diff mbox series

[v5,03/10] fs/ntfs3: Add bitmap

Message ID 20200911141018.2457639-4-almaz.alexandrovich@paragon-software.com (mailing list archive)
State New, archived
Headers show
Series NTFS read-write driver GPL implementation by Paragon Software | expand

Commit Message

Konstantin Komarov Sept. 11, 2020, 2:10 p.m. UTC
This adds bitmap

Signed-off-by: Konstantin Komarov <almaz.alexandrovich@paragon-software.com>
---
 fs/ntfs3/bitfunc.c |  137 ++++
 fs/ntfs3/bitmap.c  | 1516 ++++++++++++++++++++++++++++++++++++++++++++
 2 files changed, 1653 insertions(+)
 create mode 100644 fs/ntfs3/bitfunc.c
 create mode 100644 fs/ntfs3/bitmap.c

Comments

Joe Perches Sept. 13, 2020, 6:43 p.m. UTC | #1
On Fri, 2020-09-11 at 17:10 +0300, Konstantin Komarov wrote:
> This adds bitmap

$ make fs/ntfs3/
  SYNC    include/config/auto.conf.cmd
  CALL    scripts/checksyscalls.sh
  CALL    scripts/atomic/check-atomics.sh
  DESCEND  objtool
  CC      fs/ntfs3/bitfunc.o
  CC      fs/ntfs3/bitmap.o
fs/ntfs3/bitmap.c: In function ‘wnd_rescan’:
fs/ntfs3/bitmap.c:556:4: error: implicit declaration of function ‘page_cache_readahead_unbounded’; did you mean ‘page_cache_ra_unbounded’? [-Werror=implicit-function-declaration]
  556 |    page_cache_readahead_unbounded(
      |    ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
      |    page_cache_ra_unbounded
cc1: some warnings being treated as errors
make[2]: *** [scripts/Makefile.build:283: fs/ntfs3/bitmap.o] Error 1
make[1]: *** [scripts/Makefile.build:500: fs/ntfs3] Error 2
make: *** [Makefile:1792: fs] Error 2
Matthew Wilcox Sept. 14, 2020, 2:38 a.m. UTC | #2
On Sun, Sep 13, 2020 at 11:43:50AM -0700, Joe Perches wrote:
> On Fri, 2020-09-11 at 17:10 +0300, Konstantin Komarov wrote:
> > This adds bitmap
> 
> $ make fs/ntfs3/
>   SYNC    include/config/auto.conf.cmd
>   CALL    scripts/checksyscalls.sh
>   CALL    scripts/atomic/check-atomics.sh
>   DESCEND  objtool
>   CC      fs/ntfs3/bitfunc.o
>   CC      fs/ntfs3/bitmap.o
> fs/ntfs3/bitmap.c: In function ‘wnd_rescan’:
> fs/ntfs3/bitmap.c:556:4: error: implicit declaration of function ‘page_cache_readahead_unbounded’; did you mean ‘page_cache_ra_unbounded’? [-Werror=implicit-function-declaration]
>   556 |    page_cache_readahead_unbounded(
>       |    ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
>       |    page_cache_ra_unbounded
> cc1: some warnings being treated as errors
> make[2]: *** [scripts/Makefile.build:283: fs/ntfs3/bitmap.o] Error 1
> make[1]: *** [scripts/Makefile.build:500: fs/ntfs3] Error 2
> make: *** [Makefile:1792: fs] Error 2

That was only just renamed.  More concerningly, the documentation is
quite unambiguous:

 * This function is for filesystems to call when they want to start
 * readahead beyond a file's stated i_size.  This is almost certainly
 * not the function you want to call.  Use page_cache_async_readahead()
 * or page_cache_sync_readahead() instead.
Konstantin Komarov Sept. 18, 2020, 4:29 p.m. UTC | #3
From: Joe Perches <joe@perches.com>
Sent: Sunday, September 13, 2020 9:44 PM
> 
> On Fri, 2020-09-11 at 17:10 +0300, Konstantin Komarov wrote:
> > This adds bitmap
> 
> $ make fs/ntfs3/
>   SYNC    include/config/auto.conf.cmd
>   CALL    scripts/checksyscalls.sh
>   CALL    scripts/atomic/check-atomics.sh
>   DESCEND  objtool
>   CC      fs/ntfs3/bitfunc.o
>   CC      fs/ntfs3/bitmap.o
> fs/ntfs3/bitmap.c: In function ‘wnd_rescan’:
> fs/ntfs3/bitmap.c:556:4: error: implicit declaration of function ‘page_cache_readahead_unbounded’; did you mean
> ‘page_cache_ra_unbounded’? [-Werror=implicit-function-declaration]
>   556 |    page_cache_readahead_unbounded(
>       |    ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
>       |    page_cache_ra_unbounded
> cc1: some warnings being treated as errors
> make[2]: *** [scripts/Makefile.build:283: fs/ntfs3/bitmap.o] Error 1
> make[1]: *** [scripts/Makefile.build:500: fs/ntfs3] Error 2
> make: *** [Makefile:1792: fs] Error 2
> 
Hi Joe! Doesn't seem to be an issue for 5.9_rc5. Which repo should've
been used to reproduce?

Thanks.
Konstantin Komarov Sept. 18, 2020, 4:35 p.m. UTC | #4
From: Matthew Wilcox <willy@infradead.org>
Sent: Monday, September 14, 2020 5:39 AM
> 
> On Sun, Sep 13, 2020 at 11:43:50AM -0700, Joe Perches wrote:
> > On Fri, 2020-09-11 at 17:10 +0300, Konstantin Komarov wrote:
> > > This adds bitmap
> >
> > $ make fs/ntfs3/
> >   SYNC    include/config/auto.conf.cmd
> >   CALL    scripts/checksyscalls.sh
> >   CALL    scripts/atomic/check-atomics.sh
> >   DESCEND  objtool
> >   CC      fs/ntfs3/bitfunc.o
> >   CC      fs/ntfs3/bitmap.o
> > fs/ntfs3/bitmap.c: In function ‘wnd_rescan’:
> > fs/ntfs3/bitmap.c:556:4: error: implicit declaration of function ‘page_cache_readahead_unbounded’; did you mean
> ‘page_cache_ra_unbounded’? [-Werror=implicit-function-declaration]
> >   556 |    page_cache_readahead_unbounded(
> >       |    ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
> >       |    page_cache_ra_unbounded
> > cc1: some warnings being treated as errors
> > make[2]: *** [scripts/Makefile.build:283: fs/ntfs3/bitmap.o] Error 1
> > make[1]: *** [scripts/Makefile.build:500: fs/ntfs3] Error 2
> > make: *** [Makefile:1792: fs] Error 2
> 
> That was only just renamed.  More concerningly, the documentation is
> quite unambiguous:
> 
>  * This function is for filesystems to call when they want to start
>  * readahead beyond a file's stated i_size.  This is almost certainly
>  * not the function you want to call.  Use page_cache_async_readahead()
>  * or page_cache_sync_readahead() instead.

Hi Matthew! it's not so clear for us by several reasons (please correct
if this is wrong):
page_cache_sync_readahead() seems applicable as a replacement, but
it doesn't seem to be reasonable as readahead in this case gives perf
improvement because of it's async nature. The 'async' function is incompatible
replacement based on the arguments list.

Thanks.
Matthew Wilcox Sept. 18, 2020, 4:38 p.m. UTC | #5
On Fri, Sep 18, 2020 at 04:29:34PM +0000, Konstantin Komarov wrote:
> From: Joe Perches <joe@perches.com>
> Sent: Sunday, September 13, 2020 9:44 PM
> > 
> > On Fri, 2020-09-11 at 17:10 +0300, Konstantin Komarov wrote:
> > > This adds bitmap
> > 
> > $ make fs/ntfs3/
> >   SYNC    include/config/auto.conf.cmd
> >   CALL    scripts/checksyscalls.sh
> >   CALL    scripts/atomic/check-atomics.sh
> >   DESCEND  objtool
> >   CC      fs/ntfs3/bitfunc.o
> >   CC      fs/ntfs3/bitmap.o
> > fs/ntfs3/bitmap.c: In function ‘wnd_rescan’:
> > fs/ntfs3/bitmap.c:556:4: error: implicit declaration of function ‘page_cache_readahead_unbounded’; did you mean
> > ‘page_cache_ra_unbounded’? [-Werror=implicit-function-declaration]
> >   556 |    page_cache_readahead_unbounded(
> >       |    ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
> >       |    page_cache_ra_unbounded
> > cc1: some warnings being treated as errors
> > make[2]: *** [scripts/Makefile.build:283: fs/ntfs3/bitmap.o] Error 1
> > make[1]: *** [scripts/Makefile.build:500: fs/ntfs3] Error 2
> > make: *** [Makefile:1792: fs] Error 2
> > 
> Hi Joe! Doesn't seem to be an issue for 5.9_rc5. Which repo should've
> been used to reproduce?

It's in the -mm tree, which is included in linux-next.
Matthew Wilcox Sept. 18, 2020, 4:54 p.m. UTC | #6
On Fri, Sep 18, 2020 at 04:35:11PM +0000, Konstantin Komarov wrote:
> > That was only just renamed.  More concerningly, the documentation is
> > quite unambiguous:
> > 
> >  * This function is for filesystems to call when they want to start
> >  * readahead beyond a file's stated i_size.  This is almost certainly
> >  * not the function you want to call.  Use page_cache_async_readahead()
> >  * or page_cache_sync_readahead() instead.
> 
> Hi Matthew! it's not so clear for us by several reasons (please correct
> if this is wrong):
> page_cache_sync_readahead() seems applicable as a replacement, but
> it doesn't seem to be reasonable as readahead in this case gives perf
> improvement because of it's async nature. The 'async' function is incompatible
> replacement based on the arguments list.

I think the naming has confused you (so I need to clarify the docs).
The sync function is to be called when you need the page which is being
read, and you might want to take the opportunity to read more pages.
The async version is to be called when the page you need is in cache,
but you've noticed that you're getting towards the end of the readahead
window.  Neither version waits for I/O to complete; you have to wait for
the page to become unlocked and then you can check PageUptodate.

Looking at what you're doing, you don't have a file_ra_state because
you're just trying to readahead fs metadata, right?  I think you want
to call force_page_cache_readahead(mapping, NULL, start, nr_pages);
The prototype for it is in mm/internal.h, but I think moving it to
include/linux/pagemap.h is justifiable.
Konstantin Komarov Sept. 25, 2020, 4:04 p.m. UTC | #7
From: Matthew Wilcox <willy@infradead.org>
Sent: Friday, September 18, 2020 7:55 PM
> Subject: Re: [PATCH v5 03/10] fs/ntfs3: Add bitmap
> 
> On Fri, Sep 18, 2020 at 04:35:11PM +0000, Konstantin Komarov wrote:
> > > That was only just renamed.  More concerningly, the documentation is
> > > quite unambiguous:
> > >
> > >  * This function is for filesystems to call when they want to start
> > >  * readahead beyond a file's stated i_size.  This is almost certainly
> > >  * not the function you want to call.  Use page_cache_async_readahead()
> > >  * or page_cache_sync_readahead() instead.
> >
> > Hi Matthew! it's not so clear for us by several reasons (please correct
> > if this is wrong):
> > page_cache_sync_readahead() seems applicable as a replacement, but
> > it doesn't seem to be reasonable as readahead in this case gives perf
> > improvement because of it's async nature. The 'async' function is incompatible
> > replacement based on the arguments list.
> 
> I think the naming has confused you (so I need to clarify the docs).
> The sync function is to be called when you need the page which is being
> read, and you might want to take the opportunity to read more pages.
> The async version is to be called when the page you need is in cache,
> but you've noticed that you're getting towards the end of the readahead
> window.  Neither version waits for I/O to complete; you have to wait for
> the page to become unlocked and then you can check PageUptodate.
> 
> Looking at what you're doing, you don't have a file_ra_state because
> you're just trying to readahead fs metadata, right?

Hi Matthew! Yes, correct.

> I think you want
> to call force_page_cache_readahead(mapping, NULL, start, nr_pages);
> The prototype for it is in mm/internal.h, but I think moving it to
> include/linux/pagemap.h is justifiable.

Seems like this. We decided to temporarily remove the ra usage iv V7, while it's
not clear which of the public options is "our" case. Is there anything we can
assist regarding the force_page_cache_readahead() move to
include/linux/pagemap.h?

Thanks.
diff mbox series

Patch

diff --git a/fs/ntfs3/bitfunc.c b/fs/ntfs3/bitfunc.c
new file mode 100644
index 000000000000..b19972177535
--- /dev/null
+++ b/fs/ntfs3/bitfunc.c
@@ -0,0 +1,137 @@ 
+// SPDX-License-Identifier: GPL-2.0
+/*
+ *  linux/fs/ntfs3/bitfunc.c
+ *
+ * Copyright (C) 2019-2020 Paragon Software GmbH, All rights reserved.
+ *
+ */
+#include <linux/blkdev.h>
+#include <linux/buffer_head.h>
+#include <linux/fs.h>
+#include <linux/nls.h>
+#include <linux/sched/signal.h>
+
+#include "debug.h"
+#include "ntfs.h"
+#include "ntfs_fs.h"
+
+#define BITS_IN_SIZE_T (sizeof(size_t) * 8)
+
+/*
+ * fill_mask[i] - first i bits are '1' , i = 0,1,2,3,4,5,6,7,8
+ * fill_mask[i] = 0xFF >> (8-i)
+ */
+static const u8 fill_mask[] = { 0x00, 0x01, 0x03, 0x07, 0x0F,
+				0x1F, 0x3F, 0x7F, 0xFF };
+
+/*
+ * zero_mask[i] - first i bits are '0' , i = 0,1,2,3,4,5,6,7,8
+ * zero_mask[i] = 0xFF << i
+ */
+static const u8 zero_mask[] = { 0xFF, 0xFE, 0xFC, 0xF8, 0xF0,
+				0xE0, 0xC0, 0x80, 0x00 };
+
+/*
+ * are_bits_clear
+ *
+ * Returns true if all bits [bit, bit+nbits) are zeros "0"
+ */
+bool are_bits_clear(const ulong *lmap, size_t bit, size_t nbits)
+{
+	size_t pos = bit & 7;
+	const u8 *map = (u8 *)lmap + (bit >> 3);
+
+	if (pos) {
+		if (8 - pos >= nbits)
+			return !nbits || !(*map & fill_mask[pos + nbits] &
+					   zero_mask[pos]);
+
+		if (*map++ & zero_mask[pos])
+			return false;
+		nbits -= 8 - pos;
+	}
+
+	pos = ((size_t)map) & (sizeof(size_t) - 1);
+	if (pos) {
+		pos = sizeof(size_t) - pos;
+		if (nbits >= pos * 8) {
+			for (nbits -= pos * 8; pos; pos--, map++) {
+				if (*map)
+					return false;
+			}
+		}
+	}
+
+	for (pos = nbits / BITS_IN_SIZE_T; pos; pos--, map += sizeof(size_t)) {
+		if (*((size_t *)map))
+			return false;
+	}
+
+	for (pos = (nbits % BITS_IN_SIZE_T) >> 3; pos; pos--, map++) {
+		if (*map)
+			return false;
+	}
+
+	pos = nbits & 7;
+	if (pos && (*map & fill_mask[pos]))
+		return false;
+
+	// All bits are zero
+	return true;
+}
+
+/*
+ * are_bits_set
+ *
+ * Returns true if all bits [bit, bit+nbits) are ones "1"
+ */
+bool are_bits_set(const ulong *lmap, size_t bit, size_t nbits)
+{
+	u8 mask;
+	size_t pos = bit & 7;
+	const u8 *map = (u8 *)lmap + (bit >> 3);
+
+	if (pos) {
+		if (8 - pos >= nbits) {
+			mask = fill_mask[pos + nbits] & zero_mask[pos];
+			return !nbits || (*map & mask) == mask;
+		}
+
+		mask = zero_mask[pos];
+		if ((*map++ & mask) != mask)
+			return false;
+		nbits -= 8 - pos;
+	}
+
+	pos = ((size_t)map) & (sizeof(size_t) - 1);
+	if (pos) {
+		pos = sizeof(size_t) - pos;
+		if (nbits >= pos * 8) {
+			for (nbits -= pos * 8; pos; pos--, map++) {
+				if (*map != 0xFF)
+					return false;
+			}
+		}
+	}
+
+	for (pos = nbits / BITS_IN_SIZE_T; pos; pos--, map += sizeof(size_t)) {
+		if (*((size_t *)map) != MINUS_ONE_T)
+			return false;
+	}
+
+	for (pos = (nbits % BITS_IN_SIZE_T) >> 3; pos; pos--, map++) {
+		if (*map != 0xFF)
+			return false;
+	}
+
+	pos = nbits & 7;
+	if (pos) {
+		u8 mask = fill_mask[pos];
+
+		if ((*map & mask) != mask)
+			return false;
+	}
+
+	// All bits are ones
+	return true;
+}
diff --git a/fs/ntfs3/bitmap.c b/fs/ntfs3/bitmap.c
new file mode 100644
index 000000000000..82ab2d70bfa8
--- /dev/null
+++ b/fs/ntfs3/bitmap.c
@@ -0,0 +1,1516 @@ 
+// SPDX-License-Identifier: GPL-2.0
+/*
+ *  linux/fs/ntfs3/bitmap.c
+ *
+ * Copyright (C) 2019-2020 Paragon Software GmbH, All rights reserved.
+ *
+ */
+
+#include <linux/blkdev.h>
+#include <linux/buffer_head.h>
+#include <linux/fs.h>
+#include <linux/nls.h>
+#include <linux/sched/signal.h>
+
+#include "debug.h"
+#include "ntfs.h"
+#include "ntfs_fs.h"
+
+struct rb_node_key {
+	struct rb_node node;
+	size_t key;
+};
+
+/*
+ * Tree is sorted by start (key)
+ */
+struct e_node {
+	struct rb_node_key start; /* Tree sorted by start */
+	struct rb_node_key count; /* Tree sorted by len*/
+};
+
+static int wnd_rescan(struct wnd_bitmap *wnd);
+static struct buffer_head *wnd_map(struct wnd_bitmap *wnd, size_t iw);
+static bool wnd_is_free_hlp(struct wnd_bitmap *wnd, size_t bit, size_t bits);
+
+static inline u32 wnd_bits(const struct wnd_bitmap *wnd, size_t i)
+{
+	return i + 1 == wnd->nwnd ? wnd->bits_last : wnd->sb->s_blocksize * 8;
+}
+
+/*
+ * b_pos + b_len - biggest fragment
+ * Scan range [wpos wbits) window 'buf'
+ * Returns -1 if not found
+ */
+static size_t wnd_scan(const ulong *buf, size_t wbit, u32 wpos, u32 wend,
+		       size_t to_alloc, size_t *prev_tail, size_t *b_pos,
+		       size_t *b_len)
+{
+	while (wpos < wend) {
+		size_t free_len;
+		u32 free_bits, end;
+		u32 used = find_next_zero_bit(buf, wend, wpos);
+
+		if (used >= wend) {
+			if (*b_len < *prev_tail) {
+				*b_pos = wbit - *prev_tail;
+				*b_len = *prev_tail;
+			}
+
+			*prev_tail = 0;
+			return -1;
+		}
+
+		if (used > wpos) {
+			wpos = used;
+			if (*b_len < *prev_tail) {
+				*b_pos = wbit - *prev_tail;
+				*b_len = *prev_tail;
+			}
+
+			*prev_tail = 0;
+		}
+
+		/*
+		 * Now we have a fragment [wpos, wend) staring with 0
+		 */
+		end = wpos + to_alloc - *prev_tail;
+		free_bits = find_next_bit(buf, min(end, wend), wpos);
+
+		free_len = *prev_tail + free_bits - wpos;
+
+		if (*b_len < free_len) {
+			*b_pos = wbit + wpos - *prev_tail;
+			*b_len = free_len;
+		}
+
+		if (free_len >= to_alloc)
+			return wbit + wpos - *prev_tail;
+
+		if (free_bits >= wend) {
+			*prev_tail += free_bits - wpos;
+			return -1;
+		}
+
+		wpos = free_bits + 1;
+
+		*prev_tail = 0;
+	}
+
+	return -1;
+}
+
+/*
+ * wnd_close
+ *
+ *
+ */
+void wnd_close(struct wnd_bitmap *wnd)
+{
+	struct rb_node *node, *next;
+
+	if (wnd->free_bits != wnd->free_holder)
+		ntfs_free(wnd->free_bits);
+	run_close(&wnd->run);
+
+	node = rb_first(&wnd->start_tree);
+
+	while (node) {
+		next = rb_next(node);
+		rb_erase(node, &wnd->start_tree);
+		ntfs_free(rb_entry(node, struct e_node, start.node));
+		node = next;
+	}
+}
+
+static struct rb_node *rb_lookup(struct rb_root *root, size_t v)
+{
+	struct rb_node **p = &root->rb_node;
+	struct rb_node *r = NULL;
+
+	while (*p) {
+		struct rb_node_key *k;
+
+		k = rb_entry(*p, struct rb_node_key, node);
+		if (v < k->key) {
+			p = &(*p)->rb_left;
+		} else if (v > k->key) {
+			r = &k->node;
+			p = &(*p)->rb_right;
+		} else {
+			return &k->node;
+		}
+	}
+
+	return r;
+}
+
+/*
+ * rb_insert_count
+ *
+ * Helper function to insert special kind of 'count' tree
+ */
+static inline bool rb_insert_count(struct rb_root *root, struct e_node *e)
+{
+	struct rb_node **p = &root->rb_node;
+	struct rb_node *parent = NULL;
+	size_t e_ckey = e->count.key;
+	size_t e_skey = e->start.key;
+
+	while (*p) {
+		struct e_node *k =
+			rb_entry(parent = *p, struct e_node, count.node);
+
+		if (e_ckey > k->count.key) {
+			p = &(*p)->rb_left;
+		} else if (e_ckey < k->count.key) {
+			p = &(*p)->rb_right;
+		} else if (e_skey < k->start.key) {
+			p = &(*p)->rb_left;
+		} else if (e_skey > k->start.key) {
+			p = &(*p)->rb_right;
+		} else {
+			WARN_ON(1);
+			return false;
+		}
+	}
+
+	rb_link_node(&e->count.node, parent, p);
+	rb_insert_color(&e->count.node, root);
+	return true;
+}
+
+/*
+ * inline bool rb_insert_start
+ *
+ * Helper function to insert special kind of 'count' tree
+ */
+static inline bool rb_insert_start(struct rb_root *root, struct e_node *e)
+{
+	struct rb_node **p = &root->rb_node;
+	struct rb_node *parent = NULL;
+	size_t e_skey = e->start.key;
+
+	while (*p) {
+		struct e_node *k;
+
+		parent = *p;
+
+		k = rb_entry(parent, struct e_node, start.node);
+		if (e_skey < k->start.key) {
+			p = &(*p)->rb_left;
+		} else if (e_skey > k->start.key) {
+			p = &(*p)->rb_right;
+		} else {
+			WARN_ON(1);
+			return false;
+		}
+	}
+
+	rb_link_node(&e->start.node, parent, p);
+	rb_insert_color(&e->start.node, root);
+	return true;
+}
+
+#define NTFS_MAX_WND_EXTENTS (32u * 1024u)
+
+/*
+ * wnd_add_free_ext
+ *
+ * adds a new extent of free space
+ * build = 1 when building tree
+ */
+static void wnd_add_free_ext(struct wnd_bitmap *wnd, size_t bit, size_t len,
+			     bool build)
+{
+	struct e_node *e, *e0 = NULL;
+	size_t ib, end_in = bit + len;
+	struct rb_node *n;
+
+	if (build) {
+		/* Use extent_min to filter too short extents */
+		if (wnd->count >= NTFS_MAX_WND_EXTENTS &&
+		    len <= wnd->extent_min) {
+			wnd->uptodated = -1;
+			return;
+		}
+	} else {
+		/* Try to find extent before 'bit' */
+		n = rb_lookup(&wnd->start_tree, bit);
+
+		if (!n) {
+			n = rb_first(&wnd->start_tree);
+		} else {
+			e = rb_entry(n, struct e_node, start.node);
+			n = rb_next(n);
+			if (e->start.key + e->count.key == bit) {
+				/* Remove left */
+				bit = e->start.key;
+				len += e->count.key;
+				rb_erase(&e->start.node, &wnd->start_tree);
+				rb_erase(&e->count.node, &wnd->count_tree);
+				wnd->count -= 1;
+				e0 = e;
+			}
+		}
+
+		while (n) {
+			size_t next_end;
+
+			e = rb_entry(n, struct e_node, start.node);
+			next_end = e->start.key + e->count.key;
+			if (e->start.key > end_in)
+				break;
+
+			/* Remove right */
+			n = rb_next(n);
+			len += next_end - end_in;
+			end_in = next_end;
+			rb_erase(&e->start.node, &wnd->start_tree);
+			rb_erase(&e->count.node, &wnd->count_tree);
+			wnd->count -= 1;
+
+			if (!e0)
+				e0 = e;
+			else
+				ntfs_free(e);
+		}
+
+		if (wnd->uptodated != 1) {
+			/* Check bits before 'bit' */
+			ib = wnd->zone_bit == wnd->zone_end ||
+					     bit < wnd->zone_end ?
+				     0 :
+				     wnd->zone_end;
+
+			while (bit > ib && wnd_is_free_hlp(wnd, bit - 1, 1)) {
+				bit -= 1;
+				len += 1;
+			}
+
+			/* Check bits after 'end_in' */
+			ib = wnd->zone_bit == wnd->zone_end ||
+					     end_in > wnd->zone_bit ?
+				     wnd->nbits :
+				     wnd->zone_bit;
+
+			while (end_in < ib && wnd_is_free_hlp(wnd, end_in, 1)) {
+				end_in += 1;
+				len += 1;
+			}
+		}
+	}
+	/* Insert new fragment */
+	if (wnd->count >= NTFS_MAX_WND_EXTENTS) {
+		if (e0)
+			ntfs_free(e0);
+
+		wnd->uptodated = -1;
+
+		/* Compare with smallest fragment */
+		n = rb_last(&wnd->count_tree);
+		e = rb_entry(n, struct e_node, count.node);
+		if (len <= e->count.key)
+			goto out; /* Do not insert small fragments */
+
+		if (build) {
+			struct e_node *e2;
+
+			n = rb_prev(n);
+			e2 = rb_entry(n, struct e_node, count.node);
+			/* smallest fragment will be 'e2->count.key' */
+			wnd->extent_min = e2->count.key;
+		}
+
+		/* Replace smallest fragment by new one */
+		rb_erase(&e->start.node, &wnd->start_tree);
+		rb_erase(&e->count.node, &wnd->count_tree);
+		wnd->count -= 1;
+	} else {
+		e = e0 ? e0 : ntfs_alloc(sizeof(struct e_node), 0);
+		if (!e) {
+			wnd->uptodated = -1;
+			goto out;
+		}
+
+		if (build && len <= wnd->extent_min)
+			wnd->extent_min = len;
+	}
+	e->start.key = bit;
+	e->count.key = len;
+	if (len > wnd->extent_max)
+		wnd->extent_max = len;
+
+	rb_insert_start(&wnd->start_tree, e);
+	rb_insert_count(&wnd->count_tree, e);
+	wnd->count += 1;
+
+out:;
+}
+
+/*
+ * wnd_remove_free_ext
+ *
+ * removes a run from the cached free space
+ */
+static void wnd_remove_free_ext(struct wnd_bitmap *wnd, size_t bit, size_t len)
+{
+	struct rb_node *n, *n3;
+	struct e_node *e, *e3;
+	size_t end_in = bit + len;
+	size_t end3, end, new_key, new_len, max_new_len;
+
+	/* Try to find extent before 'bit' */
+	n = rb_lookup(&wnd->start_tree, bit);
+
+	if (!n)
+		return;
+
+	e = rb_entry(n, struct e_node, start.node);
+	end = e->start.key + e->count.key;
+
+	new_key = new_len = 0;
+	len = e->count.key;
+
+	/* Range [bit,end_in) must be inside 'e' or outside 'e' and 'n' */
+	if (e->start.key > bit)
+		;
+	else if (end_in <= end) {
+		/* Range [bit,end_in) inside 'e' */
+		new_key = end_in;
+		new_len = end - end_in;
+		len = bit - e->start.key;
+	} else if (bit > end) {
+		bool bmax = false;
+
+		n3 = rb_next(n);
+
+		while (n3) {
+			e3 = rb_entry(n3, struct e_node, start.node);
+			if (e3->start.key >= end_in)
+				break;
+
+			if (e3->count.key == wnd->extent_max)
+				bmax = true;
+
+			end3 = e3->start.key + e3->count.key;
+			if (end3 > end_in) {
+				e3->start.key = end_in;
+				rb_erase(&e3->count.node, &wnd->count_tree);
+				e3->count.key = end3 - end_in;
+				rb_insert_count(&wnd->count_tree, e3);
+				break;
+			}
+
+			n3 = rb_next(n3);
+			rb_erase(&e3->start.node, &wnd->start_tree);
+			rb_erase(&e3->count.node, &wnd->count_tree);
+			wnd->count -= 1;
+			ntfs_free(e3);
+		}
+		if (!bmax)
+			return;
+		n3 = rb_first(&wnd->count_tree);
+		wnd->extent_max =
+			n3 ? rb_entry(n3, struct e_node, count.node)->count.key :
+			     0;
+		return;
+	}
+
+	if (e->count.key != wnd->extent_max) {
+		;
+	} else if (rb_prev(&e->count.node)) {
+		;
+	} else {
+		n3 = rb_next(&e->count.node);
+		max_new_len = len > new_len ? len : new_len;
+		if (!n3) {
+			wnd->extent_max = max_new_len;
+		} else {
+			e3 = rb_entry(n3, struct e_node, count.node);
+			wnd->extent_max = max(e3->count.key, max_new_len);
+		}
+	}
+
+	if (!len) {
+		if (new_len) {
+			e->start.key = new_key;
+			rb_erase(&e->count.node, &wnd->count_tree);
+			e->count.key = new_len;
+			rb_insert_count(&wnd->count_tree, e);
+		} else {
+			rb_erase(&e->start.node, &wnd->start_tree);
+			rb_erase(&e->count.node, &wnd->count_tree);
+			wnd->count -= 1;
+			ntfs_free(e);
+		}
+		goto out;
+	}
+	rb_erase(&e->count.node, &wnd->count_tree);
+	e->count.key = len;
+	rb_insert_count(&wnd->count_tree, e);
+
+	if (!new_len)
+		goto out;
+
+	if (wnd->count >= NTFS_MAX_WND_EXTENTS) {
+		wnd->uptodated = -1;
+
+		/* Get minimal extent */
+		e = rb_entry(rb_last(&wnd->count_tree), struct e_node,
+			     count.node);
+		if (e->count.key > new_len)
+			goto out;
+
+		/* Replace minimum */
+		rb_erase(&e->start.node, &wnd->start_tree);
+		rb_erase(&e->count.node, &wnd->count_tree);
+		wnd->count -= 1;
+	} else {
+		e = ntfs_alloc(sizeof(struct e_node), 0);
+		if (!e)
+			wnd->uptodated = -1;
+	}
+
+	if (e) {
+		e->start.key = new_key;
+		e->count.key = new_len;
+		rb_insert_start(&wnd->start_tree, e);
+		rb_insert_count(&wnd->count_tree, e);
+		wnd->count += 1;
+	}
+
+out:
+	if (!wnd->count && 1 != wnd->uptodated)
+		wnd_rescan(wnd);
+}
+
+/*
+ * wnd_rescan
+ *
+ * Scan all bitmap. used while initialization.
+ */
+static int wnd_rescan(struct wnd_bitmap *wnd)
+{
+	int err = 0;
+	size_t prev_tail = 0;
+	struct super_block *sb = wnd->sb;
+	struct ntfs_sb_info *sbi = sb->s_fs_info;
+	u64 lbo, len = 0;
+	u32 blocksize = sb->s_blocksize;
+	u8 cluster_bits = sbi->cluster_bits;
+	const u32 ra_bytes = 512 * 1024;
+	const u32 ra_pages = ra_bytes >> PAGE_SHIFT;
+	u32 wbits = 8 * sb->s_blocksize;
+	u32 ra_mask = (ra_bytes >> sb->s_blocksize_bits) - 1;
+	struct address_space *mapping = sb->s_bdev->bd_inode->i_mapping;
+	u32 used, frb;
+	const ulong *buf;
+	size_t wpos, wbit, iw, vbo;
+	struct buffer_head *bh = NULL;
+	CLST lcn, clen;
+
+	wnd->uptodated = 0;
+	wnd->extent_max = 0;
+	wnd->extent_min = MINUS_ONE_T;
+	wnd->total_zeroes = 0;
+
+	vbo = 0;
+
+	for (iw = 0; iw < wnd->nwnd; iw++) {
+		if (iw + 1 == wnd->nwnd)
+			wbits = wnd->bits_last;
+
+		if (wnd->inited) {
+			if (!wnd->free_bits[iw]) {
+				/* all ones */
+				if (prev_tail) {
+					wnd_add_free_ext(wnd,
+							 vbo * 8 - prev_tail,
+							 prev_tail, true);
+					prev_tail = 0;
+				}
+				goto next_wnd;
+			}
+			if (wbits == wnd->free_bits[iw]) {
+				/* all zeroes */
+				prev_tail += wbits;
+				wnd->total_zeroes += wbits;
+				goto next_wnd;
+			}
+		}
+
+		if (!len) {
+			if (!run_lookup_entry(&wnd->run, vbo >> cluster_bits,
+					      &lcn, &clen, NULL)) {
+				err = -ENOENT;
+				goto out;
+			}
+
+			lbo = (u64)lcn << cluster_bits;
+			len = (u64)clen << cluster_bits;
+		}
+
+		if (!(iw & ra_mask))
+			page_cache_readahead_unbounded(
+				mapping, NULL, lbo >> PAGE_SHIFT, ra_pages, 0);
+
+		bh = ntfs_bread(sb, lbo >> sb->s_blocksize_bits);
+		if (!bh) {
+			err = -EIO;
+			goto out;
+		}
+
+		buf = (ulong *)bh->b_data;
+
+		used = __bitmap_weight(buf, wbits);
+		if (used < wbits) {
+			frb = wbits - used;
+			wnd->free_bits[iw] = frb;
+			wnd->total_zeroes += frb;
+		}
+
+		wpos = 0;
+		wbit = vbo * 8;
+
+		if (wbit + wbits > wnd->nbits)
+			wbits = wnd->nbits - wbit;
+
+		do {
+			used = find_next_zero_bit(buf, wbits, wpos);
+
+			if (used > wpos && prev_tail) {
+				wnd_add_free_ext(wnd, wbit + wpos - prev_tail,
+						 prev_tail, true);
+				prev_tail = 0;
+			}
+
+			wpos = used;
+
+			if (wpos >= wbits) {
+				/* No free blocks */
+				prev_tail = 0;
+				break;
+			}
+
+			frb = find_next_bit(buf, wbits, wpos);
+			if (frb >= wbits) {
+				/* keep last free block */
+				prev_tail += frb - wpos;
+				break;
+			}
+
+			wnd_add_free_ext(wnd, wbit + wpos - prev_tail,
+					 frb + prev_tail - wpos, true);
+
+			/* Skip free block and first '1' */
+			wpos = frb + 1;
+			/* Reset previous tail */
+			prev_tail = 0;
+		} while (wpos < wbits);
+
+next_wnd:
+
+		if (bh)
+			put_bh(bh);
+		bh = NULL;
+
+		vbo += blocksize;
+		if (len) {
+			len -= blocksize;
+			lbo += blocksize;
+		}
+	}
+
+	/* Add last block */
+	if (prev_tail)
+		wnd_add_free_ext(wnd, wnd->nbits - prev_tail, prev_tail, true);
+
+	/*
+	 * Before init cycle wnd->uptodated was 0
+	 * If any errors or limits occurs while initialization then
+	 * wnd->uptodated will be -1
+	 * If 'uptodated' is still 0 then Tree is really updated
+	 */
+	if (!wnd->uptodated)
+		wnd->uptodated = 1;
+
+	if (wnd->zone_bit != wnd->zone_end) {
+		size_t zlen = wnd->zone_end - wnd->zone_bit;
+
+		wnd->zone_end = wnd->zone_bit;
+		wnd_zone_set(wnd, wnd->zone_bit, zlen);
+	}
+
+out:
+	return err;
+}
+
+/*
+ * wnd_init
+ */
+int wnd_init(struct wnd_bitmap *wnd, struct super_block *sb, size_t nbits)
+{
+	int err;
+	u32 blocksize = sb->s_blocksize;
+	u32 wbits = blocksize * 8;
+
+	init_rwsem(&wnd->rw_lock);
+
+	wnd->sb = sb;
+	wnd->nbits = nbits;
+	wnd->total_zeroes = nbits;
+	wnd->extent_max = MINUS_ONE_T;
+	wnd->zone_bit = wnd->zone_end = 0;
+	wnd->nwnd = bytes_to_block(sb, bitmap_size(nbits));
+	wnd->bits_last = nbits & (wbits - 1);
+	if (!wnd->bits_last)
+		wnd->bits_last = wbits;
+
+	if (wnd->nwnd <= ARRAY_SIZE(wnd->free_holder)) {
+		wnd->free_bits = wnd->free_holder;
+	} else {
+		wnd->free_bits = ntfs_alloc(wnd->nwnd * sizeof(u16), 1);
+		if (!wnd->free_bits)
+			return -ENOMEM;
+	}
+
+	err = wnd_rescan(wnd);
+	if (err)
+		return err;
+
+	wnd->inited = true;
+
+	return 0;
+}
+
+/*
+ * wnd_map
+ *
+ * call sb_bread for requested window
+ */
+static struct buffer_head *wnd_map(struct wnd_bitmap *wnd, size_t iw)
+{
+	size_t vbo;
+	CLST lcn, clen;
+	struct super_block *sb = wnd->sb;
+	struct ntfs_sb_info *sbi;
+	struct buffer_head *bh;
+	u64 lbo;
+
+	sbi = sb->s_fs_info;
+	vbo = (u64)iw << sb->s_blocksize_bits;
+
+	if (!run_lookup_entry(&wnd->run, vbo >> sbi->cluster_bits, &lcn, &clen,
+			      NULL)) {
+		return ERR_PTR(-ENOENT);
+	}
+
+	lbo = ((u64)lcn << sbi->cluster_bits) + (vbo & sbi->cluster_mask);
+
+	bh = ntfs_bread(wnd->sb, lbo >> sb->s_blocksize_bits);
+
+	if (!bh)
+		return ERR_PTR(-EIO);
+
+	return bh;
+}
+
+/*
+ * wnd_set_free
+ *
+ * Marks the bits range from bit to bit + bits as free
+ */
+int wnd_set_free(struct wnd_bitmap *wnd, size_t bit, size_t bits)
+{
+	int err = 0;
+	struct super_block *sb = wnd->sb;
+	size_t bits0 = bits;
+	u32 wbits = 8 * sb->s_blocksize;
+	size_t iw = bit >> (sb->s_blocksize_bits + 3);
+	u32 wbit = bit & (wbits - 1);
+	struct buffer_head *bh;
+
+	while (iw < wnd->nwnd && bits) {
+		u32 tail, op;
+		ulong *buf;
+
+		if (iw + 1 == wnd->nwnd)
+			wbits = wnd->bits_last;
+
+		tail = wbits - wbit;
+		op = tail < bits ? tail : bits;
+
+		bh = wnd_map(wnd, iw);
+		if (IS_ERR(bh)) {
+			err = PTR_ERR(bh);
+			break;
+		}
+
+		buf = (ulong *)bh->b_data;
+
+		lock_buffer(bh);
+
+		__bitmap_clear(buf, wbit, op);
+
+		wnd->free_bits[iw] += op;
+
+		set_buffer_uptodate(bh);
+		mark_buffer_dirty(bh);
+		unlock_buffer(bh);
+		put_bh(bh);
+
+		wnd->total_zeroes += op;
+		bits -= op;
+		wbit = 0;
+		iw += 1;
+	}
+
+	wnd_add_free_ext(wnd, bit, bits0, false);
+
+	return err;
+}
+
+/*
+ * wnd_set_used
+ *
+ * Marks the bits range from bit to bit + bits as used
+ */
+int wnd_set_used(struct wnd_bitmap *wnd, size_t bit, size_t bits)
+{
+	int err = 0;
+	struct super_block *sb = wnd->sb;
+	size_t bits0 = bits;
+	size_t iw = bit >> (sb->s_blocksize_bits + 3);
+	u32 wbits = 8 * sb->s_blocksize;
+	u32 wbit = bit & (wbits - 1);
+	struct buffer_head *bh;
+
+	while (iw < wnd->nwnd && bits) {
+		u32 tail, op;
+		ulong *buf;
+
+		if (unlikely(iw + 1 == wnd->nwnd))
+			wbits = wnd->bits_last;
+
+		tail = wbits - wbit;
+		op = tail < bits ? tail : bits;
+
+		bh = wnd_map(wnd, iw);
+		if (IS_ERR(bh)) {
+			err = PTR_ERR(bh);
+			break;
+		}
+		buf = (ulong *)bh->b_data;
+
+		lock_buffer(bh);
+
+		__bitmap_set(buf, wbit, op);
+		wnd->free_bits[iw] -= op;
+
+		set_buffer_uptodate(bh);
+		mark_buffer_dirty(bh);
+		unlock_buffer(bh);
+		put_bh(bh);
+
+		wnd->total_zeroes -= op;
+		bits -= op;
+		wbit = 0;
+		iw += 1;
+	}
+
+	if (!RB_EMPTY_ROOT(&wnd->start_tree))
+		wnd_remove_free_ext(wnd, bit, bits0);
+
+	return err;
+}
+
+/*
+ * wnd_is_free_hlp
+ *
+ * Returns true if all clusters [bit, bit+bits) are free (bitmap only)
+ */
+static bool wnd_is_free_hlp(struct wnd_bitmap *wnd, size_t bit, size_t bits)
+{
+	struct super_block *sb = wnd->sb;
+	size_t iw = bit >> (sb->s_blocksize_bits + 3);
+	u32 wbits = 8 * sb->s_blocksize;
+	u32 wbit = bit & (wbits - 1);
+
+	while (iw < wnd->nwnd && bits) {
+		u32 tail, op;
+
+		if (unlikely(iw + 1 == wnd->nwnd))
+			wbits = wnd->bits_last;
+
+		tail = wbits - wbit;
+		op = tail < bits ? tail : bits;
+
+		if (wbits != wnd->free_bits[iw]) {
+			bool ret;
+			struct buffer_head *bh = wnd_map(wnd, iw);
+
+			if (IS_ERR(bh))
+				return false;
+
+			ret = are_bits_clear((ulong *)bh->b_data, wbit, op);
+
+			put_bh(bh);
+			if (!ret)
+				return false;
+		}
+
+		bits -= op;
+		wbit = 0;
+		iw += 1;
+	}
+
+	return true;
+}
+
+/*
+ * wnd_is_free
+ *
+ * Returns true if all clusters [bit, bit+bits) are free
+ */
+bool wnd_is_free(struct wnd_bitmap *wnd, size_t bit, size_t bits)
+{
+	bool ret;
+	struct rb_node *n;
+	size_t end;
+	struct e_node *e;
+
+	if (RB_EMPTY_ROOT(&wnd->start_tree))
+		goto use_wnd;
+
+	n = rb_lookup(&wnd->start_tree, bit);
+	if (!n)
+		goto use_wnd;
+
+	e = rb_entry(n, struct e_node, start.node);
+
+	end = e->start.key + e->count.key;
+
+	if (bit < end && bit + bits <= end)
+		return true;
+
+use_wnd:
+	ret = wnd_is_free_hlp(wnd, bit, bits);
+
+	return ret;
+}
+
+/*
+ * wnd_is_used
+ *
+ * Returns true if all clusters [bit, bit+bits) are used
+ */
+bool wnd_is_used(struct wnd_bitmap *wnd, size_t bit, size_t bits)
+{
+	bool ret = false;
+	struct super_block *sb = wnd->sb;
+	size_t iw = bit >> (sb->s_blocksize_bits + 3);
+	u32 wbits = 8 * sb->s_blocksize;
+	u32 wbit = bit & (wbits - 1);
+	size_t end;
+	struct rb_node *n;
+	struct e_node *e;
+
+	if (RB_EMPTY_ROOT(&wnd->start_tree))
+		goto use_wnd;
+
+	end = bit + bits;
+	n = rb_lookup(&wnd->start_tree, end - 1);
+	if (!n)
+		goto use_wnd;
+
+	e = rb_entry(n, struct e_node, start.node);
+	if (e->start.key + e->count.key > bit)
+		return false;
+
+use_wnd:
+	while (iw < wnd->nwnd && bits) {
+		u32 tail, op;
+
+		if (unlikely(iw + 1 == wnd->nwnd))
+			wbits = wnd->bits_last;
+
+		tail = wbits - wbit;
+		op = tail < bits ? tail : bits;
+
+		if (wnd->free_bits[iw]) {
+			bool ret;
+			struct buffer_head *bh = wnd_map(wnd, iw);
+
+			if (IS_ERR(bh))
+				goto out;
+
+			ret = are_bits_set((ulong *)bh->b_data, wbit, op);
+			put_bh(bh);
+			if (!ret)
+				goto out;
+		}
+
+		bits -= op;
+		wbit = 0;
+		iw += 1;
+	}
+	ret = true;
+
+out:
+	return ret;
+}
+
+/*
+ * wnd_find
+ * - flags - BITMAP_FIND_XXX flags
+ *
+ * looks for free space
+ * Returns 0 if not found
+ */
+size_t wnd_find(struct wnd_bitmap *wnd, size_t to_alloc, size_t hint,
+		size_t flags, size_t *allocated)
+{
+	struct super_block *sb;
+	u32 wbits, wpos, wzbit, wzend;
+	size_t fnd, max_alloc, b_len, b_pos;
+	size_t iw, prev_tail, nwnd, wbit, ebit, zbit, zend;
+	size_t to_alloc0 = to_alloc;
+	const ulong *buf;
+	const struct e_node *e;
+	const struct rb_node *pr, *cr;
+	u8 log2_bits;
+	bool fbits_valid;
+	struct buffer_head *bh;
+
+	/* fast checking for available free space */
+	if (flags & BITMAP_FIND_FULL) {
+		size_t zeroes = wnd_zeroes(wnd);
+
+		zeroes -= wnd->zone_end - wnd->zone_bit;
+		if (zeroes < to_alloc0)
+			goto no_space;
+
+		if (to_alloc0 > wnd->extent_max)
+			goto no_space;
+	} else {
+		if (to_alloc > wnd->extent_max)
+			to_alloc = wnd->extent_max;
+	}
+
+	if (wnd->zone_bit <= hint && hint < wnd->zone_end)
+		hint = wnd->zone_end;
+
+	max_alloc = wnd->nbits;
+	b_len = b_pos = 0;
+
+	if (hint >= max_alloc)
+		hint = 0;
+
+	if (RB_EMPTY_ROOT(&wnd->start_tree)) {
+		if (wnd->uptodated == 1) {
+			/* extents tree is updated -> no free space */
+			goto no_space;
+		}
+		goto scan_bitmap;
+	}
+
+	e = NULL;
+	if (!hint)
+		goto allocate_biggest;
+
+	/* Use hint: enumerate extents by start >= hint */
+	pr = NULL;
+	cr = wnd->start_tree.rb_node;
+
+	for (;;) {
+		e = rb_entry(cr, struct e_node, start.node);
+
+		if (e->start.key == hint)
+			break;
+
+		if (e->start.key < hint) {
+			pr = cr;
+			cr = cr->rb_right;
+			if (!cr)
+				break;
+			continue;
+		}
+
+		cr = cr->rb_left;
+		if (!cr) {
+			e = pr ? rb_entry(pr, struct e_node, start.node) : NULL;
+			break;
+		}
+	}
+
+	if (!e)
+		goto allocate_biggest;
+
+	if (e->start.key + e->count.key > hint) {
+		/* We have found extension with 'hint' inside */
+		size_t len = e->start.key + e->count.key - hint;
+
+		if (len >= to_alloc && hint + to_alloc <= max_alloc) {
+			fnd = hint;
+			goto found;
+		}
+
+		if (!(flags & BITMAP_FIND_FULL)) {
+			if (len > to_alloc)
+				len = to_alloc;
+
+			if (hint + len <= max_alloc) {
+				fnd = hint;
+				to_alloc = len;
+				goto found;
+			}
+		}
+	}
+
+allocate_biggest:
+
+	/* Allocate from biggest free extent */
+	e = rb_entry(rb_first(&wnd->count_tree), struct e_node, count.node);
+	if (e->count.key != wnd->extent_max)
+		wnd->extent_max = e->count.key;
+
+	if (e->count.key < max_alloc) {
+		if (e->count.key >= to_alloc) {
+			;
+		} else if (flags & BITMAP_FIND_FULL) {
+			if (e->count.key < to_alloc0) {
+				/* Biggest free block is less then requested */
+				goto no_space;
+			}
+			to_alloc = e->count.key;
+		} else if (-1 != wnd->uptodated) {
+			to_alloc = e->count.key;
+		} else {
+			/* Check if we can use more bits */
+			size_t op, max_check;
+			struct rb_root start_tree;
+
+			memcpy(&start_tree, &wnd->start_tree,
+			       sizeof(struct rb_root));
+			memset(&wnd->start_tree, 0, sizeof(struct rb_root));
+
+			max_check = e->start.key + to_alloc;
+			if (max_check > max_alloc)
+				max_check = max_alloc;
+			for (op = e->start.key + e->count.key; op < max_check;
+			     op++) {
+				if (!wnd_is_free(wnd, op, 1))
+					break;
+			}
+			memcpy(&wnd->start_tree, &start_tree,
+			       sizeof(struct rb_root));
+			to_alloc = op - e->start.key;
+		}
+
+		/* Prepare to return */
+		fnd = e->start.key;
+		if (e->start.key + to_alloc > max_alloc)
+			to_alloc = max_alloc - e->start.key;
+		goto found;
+	}
+
+	if (wnd->uptodated == 1) {
+		/* extents tree is updated -> no free space */
+		goto no_space;
+	}
+
+	b_len = e->count.key;
+	b_pos = e->start.key;
+
+scan_bitmap:
+	sb = wnd->sb;
+	log2_bits = sb->s_blocksize_bits + 3;
+
+	/* At most two ranges [hint, max_alloc) + [0, hint) */
+Again:
+
+	/* TODO: optimize request for case nbits > wbits */
+	iw = hint >> log2_bits;
+	wbits = sb->s_blocksize * 8;
+	wpos = hint & (wbits - 1);
+	prev_tail = 0;
+	fbits_valid = true;
+
+	if (max_alloc == wnd->nbits) {
+		nwnd = wnd->nwnd;
+	} else {
+		size_t t = max_alloc + wbits - 1;
+
+		nwnd = likely(t > max_alloc) ? (t >> log2_bits) : wnd->nwnd;
+	}
+
+	/* Enumerate all windows */
+	for (; iw < nwnd; iw++) {
+		wbit = iw << log2_bits;
+
+		if (!wnd->free_bits[iw]) {
+			if (prev_tail > b_len) {
+				b_pos = wbit - prev_tail;
+				b_len = prev_tail;
+			}
+
+			/* Skip full used window */
+			prev_tail = 0;
+			wpos = 0;
+			continue;
+		}
+
+		if (unlikely(iw + 1 == nwnd)) {
+			if (max_alloc == wnd->nbits) {
+				wbits = wnd->bits_last;
+			} else {
+				size_t t = max_alloc & (wbits - 1);
+
+				if (t) {
+					wbits = t;
+					fbits_valid = false;
+				}
+			}
+		}
+
+		if (wnd->zone_end > wnd->zone_bit) {
+			ebit = wbit + wbits;
+			zbit = max(wnd->zone_bit, wbit);
+			zend = min(wnd->zone_end, ebit);
+
+			/* Here we have a window [wbit, ebit) and zone [zbit, zend) */
+			if (zend <= zbit) {
+				/* Zone does not overlap window */
+			} else {
+				wzbit = zbit - wbit;
+				wzend = zend - wbit;
+
+				/* Zone overlaps window */
+				if (wnd->free_bits[iw] == wzend - wzbit) {
+					prev_tail = 0;
+					wpos = 0;
+					continue;
+				}
+
+				/* Scan two ranges window: [wbit, zbit) and [zend, ebit) */
+				bh = wnd_map(wnd, iw);
+
+				if (IS_ERR(bh)) {
+					/* TODO: error */
+					prev_tail = 0;
+					wpos = 0;
+					continue;
+				}
+
+				buf = (ulong *)bh->b_data;
+
+				/* Scan range [wbit, zbit) */
+				if (wpos < wzbit) {
+					/* Scan range [wpos, zbit) */
+					fnd = wnd_scan(buf, wbit, wpos, wzbit,
+						       to_alloc, &prev_tail,
+						       &b_pos, &b_len);
+					if (fnd != MINUS_ONE_T) {
+						put_bh(bh);
+						goto found;
+					}
+				}
+
+				prev_tail = 0;
+
+				/* Scan range [zend, ebit) */
+				if (wzend < wbits) {
+					fnd = wnd_scan(buf, wbit,
+						       max(wzend, wpos), wbits,
+						       to_alloc, &prev_tail,
+						       &b_pos, &b_len);
+					if (fnd != MINUS_ONE_T) {
+						put_bh(bh);
+						goto found;
+					}
+				}
+
+				wpos = 0;
+				put_bh(bh);
+				continue;
+			}
+		}
+
+		/* Current window does not overlap zone */
+		if (!wpos && fbits_valid && wnd->free_bits[iw] == wbits) {
+			/* window is empty */
+			if (prev_tail + wbits >= to_alloc) {
+				fnd = wbit + wpos - prev_tail;
+				goto found;
+			}
+
+			/* Increase 'prev_tail' and process next window */
+			prev_tail += wbits;
+			wpos = 0;
+			continue;
+		}
+
+		/* read window */
+		bh = wnd_map(wnd, iw);
+		if (IS_ERR(bh)) {
+			// TODO: error
+			prev_tail = 0;
+			wpos = 0;
+			continue;
+		}
+
+		buf = (ulong *)bh->b_data;
+
+		/* Scan range [wpos, eBits) */
+		fnd = wnd_scan(buf, wbit, wpos, wbits, to_alloc, &prev_tail,
+			       &b_pos, &b_len);
+		put_bh(bh);
+		if (fnd != MINUS_ONE_T)
+			goto found;
+	}
+
+	if (b_len < prev_tail) {
+		/* The last fragment */
+		b_len = prev_tail;
+		b_pos = max_alloc - prev_tail;
+	}
+
+	if (hint) {
+		/*
+		 * We have scanned range [hint max_alloc)
+		 * Prepare to scan range [0 hint + to_alloc)
+		 */
+		size_t nextmax = hint + to_alloc;
+
+		if (likely(nextmax >= hint) && nextmax < max_alloc)
+			max_alloc = nextmax;
+		hint = 0;
+		goto Again;
+	}
+
+	if (!b_len)
+		goto no_space;
+
+	wnd->extent_max = b_len;
+
+	if (flags & BITMAP_FIND_FULL)
+		goto no_space;
+
+	fnd = b_pos;
+	to_alloc = b_len;
+
+found:
+	if (flags & BITMAP_FIND_MARK_AS_USED) {
+		/* TODO optimize remove extent (pass 'e'?) */
+		if (wnd_set_used(wnd, fnd, to_alloc))
+			goto no_space;
+	} else if (wnd->extent_max != MINUS_ONE_T &&
+		   to_alloc > wnd->extent_max) {
+		wnd->extent_max = to_alloc;
+	}
+
+	*allocated = fnd;
+	return to_alloc;
+
+no_space:
+	return 0;
+}
+
+/*
+ * wnd_extend
+ *
+ * Extend bitmap ($MFT bitmap)
+ */
+int wnd_extend(struct wnd_bitmap *wnd, size_t new_bits)
+{
+	int err;
+	struct super_block *sb = wnd->sb;
+	struct ntfs_sb_info *sbi = sb->s_fs_info;
+	u32 blocksize = sb->s_blocksize;
+	u32 wbits = blocksize * 8;
+	u32 b0, new_last;
+	size_t bits, iw, new_wnd;
+	size_t old_bits = wnd->nbits;
+	u16 *new_free;
+
+	if (new_bits <= old_bits)
+		return -EINVAL;
+
+	/* align to 8 byte boundary */
+	new_wnd = bytes_to_block(sb, bitmap_size(new_bits));
+	new_last = new_bits & (wbits - 1);
+	if (!new_last)
+		new_last = wbits;
+
+	if (new_wnd != wnd->nwnd) {
+		if (new_wnd <= ARRAY_SIZE(wnd->free_holder)) {
+			new_free = wnd->free_holder;
+		} else {
+			new_free = ntfs_alloc(new_wnd * sizeof(u16), 0);
+			if (!new_free)
+				return -ENOMEM;
+		}
+
+		if (new_free != wnd->free_bits)
+			memcpy(new_free, wnd->free_bits,
+			       wnd->nwnd * sizeof(short));
+		memset(new_free + wnd->nwnd, 0,
+		       (new_wnd - wnd->nwnd) * sizeof(short));
+		if (wnd->free_bits != wnd->free_holder)
+			ntfs_free(wnd->free_bits);
+
+		wnd->free_bits = new_free;
+	}
+
+	/* Zero bits [old_bits,new_bits) */
+	bits = new_bits - old_bits;
+	b0 = old_bits & (wbits - 1);
+
+	for (iw = old_bits >> (sb->s_blocksize_bits + 3); bits; iw += 1) {
+		u32 op;
+		size_t frb;
+		u64 vbo, lbo, bytes;
+		struct buffer_head *bh;
+		ulong *buf;
+
+		if (iw + 1 == new_wnd)
+			wbits = new_last;
+
+		op = b0 + bits > wbits ? wbits - b0 : bits;
+		vbo = (u64)iw * blocksize;
+
+		err = ntfs_vbo_to_lbo(sbi, &wnd->run, vbo, &lbo, &bytes);
+		if (err)
+			break;
+
+		bh = ntfs_bread(sb, lbo >> sb->s_blocksize_bits);
+		if (!bh)
+			return -EIO;
+
+		lock_buffer(bh);
+		buf = (ulong *)bh->b_data;
+
+		__bitmap_clear(buf, b0, blocksize * 8 - b0);
+		frb = wbits - __bitmap_weight(buf, wbits);
+		wnd->total_zeroes += frb - wnd->free_bits[iw];
+		wnd->free_bits[iw] = frb;
+
+		set_buffer_uptodate(bh);
+		mark_buffer_dirty(bh);
+		unlock_buffer(bh);
+		/*err = sync_dirty_buffer(bh);*/
+
+		b0 = 0;
+		bits -= op;
+	}
+
+	wnd->nbits = new_bits;
+	wnd->nwnd = new_wnd;
+	wnd->bits_last = new_last;
+
+	wnd_add_free_ext(wnd, old_bits, new_bits - old_bits, false);
+
+	return 0;
+}
+
+/*
+ * wnd_zone_set
+ */
+void wnd_zone_set(struct wnd_bitmap *wnd, size_t lcn, size_t len)
+{
+	size_t zlen;
+
+	zlen = wnd->zone_end - wnd->zone_bit;
+	if (zlen)
+		wnd_add_free_ext(wnd, wnd->zone_bit, zlen, false);
+
+	if (!RB_EMPTY_ROOT(&wnd->start_tree) && len)
+		wnd_remove_free_ext(wnd, lcn, len);
+
+	wnd->zone_bit = lcn;
+	wnd->zone_end = lcn + len;
+}
+
+int ntfs_trim_fs(struct ntfs_sb_info *sbi, struct fstrim_range *range)
+{
+	int err = 0;
+	struct super_block *sb = sbi->sb;
+	struct wnd_bitmap *wnd = &sbi->used.bitmap;
+	u32 wbits = 8 * sb->s_blocksize;
+	CLST len = 0, lcn = 0, done = 0;
+	CLST minlen = bytes_to_cluster(sbi, range->minlen);
+	CLST lcn_from = bytes_to_cluster(sbi, range->start);
+	size_t iw = lcn_from >> (sb->s_blocksize_bits + 3);
+	u32 wbit = lcn_from & (wbits - 1);
+	const ulong *buf;
+	CLST lcn_to;
+
+	if (!minlen)
+		minlen = 1;
+
+	if (range->len == (u64)-1)
+		lcn_to = wnd->nbits;
+	else
+		lcn_to = bytes_to_cluster(sbi, range->start + range->len);
+
+	down_read_nested(&wnd->rw_lock, BITMAP_MUTEX_CLUSTERS);
+
+	for (; iw < wnd->nbits; iw++, wbit = 0) {
+		CLST lcn_wnd = iw * wbits;
+		struct buffer_head *bh;
+
+		if (lcn_wnd > lcn_to)
+			break;
+
+		if (!wnd->free_bits[iw])
+			continue;
+
+		if (iw + 1 == wnd->nwnd)
+			wbits = wnd->bits_last;
+
+		if (lcn_wnd + wbits > lcn_to)
+			wbits = lcn_to - lcn_wnd;
+
+		bh = wnd_map(wnd, iw);
+		if (IS_ERR(bh)) {
+			err = PTR_ERR(bh);
+			break;
+		}
+
+		buf = (ulong *)bh->b_data;
+
+		for (; wbit < wbits; wbit++) {
+			if (!test_bit(wbit, buf)) {
+				if (!len)
+					lcn = lcn_wnd + wbit;
+				len += 1;
+				continue;
+			}
+			if (len >= minlen) {
+				err = ntfs_discard(sbi, lcn, len);
+				if (err)
+					goto out;
+				done += len;
+			}
+			len = 0;
+		}
+		put_bh(bh);
+	}
+
+	/* Process the last fragment */
+	if (len >= minlen) {
+		err = ntfs_discard(sbi, lcn, len);
+		if (err)
+			goto out;
+		done += len;
+	}
+
+out:
+	range->len = (u64)done << sbi->cluster_bits;
+
+	up_read(&wnd->rw_lock);
+
+	return err;
+}