1 files changed, 673 insertions, 475 deletions

diff --git a/drivers/md/bcache/bset.c b/drivers/md/bcache/bset.c
index 1d27d3af3251..463eb13bd0b2 100644
--- a/drivers/md/bcache/bset.c
+++ b/drivers/md/bcache/bset.c
@@ -1,3 +1,4 @@
+// SPDX-License-Identifier: GPL-2.0
 /*
  * Code for working with individual keys, and sorted sets of keys with in a
  * btree node
@@ -5,175 +6,183 @@
  * Copyright 2012 Google, Inc.
  */
 
-#include "bcache.h"
-#include "btree.h"
-#include "debug.h"
+#define pr_fmt(fmt) "bcache: %s() " fmt, __func__
 
+#include "util.h"
+#include "bset.h"
+
+#include <linux/console.h>
+#include <linux/sched/clock.h>
 #include <linux/random.h>
 #include <linux/prefetch.h>
 
-/* Keylists */
+#ifdef CONFIG_BCACHE_DEBUG
 
-void bch_keylist_copy(struct keylist *dest, struct keylist *src)
+void bch_dump_bset(struct btree_keys *b, struct bset *i, unsigned int set)
 {
-	*dest = *src;
+	struct bkey *k, *next;
+
+	for (k = i->start; k < bset_bkey_last(i); k = next) {
+		next = bkey_next(k);
+
+		pr_err("block %u key %u/%u: ", set,
+		       (unsigned int) ((u64 *) k - i->d), i->keys);
 
-	if (src->list == src->d) {
-		size_t n = (uint64_t *) src->top - src->d;
-		dest->top = (struct bkey *) &dest->d[n];
-		dest->list = dest->d;
+		if (b->ops->key_dump)
+			b->ops->key_dump(b, k);
+		else
+			pr_cont("%llu:%llu\n", KEY_INODE(k), KEY_OFFSET(k));
+
+		if (next < bset_bkey_last(i) &&
+		    bkey_cmp(k, b->ops->is_extents ?
+			     &START_KEY(next) : next) > 0)
+			pr_err("Key skipped backwards\n");
 	}
 }
 
-int bch_keylist_realloc(struct keylist *l, int nptrs, struct cache_set *c)
+void bch_dump_bucket(struct btree_keys *b)
 {
-	unsigned oldsize = (uint64_t *) l->top - l->list;
-	unsigned newsize = oldsize + 2 + nptrs;
-	uint64_t *new;
-
-	/* The journalling code doesn't handle the case where the keys to insert
-	 * is bigger than an empty write: If we just return -ENOMEM here,
-	 * bio_insert() and bio_invalidate() will insert the keys created so far
-	 * and finish the rest when the keylist is empty.
-	 */
-	if (newsize * sizeof(uint64_t) > block_bytes(c) - sizeof(struct jset))
-		return -ENOMEM;
+	unsigned int i;
 
-	newsize = roundup_pow_of_two(newsize);
+	console_lock();
+	for (i = 0; i <= b->nsets; i++)
+		bch_dump_bset(b, b->set[i].data,
+			      bset_sector_offset(b, b->set[i].data));
+	console_unlock();
+}
 
-	if (newsize <= KEYLIST_INLINE ||
-	    roundup_pow_of_two(oldsize) == newsize)
-		return 0;
+int __bch_count_data(struct btree_keys *b)
+{
+	unsigned int ret = 0;
+	struct btree_iter_stack iter;
+	struct bkey *k;
+
+	if (b->ops->is_extents)
+		for_each_key(b, k, &iter)
+			ret += KEY_SIZE(k);
+	return ret;
+}
 
-	new = krealloc(l->list == l->d ? NULL : l->list,
-		       sizeof(uint64_t) * newsize, GFP_NOIO);
+void __bch_check_keys(struct btree_keys *b, const char *fmt, ...)
+{
+	va_list args;
+	struct bkey *k, *p = NULL;
+	struct btree_iter_stack iter;
+	const char *err;
+
+	for_each_key(b, k, &iter) {
+		if (b->ops->is_extents) {
+			err = "Keys out of order";
+			if (p && bkey_cmp(&START_KEY(p), &START_KEY(k)) > 0)
+				goto bug;
+
+			if (bch_ptr_invalid(b, k))
+				continue;
 
-	if (!new)
-		return -ENOMEM;
+			err =  "Overlapping keys";
+			if (p && bkey_cmp(p, &START_KEY(k)) > 0)
+				goto bug;
+		} else {
+			if (bch_ptr_bad(b, k))
+				continue;
 
-	if (l->list == l->d)
-		memcpy(new, l->list, sizeof(uint64_t) * KEYLIST_INLINE);
+			err = "Duplicate keys";
+			if (p && !bkey_cmp(p, k))
+				goto bug;
+		}
+		p = k;
+	}
+#if 0
+	err = "Key larger than btree node key";
+	if (p && bkey_cmp(p, &b->key) > 0)
+		goto bug;
+#endif
+	return;
+bug:
+	bch_dump_bucket(b);
 
-	l->list = new;
-	l->top = (struct bkey *) (&l->list[oldsize]);
+	va_start(args, fmt);
+	vprintk(fmt, args);
+	va_end(args);
 
-	return 0;
+	panic("bch_check_keys error:  %s:\n", err);
 }
 
-struct bkey *bch_keylist_pop(struct keylist *l)
+static void bch_btree_iter_next_check(struct btree_iter *iter)
 {
-	struct bkey *k = l->bottom;
+	struct bkey *k = iter->data->k, *next = bkey_next(k);
 
-	if (k == l->top)
-		return NULL;
-
-	while (bkey_next(k) != l->top)
-		k = bkey_next(k);
-
-	return l->top = k;
+	if (next < iter->data->end &&
+	    bkey_cmp(k, iter->b->ops->is_extents ?
+		     &START_KEY(next) : next) > 0) {
+		bch_dump_bucket(iter->b);
+		panic("Key skipped backwards\n");
+	}
 }
 
-/* Pointer validation */
+#else
 
-bool __bch_ptr_invalid(struct cache_set *c, int level, const struct bkey *k)
-{
-	unsigned i;
+static inline void bch_btree_iter_next_check(struct btree_iter *iter) {}
 
-	if (level && (!KEY_PTRS(k) || !KEY_SIZE(k) || KEY_DIRTY(k)))
-		goto bad;
+#endif
 
-	if (!level && KEY_SIZE(k) > KEY_OFFSET(k))
-		goto bad;
+/* Keylists */
 
-	if (!KEY_SIZE(k))
-		return true;
+int __bch_keylist_realloc(struct keylist *l, unsigned int u64s)
+{
+	size_t oldsize = bch_keylist_nkeys(l);
+	size_t newsize = oldsize + u64s;
+	uint64_t *old_keys = l->keys_p == l->inline_keys ? NULL : l->keys_p;
+	uint64_t *new_keys;
 
-	for (i = 0; i < KEY_PTRS(k); i++)
-		if (ptr_available(c, k, i)) {
-			struct cache *ca = PTR_CACHE(c, k, i);
-			size_t bucket = PTR_BUCKET_NR(c, k, i);
-			size_t r = bucket_remainder(c, PTR_OFFSET(k, i));
-
-			if (KEY_SIZE(k) + r > c->sb.bucket_size ||
-			    bucket <  ca->sb.first_bucket ||
-			    bucket >= ca->sb.nbuckets)
-				goto bad;
-		}
+	newsize = roundup_pow_of_two(newsize);
 
-	return false;
-bad:
-	cache_bug(c, "spotted bad key %s: %s", pkey(k), bch_ptr_status(c, k));
-	return true;
-}
+	if (newsize <= KEYLIST_INLINE ||
+	    roundup_pow_of_two(oldsize) == newsize)
+		return 0;
 
-bool bch_ptr_bad(struct btree *b, const struct bkey *k)
-{
-	struct bucket *g;
-	unsigned i, stale;
+	new_keys = krealloc(old_keys, sizeof(uint64_t) * newsize, GFP_NOIO);
 
-	if (!bkey_cmp(k, &ZERO_KEY) ||
-	    !KEY_PTRS(k) ||
-	    bch_ptr_invalid(b, k))
-		return true;
+	if (!new_keys)
+		return -ENOMEM;
 
-	if (KEY_PTRS(k) && PTR_DEV(k, 0) == PTR_CHECK_DEV)
-		return true;
+	if (!old_keys)
+		memcpy(new_keys, l->inline_keys, sizeof(uint64_t) * oldsize);
 
-	for (i = 0; i < KEY_PTRS(k); i++)
-		if (ptr_available(b->c, k, i)) {
-			g = PTR_BUCKET(b->c, k, i);
-			stale = ptr_stale(b->c, k, i);
+	l->keys_p = new_keys;
+	l->top_p = new_keys + oldsize;
 
-			btree_bug_on(stale > 96, b,
-				     "key too stale: %i, need_gc %u",
-				     stale, b->c->need_gc);
+	return 0;
+}
 
-			btree_bug_on(stale && KEY_DIRTY(k) && KEY_SIZE(k),
-				     b, "stale dirty pointer");
+/* Pop the top key of keylist by pointing l->top to its previous key */
+struct bkey *bch_keylist_pop(struct keylist *l)
+{
+	struct bkey *k = l->keys;
 
-			if (stale)
-				return true;
+	if (k == l->top)
+		return NULL;
 
-#ifdef CONFIG_BCACHE_EDEBUG
-			if (!mutex_trylock(&b->c->bucket_lock))
-				continue;
+	while (bkey_next(k) != l->top)
+		k = bkey_next(k);
 
-			if (b->level) {
-				if (KEY_DIRTY(k) ||
-				    g->prio != BTREE_PRIO ||
-				    (b->c->gc_mark_valid &&
-				     GC_MARK(g) != GC_MARK_METADATA))
-					goto bug;
-
-			} else {
-				if (g->prio == BTREE_PRIO)
-					goto bug;
-
-				if (KEY_DIRTY(k) &&
-				    b->c->gc_mark_valid &&
-				    GC_MARK(g) != GC_MARK_DIRTY)
-					goto bug;
-			}
-			mutex_unlock(&b->c->bucket_lock);
-#endif
-		}
+	return l->top = k;
+}
 
-	return false;
-#ifdef CONFIG_BCACHE_EDEBUG
-bug:
-	mutex_unlock(&b->c->bucket_lock);
-	btree_bug(b,
-"inconsistent pointer %s: bucket %zu pin %i prio %i gen %i last_gc %i mark %llu gc_gen %i",
-		  pkey(k), PTR_BUCKET_NR(b->c, k, i), atomic_read(&g->pin),
-		  g->prio, g->gen, g->last_gc, GC_MARK(g), g->gc_gen);
-	return true;
-#endif
+/* Pop the bottom key of keylist and update l->top_p */
+void bch_keylist_pop_front(struct keylist *l)
+{
+	l->top_p -= bkey_u64s(l->keys);
+
+	memmove(l->keys,
+		bkey_next(l->keys),
+		bch_keylist_bytes(l));
 }
 
 /* Key/pointer manipulation */
 
 void bch_bkey_copy_single_ptr(struct bkey *dest, const struct bkey *src,
-			      unsigned i)
+			      unsigned int i)
 {
 	BUG_ON(i > KEY_PTRS(src));
 
@@ -187,7 +196,7 @@ void bch_bkey_copy_single_ptr(struct bkey *dest, const struct bkey *src,
 
 bool __bch_cut_front(const struct bkey *where, struct bkey *k)
 {
-	unsigned i, len = 0;
+	unsigned int i, len = 0;
 
 	if (bkey_cmp(where, &START_KEY(k)) <= 0)
 		return false;
@@ -207,7 +216,7 @@ bool __bch_cut_front(const struct bkey *where, struct bkey *k)
 
 bool __bch_cut_back(const struct bkey *where, struct bkey *k)
 {
-	unsigned len = 0;
+	unsigned int len = 0;
 
 	if (bkey_cmp(where, k) >= 0)
 		return false;
@@ -224,60 +233,142 @@ bool __bch_cut_back(const struct bkey *where, struct bkey *k)
 	return true;
 }
 
-static uint64_t merge_chksums(struct bkey *l, struct bkey *r)
+/* Auxiliary search trees */
+
+/* 32 bits total: */
+#define BKEY_MID_BITS		3
+#define BKEY_EXPONENT_BITS	7
+#define BKEY_MANTISSA_BITS	(32 - BKEY_MID_BITS - BKEY_EXPONENT_BITS)
+#define BKEY_MANTISSA_MASK	((1 << BKEY_MANTISSA_BITS) - 1)
+
+struct bkey_float {
+	unsigned int	exponent:BKEY_EXPONENT_BITS;
+	unsigned int	m:BKEY_MID_BITS;
+	unsigned int	mantissa:BKEY_MANTISSA_BITS;
+} __packed;
+
+/*
+ * BSET_CACHELINE was originally intended to match the hardware cacheline size -
+ * it used to be 64, but I realized the lookup code would touch slightly less
+ * memory if it was 128.
+ *
+ * It definites the number of bytes (in struct bset) per struct bkey_float in
+ * the auxiliar search tree - when we're done searching the bset_float tree we
+ * have this many bytes left that we do a linear search over.
+ *
+ * Since (after level 5) every level of the bset_tree is on a new cacheline,
+ * we're touching one fewer cacheline in the bset tree in exchange for one more
+ * cacheline in the linear search - but the linear search might stop before it
+ * gets to the second cacheline.
+ */
+
+#define BSET_CACHELINE		128
+
+/* Space required for the btree node keys */
+static inline size_t btree_keys_bytes(struct btree_keys *b)
 {
-	return (l->ptr[KEY_PTRS(l)] + r->ptr[KEY_PTRS(r)]) &
-		~((uint64_t)1 << 63);
+	return PAGE_SIZE << b->page_order;
 }
 
-/* Tries to merge l and r: l should be lower than r
- * Returns true if we were able to merge. If we did merge, l will be the merged
- * key, r will be untouched.
- */
-bool bch_bkey_try_merge(struct btree *b, struct bkey *l, struct bkey *r)
+static inline size_t btree_keys_cachelines(struct btree_keys *b)
 {
-	unsigned i;
+	return btree_keys_bytes(b) / BSET_CACHELINE;
+}
 
-	if (key_merging_disabled(b->c))
-		return false;
+/* Space required for the auxiliary search trees */
+static inline size_t bset_tree_bytes(struct btree_keys *b)
+{
+	return btree_keys_cachelines(b) * sizeof(struct bkey_float);
+}
 
-	if (KEY_PTRS(l) != KEY_PTRS(r) ||
-	    KEY_DIRTY(l) != KEY_DIRTY(r) ||
-	    bkey_cmp(l, &START_KEY(r)))
-		return false;
+/* Space required for the prev pointers */
+static inline size_t bset_prev_bytes(struct btree_keys *b)
+{
+	return btree_keys_cachelines(b) * sizeof(uint8_t);
+}
 
-	for (i = 0; i < KEY_PTRS(l); i++)
-		if (l->ptr[i] + PTR(0, KEY_SIZE(l), 0) != r->ptr[i] ||
-		    PTR_BUCKET_NR(b->c, l, i) != PTR_BUCKET_NR(b->c, r, i))
-			return false;
+/* Memory allocation */
 
-	/* Keys with no pointers aren't restricted to one bucket and could
-	 * overflow KEY_SIZE
-	 */
-	if (KEY_SIZE(l) + KEY_SIZE(r) > USHRT_MAX) {
-		SET_KEY_OFFSET(l, KEY_OFFSET(l) + USHRT_MAX - KEY_SIZE(l));
-		SET_KEY_SIZE(l, USHRT_MAX);
+void bch_btree_keys_free(struct btree_keys *b)
+{
+	struct bset_tree *t = b->set;
 
-		bch_cut_front(l, r);
-		return false;
-	}
+	if (bset_prev_bytes(b) < PAGE_SIZE)
+		kfree(t->prev);
+	else
+		free_pages((unsigned long) t->prev,
+			   get_order(bset_prev_bytes(b)));
 
-	if (KEY_CSUM(l)) {
-		if (KEY_CSUM(r))
-			l->ptr[KEY_PTRS(l)] = merge_chksums(l, r);
-		else
-			SET_KEY_CSUM(l, 0);
-	}
+	if (bset_tree_bytes(b) < PAGE_SIZE)
+		kfree(t->tree);
+	else
+		free_pages((unsigned long) t->tree,
+			   get_order(bset_tree_bytes(b)));
 
-	SET_KEY_OFFSET(l, KEY_OFFSET(l) + KEY_SIZE(r));
-	SET_KEY_SIZE(l, KEY_SIZE(l) + KEY_SIZE(r));
+	free_pages((unsigned long) t->data, b->page_order);
 
-	return true;
+	t->prev = NULL;
+	t->tree = NULL;
+	t->data = NULL;
+}
+
+int bch_btree_keys_alloc(struct btree_keys *b,
+			 unsigned int page_order,
+			 gfp_t gfp)
+{
+	struct bset_tree *t = b->set;
+
+	BUG_ON(t->data);
+
+	b->page_order = page_order;
+
+	t->data = (void *) __get_free_pages(__GFP_COMP|gfp, b->page_order);
+	if (!t->data)
+		goto err;
+
+	t->tree = bset_tree_bytes(b) < PAGE_SIZE
+		? kmalloc(bset_tree_bytes(b), gfp)
+		: (void *) __get_free_pages(gfp, get_order(bset_tree_bytes(b)));
+	if (!t->tree)
+		goto err;
+
+	t->prev = bset_prev_bytes(b) < PAGE_SIZE
+		? kmalloc(bset_prev_bytes(b), gfp)
+		: (void *) __get_free_pages(gfp, get_order(bset_prev_bytes(b)));
+	if (!t->prev)
+		goto err;
+
+	return 0;
+err:
+	bch_btree_keys_free(b);
+	return -ENOMEM;
+}
+
+void bch_btree_keys_init(struct btree_keys *b, const struct btree_keys_ops *ops,
+			 bool *expensive_debug_checks)
+{
+	b->ops = ops;
+	b->expensive_debug_checks = expensive_debug_checks;
+	b->nsets = 0;
+	b->last_set_unwritten = 0;
+
+	/*
+	 * struct btree_keys in embedded in struct btree, and struct
+	 * bset_tree is embedded into struct btree_keys. They are all
+	 * initialized as 0 by kzalloc() in mca_bucket_alloc(), and
+	 * b->set[0].data is allocated in bch_btree_keys_alloc(), so we
+	 * don't have to initiate b->set[].size and b->set[].data here
+	 * any more.
+	 */
 }
 
 /* Binary tree stuff for auxiliary search trees */
 
-static unsigned inorder_next(unsigned j, unsigned size)
+/*
+ * return array index next to j when does in-order traverse
+ * of a binary tree which is stored in a linear array
+ */
+static unsigned int inorder_next(unsigned int j, unsigned int size)
 {
 	if (j * 2 + 1 < size) {
 		j = j * 2 + 1;
@@ -290,7 +381,11 @@ static unsigned inorder_next(unsigned j, unsigned size)
 	return j;
 }
 
-static unsigned inorder_prev(unsigned j, unsigned size)
+/*
+ * return array index previous to j when does in-order traverse
+ * of a binary tree which is stored in a linear array
+ */
+static unsigned int inorder_prev(unsigned int j, unsigned int size)
 {
 	if (j * 2 < size) {
 		j = j * 2;
@@ -303,7 +398,8 @@ static unsigned inorder_prev(unsigned j, unsigned size)
 	return j;
 }
 
-/* I have no idea why this code works... and I'm the one who wrote it
+/*
+ * I have no idea why this code works... and I'm the one who wrote it
  *
  * However, I do know what it does:
  * Given a binary tree constructed in an array (i.e. how you normally implement
@@ -316,10 +412,12 @@ static unsigned inorder_prev(unsigned j, unsigned size)
  * extra is a function of size:
  *   extra = (size - rounddown_pow_of_two(size - 1)) << 1;
  */
-static unsigned __to_inorder(unsigned j, unsigned size, unsigned extra)
+static unsigned int __to_inorder(unsigned int j,
+				  unsigned int size,
+				  unsigned int extra)
 {
-	unsigned b = fls(j);
-	unsigned shift = fls(size - 1) - b;
+	unsigned int b = fls(j);
+	unsigned int shift = fls(size - 1) - b;
 
 	j  ^= 1U << (b - 1);
 	j <<= 1;
@@ -332,14 +430,20 @@ static unsigned __to_inorder(unsigned j, unsigned size, unsigned extra)
 	return j;
 }
 
-static unsigned to_inorder(unsigned j, struct bset_tree *t)
+/*
+ * Return the cacheline index in bset_tree->data, where j is index
+ * from a linear array which stores the auxiliar binary tree
+ */
+static unsigned int to_inorder(unsigned int j, struct bset_tree *t)
 {
 	return __to_inorder(j, t->size, t->extra);
 }
 
-static unsigned __inorder_to_tree(unsigned j, unsigned size, unsigned extra)
+static unsigned int __inorder_to_tree(unsigned int j,
+				      unsigned int size,
+				      unsigned int extra)
 {
-	unsigned shift;
+	unsigned int shift;
 
 	if (j > extra)
 		j += j - extra;
@@ -352,7 +456,11 @@ static unsigned __inorder_to_tree(unsigned j, unsigned size, unsigned extra)
 	return j;
 }
 
-static unsigned inorder_to_tree(unsigned j, struct bset_tree *t)
+/*
+ * Return an index from a linear array which stores the auxiliar binary
+ * tree, j is the cacheline index of t->data.
+ */
+static unsigned int inorder_to_tree(unsigned int j, struct bset_tree *t)
 {
 	return __inorder_to_tree(j, t->size, t->extra);
 }
@@ -363,14 +471,15 @@ void inorder_test(void)
 	unsigned long done = 0;
 	ktime_t start = ktime_get();
 
-	for (unsigned size = 2;
+	for (unsigned int size = 2;
 	     size < 65536000;
 	     size++) {
-		unsigned extra = (size - rounddown_pow_of_two(size - 1)) << 1;
-		unsigned i = 1, j = rounddown_pow_of_two(size - 1);
+		unsigned int extra =
+			(size - rounddown_pow_of_two(size - 1)) << 1;
+		unsigned int i = 1, j = rounddown_pow_of_two(size - 1);
 
 		if (!(size % 4096))
-			printk(KERN_NOTICE "loop %u, %llu per us\n", size,
+			pr_notice("loop %u, %llu per us\n", size,
 			       done / ktime_us_delta(ktime_get(), start));
 
 		while (1) {
@@ -413,28 +522,31 @@ void inorder_test(void)
  * of the previous key so we can walk backwards to it from t->tree[j]'s key.
  */
 
-static struct bkey *cacheline_to_bkey(struct bset_tree *t, unsigned cacheline,
-				      unsigned offset)
+static struct bkey *cacheline_to_bkey(struct bset_tree *t,
+				      unsigned int cacheline,
+				      unsigned int offset)
 {
 	return ((void *) t->data) + cacheline * BSET_CACHELINE + offset * 8;
 }
 
-static unsigned bkey_to_cacheline(struct bset_tree *t, struct bkey *k)
+static unsigned int bkey_to_cacheline(struct bset_tree *t, struct bkey *k)
 {
 	return ((void *) k - (void *) t->data) / BSET_CACHELINE;
 }
 
-static unsigned bkey_to_cacheline_offset(struct bkey *k)
+static unsigned int bkey_to_cacheline_offset(struct bset_tree *t,
+					 unsigned int cacheline,
+					 struct bkey *k)
 {
-	return ((size_t) k & (BSET_CACHELINE - 1)) / sizeof(uint64_t);
+	return (u64 *) k - (u64 *) cacheline_to_bkey(t, cacheline, 0);
 }
 
-static struct bkey *tree_to_bkey(struct bset_tree *t, unsigned j)
+static struct bkey *tree_to_bkey(struct bset_tree *t, unsigned int j)
 {
 	return cacheline_to_bkey(t, to_inorder(j, t), t->tree[j].m);
 }
 
-static struct bkey *tree_to_prev_bkey(struct bset_tree *t, unsigned j)
+static struct bkey *tree_to_prev_bkey(struct bset_tree *t, unsigned int j)
 {
 	return (void *) (((uint64_t *) tree_to_bkey(t, j)) - t->prev[j]);
 }
@@ -443,34 +555,41 @@ static struct bkey *tree_to_prev_bkey(struct bset_tree *t, unsigned j)
  * For the write set - the one we're currently inserting keys into - we don't
  * maintain a full search tree, we just keep a simple lookup table in t->prev.
  */
-static struct bkey *table_to_bkey(struct bset_tree *t, unsigned cacheline)
+static struct bkey *table_to_bkey(struct bset_tree *t, unsigned int cacheline)
 {
 	return cacheline_to_bkey(t, cacheline, t->prev[cacheline]);
 }
 
 static inline uint64_t shrd128(uint64_t high, uint64_t low, uint8_t shift)
 {
-#ifdef CONFIG_X86_64
-	asm("shrd %[shift],%[high],%[low]"
-	    : [low] "+Rm" (low)
-	    : [high] "R" (high),
-	    [shift] "ci" (shift)
-	    : "cc");
-#else
 	low >>= shift;
 	low  |= (high << 1) << (63U - shift);
-#endif
 	return low;
 }
 
-static inline unsigned bfloat_mantissa(const struct bkey *k,
+/*
+ * Calculate mantissa value for struct bkey_float.
+ * If most significant bit of f->exponent is not set, then
+ *  - f->exponent >> 6 is 0
+ *  - p[0] points to bkey->low
+ *  - p[-1] borrows bits from KEY_INODE() of bkey->high
+ * if most isgnificant bits of f->exponent is set, then
+ *  - f->exponent >> 6 is 1
+ *  - p[0] points to bits from KEY_INODE() of bkey->high
+ *  - p[-1] points to other bits from KEY_INODE() of
+ *    bkey->high too.
+ * See make_bfloat() to check when most significant bit of f->exponent
+ * is set or not.
+ */
+static inline unsigned int bfloat_mantissa(const struct bkey *k,
 				       struct bkey_float *f)
 {
 	const uint64_t *p = &k->low - (f->exponent >> 6);
+
 	return shrd128(p[-1], p[0], f->exponent & 63) & BKEY_MANTISSA_MASK;
 }
 
-static void make_bfloat(struct bset_tree *t, unsigned j)
+static void make_bfloat(struct bset_tree *t, unsigned int j)
 {
 	struct bkey_float *f = &t->tree[j];
 	struct bkey *m = tree_to_bkey(t, j);
@@ -481,12 +600,22 @@ static void make_bfloat(struct bset_tree *t, unsigned j)
 		: tree_to_prev_bkey(t, j >> ffs(j));
 
 	struct bkey *r = is_power_of_2(j + 1)
-		? node(t->data, t->data->keys - bkey_u64s(&t->end))
+		? bset_bkey_idx(t->data, t->data->keys - bkey_u64s(&t->end))
 		: tree_to_bkey(t, j >> (ffz(j) + 1));
 
 	BUG_ON(m < l || m > r);
 	BUG_ON(bkey_next(p) != m);
 
+	/*
+	 * If l and r have different KEY_INODE values (different backing
+	 * device), f->exponent records how many least significant bits
+	 * are different in KEY_INODE values and sets most significant
+	 * bits to 1 (by +64).
+	 * If l and r have same KEY_INODE value, f->exponent records
+	 * how many different bits in least significant bits of bkey->low.
+	 * See bfloat_mantiss() how the most significant bit of
+	 * f->exponent is used to calculate bfloat mantissa value.
+	 */
 	if (KEY_INODE(l) != KEY_INODE(r))
 		f->exponent = fls64(KEY_INODE(r) ^ KEY_INODE(l)) + 64;
 	else
@@ -505,43 +634,72 @@ static void make_bfloat(struct bset_tree *t, unsigned j)
 		f->exponent = 127;
 }
 
-static void bset_alloc_tree(struct btree *b, struct bset_tree *t)
+static void bset_alloc_tree(struct btree_keys *b, struct bset_tree *t)
 {
-	if (t != b->sets) {
-		unsigned j = roundup(t[-1].size,
+	if (t != b->set) {
+		unsigned int j = roundup(t[-1].size,
 				     64 / sizeof(struct bkey_float));
 
 		t->tree = t[-1].tree + j;
 		t->prev = t[-1].prev + j;
 	}
 
-	while (t < b->sets + MAX_BSETS)
+	while (t < b->set + MAX_BSETS)
 		t++->size = 0;
 }
 
-static void bset_build_unwritten_tree(struct btree *b)
+static void bch_bset_build_unwritten_tree(struct btree_keys *b)
 {
-	struct bset_tree *t = b->sets + b->nsets;
+	struct bset_tree *t = bset_tree_last(b);
+
+	BUG_ON(b->last_set_unwritten);
+	b->last_set_unwritten = 1;
 
 	bset_alloc_tree(b, t);
 
-	if (t->tree != b->sets->tree + bset_tree_space(b)) {
-		t->prev[0] = bkey_to_cacheline_offset(t->data->start);
+	if (t->tree != b->set->tree + btree_keys_cachelines(b)) {
+		t->prev[0] = bkey_to_cacheline_offset(t, 0, t->data->start);
 		t->size = 1;
 	}
 }
 
-static void bset_build_written_tree(struct btree *b)
+void bch_bset_init_next(struct btree_keys *b, struct bset *i, uint64_t magic)
+{
+	if (i != b->set->data) {
+		b->set[++b->nsets].data = i;
+		i->seq = b->set->data->seq;
+	} else
+		get_random_bytes(&i->seq, sizeof(uint64_t));
+
+	i->magic	= magic;
+	i->version	= 0;
+	i->keys		= 0;
+
+	bch_bset_build_unwritten_tree(b);
+}
+
+/*
+ * Build auxiliary binary tree 'struct bset_tree *t', this tree is used to
+ * accelerate bkey search in a btree node (pointed by bset_tree->data in
+ * memory). After search in the auxiliar tree by calling bset_search_tree(),
+ * a struct bset_search_iter is returned which indicates range [l, r] from
+ * bset_tree->data where the searching bkey might be inside. Then a followed
+ * linear comparison does the exact search, see __bch_bset_search() for how
+ * the auxiliary tree is used.
+ */
+void bch_bset_build_written_tree(struct btree_keys *b)
 {
-	struct bset_tree *t = b->sets + b->nsets;
-	struct bkey *k = t->data->start;
-	unsigned j, cacheline = 1;
+	struct bset_tree *t = bset_tree_last(b);
+	struct bkey *prev = NULL, *k = t->data->start;
+	unsigned int j, cacheline = 1;
+
+	b->last_set_unwritten = 0;
 
 	bset_alloc_tree(b, t);
 
-	t->size = min_t(unsigned,
-			bkey_to_cacheline(t, end(t->data)),
-			b->sets->tree + bset_tree_space(b) - t->tree);
+	t->size = min_t(unsigned int,
+			bkey_to_cacheline(t, bset_bkey_last(t->data)),
+			b->set->tree + btree_keys_cachelines(b) - t->tree);
 
 	if (t->size < 2) {
 		t->size = 0;
@@ -554,16 +712,16 @@ static void bset_build_written_tree(struct btree *b)
 	for (j = inorder_next(0, t->size);
 	     j;
 	     j = inorder_next(j, t->size)) {
-		while (bkey_to_cacheline(t, k) != cacheline)
+		while (bkey_to_cacheline(t, k) < cacheline) {
+			prev = k;
 			k = bkey_next(k);
+		}
 
-		t->prev[j] = bkey_u64s(k);
-		k = bkey_next(k);
-		cacheline++;
-		t->tree[j].m = bkey_to_cacheline_offset(k);
+		t->prev[j] = bkey_u64s(prev);
+		t->tree[j].m = bkey_to_cacheline_offset(t, cacheline++, k);
 	}
 
-	while (bkey_next(k) != end(t->data))
+	while (bkey_next(k) != bset_bkey_last(t->data))
 		k = bkey_next(k);
 
 	t->end = *k;
@@ -575,13 +733,15 @@ static void bset_build_written_tree(struct btree *b)
 		make_bfloat(t, j);
 }
 
-void bch_bset_fix_invalidated_key(struct btree *b, struct bkey *k)
+/* Insert */
+
+void bch_bset_fix_invalidated_key(struct btree_keys *b, struct bkey *k)
 {
 	struct bset_tree *t;
-	unsigned inorder, j = 1;
+	unsigned int inorder, j = 1;
 
-	for (t = b->sets; t <= &b->sets[b->nsets]; t++)
-		if (k < end(t->data))
+	for (t = b->set; t <= bset_tree_last(b); t++)
+		if (k < bset_bkey_last(t->data))
 			goto found_set;
 
 	BUG();
@@ -594,7 +754,7 @@ found_set:
 	if (k == t->data->start)
 		goto fix_left;
 
-	if (bkey_next(k) == end(t->data)) {
+	if (bkey_next(k) == bset_bkey_last(t->data)) {
 		t->end = *k;
 		goto fix_right;
 	}
@@ -620,25 +780,28 @@ fix_right:	do {
 		} while (j < t->size);
 }
 
-void bch_bset_fix_lookup_table(struct btree *b, struct bkey *k)
+static void bch_bset_fix_lookup_table(struct btree_keys *b,
+				      struct bset_tree *t,
+				      struct bkey *k)
 {
-	struct bset_tree *t = &b->sets[b->nsets];
-	unsigned shift = bkey_u64s(k);
-	unsigned j = bkey_to_cacheline(t, k);
+	unsigned int shift = bkey_u64s(k);
+	unsigned int j = bkey_to_cacheline(t, k);
 
 	/* We're getting called from btree_split() or btree_gc, just bail out */
 	if (!t->size)
 		return;
 
-	/* k is the key we just inserted; we need to find the entry in the
+	/*
+	 * k is the key we just inserted; we need to find the entry in the
 	 * lookup table for the first key that is strictly greater than k:
 	 * it's either k's cacheline or the next one
 	 */
-	if (j < t->size &&
-	    table_to_bkey(t, j) <= k)
+	while (j < t->size &&
+	       table_to_bkey(t, j) <= k)
 		j++;
 
-	/* Adjust all the lookup table entries, and find a new key for any that
+	/*
+	 * Adjust all the lookup table entries, and find a new key for any that
 	 * have gotten too big
 	 */
 	for (; j < t->size; j++) {
@@ -650,56 +813,136 @@ void bch_bset_fix_lookup_table(struct btree *b, struct bkey *k)
 			while (k < cacheline_to_bkey(t, j, 0))
 				k = bkey_next(k);
 
-			t->prev[j] = bkey_to_cacheline_offset(k);
+			t->prev[j] = bkey_to_cacheline_offset(t, j, k);
 		}
 	}
 
-	if (t->size == b->sets->tree + bset_tree_space(b) - t->tree)
+	if (t->size == b->set->tree + btree_keys_cachelines(b) - t->tree)
 		return;
 
 	/* Possibly add a new entry to the end of the lookup table */
 
 	for (k = table_to_bkey(t, t->size - 1);
-	     k != end(t->data);
+	     k != bset_bkey_last(t->data);
 	     k = bkey_next(k))
 		if (t->size == bkey_to_cacheline(t, k)) {
-			t->prev[t->size] = bkey_to_cacheline_offset(k);
+			t->prev[t->size] =
+				bkey_to_cacheline_offset(t, t->size, k);
 			t->size++;
 		}
 }
 
-void bch_bset_init_next(struct btree *b)
+/*
+ * Tries to merge l and r: l should be lower than r
+ * Returns true if we were able to merge. If we did merge, l will be the merged
+ * key, r will be untouched.
+ */
+bool bch_bkey_try_merge(struct btree_keys *b, struct bkey *l, struct bkey *r)
 {
-	struct bset *i = write_block(b);
+	if (!b->ops->key_merge)
+		return false;
 
-	if (i != b->sets[0].data) {
-		b->sets[++b->nsets].data = i;
-		i->seq = b->sets[0].data->seq;
-	} else
-		get_random_bytes(&i->seq, sizeof(uint64_t));
+	/*
+	 * Generic header checks
+	 * Assumes left and right are in order
+	 * Left and right must be exactly aligned
+	 */
+	if (!bch_bkey_equal_header(l, r) ||
+	     bkey_cmp(l, &START_KEY(r)))
+		return false;
 
-	i->magic	= bset_magic(b->c);
-	i->version	= 0;
-	i->keys		= 0;
+	return b->ops->key_merge(b, l, r);
+}
+
+void bch_bset_insert(struct btree_keys *b, struct bkey *where,
+		     struct bkey *insert)
+{
+	struct bset_tree *t = bset_tree_last(b);
+
+	BUG_ON(!b->last_set_unwritten);
+	BUG_ON(bset_byte_offset(b, t->data) +
+	       __set_bytes(t->data, t->data->keys + bkey_u64s(insert)) >
+	       PAGE_SIZE << b->page_order);
 
-	bset_build_unwritten_tree(b);
+	memmove((uint64_t *) where + bkey_u64s(insert),
+		where,
+		(void *) bset_bkey_last(t->data) - (void *) where);
+
+	t->data->keys += bkey_u64s(insert);
+	bkey_copy(where, insert);
+	bch_bset_fix_lookup_table(b, t, where);
 }
 
+unsigned int bch_btree_insert_key(struct btree_keys *b, struct bkey *k,
+			      struct bkey *replace_key)
+{
+	unsigned int status = BTREE_INSERT_STATUS_NO_INSERT;
+	struct bset *i = bset_tree_last(b)->data;
+	struct bkey *m, *prev = NULL;
+	struct btree_iter_stack iter;
+	struct bkey preceding_key_on_stack = ZERO_KEY;
+	struct bkey *preceding_key_p = &preceding_key_on_stack;
+
+	BUG_ON(b->ops->is_extents && !KEY_SIZE(k));
+
+	/*
+	 * If k has preceding key, preceding_key_p will be set to address
+	 *  of k's preceding key; otherwise preceding_key_p will be set
+	 * to NULL inside preceding_key().
+	 */
+	if (b->ops->is_extents)
+		preceding_key(&START_KEY(k), &preceding_key_p);
+	else
+		preceding_key(k, &preceding_key_p);
+
+	m = bch_btree_iter_stack_init(b, &iter, preceding_key_p);
+
+	if (b->ops->insert_fixup(b, k, &iter.iter, replace_key))
+		return status;
+
+	status = BTREE_INSERT_STATUS_INSERT;
+
+	while (m != bset_bkey_last(i) &&
+	       bkey_cmp(k, b->ops->is_extents ? &START_KEY(m) : m) > 0) {
+		prev = m;
+		m = bkey_next(m);
+	}
+
+	/* prev is in the tree, if we merge we're done */
+	status = BTREE_INSERT_STATUS_BACK_MERGE;
+	if (prev &&
+	    bch_bkey_try_merge(b, prev, k))
+		goto merged;
+#if 0
+	status = BTREE_INSERT_STATUS_OVERWROTE;
+	if (m != bset_bkey_last(i) &&
+	    KEY_PTRS(m) == KEY_PTRS(k) && !KEY_SIZE(m))
+		goto copy;
+#endif
+	status = BTREE_INSERT_STATUS_FRONT_MERGE;
+	if (m != bset_bkey_last(i) &&
+	    bch_bkey_try_merge(b, k, m))
+		goto copy;
+
+	bch_bset_insert(b, m, k);
+copy:	bkey_copy(m, k);
+merged:
+	return status;
+}
+
+/* Lookup */
+
 struct bset_search_iter {
 	struct bkey *l, *r;
 };
 
-static struct bset_search_iter bset_search_write_set(struct btree *b,
-						     struct bset_tree *t,
+static struct bset_search_iter bset_search_write_set(struct bset_tree *t,
 						     const struct bkey *search)
 {
-	unsigned li = 0, ri = t->size;
-
-	BUG_ON(!b->nsets &&
-	       t->size < bkey_to_cacheline(t, end(t->data)));
+	unsigned int li = 0, ri = t->size;
 
 	while (li + 1 != ri) {
-		unsigned m = (li + ri) >> 1;
+		unsigned int m = (li + ri) >> 1;
 
 		if (bkey_cmp(table_to_bkey(t, m), search) > 0)
 			ri = m;
@@ -709,43 +952,37 @@ static struct bset_search_iter bset_search_write_set(struct btree *b,
 
 	return (struct bset_search_iter) {
 		table_to_bkey(t, li),
-		ri < t->size ? table_to_bkey(t, ri) : end(t->data)
+		ri < t->size ? table_to_bkey(t, ri) : bset_bkey_last(t->data)
 	};
 }
 
-static struct bset_search_iter bset_search_tree(struct btree *b,
-						struct bset_tree *t,
+static struct bset_search_iter bset_search_tree(struct bset_tree *t,
 						const struct bkey *search)
 {
 	struct bkey *l, *r;
 	struct bkey_float *f;
-	unsigned inorder, j, n = 1;
+	unsigned int inorder, j, n = 1;
 
 	do {
-		unsigned p = n << 4;
-		p &= ((int) (p - t->size)) >> 31;
+		unsigned int p = n << 4;
 
-		prefetch(&t->tree[p]);
+		if (p < t->size)
+			prefetch(&t->tree[p]);
 
 		j = n;
 		f = &t->tree[j];
 
-		/*
-		 * n = (f->mantissa > bfloat_mantissa())
-		 *	? j * 2
-		 *	: j * 2 + 1;
-		 *
-		 * We need to subtract 1 from f->mantissa for the sign bit trick
-		 * to work  - that's done in make_bfloat()
-		 */
-		if (likely(f->exponent != 127))
-			n = j * 2 + (((unsigned)
-				      (f->mantissa -
-				       bfloat_mantissa(search, f))) >> 31);
-		else
-			n = (bkey_cmp(tree_to_bkey(t, j), search) > 0)
-				? j * 2
-				: j * 2 + 1;
+		if (likely(f->exponent != 127)) {
+			if (f->mantissa >= bfloat_mantissa(search, f))
+				n = j * 2;
+			else
+				n = j * 2 + 1;
+		} else {
+			if (bkey_cmp(tree_to_bkey(t, j), search) > 0)
+				n = j * 2;
+			else
+				n = j * 2 + 1;
+		}
 	} while (n < t->size);
 
 	inorder = to_inorder(j, t);
@@ -761,7 +998,7 @@ static struct bset_search_iter bset_search_tree(struct btree *b,
 			f = &t->tree[inorder_next(j, t->size)];
 			r = cacheline_to_bkey(t, inorder, f->m);
 		} else
-			r = end(t->data);
+			r = bset_bkey_last(t->data);
 	} else {
 		r = cacheline_to_bkey(t, inorder, f->m);
 
@@ -775,7 +1012,7 @@ static struct bset_search_iter bset_search_tree(struct btree *b,
 	return (struct bset_search_iter) {l, r};
 }
 
-struct bkey *__bch_bset_search(struct btree *b, struct bset_tree *t,
+struct bkey *__bch_bset_search(struct btree_keys *b, struct bset_tree *t,
 			       const struct bkey *search)
 {
 	struct bset_search_iter i;
@@ -797,7 +1034,7 @@ struct bkey *__bch_bset_search(struct btree *b, struct bset_tree *t,
 
 	if (unlikely(!t->size)) {
 		i.l = t->data->start;
-		i.r = end(t->data);
+		i.r = bset_bkey_last(t->data);
 	} else if (bset_written(b, t)) {
 		/*
 		 * Each node in the auxiliary search tree covers a certain range
@@ -807,25 +1044,29 @@ struct bkey *__bch_bset_search(struct btree *b, struct bset_tree *t,
 		 */
 
 		if (unlikely(bkey_cmp(search, &t->end) >= 0))
-			return end(t->data);
+			return bset_bkey_last(t->data);
 
 		if (unlikely(bkey_cmp(search, t->data->start) < 0))
 			return t->data->start;
 
-		i = bset_search_tree(b, t, search);
-	} else
-		i = bset_search_write_set(b, t, search);
+		i = bset_search_tree(t, search);
+	} else {
+		BUG_ON(!b->nsets &&
+		       t->size < bkey_to_cacheline(t, bset_bkey_last(t->data)));
 
-#ifdef CONFIG_BCACHE_EDEBUG
-	BUG_ON(bset_written(b, t) &&
-	       i.l != t->data->start &&
-	       bkey_cmp(tree_to_prev_bkey(t,
-		  inorder_to_tree(bkey_to_cacheline(t, i.l), t)),
-			search) > 0);
+		i = bset_search_write_set(t, search);
+	}
 
-	BUG_ON(i.r != end(t->data) &&
-	       bkey_cmp(i.r, search) <= 0);
-#endif
+	if (btree_keys_expensive_checks(b)) {
+		BUG_ON(bset_written(b, t) &&
+		       i.l != t->data->start &&
+		       bkey_cmp(tree_to_prev_bkey(t,
+			  inorder_to_tree(bkey_to_cacheline(t, i.l), t)),
+				search) > 0);
+
+		BUG_ON(i.r != bset_bkey_last(t->data) &&
+		       bkey_cmp(i.r, search) <= 0);
+	}
 
 	while (likely(i.l != i.r) &&
 	       bkey_cmp(i.l, search) <= 0)
@@ -836,12 +1077,13 @@ struct bkey *__bch_bset_search(struct btree *b, struct bset_tree *t,
 
 /* Btree iterator */
 
+typedef bool (btree_iter_cmp_fn)(struct btree_iter_set,
+				 struct btree_iter_set);
+
 static inline bool btree_iter_cmp(struct btree_iter_set l,
 				  struct btree_iter_set r)
 {
-	int64_t c = bkey_cmp(&START_KEY(l.k), &START_KEY(r.k));
-
-	return c ? c > 0 : l.k < r.k;
+	return bkey_cmp(l.k, r.k) > 0;
 }
 
 static inline bool btree_iter_end(struct btree_iter *iter)
@@ -858,27 +1100,44 @@ void bch_btree_iter_push(struct btree_iter *iter, struct bkey *k,
 				 btree_iter_cmp));
 }
 
-struct bkey *__bch_btree_iter_init(struct btree *b, struct btree_iter *iter,
-			       struct bkey *search, struct bset_tree *start)
+static struct bkey *__bch_btree_iter_stack_init(struct btree_keys *b,
+						struct btree_iter_stack *iter,
+						struct bkey *search,
+						struct bset_tree *start)
 {
 	struct bkey *ret = NULL;
-	iter->size = ARRAY_SIZE(iter->data);
-	iter->used = 0;
 
-	for (; start <= &b->sets[b->nsets]; start++) {
+	iter->iter.size = ARRAY_SIZE(iter->stack_data);
+	iter->iter.used = 0;
+
+#ifdef CONFIG_BCACHE_DEBUG
+	iter->iter.b = b;
+#endif
+
+	for (; start <= bset_tree_last(b); start++) {
 		ret = bch_bset_search(b, start, search);
-		bch_btree_iter_push(iter, ret, end(start->data));
+		bch_btree_iter_push(&iter->iter, ret, bset_bkey_last(start->data));
 	}
 
 	return ret;
 }
 
-struct bkey *bch_btree_iter_next(struct btree_iter *iter)
+struct bkey *bch_btree_iter_stack_init(struct btree_keys *b,
+				 struct btree_iter_stack *iter,
+				 struct bkey *search)
+{
+	return __bch_btree_iter_stack_init(b, iter, search, b->set);
+}
+
+static inline struct bkey *__bch_btree_iter_next(struct btree_iter *iter,
+						 btree_iter_cmp_fn *cmp)
 {
-	struct btree_iter_set unused;
+	struct btree_iter_set b __maybe_unused;
 	struct bkey *ret = NULL;
 
 	if (!btree_iter_end(iter)) {
+		bch_btree_iter_next_check(iter);
+
 		ret = iter->data->k;
 		iter->data->k = bkey_next(iter->data->k);
 
@@ -888,16 +1147,22 @@ struct bkey *bch_btree_iter_next(struct btree_iter *iter)
 		}
 
 		if (iter->data->k == iter->data->end)
-			heap_pop(iter, unused, btree_iter_cmp);
+			heap_pop(iter, b, cmp);
 		else
-			heap_sift(iter, 0, btree_iter_cmp);
+			heap_sift(iter, 0, cmp);
 	}
 
 	return ret;
 }
 
+struct bkey *bch_btree_iter_next(struct btree_iter *iter)
+{
+	return __bch_btree_iter_next(iter, btree_iter_cmp);
+
+}
+
 struct bkey *bch_btree_iter_next_filter(struct btree_iter *iter,
-					struct btree *b, ptr_filter_fn fn)
+					struct btree_keys *b, ptr_filter_fn fn)
 {
 	struct bkey *ret;
 
@@ -908,63 +1173,55 @@ struct bkey *bch_btree_iter_next_filter(struct btree_iter *iter,
 	return ret;
 }
 
-struct bkey *bch_next_recurse_key(struct btree *b, struct bkey *search)
-{
-	struct btree_iter iter;
-
-	bch_btree_iter_init(b, &iter, search);
-	return bch_btree_iter_next_filter(&iter, b, bch_ptr_bad);
-}
-
 /* Mergesort */
 
-static void btree_sort_fixup(struct btree_iter *iter)
+void bch_bset_sort_state_free(struct bset_sort_state *state)
 {
-	while (iter->used > 1) {
-		struct btree_iter_set *top = iter->data, *i = top + 1;
-		struct bkey *k;
-
-		if (iter->used > 2 &&
-		    btree_iter_cmp(i[0], i[1]))
-			i++;
+	mempool_exit(&state->pool);
+}
 
-		for (k = i->k;
-		     k != i->end && bkey_cmp(top->k, &START_KEY(k)) > 0;
-		     k = bkey_next(k))
-			if (top->k > i->k)
-				__bch_cut_front(top->k, k);
-			else if (KEY_SIZE(k))
-				bch_cut_back(&START_KEY(k), top->k);
+int bch_bset_sort_state_init(struct bset_sort_state *state,
+			     unsigned int page_order)
+{
+	spin_lock_init(&state->time.lock);
 
-		if (top->k < i->k || k == i->k)
-			break;
+	state->page_order = page_order;
+	state->crit_factor = int_sqrt(1 << page_order);
 
-		heap_sift(iter, i - top, btree_iter_cmp);
-	}
+	return mempool_init_page_pool(&state->pool, 1, page_order);
 }
 
-static void btree_mergesort(struct btree *b, struct bset *out,
+static void btree_mergesort(struct btree_keys *b, struct bset *out,
 			    struct btree_iter *iter,
 			    bool fixup, bool remove_stale)
 {
+	int i;
 	struct bkey *k, *last = NULL;
-	bool (*bad)(struct btree *, const struct bkey *) = remove_stale
+	BKEY_PADDED(k) tmp;
+	bool (*bad)(struct btree_keys *, const struct bkey *) = remove_stale
 		? bch_ptr_bad
 		: bch_ptr_invalid;
 
+	/* Heapify the iterator, using our comparison function */
+	for (i = iter->used / 2 - 1; i >= 0; --i)
+		heap_sift(iter, i, b->ops->sort_cmp);
+
 	while (!btree_iter_end(iter)) {
-		if (fixup && !b->level)
-			btree_sort_fixup(iter);
+		if (b->ops->sort_fixup && fixup)
+			k = b->ops->sort_fixup(iter, &tmp.k);
+		else
+			k = NULL;
+
+		if (!k)
+			k = __bch_btree_iter_next(iter, b->ops->sort_cmp);
 
-		k = bch_btree_iter_next(iter);
 		if (bad(b, k))
 			continue;
 
 		if (!last) {
 			last = out->start;
 			bkey_copy(last, k);
-		} else if (b->level ||
-			   !bch_bkey_try_merge(b, last, k)) {
+		} else if (!bch_bkey_try_merge(b, last, k)) {
 			last = bkey_next(last);
 			bkey_copy(last, k);
 		}
@@ -972,167 +1229,147 @@ static void btree_mergesort(struct btree *b, struct bset *out,
 
 	out->keys = last ? (uint64_t *) bkey_next(last) - out->d : 0;
 
-	pr_debug("sorted %i keys", out->keys);
-	bch_check_key_order(b, out);
+	pr_debug("sorted %i keys\n", out->keys);
 }
 
-static void __btree_sort(struct btree *b, struct btree_iter *iter,
-			 unsigned start, unsigned order, bool fixup)
+static void __btree_sort(struct btree_keys *b, struct btree_iter *iter,
+			 unsigned int start, unsigned int order, bool fixup,
+			 struct bset_sort_state *state)
 {
 	uint64_t start_time;
-	bool remove_stale = !b->written;
-	struct bset *out = (void *) __get_free_pages(__GFP_NOWARN|GFP_NOIO,
+	bool used_mempool = false;
+	struct bset *out = (void *) __get_free_pages(__GFP_NOWARN|GFP_NOWAIT,
 						     order);
 	if (!out) {
-		mutex_lock(&b->c->sort_lock);
-		out = b->c->sort;
-		order = ilog2(bucket_pages(b->c));
+		struct page *outp;
+
+		BUG_ON(order > state->page_order);
+
+		outp = mempool_alloc(&state->pool, GFP_NOIO);
+		out = page_address(outp);
+		used_mempool = true;
+		order = state->page_order;
 	}
 
 	start_time = local_clock();
 
-	btree_mergesort(b, out, iter, fixup, remove_stale);
+	btree_mergesort(b, out, iter, fixup, false);
 	b->nsets = start;
 
-	if (!fixup && !start && b->written)
-		bch_btree_verify(b, out);
-
 	if (!start && order == b->page_order) {
 		/*
 		 * Our temporary buffer is the same size as the btree node's
 		 * buffer, we can just swap buffers instead of doing a big
 		 * memcpy()
+		 *
+		 * Don't worry event 'out' is allocated from mempool, it can
+		 * still be swapped here. Because state->pool is a page mempool
+		 * created by mempool_init_page_pool(), which allocates
+		 * pages by alloc_pages() indeed.
 		 */
 
-		out->magic	= bset_magic(b->c);
-		out->seq	= b->sets[0].data->seq;
-		out->version	= b->sets[0].data->version;
-		swap(out, b->sets[0].data);
-
-		if (b->c->sort == b->sets[0].data)
-			b->c->sort = out;
+		out->magic	= b->set->data->magic;
+		out->seq	= b->set->data->seq;
+		out->version	= b->set->data->version;
+		swap(out, b->set->data);
 	} else {
-		b->sets[start].data->keys = out->keys;
-		memcpy(b->sets[start].data->start, out->start,
-		       (void *) end(out) - (void *) out->start);
+		b->set[start].data->keys = out->keys;
+		memcpy(b->set[start].data->start, out->start,
+		       (void *) bset_bkey_last(out) - (void *) out->start);
 	}
 
-	if (out == b->c->sort)
-		mutex_unlock(&b->c->sort_lock);
+	if (used_mempool)
+		mempool_free(virt_to_page(out), &state->pool);
 	else
 		free_pages((unsigned long) out, order);
 
-	if (b->written)
-		bset_build_written_tree(b);
+	bch_bset_build_written_tree(b);
 
-	if (!start) {
-		spin_lock(&b->c->sort_time_lock);
-		bch_time_stats_update(&b->c->sort_time, start_time);
-		spin_unlock(&b->c->sort_time_lock);
-	}
+	if (!start)
+		bch_time_stats_update(&state->time, start_time);
 }
 
-void bch_btree_sort_partial(struct btree *b, unsigned start)
+void bch_btree_sort_partial(struct btree_keys *b, unsigned int start,
+			    struct bset_sort_state *state)
 {
-	size_t oldsize = 0, order = b->page_order, keys = 0;
-	struct btree_iter iter;
-	__bch_btree_iter_init(b, &iter, NULL, &b->sets[start]);
-
-	BUG_ON(b->sets[b->nsets].data == write_block(b) &&
-	       (b->sets[b->nsets].size || b->nsets));
+	size_t order = b->page_order, keys = 0;
+	struct btree_iter_stack iter;
+	int oldsize = bch_count_data(b);
 
-	if (b->written)
-		oldsize = bch_count_data(b);
+	__bch_btree_iter_stack_init(b, &iter, NULL, &b->set[start]);
 
 	if (start) {
-		unsigned i;
+		unsigned int i;
 
 		for (i = start; i <= b->nsets; i++)
-			keys += b->sets[i].data->keys;
+			keys += b->set[i].data->keys;
 
-		order = roundup_pow_of_two(__set_bytes(b->sets->data,
-						       keys)) / PAGE_SIZE;
-		if (order)
-			order = ilog2(order);
+		order = get_order(__set_bytes(b->set->data, keys));
 	}
 
-	__btree_sort(b, &iter, start, order, false);
+	__btree_sort(b, &iter.iter, start, order, false, state);
 
-	EBUG_ON(b->written && bch_count_data(b) != oldsize);
+	EBUG_ON(oldsize >= 0 && bch_count_data(b) != oldsize);
 }
 
-void bch_btree_sort_and_fix_extents(struct btree *b, struct btree_iter *iter)
+void bch_btree_sort_and_fix_extents(struct btree_keys *b,
+				    struct btree_iter *iter,
+				    struct bset_sort_state *state)
 {
-	BUG_ON(!b->written);
-	__btree_sort(b, iter, 0, b->page_order, true);
+	__btree_sort(b, iter, 0, b->page_order, true, state);
 }
 
-void bch_btree_sort_into(struct btree *b, struct btree *new)
+void bch_btree_sort_into(struct btree_keys *b, struct btree_keys *new,
+			 struct bset_sort_state *state)
 {
 	uint64_t start_time = local_clock();
+	struct btree_iter_stack iter;
 
-	struct btree_iter iter;
-	bch_btree_iter_init(b, &iter, NULL);
+	bch_btree_iter_stack_init(b, &iter, NULL);
 
-	btree_mergesort(b, new->sets->data, &iter, false, true);
+	btree_mergesort(b, new->set->data, &iter.iter, false, true);
 
-	spin_lock(&b->c->sort_time_lock);
-	bch_time_stats_update(&b->c->sort_time, start_time);
-	spin_unlock(&b->c->sort_time_lock);
+	bch_time_stats_update(&state->time, start_time);
 
-	bkey_copy_key(&new->key, &b->key);
-	new->sets->size = 0;
+	new->set->size = 0; // XXX: why?
 }
 
-void bch_btree_sort_lazy(struct btree *b)
-{
-	if (b->nsets) {
-		unsigned i, j, keys = 0, total;
-
-		for (i = 0; i <= b->nsets; i++)
-			keys += b->sets[i].data->keys;
+#define SORT_CRIT	(4096 / sizeof(uint64_t))
 
-		total = keys;
+void bch_btree_sort_lazy(struct btree_keys *b, struct bset_sort_state *state)
+{
+	unsigned int crit = SORT_CRIT;
+	int i;
 
-		for (j = 0; j < b->nsets; j++) {
-			if (keys * 2 < total ||
-			    keys < 1000) {
-				bch_btree_sort_partial(b, j);
-				return;
-			}
+	/* Don't sort if nothing to do */
+	if (!b->nsets)
+		goto out;
 
-			keys -= b->sets[j].data->keys;
-		}
+	for (i = b->nsets - 1; i >= 0; --i) {
+		crit *= state->crit_factor;
 
-		/* Must sort if b->nsets == 3 or we'll overflow */
-		if (b->nsets >= (MAX_BSETS - 1) - b->level) {
-			bch_btree_sort(b);
+		if (b->set[i].data->keys < crit) {
+			bch_btree_sort_partial(b, i, state);
 			return;
 		}
 	}
 
-	bset_build_written_tree(b);
-}
-
-/* Sysfs stuff */
+	/* Sort if we'd overflow */
+	if (b->nsets + 1 == MAX_BSETS) {
+		bch_btree_sort(b, state);
+		return;
+	}
 
-struct bset_stats {
-	size_t nodes;
-	size_t sets_written, sets_unwritten;
-	size_t bytes_written, bytes_unwritten;
-	size_t floats, failed;
-};
+out:
+	bch_bset_build_written_tree(b);
+}
 
-static int bch_btree_bset_stats(struct btree *b, struct btree_op *op,
-			    struct bset_stats *stats)
+void bch_btree_keys_stats(struct btree_keys *b, struct bset_stats *stats)
 {
-	struct bkey *k;
-	unsigned i;
-
-	stats->nodes++;
+	unsigned int i;
 
 	for (i = 0; i <= b->nsets; i++) {
-		struct bset_tree *t = &b->sets[i];
+		struct bset_tree *t = &b->set[i];
 		size_t bytes = t->data->keys * sizeof(uint64_t);
 		size_t j;
 
@@ -1150,43 +1387,4 @@ static int bch_btree_bset_stats(struct btree *b, struct btree_op *op,
 			stats->bytes_unwritten += bytes;
 		}
 	}
-
-	if (b->level) {
-		struct btree_iter iter;
-
-		for_each_key_filter(b, k, &iter, bch_ptr_bad) {
-			int ret = btree(bset_stats, k, b, op, stats);
-			if (ret)
-				return ret;
-		}
-	}
-
-	return 0;
-}
-
-int bch_bset_print_stats(struct cache_set *c, char *buf)
-{
-	struct btree_op op;
-	struct bset_stats t;
-	int ret;
-
-	bch_btree_op_init_stack(&op);
-	memset(&t, 0, sizeof(struct bset_stats));
-
-	ret = btree_root(bset_stats, c, &op, &t);
-	if (ret)
-		return ret;
-
-	return snprintf(buf, PAGE_SIZE,
-			"btree nodes:		%zu\n"
-			"written sets:		%zu\n"
-			"unwritten sets:		%zu\n"
-			"written key bytes:	%zu\n"
-			"unwritten key bytes:	%zu\n"
-			"floats:			%zu\n"
-			"failed:			%zu\n",
-			t.nodes,
-			t.sets_written, t.sets_unwritten,
-			t.bytes_written, t.bytes_unwritten,
-			t.floats, t.failed);
 }