metadata, sm);
+	dm_block_t old_len = smm->ll.nr_blocks;
+
+	/*
+	 * Flick into a mode where all blocks get allocated in the new area.
+	 */
+	smm->begin = old_len;
+	memcpy(sm, &bootstrap_ops, sizeof(*sm));
+
+	/*
+	 * Extend.
+	 */
+	r = sm_ll_extend(&smm->ll, extra_blocks);
+	if (r)
+		goto out;
+
+	/*
+	 * We repeatedly increment then commit until the commit doesn't
+	 * allocate any new blocks.
+	 */
+	do {
+		r = add_bop(smm, BOP_INC, old_len, smm->begin);
+		if (r)
+			goto out;
+
+		old_len = smm->begin;
+
+		r = apply_bops(smm);
+		if (r) {
+			DMERR("%s: apply_bops failed", __func__);
+			goto out;
+		}
+
+		r = sm_ll_commit(&smm->ll);
+		if (r)
+			goto out;
+
+	} while (old_len != smm->begin);
+
+out:
+	/*
+	 * Switch back to normal behaviour.
+	 */
+	memcpy(sm, &ops, sizeof(*sm));
+	return r;
+}
+
+/*----------------------------------------------------------------*/
+
+struct dm_space_map *dm_sm_metadata_init(void)
+{
+	struct sm_metadata *smm;
+
+	smm = kvmalloc_obj(*smm);
+	if (!smm)
+		return ERR_PTR(-ENOMEM);
+
+	memcpy(&smm->sm, &ops, sizeof(smm->sm));
+
+	return &smm->sm;
+}
+
+int dm_sm_metadata_create(struct dm_space_map *sm,
+			  struct dm_transaction_manager *tm,
+			  dm_block_t nr_blocks,
+			  dm_block_t superblock)
+{
+	int r;
+	struct sm_metadata *smm = container_of(sm, struct sm_metadata, sm);
+
+	smm->begin = superblock + 1;
+	smm->recursion_count = 0;
+	smm->allocated_this_transaction = 0;
+	brb_init(&smm->uncommitted);
+	threshold_init(&smm->threshold);
+
+	memcpy(&smm->sm, &bootstrap_ops, sizeof(smm->sm));
+
+	r = sm_ll_new_metadata(&smm->ll, tm);
+	if (!r) {
+		if (nr_blocks > DM_SM_METADATA_MAX_BLOCKS)
+			nr_blocks = DM_SM_METADATA_MAX_BLOCKS;
+		r = sm_ll_extend(&smm->ll, nr_blocks);
+	}
+	memcpy(&smm->sm, &ops, sizeof(smm->sm));
+	if (r)
+		return r;
+
+	/*
+	 * Now we need to update the newly created data structures with the
+	 * allocated blocks that they were built from.
+	 */
+	r = add_bop(smm, BOP_INC, superblock, smm->begin);
+	if (r)
+		return r;
+
+	r = apply_bops(smm);
+	if (r) {
+		DMERR("%s: apply_bops failed", __func__);
+		return r;
+	}
+
+	return sm_metadata_commit(sm);
+}
+
+int dm_sm_metadata_open(struct dm_space_map *sm,
+			struct dm_transaction_manager *tm,
+			void *root_le, size_t len)
+{
+	int r;
+	struct sm_metadata *smm = container_of(sm, struct sm_metadata, sm);
+
+	r = sm_ll_open_metadata(&smm->ll, tm, root_le, len);
+	if (r)
+		return r;
+
+	smm->begin = 0;
+	smm->recursion_count = 0;
+	smm->allocated_this_transaction = 0;
+	brb_init(&smm->uncommitted);
+	threshold_init(&smm->threshold);
+
+	memcpy(&smm->old_ll, &smm->ll, sizeof(smm->old_ll));
+	return 0;
+}

linux-7.2.3/drivers/md/persistent-data/dm-space-map-metadata.h

@@ -0,0 +1,45 @@
+/* SPDX-License-Identifier: GPL-2.0-only */
+/*
+ * Copyright (C) 2011 Red Hat, Inc.
+ *
+ * This file is released under the GPL.
+ */
+
+#ifndef DM_SPACE_MAP_METADATA_H
+#define DM_SPACE_MAP_METADATA_H
+
+#include "dm-transaction-manager.h"
+
+#define DM_SM_METADATA_BLOCK_SIZE (4096 >> SECTOR_SHIFT)
+
+/*
+ * The metadata device is currently limited in size.
+ *
+ * We have one block of index, which can hold 255 index entries.  Each
+ * index entry contains allocation info about ~16k metadata blocks.
+ */
+#define DM_SM_METADATA_MAX_BLOCKS (255 * ((1 << 14) - 64))
+#define DM_SM_METADATA_MAX_SECTORS (DM_SM_METADATA_MAX_BLOCKS * DM_SM_METADATA_BLOCK_SIZE)
+
+/*
+ * Unfortunately we have to use two-phase construction due to the cycle
+ * between the tm and sm.
+ */
+struct dm_space_map *dm_sm_metadata_init(void);
+
+/*
+ * Create a fresh space map.
+ */
+int dm_sm_metadata_create(struct dm_space_map *sm,
+			  struct dm_transaction_manager *tm,
+			  dm_block_t nr_blocks,
+			  dm_block_t superblock);
+
+/*
+ * Open from a previously-recorded root.
+ */
+int dm_sm_metadata_open(struct dm_space_map *sm,
+			struct dm_transaction_manager *tm,
+			void *root_le, size_t len);
+
+#endif	/* DM_SPACE_MAP_METADATA_H */

linux-7.2.3/drivers/md/persistent-data/dm-space-map.h

@@ -0,0 +1,169 @@
+/* SPDX-License-Identifier: GPL-2.0-only */
+/*
+ * Copyright (C) 2011 Red Hat, Inc.
+ *
+ * This file is released under the GPL.
+ */
+
+#ifndef _LINUX_DM_SPACE_MAP_H
+#define _LINUX_DM_SPACE_MAP_H
+
+#include "dm-block-manager.h"
+
+typedef void (*dm_sm_threshold_fn)(void *context);
+
+/*
+ * struct dm_space_map keeps a record of how many times each block in a device
+ * is referenced.  It needs to be fixed on disk as part of the transaction.
+ */
+struct dm_space_map {
+	void (*destroy)(struct dm_space_map *sm);
+
+	/*
+	 * You must commit before allocating the newly added space.
+	 */
+	int (*extend)(struct dm_space_map *sm, dm_block_t extra_blocks);
+
+	/*
+	 * Extensions do not appear in this count until after commit has
+	 * been called.
+	 */
+	int (*get_nr_blocks)(struct dm_space_map *sm, dm_block_t *count);
+
+	/*
+	 * Space maps must never allocate a block from the previous
+	 * transaction, in case we need to rollback.  This complicates the
+	 * semantics of get_nr_free(), it should return the number of blocks
+	 * that are available for allocation _now_.  For instance you may
+	 * have blocks with a zero reference count that will not be
+	 * available for allocation until after the next commit.
+	 */
+	int (*get_nr_free)(struct dm_space_map *sm, dm_block_t *count);
+
+	int (*get_count)(struct dm_space_map *sm, dm_block_t b, uint32_t *result);
+	int (*count_is_more_than_one)(struct dm_space_map *sm, dm_block_t b,
+				      int *result);
+	int (*set_count)(struct dm_space_map *sm, dm_block_t b, uint32_t count);
+
+	int (*commit)(struct dm_space_map *sm);
+
+	int (*inc_blocks)(struct dm_space_map *sm, dm_block_t b, dm_block_t e);
+	int (*dec_blocks)(struct dm_space_map *sm, dm_block_t b, dm_block_t e);
+
+	/*
+	 * new_block will increment the returned block.
+	 */
+	int (*new_block)(struct dm_space_map *sm, dm_block_t *b);
+
+	/*
+	 * The root contains all the information needed to fix the space map.
+	 * Generally this info is small, so squirrel it away in a disk block
+	 * along with other info.
+	 */
+	int (*root_size)(struct dm_space_map *sm, size_t *result);
+	int (*copy_root)(struct dm_space_map *sm, void *copy_to_here_le, size_t len);
+
+	/*
+	 * You can register one threshold callback which is edge-triggered
+	 * when the free space in the space map drops below the threshold.
+	 */
+	int (*register_threshold_callback)(struct dm_space_map *sm,
+					   dm_block_t threshold,
+					   dm_sm_threshold_fn fn,
+					   void *context);
+};
+
+/*----------------------------------------------------------------*/
+
+static inline void dm_sm_destroy(struct dm_space_map *sm)
+{
+	if (sm)
+		sm->destroy(sm);
+}
+
+static inline int dm_sm_extend(struct dm_space_map *sm, dm_block_t extra_blocks)
+{
+	return sm->extend(sm, extra_blocks);
+}
+
+static inline int dm_sm_get_nr_blocks(struct dm_space_map *sm, dm_block_t *count)
+{
+	return sm->get_nr_blocks(sm, count);
+}
+
+static inline int dm_sm_get_nr_free(struct dm_space_map *sm, dm_block_t *count)
+{
+	return sm->get_nr_free(sm, count);
+}
+
+static inline int dm_sm_get_count(struct dm_space_map *sm, dm_block_t b,
+				  uint32_t *result)
+{
+	return sm->get_count(sm, b, result);
+}
+
+static inline int dm_sm_count_is_more_than_one(struct dm_space_map *sm,
+					       dm_block_t b, int *result)
+{
+	return sm->count_is_more_than_one(sm, b, result);
+}
+
+static inline int dm_sm_set_count(struct dm_space_map *sm, dm_block_t b,
+				  uint32_t count)
+{
+	return sm->set_count(sm, b, count);
+}
+
+static inline int dm_sm_commit(struct dm_space_map *sm)
+{
+	return sm->commit(sm);
+}
+
+static inline int dm_sm_inc_blocks(struct dm_space_map *sm, dm_block_t b, dm_block_t e)
+{
+	return sm->inc_blocks(sm, b, e);
+}
+
+static inline int dm_sm_inc_block(struct dm_space_map *sm, dm_block_t b)
+{
+	return dm_sm_inc_blocks(sm, b, b + 1);
+}
+
+static inline int dm_sm_dec_blocks(struct dm_space_map *sm, dm_block_t b, dm_block_t e)
+{
+	return sm->dec_blocks(sm, b, e);
+}
+
+static inline int dm_sm_dec_block(struct dm_space_map *sm, dm_block_t b)
+{
+	return dm_sm_dec_blocks(sm, b, b + 1);
+}
+
+static inline int dm_sm_new_block(struct dm_space_map *sm, dm_block_t *b)
+{
+	return sm->new_block(sm, b);
+}
+
+static inline int dm_sm_root_size(struct dm_space_map *sm, size_t *result)
+{
+	return sm->root_size(sm, result);
+}
+
+static inline int dm_sm_copy_root(struct dm_space_map *sm, void *copy_to_here_le, size_t len)
+{
+	return sm->copy_root(sm, copy_to_here_le, len);
+}
+
+static inline int dm_sm_register_threshold_callback(struct dm_space_map *sm,
+						    dm_block_t threshold,
+						    dm_sm_threshold_fn fn,
+						    void *context)
+{
+	if (sm->register_threshold_callback)
+		return sm->register_threshold_callback(sm, threshold, fn, context);
+
+	return -EINVAL;
+}
+
+
+#endif	/* _LINUX_DM_SPACE_MAP_H */

linux-7.2.3/drivers/md/persistent-data/dm-transaction-manager.c

@@ -0,0 +1,541 @@
+// SPDX-License-Identifier: GPL-2.0-only
+/*
+ * Copyright (C) 2011 Red Hat, Inc.
+ *
+ * This file is released under the GPL.
+ */
+#include "dm-transaction-manager.h"
+#include "dm-space-map.h"
+#include "dm-space-map-disk.h"
+#include "dm-space-map-metadata.h"
+#include "dm-persistent-data-internal.h"
+
+#include <linux/export.h>
+#include <linux/mutex.h>
+#include <linux/hash.h>
+#include <linux/rbtree.h>
+#include <linux/slab.h>
+#include <linux/device-mapper.h>
+
+#define DM_MSG_PREFIX "transaction manager"
+
+/*----------------------------------------------------------------*/
+
+#define PREFETCH_SIZE 128
+#define PREFETCH_BITS 7
+#define PREFETCH_SENTINEL ((dm_block_t) -1ULL)
+
+struct prefetch_set {
+	struct mutex lock;
+	dm_block_t blocks[PREFETCH_SIZE];
+};
+
+static unsigned int prefetch_hash(dm_block_t b)
+{
+	return hash_64(b, PREFETCH_BITS);
+}
+
+static void prefetch_wipe(struct prefetch_set *p)
+{
+	unsigned int i;
+
+	for (i = 0; i < PREFETCH_SIZE; i++)
+		p->blocks[i] = PREFETCH_SENTINEL;
+}
+
+static void prefetch_init(struct prefetch_set *p)
+{
+	mutex_init(&p->lock);
+	prefetch_wipe(p);
+}
+
+static void prefetch_add(struct prefetch_set *p, dm_block_t b)
+{
+	unsigned int h = prefetch_hash(b);
+
+	mutex_lock(&p->lock);
+	if (p->blocks[h] == PREFETCH_SENTINEL)
+		p->blocks[h] = b;
+
+	mutex_unlock(&p->lock);
+}
+
+static void prefetch_issue(struct prefetch_set *p, struct dm_block_manager *bm)
+{
+	unsigned int i;
+
+	mutex_lock(&p->lock);
+
+	for (i = 0; i < PREFETCH_SIZE; i++)
+		if (p->blocks[i] != PREFETCH_SENTINEL) {
+			dm_bm_prefetch(bm, p->blocks[i]);
+			p->blocks[i] = PREFETCH_SENTINEL;
+		}
+
+	mutex_unlock(&p->lock);
+}
+
+/*----------------------------------------------------------------*/
+
+struct shadow_info {
+	struct rb_node node;
+	dm_block_t where;
+};
+
+/*
+ * It would be nice if we scaled with the size of transaction.
+ */
+#define DM_HASH_SIZE 256
+#define DM_HASH_MASK (DM_HASH_SIZE - 1)
+
+struct dm_transaction_manager {
+	int is_clone;
+	struct dm_transaction_manager *real;
+
+	struct dm_block_manager *bm;
+	struct dm_space_map *sm;
+
+	spinlock_t lock;
+	struct rb_root buckets[DM_HASH_SIZE];
+
+	struct prefetch_set prefetches;
+};
+
+/*----------------------------------------------------------------*/
+
+static int is_shadow(struct dm_transaction_manager *tm, dm_block_t b)
+{
+	int r = 0;
+	unsigned int bucket = dm_hash_block(b, DM_HASH_MASK);
+	struct rb_node **node;
+
+	spin_lock(&tm->lock);
+	node = &tm->buckets[bucket].rb_node;
+	while (*node) {
+		struct shadow_info *si =
+			rb_entry(*node, struct shadow_info, node);
+		if (b == si->where) {
+			r = 1;
+			break;
+		}
+		if (b < si->where)
+			node = &si->node.rb_left;
+		else
+			node = &si->node.rb_right;
+	}
+	spin_unlock(&tm->lock);
+
+	return r;
+}
+
+/*
+ * This can silently fail if there's no memory.  We're ok with this since
+ * creating redundant shadows causes no harm.
+ */
+static void insert_shadow(struct dm_transaction_manager *tm, dm_block_t b)
+{
+	unsigned int bucket;
+	struct shadow_info *si;
+
+	si = kmalloc_obj(*si, GFP_NOIO);
+	if (si) {
+		struct rb_node **node, *parent;
+		si->where = b;
+		bucket = dm_hash_block(b, DM_HASH_MASK);
+
+		spin_lock(&tm->lock);
+		node = &tm->buckets[bucket].rb_node;
+		parent = NULL;
+		while (*node) {
+			struct shadow_info *si =
+				rb_entry(*node, struct shadow_info, node);
+			parent = *node;
+			if (b < si->where)
+				node = &si->node.rb_left;
+			else
+				node = &si->node.rb_right;
+		}
+		rb_link_node(&si->node, parent, node);
+		rb_insert_color(&si->node, &tm->buckets[bucket]);
+		spin_unlock(&tm->lock);
+	}
+}
+
+static void wipe_shadow_table(struct dm_transaction_manager *tm)
+{
+	unsigned int i;
+
+	spin_lock(&tm->lock);
+	for (i = 0; i < DM_HASH_SIZE; i++) {
+		while (!RB_EMPTY_ROOT(&tm->buckets[i])) {
+			struct shadow_info *si =
+				rb_entry(tm->buckets[i].rb_node, struct shadow_info, node);
+			rb_erase(&si->node, &tm->buckets[i]);
+			kfree(si);
+		}
+	}
+	spin_unlock(&tm->lock);
+}
+
+/*----------------------------------------------------------------*/
+
+static struct dm_transaction_manager *dm_tm_create(struct dm_block_manager *bm,
+						   struct dm_space_map *sm)
+{
+	unsigned int i;
+	struct dm_transaction_manager *tm;
+
+	tm = kmalloc_obj(*tm);
+	if (!tm)
+		return ERR_PTR(-ENOMEM);
+
+	tm->is_clone = 0;
+	tm->real = NULL;
+	tm->bm = bm;
+	tm->sm = sm;
+
+	spin_lock_init(&tm->lock);
+	for (i = 0; i < DM_HASH_SIZE; i++)
+		tm->buckets[i] = RB_ROOT;
+
+	prefetch_init(&tm->prefetches);
+
+	return tm;
+}
+
+struct dm_transaction_manager *dm_tm_create_non_blocking_clone(struct dm_transaction_manager *real)
+{
+	struct dm_transaction_manager *tm;
+
+	tm = kmalloc_obj(*tm);
+	if (tm) {
+		tm->is_clone = 1;
+		tm->real = real;
+	}
+
+	return tm;
+}
+EXPORT_SYMBOL_GPL(dm_tm_create_non_blocking_clone);
+
+void dm_tm_destroy(struct dm_transaction_manager *tm)
+{
+	if (!tm)
+		return;
+
+	if (!tm->is_clone)
+		wipe_shadow_table(tm);
+
+	kfree(tm);
+}
+EXPORT_SYMBOL_GPL(dm_tm_destroy);
+
+int dm_tm_pre_commit(struct dm_transaction_manager *tm)
+{
+	int r;
+
+	if (tm->is_clone)
+		return -EWOULDBLOCK;
+
+	r = dm_sm_commit(tm->sm);
+	if (r < 0)
+		return r;
+
+	return dm_bm_flush(tm->bm);
+}
+EXPORT_SYMBOL_GPL(dm_tm_pre_commit);
+
+int dm_tm_commit(struct dm_transaction_manager *tm, struct dm_block *root)
+{
+	if (tm->is_clone)
+		return -EWOULDBLOCK;
+
+	wipe_shadow_table(tm);
+	dm_bm_unlock(root);
+
+	return dm_bm_flush(tm->bm);
+}
+EXPORT_SYMBOL_GPL(dm_tm_commit);
+
+int dm_tm_new_block(struct dm_transaction_manager *tm,
+		    const struct dm_block_validator *v,
+		    struct dm_block **result)
+{
+	int r;
+	dm_block_t new_block;
+
+	if (tm->is_clone)
+		return -EWOULDBLOCK;
+
+	r = dm_sm_new_block(tm->sm, &new_block);
+	if (r < 0)
+		return r;
+
+	r = dm_bm_write_lock_zero(tm->bm, new_block, v, result);
+	if (r < 0) {
+		dm_sm_dec_block(tm->sm, new_block);
+		return r;
+	}
+
+	/*
+	 * New blocks count as shadows in that they don't need to be
+	 * shadowed again.
+	 */
+	insert_shadow(tm, new_block);
+
+	return 0;
+}
+
+static int __shadow_block(struct dm_transaction_manager *tm, dm_block_t orig,
+			  const struct dm_block_validator *v,
+			  struct dm_block **result)
+{
+	int r;
+	dm_block_t new;
+	struct dm_block *orig_block;
+
+	r = dm_sm_new_block(tm->sm, &new);
+	if (r < 0)
+		return r;
+
+	r = dm_sm_dec_block(tm->sm, orig);
+	if (r < 0)
+		return r;
+
+	r = dm_bm_read_lock(tm->bm, orig, v, &orig_block);
+	if (r < 0)
+		return r;
+
+	/*
+	 * It would be tempting to use dm_bm_unlock_move here, but some
+	 * code, such as the space maps, keeps using the old data structures
+	 * secure in the knowledge they won't be changed until the next
+	 * transaction.  Using unlock_move would force a synchronous read
+	 * since the old block would no longer be in the cache.
+	 */
+	r = dm_bm_write_lock_zero(tm->bm, new, v, result);
+	if (r) {
+		dm_bm_unlock(orig_block);
+		return r;
+	}
+
+	memcpy(dm_block_data(*result), dm_block_data(orig_block),
+	       dm_bm_block_size(tm->bm));
+
+	dm_bm_unlock(orig_block);
+	return r;
+}
+
+int dm_tm_shadow_block(struct dm_transaction_manager *tm, dm_block_t orig,
+		       const struct dm_block_validator *v, struct dm_block **result,
+		       int *inc_children)
+{
+	int r;
+
+	if (tm->is_clone)
+		return -EWOULDBLOCK;
+
+	r = dm_sm_count_is_more_than_one(tm->sm, orig, inc_children);
+	if (r < 0)
+		return r;
+
+	if (is_shadow(tm, orig) && !*inc_children)
+		return dm_bm_write_lock(tm->bm, orig, v, result);
+
+	r = __shadow_block(tm, orig, v, result);
+	if (r < 0)
+		return r;
+	insert_shadow(tm, dm_block_location(*result));
+
+	return r;
+}
+EXPORT_SYMBOL_GPL(dm_tm_shadow_block);
+
+int dm_tm_read_lock(struct dm_transaction_manager *tm, dm_block_t b,
+		    const struct dm_block_validator *v,
+		    struct dm_block **blk)
+{
+	if (tm->is_clone) {
+		int r = dm_bm_read_try_lock(tm->real->bm, b, v, blk);
+
+		if (r == -EWOULDBLOCK)
+			prefetch_add(&tm->real->prefetches, b);
+
+		return r;
+	}
+
+	return dm_bm_read_lock(tm->bm, b, v, blk);
+}
+EXPORT_SYMBOL_GPL(dm_tm_read_lock);
+
+void dm_tm_unlock(struct dm_transaction_manager *tm, struct dm_block *b)
+{
+	dm_bm_unlock(b);
+}
+EXPORT_SYMBOL_GPL(dm_tm_unlock);
+
+void dm_tm_inc(struct dm_transaction_manager *tm, dm_block_t b)
+{
+	/*
+	 * The non-blocking clone doesn't support this.
+	 */
+	BUG_ON(tm->is_clone);
+
+	dm_sm_inc_block(tm->sm, b);
+}
+EXPORT_SYMBOL_GPL(dm_tm_inc);
+
+void dm_tm_inc_range(struct dm_transaction_manager *tm, dm_block_t b, dm_block_t e)
+{
+	/*
+	 * The non-blocking clone doesn't support this.
+	 */
+	BUG_ON(tm->is_clone);
+
+	dm_sm_inc_blocks(tm->sm, b, e);
+}
+EXPORT_SYMBOL_GPL(dm_tm_inc_range);
+
+void dm_tm_dec(struct dm_transaction_manager *tm, dm_block_t b)
+{
+	/*
+	 * The non-blocking clone doesn't support this.
+	 */
+	BUG_ON(tm->is_clone);
+
+	dm_sm_dec_block(tm->sm, b);
+}
+EXPORT_SYMBOL_GPL(dm_tm_dec);
+
+void dm_tm_dec_range(struct dm_transaction_manager *tm, dm_block_t b, dm_block_t e)
+{
+	/*
+	 * The non-blocking clone doesn't support this.
+	 */
+	BUG_ON(tm->is_clone);
+
+	dm_sm_dec_blocks(tm->sm, b, e);
+}
+EXPORT_SYMBOL_GPL(dm_tm_dec_range);
+
+void dm_tm_with_runs(struct dm_transaction_manager *tm,
+		     const __le64 *value_le, unsigned int count, dm_tm_run_fn fn)
+{
+	uint64_t b, begin, end;
+	bool in_run = false;
+	unsigned int i;
+
+	for (i = 0; i < count; i++, value_le++) {
+		b = le64_to_cpu(*value_le);
+
+		if (in_run) {
+			if (b == end)
+				end++;
+			else {
+				fn(tm, begin, end);
+				begin = b;
+				end = b + 1;
+			}
+		} else {
+			in_run = true;
+			begin = b;
+			end = b + 1;
+		}
+	}
+
+	if (in_run)
+		fn(tm, begin, end);
+}
+EXPORT_SYMBOL_GPL(dm_tm_with_runs);
+
+int dm_tm_ref(struct dm_transaction_manager *tm, dm_block_t b,
+	      uint32_t *result)
+{
+	if (tm->is_clone)
+		return -EWOULDBLOCK;
+
+	return dm_sm_get_count(tm->sm, b, result);
+}
+
+int dm_tm_block_is_shared(struct dm_transaction_manager *tm, dm_block_t b,
+			  int *result)
+{
+	if (tm->is_clone)
+		return -EWOULDBLOCK;
+
+	return dm_sm_count_is_more_than_one(tm->sm, b, result);
+}
+
+struct dm_block_manager *dm_tm_get_bm(struct dm_transaction_manager *tm)
+{
+	return tm->bm;
+}
+
+void dm_tm_issue_prefetches(struct dm_transaction_manager *tm)
+{
+	prefetch_issue(&tm->prefetches, tm->bm);
+}
+EXPORT_SYMBOL_GPL(dm_tm_issue_prefetches);
+
+/*----------------------------------------------------------------*/
+
+static int dm_tm_create_internal(struct dm_block_manager *bm,
+				 dm_block_t sb_location,
+				 struct dm_transaction_manager **tm,
+				 struct dm_space_map **sm,
+				 int create,
+				 void *sm_root, size_t sm_len)
+{
+	int r;
+
+	*sm = dm_sm_metadata_init();
+	if (IS_ERR(*sm))
+		return PTR_ERR(*sm);
+
+	*tm = dm_tm_create(bm, *sm);
+	if (IS_ERR(*tm)) {
+		dm_sm_destroy(*sm);
+		return PTR_ERR(*tm);
+	}
+
+	if (create) {
+		r = dm_sm_metadata_create(*sm, *tm, dm_bm_nr_blocks(bm),
+					  sb_location);
+		if (r) {
+			DMERR("couldn't create metadata space map");
+			goto bad;
+		}
+
+	} else {
+		r = dm_sm_metadata_open(*sm, *tm, sm_root, sm_len);
+		if (r) {
+			DMERR("couldn't open metadata space map");
+			goto bad;
+		}
+	}
+
+	return 0;
+
+bad:
+	dm_tm_destroy(*tm);
+	dm_sm_destroy(*sm);
+	return r;
+}
+
+int dm_tm_create_with_sm(struct dm_block_manager *bm, dm_block_t sb_location,
+			 struct dm_transaction_manager **tm,
+			 struct dm_space_map **sm)
+{
+	return dm_tm_create_internal(bm, sb_location, tm, sm, 1, NULL, 0);
+}
+EXPORT_SYMBOL_GPL(dm_tm_create_with_sm);
+
+int dm_tm_open_with_sm(struct dm_block_manager *bm, dm_block_t sb_location,
+		       void *sm_root, size_t root_len,
+		       struct dm_transaction_manager **tm,
+		       struct dm_space_map **sm)
+{
+	return dm_tm_create_internal(bm, sb_location, tm, sm, 0, sm_root, root_len);
+}
+EXPORT_SYMBOL_GPL(dm_tm_open_with_sm);
+
+/*----------------------------------------------------------------*/

linux-7.2.3/drivers/md/persistent-data/dm-transaction-manager.h

@@ -0,0 +1,154 @@
+/* SPDX-License-Identifier: GPL-2.0-only */
+/*
+ * Copyright (C) 2011 Red Hat, Inc.
+ *
+ * This file is released under the GPL.
+ */
+
+#ifndef _LINUX_DM_TRANSACTION_MANAGER_H
+#define _LINUX_DM_TRANSACTION_MANAGER_H
+
+#include "dm-block-manager.h"
+
+struct dm_transaction_manager;
+struct dm_space_map;
+
+/*----------------------------------------------------------------*/
+
+/*
+ * This manages the scope of a transaction.  It also enforces immutability
+ * of the on-disk data structures by limiting access to writeable blocks.
+ *
+ * Clients should not fiddle with the block manager directly.
+ */
+
+void dm_tm_destroy(struct dm_transaction_manager *tm);
+
+/*
+ * The non-blocking version of a transaction manager is intended for use in
+ * fast path code that needs to do lookups e.g. a dm mapping function.
+ * You create the non-blocking variant from a normal tm.  The interface is
+ * the same, except that most functions will just return -EWOULDBLOCK.
+ * Methods that return void yet may block should not be called on a clone
+ * viz. dm_tm_inc, dm_tm_dec.  Call dm_tm_destroy() as you would with a normal
+ * tm when you've finished with it.  You may not destroy the original prior
+ * to clones.
+ */
+struct dm_transaction_manager *dm_tm_create_non_blocking_clone(struct dm_transaction_manager *real);
+
+/*
+ * We use a 2-phase commit here.
+ *
+ * i) Make all changes for the transaction *except* for the superblock.
+ * Then call dm_tm_pre_commit() to flush them to disk.
+ *
+ * ii) Lock your superblock.  Update.  Then call dm_tm_commit() which will
+ * unlock the superblock and flush it.  No other blocks should be updated
+ * during this period.  Care should be taken to never unlock a partially
+ * updated superblock; perform any operations that could fail *before* you
+ * take the superblock lock.
+ */
+int dm_tm_pre_commit(struct dm_transaction_manager *tm);
+int dm_tm_commit(struct dm_transaction_manager *tm, struct dm_block *superblock);
+
+/*
+ * These methods are the only way to get hold of a writeable block.
+ */
+
+/*
+ * dm_tm_new_block() is pretty self-explanatory.  Make sure you do actually
+ * write to the whole of @data before you unlock, otherwise you could get
+ * a data leak.  (The other option is for tm_new_block() to zero new blocks
+ * before handing them out, which will be redundant in most, if not all,
+ * cases).
+ * Zeroes the new block and returns with write lock held.
+ */
+int dm_tm_new_block(struct dm_transaction_manager *tm,
+		    const struct dm_block_validator *v,
+		    struct dm_block **result);
+
+/*
+ * dm_tm_shadow_block() allocates a new block and copies the data from @orig
+ * to it.  It then decrements the reference count on original block.  Use
+ * this to update the contents of a block in a data structure, don't
+ * confuse this with a clone - you shouldn't access the orig block after
+ * this operation.  Because the tm knows the scope of the transaction it
+ * can optimise requests for a shadow of a shadow to a no-op.  Don't forget
+ * to unlock when you've finished with the shadow.
+ *
+ * The @inc_children flag is used to tell the caller whether it needs to
+ * adjust reference counts for children.  (Data in the block may refer to
+ * other blocks.)
+ *
+ * Shadowing implicitly drops a reference on @orig so you must not have
+ * it locked when you call this.
+ */
+int dm_tm_shadow_block(struct dm_transaction_manager *tm, dm_block_t orig,
+		       const struct dm_block_validator *v,
+		       struct dm_block **result, int *inc_children);
+
+/*
+ * Read access.  You can lock any block you want.  If there's a write lock
+ * on it outstanding then it'll block.
+ */
+int dm_tm_read_lock(struct dm_transaction_manager *tm, dm_block_t b,
+		    const struct dm_block_validator *v,
+		    struct dm_block **result);
+
+void dm_tm_unlock(struct dm_transaction_manager *tm, struct dm_block *b);
+
+/*
+ * Functions for altering the reference count of a block directly.
+ */
+void dm_tm_inc(struct dm_transaction_manager *tm, dm_block_t b);
+void dm_tm_inc_range(struct dm_transaction_manager *tm, dm_block_t b, dm_block_t e);
+void dm_tm_dec(struct dm_transaction_manager *tm, dm_block_t b);
+void dm_tm_dec_range(struct dm_transaction_manager *tm, dm_block_t b, dm_block_t e);
+
+/*
+ * Builds up runs of adjacent blocks, and then calls the given fn
+ * (typically dm_tm_inc/dec).  Very useful when you have to perform
+ * the same tm operation on all values in a btree leaf.
+ */
+typedef void (*dm_tm_run_fn)(struct dm_transaction_manager *, dm_block_t, dm_block_t);
+void dm_tm_with_runs(struct dm_transaction_manager *tm,
+		     const __le64 *value_le, unsigned int count, dm_tm_run_fn fn);
+
+int dm_tm_ref(struct dm_transaction_manager *tm, dm_block_t b, uint32_t *result);
+
+/*
+ * Finds out if a given block is shared (ie. has a reference count higher
+ * than one).
+ */
+int dm_tm_block_is_shared(struct dm_transaction_manager *tm, dm_block_t b,
+			  int *result);
+
+struct dm_block_manager *dm_tm_get_bm(struct dm_transaction_manager *tm);
+
+/*
+ * If you're using a non-blocking clone the tm will build up a list of
+ * requested blocks that weren't in core.  This call will request those
+ * blocks to be prefetched.
+ */
+void dm_tm_issue_prefetches(struct dm_transaction_manager *tm);
+
+/*
+ * A little utility that ties the knot by producing a transaction manager
+ * that has a space map managed by the transaction manager...
+ *
+ * Returns a tm that has an open transaction to write the new disk sm.
+ * Caller should store the new sm root and commit.
+ *
+ * The superblock location is passed so the metadata space map knows it
+ * shouldn't be used.
+ */
+int dm_tm_create_with_sm(struct dm_block_manager *bm, dm_block_t sb_location,
+			 struct dm_transaction_manager **tm,
+			 struct dm_space_map **sm);
+
+int dm_tm_open_with_sm(struct dm_block_manager *bm, dm_block_t sb_location,
+		       void *sm_root, size_t root_len,
+		       struct dm_transaction_manager **tm,
+		       struct dm_space_map **sm);
+
+#endif	/* _LINUX_DM_TRANSACTION_MANAGER_H */