diff mbox

RFC: direct MTD support for SquashFS

Message ID 8739zqpxxw.fsf@tac.ki.iif.hu
State RFC
Headers show

Commit Message

Ferenc Wagner March 23, 2010, 8:47 p.m. UTC
Now with the patch series, sorry.

Ferenc Wagner <wferi@niif.hu> writes:

> I've got one more patch, which I forgot to export, to pull out the
> common logic from the backend init functions back into squashfs_read_data().
> With the bdev backend, that entails reading the first block twice in a
> row most of the time.  This again could be worked around by extending
> the backend interface, but I'm not sure if it's worth it.

Here it is.  I also corrected the name of SQUASHFS_METADATA_SIZE, so it
may as well compile now.

Comments

Phillip Lougher March 24, 2010, 5:23 a.m. UTC | #1
Ferenc Wagner wrote:
> Now with the patch series, sorry.
> 
> Ferenc Wagner <wferi@niif.hu> writes:
> 
>> I've got one more patch, which I forgot to export, to pull out the
>> common logic from the backend init functions back into squashfs_read_data().
>> With the bdev backend, that entails reading the first block twice in a
>> row most of the time.  This again could be worked around by extending
>> the backend interface, but I'm not sure if it's worth it.
> 
> Here it is.  I also corrected the name of SQUASHFS_METADATA_SIZE, so it
> may as well compile now.
> 


Thanks for your patches.

First things first.  Patch sending etiquette :-)

1. Please don't send separate git commits in one big email,  it makes them
    hard to cross-reference (I lost count of the times I had to repeatedly
    scroll though the email to check an earlier commit).

2. Please don't send a patch series consisting of your development process.
    Reviewing takes effort, having reviewed one commit to find it reworked
    in a second commit, and then reworked again, as the code evolves from
    a -> b -> c-> d is irritating, it makes the final solution harder to
    evaluate (because you have to keep all the changes in your head), and
    wastes my time.

3. Incremental patches are valid if they break up a patch series into easily
    digestible changes which can be separately understood, but not if they're
    just reworking your code as development progresses...

OK, now for the specific comments.

Frameworks are supposed to make the code cleaner, more understandable,
and generally better.  Unfortunately, I see no evidence of that in your
patch.  Sorry to be blunt, but there it is.

The backend has seven functions!

+       void	*(*init)(struct squashfs_sb_info *, u64, size_t);
+	void	(*free)(struct squashfs_sb_info *);
+	ssize_t	(*read)(struct squashfs_sb_info *, void **, size_t);
+	int	(*probe)(struct file_system_type *, int, const char *,
+				void*, struct vfsmount *);
+	void	(*kill)(struct squashfs_sb_info *);
+	loff_t  (*size)(const struct super_block *);
+	const char *(*devname)(const struct super_block *, char *);

We move from a single read() function to requiring seven functions to do
the same work, just to add mtd support...

Reading the superblock changes into a 6 function deep nightmare
squashfs_find_backend -> probe -> get_sb_dev -> fill_bdev_super ->
squashfs_fill_super -> add_backend

I find the current 3 function deep situation forced by VFS already a mess,
squashfs_get_sb -> get_sb_bdev -> squashfs_fill_super

Reading a block now takes 3 function calls, init, read, and free.  Plus in
your last commit you make the huge mistake of preferring code elegance over
performance, and resort to calling init, read, and free *twice*, the first
just to read *two* bytes (yes, 2 bytes).  Why do you think the code was
coded that way in the first place?  A fit of absent mindedness perhaps?  No, for
performance reasons.

Secondly you make the classic mistake of concentrating on what you want to do
(add MTD), and not giving a damn about anything else.  You've totally broken all
the read optimisations in zlib decompression for block devices.  Buffer_heads
are deliberately passed directly to the zlib decompressor to allow it to
optimise performance (pass the buffer_head directly to zlib etc.).  Buffer_heads
are exposed to the decompressors without crappy go-between wrappers deliberately.

Unfortunately there are lots of other instances where you've eliminated carefully
optimised code.

That's another of my major issues, the patch seems hugely gratuitous, it
touches a huge amount of files/code it shouldn't do, much of this slicing up
optimisied heavily tested and fragile code into other files/functions hither
and thither.  Unfortunately much of this code took a long time to get
right, and suffered numerous unanticipated edge conditions/fsfuzzer
triggered bugs.  I only touch this code with caution and nothing like to
the extent you've subjected it to.

In short this doesn't pass my quality threshold or the make the minimum changes
necessary threshold.  I wouldn't consider asking Linus to merge this.

BTW as a comparison, I have added MTD support to Squashfs twice in the
past, *with* bad block support.  The latest has a diff patch of ~500 lines,
with far less complexity and code changes necessary.

BTW2 as evidenced by your FIXME comments against my code in your patch,
you really have to get over the notion that if you don't understand it,
the code must be wrong.

Cheers

Phillip
Peter Korsgaard March 24, 2010, 6:35 a.m. UTC | #2
>>>>> "Phillip" == Phillip Lougher <phillip@lougher.demon.co.uk> writes:

Hi,

 Phillip> BTW as a comparison, I have added MTD support to Squashfs
 Phillip> twice in the past, *with* bad block support.  The latest has a
 Phillip> diff patch of ~500 lines, with far less complexity and code
 Phillip> changes necessary.

Could we merge that then please?

Again, what kind of size reduction can we expect from getting rid of the
block layer?
Ferenc Wagner March 24, 2010, 11:28 a.m. UTC | #3
Peter Korsgaard <jacmet@sunsite.dk> writes:

>>>>>> "Phillip" == Phillip Lougher <phillip@lougher.demon.co.uk> writes:
>
>  Phillip> BTW as a comparison, I have added MTD support to Squashfs
>  Phillip> twice in the past, *with* bad block support.  The latest has a
>  Phillip> diff patch of ~500 lines, with far less complexity and code
>  Phillip> changes necessary.
>
> Again, what kind of size reduction can we expect from getting rid of
> the block layer?

My size-optimized OpenWRT vmlinux shrank ~170 kB after disabling
CONFIG_BLOCK.  Similarly for the memory consumption.
Peter Korsgaard March 24, 2010, 11:35 a.m. UTC | #4
>>>>> "Ferenc" == Ferenc Wagner <wferi@niif.hu> writes:

Hi,

 >> Again, what kind of size reduction can we expect from getting rid of
 >> the block layer?

 Ferenc> My size-optimized OpenWRT vmlinux shrank ~170 kB after disabling
 Ferenc> CONFIG_BLOCK.  Similarly for the memory consumption.

Neat, thanks - That sounds worthwhile.
Ferenc Wagner March 24, 2010, 1:48 p.m. UTC | #5
Phillip Lougher <phillip@lougher.demon.co.uk> writes:

> Ferenc Wagner wrote:
>
>> Now with the patch series, sorry.
>>
>> Ferenc Wagner <wferi@niif.hu> writes:
>>
>>> I've got one more patch, which I forgot to export, to pull out the
>>> common logic from the backend init functions back into squashfs_read_data().
>>> With the bdev backend, that entails reading the first block twice in a
>>> row most of the time.  This again could be worked around by extending
>>> the backend interface, but I'm not sure if it's worth it.
>>
>> Here it is.  I also corrected the name of SQUASHFS_METADATA_SIZE, so it
>> may as well compile now.
>
> Thanks for your patches.
>
> First things first.  Patch sending etiquette :-) [...]

Points taken.  About the splits, they were the result of quite some
rebasing: I tried to give each patch a unique focus, but apparently
failed.  Only the last one touches substantial changed code, for the
explicitly stated reason: it hurts performance.

> Frameworks are supposed to make the code cleaner, more understandable,
> and generally better.

This framework is about added flexibility.  I agree it requires some
added complexity, but I hope we can manage it.

> Unfortunately, I see no evidence of that in your patch.  Sorry to be
> blunt, but there it is.

No problem.

> +       void	*(*init)(struct squashfs_sb_info *, u64, size_t);
> +	void	(*free)(struct squashfs_sb_info *);
> +	ssize_t	(*read)(struct squashfs_sb_info *, void **, size_t);
> +	int	(*probe)(struct file_system_type *, int, const char *,
> +				void*, struct vfsmount *);
> +	void	(*kill)(struct squashfs_sb_info *);
> +	loff_t  (*size)(const struct super_block *);
> +	const char *(*devname)(const struct super_block *, char *);
>
> We move from a single read() function to requiring seven functions to do
> the same work, just to add mtd support...

Your single read() function also had separate init, read and free
phases, with something generic in between.  I can try to invert the
structure, but see below.

> Reading the superblock changes into a 6 function deep nightmare
> squashfs_find_backend -> probe -> get_sb_dev -> fill_bdev_super ->
> squashfs_fill_super -> add_backend

It's convoluted, but only moves original code around.  If you prefer, I
can simplify that by embedding squashfs_sb_info into backend-specific
structures and allocating those separately.  This would also get rid of
some of the above 7 functions.

> Reading a block now takes 3 function calls, init, read, and free.

Actually, read must be called several times.  Are you worried by the
cost of the indirect function calls?

> Plus in your last commit you make the huge mistake of preferring code
> elegance over performance, and resort to calling init, read, and free
> *twice*, the first just to read *two* bytes (yes, 2 bytes).  Why do
> you think the code was coded that way in the first place?  A fit of
> absent mindedness perhaps?  No, for performance reasons.

I also pointed this out, and that's mosly why I created a separate last
patch.  I don't think it's optimal, but replicating the metadata length
handling in the backends isn't either.  There's RFC in the subject after
all...

> You've totally broken all the read optimisations in zlib decompression
> for block devices.  Buffer_heads are deliberately passed directly to
> the zlib decompressor to allow it to optimise performance (pass the
> buffer_head directly to zlib etc.).  Buffer_heads are exposed to the
> decompressors without crappy go-between wrappers deliberately.

At least this is something I noticed and expressed my concers about.
You probably missed that due to the lots of scrolling required by my
sub-par patch-posting technique. :)

> Unfortunately there are lots of other instances where you've
> eliminated carefully optimised code.

Could you please provide some examples?

> In short this doesn't pass my quality threshold or the make the
> minimum changes necessary threshold.  I wouldn't consider asking Linus
> to merge this.

Please don't.  It's a proof of concept, nothing more.  Even a proof of a
wrong concept.

> BTW as a comparison, I have added MTD support to Squashfs twice in the
> past, *with* bad block support.

Unfortunately you aren't allowed to share those, if I read it right.

> The latest has a diff patch of ~500 lines, with far less complexity
> and code changes necessary.

Sure, have a look at my very first patch: 3 files changed, 210
insertions, 21 deletions.  And that didn't remove bdev support.

> BTW2 as evidenced by your FIXME comments against my code in your patch,
> you really have to get over the notion that if you don't understand it,
> the code must be wrong.

Sometimes those are FIXMEs to make finding them easier, and are
questions really.  If you feel like answering them:

/* FIXME: == s->s_blocksize(_bits)? */
Why do you use devblksize and devblksize_log2 in squashfs_sb_info?
Aren't they the same as sb->s_blocksize and sb->s_blocksize_bits, as
their names suggest?

/* FIXME: squashfs_put_super has meta_index instead */
The failed_mount: part of squashfs_fill_super() is very similar to
squashfs_put_super(), but frees meta_index instead of id_table.  Why?

These are probably easy questions for people knowing the VFS layer well,
but I am not one of those.  They just triggered my built-in pattern
matcher.
Marco Stornelli March 31, 2010, 6:35 a.m. UTC | #6
2010/3/30 Ferenc Wagner <wferi@niif.hu>:
>
> Extra question: does squashfs_put_super require BKL protection?
>

We should get rid off of BKL, so if you need a lock, you should use a
local mutex.

Marco
diff mbox

Patch

From 1318e378d15008c89971f4c300738bd46d0fd2ba Mon Sep 17 00:00:00 2001
From: Ferenc Wagner <wferi@niif.hu>
Date: Thu, 18 Mar 2010 03:12:01 +0100
Subject: [PATCH] squashfs: add backend plugin framework

---
 fs/squashfs/Makefile         |    3 +-
 fs/squashfs/backend.c        |   24 +++++++++
 fs/squashfs/backend.h        |   22 ++++++++
 fs/squashfs/bdev.c           |  110 ++++++++++++++++++++++++++++++++++++++++++
 fs/squashfs/squashfs.h       |    9 +++-
 fs/squashfs/squashfs_fs_sb.h |    2 +
 6 files changed, 168 insertions(+), 2 deletions(-)
 create mode 100644 fs/squashfs/backend.c
 create mode 100644 fs/squashfs/backend.h
 create mode 100644 fs/squashfs/bdev.c

diff --git a/fs/squashfs/Makefile b/fs/squashfs/Makefile
index 45aaefd..0f02891 100644
--- a/fs/squashfs/Makefile
+++ b/fs/squashfs/Makefile
@@ -4,5 +4,6 @@ 
 
 obj-$(CONFIG_SQUASHFS) += squashfs.o
 squashfs-y += block.o cache.o dir.o export.o file.o fragment.o id.o inode.o
-squashfs-y += namei.o super.o symlink.o zlib_wrapper.o decompressor.o
+squashfs-y += namei.o super.o symlink.o zlib_wrapper.o decompressor.o backend.o
 squashfs-$(CONFIG_SQUASHFS_LZMA) += lzma_wrapper.o
+squashfs-$(CONFIG_BLOCK) += bdev.o
diff --git a/fs/squashfs/backend.c b/fs/squashfs/backend.c
new file mode 100644
index 0000000..4411c60
--- /dev/null
+++ b/fs/squashfs/backend.c
@@ -0,0 +1,24 @@ 
+#include <linux/fs.h>
+
+#include "squashfs_fs_i.h"
+#include "squashfs.h"
+#include "backend.h"
+
+const struct squashfs_backend *backends[] = {
+#ifdef CONFIG_BLOCK
+	&squashfs_bdev_ops,
+#endif
+	NULL
+};
+
+const struct squashfs_backend *
+squashfs_find_backend(struct file_system_type *fs_type, int flags,
+			const char *dev_name, void *data, struct vfsmount *mnt)
+{
+	const struct squashfs_backend **b;
+
+	for (b = backends; *b; b++)
+		if (!(*b)->probe(fs_type, flags, dev_name, data, mnt))
+			break;
+	return *b;
+}
diff --git a/fs/squashfs/backend.h b/fs/squashfs/backend.h
new file mode 100644
index 0000000..ca3ffe1
--- /dev/null
+++ b/fs/squashfs/backend.h
@@ -0,0 +1,22 @@ 
+#include <linux/fs.h>
+#include <linux/vfs.h>
+
+#include "squashfs_fs_sb.h"
+
+struct squashfs_backend {
+	void	*(*init)(struct squashfs_sb_info *, u64, size_t);
+	void	(*free)(struct squashfs_sb_info *);
+	ssize_t	(*read)(struct squashfs_sb_info *, void **, size_t);
+	int	(*probe)(struct file_system_type *, int, const char *,
+				void*, struct vfsmount *);
+	void	(*kill)(struct squashfs_sb_info *);
+	loff_t  (*size)(const struct super_block *);
+	const char *(*devname)(const struct super_block *, char *);
+};
+
+/* Dummy, until the original is nuked */
+static inline int squashfs_fill_super2(struct super_block *sb, void *data,
+				int silent, int (*add_bend)(struct super_block *))
+{
+	return -1;
+}
diff --git a/fs/squashfs/bdev.c b/fs/squashfs/bdev.c
new file mode 100644
index 0000000..c1d4389
--- /dev/null
+++ b/fs/squashfs/bdev.c
@@ -0,0 +1,110 @@ 
+#include <linux/fs.h>
+#include <linux/buffer_head.h>
+
+#include "squashfs_fs_i.h"
+#include "squashfs.h"
+#include "backend.h"
+
+struct squashfs_bdev {
+	int devblksize;			/* FIXME: == s->s_blocksize(_bits)? */
+	unsigned short devblksize_log2;
+	size_t bytes_left;
+	struct buffer_head **bh;
+	int bh_index;			/* number of consumed buffer_heads */
+	int offset;			/* offset of next byte in buffer_head */
+};
+
+/* A backend is initialized for each SquashFS block read operation,
+ * making further sequential reads possible from the block.
+ */
+static void *bdev_init(struct squashfs_sb_info *msblk, u64 index, size_t length)
+{
+	struct squashfs_bdev *bdev = msblk->backend_data;
+	struct buffer_head *bh;
+
+	bh = kcalloc((max_t(int, msblk->block_size, SQUASHFS_METADATA_SIZE)
+		      >> bdev->devblksize_log2) + 1,
+		     sizeof(*bh), GFP_KERNEL);
+	if (!bh)
+		goto nomem;
+
+	/* different preread for data blocks and metadata blocks */
+
+	bdev->bh_index = 0;
+	bdev->bytes_left = length;
+	return bdev;
+
+nomem:
+	ERROR("failed to allocate buffer_heads\n");
+	return NULL;
+}
+
+static void bdev_free(struct squashfs_sb_info *msblk)
+{
+	struct squashfs_bdev *bdev = msblk->backend_data;
+	kfree(bdev->bh);
+	bdev->bh = 0; /* FIXME: to make bdev_kill universal (see there) */
+}
+
+static ssize_t bdev_read(struct squashfs_sb_info *msblk, void **buf, size_t len)
+{
+	return -ENOSYS;
+}
+
+static int add_bdev_backend(struct super_block *sb)
+{
+	struct squashfs_sb_info *msblk = sb->s_fs_info;
+	struct squashfs_bdev *bdev = kzalloc(sizeof(*bdev), GFP_KERNEL);
+
+	if (!bdev)
+		return -ENOMEM;
+
+	bdev->devblksize = sb_min_blocksize(sb, BLOCK_SIZE);
+	bdev->devblksize_log2 = ffz(~bdev->devblksize);
+
+	msblk->backend = &squashfs_bdev_ops;
+	msblk->backend_data = bdev;
+	return 0;
+}
+
+static int fill_bdev_super(struct super_block *sb, void *data, int silent)
+{
+	return squashfs_fill_super2(sb, data, silent, add_bdev_backend);
+}
+
+static int bdev_probe(struct file_system_type *fs_type, int flags,
+			const char *dev_name, void *data,
+			struct vfsmount *mnt)
+{
+	return get_sb_bdev(fs_type, flags, dev_name, data,
+				fill_bdev_super, mnt);
+}
+
+static void bdev_kill(struct squashfs_sb_info *msblk)
+{
+	struct squashfs_bdev *bdev = msblk->backend_data;
+	if (bdev) {
+		kfree(bdev->bh); /* FIXME: can this be nonzero? cf. bdev_free */
+		kfree(bdev);
+	}
+}
+
+static loff_t bdev_size(const struct super_block *sb)
+{
+	return i_size_read(sb->s_bdev->bd_inode);
+}
+
+static const char *bdev_devname(const struct super_block *sb, char *buffer)
+{
+	return bdevname(sb->s_bdev, buffer);
+}
+
+const struct squashfs_backend squashfs_bdev_ops = {
+	.init	= bdev_init,
+	.free	= bdev_free,
+	.read	= bdev_read,
+	.probe	= bdev_probe,
+	.kill	= bdev_kill,
+	.size	= bdev_size,
+	.devname= bdev_devname
+};
diff --git a/fs/squashfs/squashfs.h b/fs/squashfs/squashfs.h
index d094886..0e74175 100644
--- a/fs/squashfs/squashfs.h
+++ b/fs/squashfs/squashfs.h
@@ -73,8 +73,12 @@  extern struct inode *squashfs_iget(struct super_block *, long long,
 				unsigned int);
 extern int squashfs_read_inode(struct inode *, long long);
 
+/* super.c
+int squashfs_fill_super(struct super_block *, void *, int,
+                        struct squashfs_backend *);
+*/
 /*
- * Inodes, files and decompressor operations
+ * Inodes, files, backend and decompressor operations
  */
 
 /* dir.c */
@@ -97,3 +101,6 @@  extern const struct squashfs_decompressor squashfs_zlib_comp_ops;
 
 /* lzma wrapper.c */
 extern const struct squashfs_decompressor squashfs_lzma_comp_ops;
+
+/* bdev.c */
+extern const struct squashfs_backend squashfs_bdev_ops;
diff --git a/fs/squashfs/squashfs_fs_sb.h b/fs/squashfs/squashfs_fs_sb.h
index 7533350..5662d9b 100644
--- a/fs/squashfs/squashfs_fs_sb.h
+++ b/fs/squashfs/squashfs_fs_sb.h
@@ -53,6 +53,8 @@  struct squashfs_cache_entry {
 
 struct squashfs_sb_info {
 	const struct squashfs_decompressor	*decompressor;
+	const struct squashfs_backend		*backend;
+	void					*backend_data;
 	int					devblksize;
 	int					devblksize_log2;
 	struct squashfs_cache			*block_cache;
-- 
1.5.6.5


From 3d469f3f3f2ecaacc8d99a142899216879099867 Mon Sep 17 00:00:00 2001
From: Ferenc Wagner <wferi@niif.hu>
Date: Sat, 20 Mar 2010 15:58:21 +0100
Subject: [PATCH] squashfs: thread initialization through the backend framework

This necessitated introducing the killsb callback.
The read path is still untouched.
---
 fs/squashfs/backend.c  |   22 +++++++++++++++-----
 fs/squashfs/backend.h  |   11 ++++-----
 fs/squashfs/bdev.c     |    3 +-
 fs/squashfs/squashfs.h |    6 ++--
 fs/squashfs/super.c    |   49 +++++++++++++++++++++--------------------------
 5 files changed, 48 insertions(+), 43 deletions(-)

diff --git a/fs/squashfs/backend.c b/fs/squashfs/backend.c
index 4411c60..49fa6fa 100644
--- a/fs/squashfs/backend.c
+++ b/fs/squashfs/backend.c
@@ -11,14 +11,24 @@  const struct squashfs_backend *backends[] = {
 	NULL
 };
 
-const struct squashfs_backend *
-squashfs_find_backend(struct file_system_type *fs_type, int flags,
+int squashfs_find_backend(struct file_system_type *fs_type, int flags,
 			const char *dev_name, void *data, struct vfsmount *mnt)
 {
 	const struct squashfs_backend **b;
+	int err;
 
-	for (b = backends; *b; b++)
-		if (!(*b)->probe(fs_type, flags, dev_name, data, mnt))
-			break;
-	return *b;
+	for (b = backends; *b; b++) {
+		err = (*b)->probe(fs_type, flags, dev_name, data, mnt);
+		if (!err)
+			return 0;
+	}
+	ERROR("No suitable backend found\n");
+	return -ENODEV;
+}
+
+void squashfs_kill_super(struct super_block *sb)
+{
+	struct squashfs_sb_info *msblk = sb->s_fs_info;
+
+	msblk->backend->killsb(sb);
 }
diff --git a/fs/squashfs/backend.h b/fs/squashfs/backend.h
index ca3ffe1..02c4bcd 100644
--- a/fs/squashfs/backend.h
+++ b/fs/squashfs/backend.h
@@ -9,14 +9,13 @@  struct squashfs_backend {
 	ssize_t	(*read)(struct squashfs_sb_info *, void **, size_t);
 	int	(*probe)(struct file_system_type *, int, const char *,
 				void*, struct vfsmount *);
+	void    (*killsb)(struct super_block *);
 	void	(*kill)(struct squashfs_sb_info *);
 	loff_t  (*size)(const struct super_block *);
 	const char *(*devname)(const struct super_block *, char *);
 };
 
-/* Dummy, until the original is nuked */
-static inline int squashfs_fill_super2(struct super_block *sb, void *data,
-				int silent, int (*add_bend)(struct super_block *))
-{
-	return -1;
-}
+int squashfs_find_backend(struct file_system_type *, int,
+			const char *, void *, struct vfsmount *);
+
+void squashfs_kill_super(struct super_block *);
diff --git a/fs/squashfs/bdev.c b/fs/squashfs/bdev.c
index c1d4389..0ee09f8 100644
--- a/fs/squashfs/bdev.c
+++ b/fs/squashfs/bdev.c
@@ -69,7 +69,7 @@  static int add_bdev_backend(struct super_block *sb)
 
 static int fill_bdev_super(struct super_block *sb, void *data, int silent)
 {
-	return squashfs_fill_super2(sb, data, silent, add_bdev_backend);
+	return squashfs_fill_super(sb, data, silent, add_bdev_backend);
 }
 
 static int bdev_probe(struct file_system_type *fs_type, int flags,
@@ -104,6 +104,7 @@  const struct squashfs_backend squashfs_bdev_ops = {
 	.free	= bdev_free,
 	.read	= bdev_read,
 	.probe	= bdev_probe,
+	.killsb	= kill_block_super,
 	.kill	= bdev_kill,
 	.size	= bdev_size,
 	.devname= bdev_devname
diff --git a/fs/squashfs/squashfs.h b/fs/squashfs/squashfs.h
index 0e74175..24c4b9e 100644
--- a/fs/squashfs/squashfs.h
+++ b/fs/squashfs/squashfs.h
@@ -73,10 +73,10 @@  extern struct inode *squashfs_iget(struct super_block *, long long,
 				unsigned int);
 extern int squashfs_read_inode(struct inode *, long long);
 
-/* super.c
+/* super.c */
 int squashfs_fill_super(struct super_block *, void *, int,
-                        struct squashfs_backend *);
-*/
+                        int (*)(struct super_block *));
+
 /*
  * Inodes, files, backend and decompressor operations
  */
diff --git a/fs/squashfs/super.c b/fs/squashfs/super.c
index 3550aec..710179f 100644
--- a/fs/squashfs/super.c
+++ b/fs/squashfs/super.c
@@ -42,6 +42,7 @@ 
 #include "squashfs_fs_i.h"
 #include "squashfs.h"
 #include "decompressor.h"
+#include "backend.h"
 
 static struct file_system_type squashfs_fs_type;
 static const struct super_operations squashfs_super_ops;
@@ -73,7 +74,8 @@  static const struct squashfs_decompressor *supported_squashfs_filesystem(short
 }
 
 
-static int squashfs_fill_super(struct super_block *sb, void *data, int silent)
+int squashfs_fill_super(struct super_block *sb, void *data, int silent,
+			int (*add_backend)(struct super_block *))
 {
 	struct squashfs_sb_info *msblk;
 	struct squashfs_super_block *sblk = NULL;
@@ -97,11 +99,15 @@  static int squashfs_fill_super(struct super_block *sb, void *data, int silent)
 	sblk = kzalloc(sizeof(*sblk), GFP_KERNEL);
 	if (sblk == NULL) {
 		ERROR("Failed to allocate squashfs_super_block\n");
-		goto failure;
+		err = -ENOMEM;
+		goto free_msblk;
 	}
 
 	msblk->devblksize = sb_min_blocksize(sb, BLOCK_SIZE);
 	msblk->devblksize_log2 = ffz(~msblk->devblksize);
+	err = add_backend(sb);
+	if (err)
+		goto free_sblk;
 
 	mutex_init(&msblk->read_data_mutex);
 	mutex_init(&msblk->meta_index_mutex);
@@ -127,7 +133,7 @@  static int squashfs_fill_super(struct super_block *sb, void *data, int silent)
 	if (sb->s_magic != SQUASHFS_MAGIC) {
 		if (!silent)
 			ERROR("Can't find a SQUASHFS superblock on %s\n",
-						bdevname(sb->s_bdev, b));
+					msblk->backend->devname(sb, b));
 		goto failed_mount;
 	}
 
@@ -146,11 +152,10 @@  static int squashfs_fill_super(struct super_block *sb, void *data, int silent)
 	if (le64_to_cpu(sblk->xattr_table_start) != SQUASHFS_INVALID_BLK)
 		ERROR("Xattrs in filesystem, these will be ignored\n");
 
-	/* Check the filesystem does not extend beyond the end of the
-	   block device */
+	/* Check the filesystem does not extend beyond the end of the device */
 	msblk->bytes_used = le64_to_cpu(sblk->bytes_used);
 	if (msblk->bytes_used < 0 || msblk->bytes_used >
-			i_size_read(sb->s_bdev->bd_inode))
+			msblk->backend->size(sb))
 		goto failed_mount;
 
 	/* Check block size for sanity */
@@ -182,7 +187,7 @@  static int squashfs_fill_super(struct super_block *sb, void *data, int silent)
 	msblk->inodes = le32_to_cpu(sblk->inodes);
 	flags = le16_to_cpu(sblk->flags);
 
-	TRACE("Found valid superblock on %s\n", bdevname(sb->s_bdev, b));
+	TRACE("Found valid superblock on %s\n", msblk->backend->devname(sb, b));
 	TRACE("Inodes are %scompressed\n", SQUASHFS_UNCOMPRESSED_INODES(flags)
 				? "un" : "");
 	TRACE("Data is %scompressed\n", SQUASHFS_UNCOMPRESSED_DATA(flags)
@@ -297,18 +302,16 @@  failed_mount:
 	squashfs_cache_delete(msblk->fragment_cache);
 	squashfs_cache_delete(msblk->read_page);
 	squashfs_decompressor_free(msblk, msblk->stream);
-	kfree(msblk->inode_lookup_table);
+	kfree(msblk->inode_lookup_table); /* FIXME: squashfs_put_super has meta_index instead */
 	kfree(msblk->fragment_index);
 	kfree(msblk->id_table);
-	kfree(sb->s_fs_info);
-	sb->s_fs_info = NULL;
+	msblk->backend->kill(msblk);
+free_sblk:
 	kfree(sblk);
-	return err;
-
-failure:
-	kfree(sb->s_fs_info);
+free_msblk:
+	kfree(msblk);
 	sb->s_fs_info = NULL;
-	return -ENOMEM;
+	return err;
 }
 
 
@@ -353,7 +356,8 @@  static void squashfs_put_super(struct super_block *sb)
 		kfree(sbi->id_table);
 		kfree(sbi->fragment_index);
 		kfree(sbi->meta_index);
-		kfree(sb->s_fs_info);
+		sbi->backend->kill(sbi);
+		kfree(sbi);
 		sb->s_fs_info = NULL;
 	}
 
@@ -361,15 +365,6 @@  static void squashfs_put_super(struct super_block *sb)
 }
 
 
-static int squashfs_get_sb(struct file_system_type *fs_type, int flags,
-				const char *dev_name, void *data,
-				struct vfsmount *mnt)
-{
-	return get_sb_bdev(fs_type, flags, dev_name, data, squashfs_fill_super,
-				mnt);
-}
-
-
 static struct kmem_cache *squashfs_inode_cachep;
 
 
@@ -442,8 +437,8 @@  static void squashfs_destroy_inode(struct inode *inode)
 static struct file_system_type squashfs_fs_type = {
 	.owner = THIS_MODULE,
 	.name = "squashfs",
-	.get_sb = squashfs_get_sb,
-	.kill_sb = kill_block_super,
+	.get_sb = squashfs_find_backend,
+	.kill_sb = squashfs_kill_super,
 	.fs_flags = FS_REQUIRES_DEV
 };
 
-- 
1.5.6.5


From 32d8c56b48b79a37a2b8204c1a14767287ea6044 Mon Sep 17 00:00:00 2001
From: Ferenc Wagner <wferi@niif.hu>
Date: Sun, 21 Mar 2010 02:08:58 +0100
Subject: [PATCH] squashfs: read uncompressed blocks through the backend framework

Avoid double free on the temporary decompression hack-path; this may
leak buffer heads if decompression exits early.
---
 fs/squashfs/backend.h        |    4 +-
 fs/squashfs/bdev.c           |  191 ++++++++++++++++++++++++++++++++++++++----
 fs/squashfs/block.c          |  167 ++++++++-----------------------------
 fs/squashfs/lzma_wrapper.c   |    2 +
 fs/squashfs/squashfs_fs_sb.h |    2 -
 fs/squashfs/super.c          |    5 +-
 6 files changed, 216 insertions(+), 155 deletions(-)

diff --git a/fs/squashfs/backend.h b/fs/squashfs/backend.h
index 02c4bcd..308fe6d 100644
--- a/fs/squashfs/backend.h
+++ b/fs/squashfs/backend.h
@@ -4,9 +4,9 @@ 
 #include "squashfs_fs_sb.h"
 
 struct squashfs_backend {
-	void	*(*init)(struct squashfs_sb_info *, u64, size_t);
+	int	(*init)(struct super_block *, u64, int, int, u64 *, int *);
 	void	(*free)(struct squashfs_sb_info *);
-	ssize_t	(*read)(struct squashfs_sb_info *, void **, size_t);
+	int	(*read)(struct super_block *, void *, int);
 	int	(*probe)(struct file_system_type *, int, const char *,
 				void*, struct vfsmount *);
 	void    (*killsb)(struct super_block *);
diff --git a/fs/squashfs/bdev.c b/fs/squashfs/bdev.c
index 0ee09f8..497029b 100644
--- a/fs/squashfs/bdev.c
+++ b/fs/squashfs/bdev.c
@@ -8,47 +8,200 @@ 
 struct squashfs_bdev {
 	int devblksize;			/* FIXME: == s->s_blocksize(_bits)? */
 	unsigned short devblksize_log2;
-	size_t bytes_left;
+	int length;
+	int bytes_read;
 	struct buffer_head **bh;
 	int bh_index;			/* number of consumed buffer_heads */
 	int offset;			/* offset of next byte in buffer_head */
+	int b;				/* total number of buffer heads */
 };
 
-/* A backend is initialized for each SquashFS block read operation,
- * making further sequential reads possible from the block.
+/*
+ * Read the metadata block length, this is stored in the first two
+ * bytes of the metadata block.
  */
-static void *bdev_init(struct squashfs_sb_info *msblk, u64 index, size_t length)
+static struct buffer_head *get_block_length(struct super_block *sb,
+			u64 *cur_index, int *offset, int *length)
 {
+	struct squashfs_sb_info *msblk = sb->s_fs_info;
 	struct squashfs_bdev *bdev = msblk->backend_data;
 	struct buffer_head *bh;
 
+	TRACE("get_block_length: cur_index=0x%llx, offset=%d\n",
+		(unsigned long long)*cur_index, *offset);
+	bh = sb_bread(sb, *cur_index);
+	if (bh == NULL)
+		return NULL;
+
+	if (bdev->devblksize - *offset == 1) {
+		*length = (unsigned char) bh->b_data[*offset];
+		put_bh(bh);
+		bh = sb_bread(sb, ++(*cur_index));
+		if (bh == NULL)
+			return NULL;
+		*length |= (unsigned char) bh->b_data[0] << 8;
+		*offset = 1;
+	} else {
+		*length = (unsigned char) bh->b_data[*offset] |
+			(unsigned char) bh->b_data[*offset + 1] << 8;
+		*offset += 2;
+	}
+
+	TRACE("get_block_length: length=%d, offset=%d\n", *length, *offset);
+	return bh;
+}
+
+/*
+ * Read a metadata block or datablock into memory.  Length is non-zero
+ * if a datablock is being read (the size is stored elsewhere in the
+ * filesystem), otherwise the length is obtained from the first two bytes of
+ * the metadata block.  A bit in the length field indicates if the block
+ * is stored uncompressed in the filesystem (usually because compression
+ * generated a larger block - this does occasionally happen with zlib).
+ */
+static int bdev_init(struct super_block *sb, u64 index, int length,
+			int srclength, u64 *next_index, int *compressed)
+{
+	struct squashfs_sb_info *msblk = sb->s_fs_info;
+	struct squashfs_bdev *bdev = msblk->backend_data;
+	struct buffer_head **bh;
+	u64 cur_index;			/* sector of next byte */
+	int bytes;			/* number of bytes handled */
+	int b;				/* for counting buffer heads */
+
+	TRACE("Entering bdev_init: index=0x%llx, length=0x%x, srclength=%d\n",
+		index, length, srclength);
 	bh = kcalloc((max_t(int, msblk->block_size, SQUASHFS_METADATA_SIZE)
 		      >> bdev->devblksize_log2) + 1,
 		     sizeof(*bh), GFP_KERNEL);
-	if (!bh)
-		goto nomem;
+	if (bh == NULL)
+		return -ENOMEM;
+
+	bdev->offset = index & ((1 << bdev->devblksize_log2) - 1);
+	cur_index = index >> bdev->devblksize_log2;
+	if (length) { /* FIXME: this logic and the comment above should be pulled out! */
+		/*
+		 * Datablock.
+		 */
+		bytes = -bdev->offset;
+		*compressed = SQUASHFS_COMPRESSED_BLOCK(length);
+		length = SQUASHFS_COMPRESSED_SIZE_BLOCK(length);
+		if (next_index)
+			*next_index = index + length;
+
+		TRACE("Block @ 0x%llx, %scompressed size %d, src size %d\n",
+			index, *compressed ? "" : "un", length, srclength);
 
-	/* different preread for data blocks and metadata blocks */
+		if (length < 0 || length > srclength ||
+				(index + length) > msblk->bytes_used)
+			goto read_failure;
 
+		for (b = 0; bytes < length; b++, cur_index++) {
+			bh[b] = sb_getblk(sb, cur_index);
+			if (bh[b] == NULL)
+				goto block_release;
+			bytes += bdev->devblksize;
+		}
+		ll_rw_block(READ, b, bh);
+	} else {
+		/*
+		 * Metadata block.
+		 */
+		if ((index + 2) > msblk->bytes_used)
+			goto read_failure;
+
+		bh[0] = get_block_length(sb, &cur_index, &bdev->offset, &length);
+		if (bh[0] == NULL)
+			goto read_failure;
+		b = 1;
+
+		bytes = bdev->devblksize - bdev->offset;
+		*compressed = SQUASHFS_COMPRESSED(length);
+		length = SQUASHFS_COMPRESSED_SIZE(length);
+		if (next_index)
+			*next_index = index + length + 2;
+
+		TRACE("Block @ 0x%llx, %scompressed size %d\n", index,
+				*compressed ? "" : "un", length);
+
+		if (length < 0 || length > srclength ||
+					(index + length) > msblk->bytes_used)
+			goto block_release;
+
+		for (; bytes < length; b++) {
+			bh[b] = sb_getblk(sb, ++cur_index);
+			if (bh[b] == NULL)
+				goto block_release;
+			bytes += bdev->devblksize;
+		}
+		ll_rw_block(READ, b - 1, bh + 1);
+	}
+
+	bdev->bh = bh;
+	bdev->length = length;
+	bdev->bytes_read = 0;
+	bdev->b = b;
 	bdev->bh_index = 0;
-	bdev->bytes_left = length;
-	return bdev;
+	TRACE("bdev_init: allocated %d buffer heads at %p, returning length=%d",
+	      b, bh, length);
+	return length;
 
-nomem:
-	ERROR("failed to allocate buffer_heads\n");
-	return NULL;
+block_release:
+	while (b--)
+		put_bh(bh[b]);
+read_failure:
+	ERROR("bdev_init failed to read block 0x%llx\n",
+					(unsigned long long) index);
+	kfree(bh);
+	return -EIO;
 }
 
 static void bdev_free(struct squashfs_sb_info *msblk)
 {
 	struct squashfs_bdev *bdev = msblk->backend_data;
+
+	TRACE("bdev_free: bh=%p, bh_index=%d, b=%d\n",
+		bdev->bh, bdev->bh_index, bdev->b);
+	while (bdev->bh_index < bdev->b)
+		put_bh(bdev->bh[bdev->bh_index++]);
 	kfree(bdev->bh);
-	bdev->bh = 0; /* FIXME: to make bdev_kill universal (see there) */
+	bdev->bh = NULL;
 }
 
-static ssize_t bdev_read(struct squashfs_sb_info *msblk, void **buf, size_t len)
+static int bdev_read(struct super_block *sb, void *buf, int len)
 {
-	return -ENOSYS;
+	struct squashfs_sb_info *msblk = sb->s_fs_info;
+	struct squashfs_bdev *bdev = msblk->backend_data;
+	struct buffer_head *bh = bdev->bh[bdev->bh_index];
+	int in_block = bdev->devblksize - bdev->offset;
+
+	TRACE("bdev_read: buf=%p, len=%d, length=%d, bytes_read=%d, in_block=%d, bh_index=%d, offset=%d\n",
+	      buf, len, bdev->length, bdev->bytes_read, in_block, bdev->bh_index, bdev->offset);
+	if (bdev->bytes_read == bdev->length) /* EOF */
+		return 0;
+	if (bdev->bytes_read + len > bdev->length)
+		len = bdev->length - bdev->bytes_read;
+	if (bdev->bytes_read == 0 || bdev->offset == 0) {
+		TRACE("bdev_read: wait_on_buffer %d\n", bdev->bh_index);
+		wait_on_buffer(bh);
+		if (!buffer_uptodate(bh))
+			return -EIO;
+	}
+	if (len < in_block) {
+		TRACE("bdev_read: copying %d bytes", len);
+		memcpy(buf, bh->b_data + bdev->offset, len);
+		bdev->offset += len;
+		bdev->bytes_read += len;
+		return len;
+	} else {
+		TRACE("bdev_read: copying %d bytes", in_block);
+		memcpy(buf, bh->b_data + bdev->offset, in_block);
+		put_bh(bh);
+		bdev->bh_index++;
+		bdev->offset = 0;
+		bdev->bytes_read += in_block;
+		return in_block;
+	}
 }
 
 static int add_bdev_backend(struct super_block *sb)
@@ -56,6 +209,7 @@  static int add_bdev_backend(struct super_block *sb)
 	struct squashfs_sb_info *msblk = sb->s_fs_info;
 	struct squashfs_bdev *bdev = kzalloc(sizeof(*bdev), GFP_KERNEL);
 
+	TRACE("add_bdev_backend\n");
 	if (!bdev)
 		return -ENOMEM;
 
@@ -83,9 +237,12 @@  static int bdev_probe(struct file_system_type *fs_type, int flags,
 static void bdev_kill(struct squashfs_sb_info *msblk)
 {
 	struct squashfs_bdev *bdev = msblk->backend_data;
+
+	TRACE("bdev_kill: bdev=%p\n", bdev);
 	if (bdev) {
-		kfree(bdev->bh); /* FIXME: can this be nonzero? cf. bdev_free */
+		bdev_free(msblk);
 		kfree(bdev);
+		bdev = NULL;
 	}
 }
 
@@ -105,7 +262,7 @@  const struct squashfs_backend squashfs_bdev_ops = {
 	.read	= bdev_read,
 	.probe	= bdev_probe,
 	.killsb	= kill_block_super,
-	.kill	= bdev_kill,
+	.kill	= bdev_kill, /* FIXME: is this needed at all? */
 	.size	= bdev_size,
 	.devname= bdev_devname
 };
diff --git a/fs/squashfs/block.c b/fs/squashfs/block.c
index 6f9914d..8787aac 100644
--- a/fs/squashfs/block.c
+++ b/fs/squashfs/block.c
@@ -37,37 +37,19 @@ 
 #include "squashfs_fs_i.h"
 #include "squashfs.h"
 #include "decompressor.h"
-/*
- * Read the metadata block length, this is stored in the first two
- * bytes of the metadata block.
- */
-static struct buffer_head *get_block_length(struct super_block *sb,
-			u64 *cur_index, int *offset, int *length)
-{
-	struct squashfs_sb_info *msblk = sb->s_fs_info;
-	struct buffer_head *bh;
-
-	bh = sb_bread(sb, *cur_index);
-	if (bh == NULL)
-		return NULL;
-
-	if (msblk->devblksize - *offset == 1) {
-		*length = (unsigned char) bh->b_data[*offset];
-		put_bh(bh);
-		bh = sb_bread(sb, ++(*cur_index));
-		if (bh == NULL)
-			return NULL;
-		*length |= (unsigned char) bh->b_data[0] << 8;
-		*offset = 1;
-	} else {
-		*length = (unsigned char) bh->b_data[*offset] |
-			(unsigned char) bh->b_data[*offset + 1] << 8;
-		*offset += 2;
-	}
-
-	return bh;
-}
-
+#include "backend.h"
+
+/* FIXME: here for a temporary cheat only */
+struct squashfs_bdev {
+	int devblksize;			/* FIXME: == s->s_blocksize(_bits)? */
+	unsigned short devblksize_log2;
+	int length;
+	int bytes_read;
+	struct buffer_head **bh;
+	int bh_index;			/* number of consumed buffer_heads */
+	int offset;			/* offset of next byte in buffer_head */
+	int b;				/* total number of buffer heads */
+};
 
 /*
  * Read and decompress a metadata block or datablock.  Length is non-zero
@@ -80,124 +62,47 @@  static struct buffer_head *get_block_length(struct super_block *sb,
 int squashfs_read_data(struct super_block *sb, void **buffer, u64 index,
 			int length, u64 *next_index, int srclength, int pages)
 {
+	int compressed;
 	struct squashfs_sb_info *msblk = sb->s_fs_info;
-	struct buffer_head **bh;
-	int offset = index & ((1 << msblk->devblksize_log2) - 1);
-	u64 cur_index = index >> msblk->devblksize_log2;
-	int bytes, compressed, b = 0, k = 0, page = 0, avail;
-
-
-	bh = kcalloc((msblk->block_size >> msblk->devblksize_log2) + 1,
-				sizeof(*bh), GFP_KERNEL);
-	if (bh == NULL)
-		return -ENOMEM;
-
-	if (length) {
-		/*
-		 * Datablock.
-		 */
-		bytes = -offset;
-		compressed = SQUASHFS_COMPRESSED_BLOCK(length);
-		length = SQUASHFS_COMPRESSED_SIZE_BLOCK(length);
-		if (next_index)
-			*next_index = index + length;
-
-		TRACE("Block @ 0x%llx, %scompressed size %d, src size %d\n",
-			index, compressed ? "" : "un", length, srclength);
 
-		if (length < 0 || length > srclength ||
-				(index + length) > msblk->bytes_used)
-			goto read_failure;
-
-		for (b = 0; bytes < length; b++, cur_index++) {
-			bh[b] = sb_getblk(sb, cur_index);
-			if (bh[b] == NULL)
-				goto block_release;
-			bytes += msblk->devblksize;
-		}
-		ll_rw_block(READ, b, bh);
-	} else {
-		/*
-		 * Metadata block.
-		 */
-		if ((index + 2) > msblk->bytes_used)
-			goto read_failure;
-
-		bh[0] = get_block_length(sb, &cur_index, &offset, &length);
-		if (bh[0] == NULL)
-			goto read_failure;
-		b = 1;
-
-		bytes = msblk->devblksize - offset;
-		compressed = SQUASHFS_COMPRESSED(length);
-		length = SQUASHFS_COMPRESSED_SIZE(length);
-		if (next_index)
-			*next_index = index + length + 2;
-
-		TRACE("Block @ 0x%llx, %scompressed size %d\n", index,
-				compressed ? "" : "un", length);
-
-		if (length < 0 || length > srclength ||
-					(index + length) > msblk->bytes_used)
-			goto block_release;
-
-		for (; bytes < length; b++) {
-			bh[b] = sb_getblk(sb, ++cur_index);
-			if (bh[b] == NULL)
-				goto block_release;
-			bytes += msblk->devblksize;
-		}
-		ll_rw_block(READ, b - 1, bh + 1);
-	}
+	length = msblk->backend->init(sb, index, length, srclength,
+					next_index, &compressed);
+	if (length < 0)
+		return length;
 
 	if (compressed) {
-		length = squashfs_decompress(msblk, buffer, bh, b, offset,
-			length, srclength, pages);
+		struct squashfs_bdev *bdev = msblk->backend_data;
+		length = squashfs_decompress(msblk, buffer, bdev->bh, bdev->b,
+				bdev->offset, length, srclength, pages);
+		bdev->bh_index = bdev->b; /* temporary hack to avoid double free */
 		if (length < 0)
 			goto read_failure;
 	} else {
 		/*
 		 * Block is uncompressed.
 		 */
-		int i, in, pg_offset = 0;
-
-		for (i = 0; i < b; i++) {
-			wait_on_buffer(bh[i]);
-			if (!buffer_uptodate(bh[i]))
-				goto block_release;
-		}
-
-		for (bytes = length; k < b; k++) {
-			in = min(bytes, msblk->devblksize - offset);
-			bytes -= in;
-			while (in) {
-				if (pg_offset == PAGE_CACHE_SIZE) {
-					page++;
-					pg_offset = 0;
-				}
-				avail = min_t(int, in, PAGE_CACHE_SIZE -
-						pg_offset);
-				memcpy(buffer[page] + pg_offset,
-						bh[k]->b_data + offset, avail);
-				in -= avail;
-				pg_offset += avail;
-				offset += avail;
+		int read, page = 0, pg_offset = 0;
+
+		while ((read = msblk->backend->read(sb, buffer[page] + pg_offset,
+						PAGE_CACHE_SIZE - pg_offset))) {
+			if (read < 0)
+				goto read_failure;
+			TRACE("copied %d bytes\n", read);
+			pg_offset += read;
+			if (pg_offset == PAGE_CACHE_SIZE) {
+				page++;
+				pg_offset = 0;
 			}
-			offset = 0;
-			put_bh(bh[k]);
 		}
 	}
 
-	kfree(bh);
+	msblk->backend->free(msblk);
+	TRACE("squashfs_read_data: returning length=%d\n", length);
 	return length;
 
-block_release:
-	for (; k < b; k++)
-		put_bh(bh[k]);
-
 read_failure:
 	ERROR("squashfs_read_data failed to read block 0x%llx\n",
 					(unsigned long long) index);
-	kfree(bh);
+	msblk->backend->free(msblk);
 	return -EIO;
 }
diff --git a/fs/squashfs/lzma_wrapper.c b/fs/squashfs/lzma_wrapper.c
index 9fa617d..439798f 100644
--- a/fs/squashfs/lzma_wrapper.c
+++ b/fs/squashfs/lzma_wrapper.c
@@ -94,6 +94,8 @@  static int lzma_uncompress(struct squashfs_sb_info *msblk, void **buffer,
 	void *buff = stream->input;
 	int avail, i, bytes = length, res;
 
+	TRACE("lzma_uncompress: bh=%p, b=%d, offset=%d, length=%d, srclength=%d,"
+		" pages=%d\n", bh, b, offset, length, srclength, pages);
 	mutex_lock(&lzma_mutex);
 
 	for (i = 0; i < b; i++) {
diff --git a/fs/squashfs/squashfs_fs_sb.h b/fs/squashfs/squashfs_fs_sb.h
index 5662d9b..87958fc 100644
--- a/fs/squashfs/squashfs_fs_sb.h
+++ b/fs/squashfs/squashfs_fs_sb.h
@@ -55,8 +55,6 @@  struct squashfs_sb_info {
 	const struct squashfs_decompressor	*decompressor;
 	const struct squashfs_backend		*backend;
 	void					*backend_data;
-	int					devblksize;
-	int					devblksize_log2;
 	struct squashfs_cache			*block_cache;
 	struct squashfs_cache			*fragment_cache;
 	struct squashfs_cache			*read_page;
diff --git a/fs/squashfs/super.c b/fs/squashfs/super.c
index 710179f..cb561bf 100644
--- a/fs/squashfs/super.c
+++ b/fs/squashfs/super.c
@@ -87,7 +87,7 @@  int squashfs_fill_super(struct super_block *sb, void *data, int silent,
 	u64 lookup_table_start;
 	int err;
 
-	TRACE("Entered squashfs_fill_superblock\n");
+	TRACE("Entered squashfs_fill_superblock, silent=%d\n", silent);
 
 	sb->s_fs_info = kzalloc(sizeof(*msblk), GFP_KERNEL);
 	if (sb->s_fs_info == NULL) {
@@ -103,8 +103,6 @@  int squashfs_fill_super(struct super_block *sb, void *data, int silent,
 		goto free_msblk;
 	}
 
-	msblk->devblksize = sb_min_blocksize(sb, BLOCK_SIZE);
-	msblk->devblksize_log2 = ffz(~msblk->devblksize);
 	err = add_backend(sb);
 	if (err)
 		goto free_sblk;
@@ -305,6 +303,7 @@  failed_mount:
 	kfree(msblk->inode_lookup_table); /* FIXME: squashfs_put_super has meta_index instead */
 	kfree(msblk->fragment_index);
 	kfree(msblk->id_table);
+	TRACE("squasfs_fill_super: killing backend, err=%d\n", err);
 	msblk->backend->kill(msblk);
 free_sblk:
 	kfree(sblk);
-- 
1.5.6.5


From 7bad4e58afdc8c5e21dafcd538a1390e30176b0a Mon Sep 17 00:00:00 2001
From: Ferenc Wagner <wferi@niif.hu>
Date: Sun, 21 Mar 2010 22:38:00 +0100
Subject: [PATCH] squashfs: move the decompressors to the backend framework

---
 fs/squashfs/block.c        |   17 +----------------
 fs/squashfs/decompressor.h |   12 +++++-------
 fs/squashfs/lzma_wrapper.c |   27 +++++++++------------------
 fs/squashfs/zlib_wrapper.c |   44 ++++++++++++++++++--------------------------
 4 files changed, 33 insertions(+), 67 deletions(-)

diff --git a/fs/squashfs/block.c b/fs/squashfs/block.c
index 8787aac..1c85063 100644
--- a/fs/squashfs/block.c
+++ b/fs/squashfs/block.c
@@ -39,18 +39,6 @@ 
 #include "decompressor.h"
 #include "backend.h"
 
-/* FIXME: here for a temporary cheat only */
-struct squashfs_bdev {
-	int devblksize;			/* FIXME: == s->s_blocksize(_bits)? */
-	unsigned short devblksize_log2;
-	int length;
-	int bytes_read;
-	struct buffer_head **bh;
-	int bh_index;			/* number of consumed buffer_heads */
-	int offset;			/* offset of next byte in buffer_head */
-	int b;				/* total number of buffer heads */
-};
-
 /*
  * Read and decompress a metadata block or datablock.  Length is non-zero
  * if a datablock is being read (the size is stored elsewhere in the
@@ -71,10 +59,7 @@  int squashfs_read_data(struct super_block *sb, void **buffer, u64 index,
 		return length;
 
 	if (compressed) {
-		struct squashfs_bdev *bdev = msblk->backend_data;
-		length = squashfs_decompress(msblk, buffer, bdev->bh, bdev->b,
-				bdev->offset, length, srclength, pages);
-		bdev->bh_index = bdev->b; /* temporary hack to avoid double free */
+		length = squashfs_decompress(sb, buffer, length, pages);
 		if (length < 0)
 			goto read_failure;
 	} else {
diff --git a/fs/squashfs/decompressor.h b/fs/squashfs/decompressor.h
index 7425f80..06b4233 100644
--- a/fs/squashfs/decompressor.h
+++ b/fs/squashfs/decompressor.h
@@ -26,8 +26,7 @@ 
 struct squashfs_decompressor {
 	void	*(*init)(struct squashfs_sb_info *);
 	void	(*free)(void *);
-	int	(*decompress)(struct squashfs_sb_info *, void **,
-		struct buffer_head **, int, int, int, int, int);
+	int	(*decompress)(struct super_block *, void **, int, int);
 	int	id;
 	char	*name;
 	int	supported;
@@ -45,11 +44,10 @@  static inline void squashfs_decompressor_free(struct squashfs_sb_info *msblk,
 		msblk->decompressor->free(s);
 }
 
-static inline int squashfs_decompress(struct squashfs_sb_info *msblk,
-	void **buffer, struct buffer_head **bh, int b, int offset, int length,
-	int srclength, int pages)
+static inline int squashfs_decompress(struct super_block *sb,
+	void **buffer, int length, int pages)
 {
-	return msblk->decompressor->decompress(msblk, buffer, bh, b, offset,
-		length, srclength, pages);
+	struct squashfs_sb_info *msblk = sb->s_fs_info;
+	return msblk->decompressor->decompress(sb, buffer, length, pages);
 }
 #endif
diff --git a/fs/squashfs/lzma_wrapper.c b/fs/squashfs/lzma_wrapper.c
index 439798f..da571ce 100644
--- a/fs/squashfs/lzma_wrapper.c
+++ b/fs/squashfs/lzma_wrapper.c
@@ -32,6 +32,7 @@ 
 #include "squashfs_fs_i.h"
 #include "squashfs.h"
 #include "decompressor.h"
+#include "backend.h"
 
 struct squashfs_lzma {
 	void	*input;
@@ -86,29 +87,23 @@  static void lzma_free(void *strm)
 }
 
 
-static int lzma_uncompress(struct squashfs_sb_info *msblk, void **buffer,
-	struct buffer_head **bh, int b, int offset, int length, int srclength,
-	int pages)
+static int lzma_uncompress(struct super_block *sb, void **buffer,
+			   int length, int pages)
 {
+	struct squashfs_sb_info *msblk = sb->s_fs_info;
 	struct squashfs_lzma *stream = msblk->stream;
 	void *buff = stream->input;
 	int avail, i, bytes = length, res;
 
-	TRACE("lzma_uncompress: bh=%p, b=%d, offset=%d, length=%d, srclength=%d,"
-		" pages=%d\n", bh, b, offset, length, srclength, pages);
+	TRACE("lzma_uncompress: length=%d, pages=%d\n", length, pages);
 	mutex_lock(&lzma_mutex);
 
-	for (i = 0; i < b; i++) {
-		wait_on_buffer(bh[i]);
-		if (!buffer_uptodate(bh[i]))
-			goto block_release;
-
-		avail = min(bytes, msblk->devblksize - offset);
-		memcpy(buff, bh[i]->b_data + offset, avail);
+	while ((avail = msblk->backend->read(sb, buff, length))) {
+		if (avail < 0)
+			goto failed;
+		TRACE("read %d bytes\n", avail);
 		buff += avail;
 		bytes -= avail;
-		offset = 0;
-		put_bh(bh[i]);
 	}
 
 	lzma_error = 0;
@@ -131,10 +126,6 @@  static int lzma_uncompress(struct squashfs_sb_info *msblk, void **buffer,
 	mutex_unlock(&lzma_mutex);
 	return res;
 
-block_release:
-	for (; i < b; i++)
-		put_bh(bh[i]);
-
 failed:
 	mutex_unlock(&lzma_mutex);
 
diff --git a/fs/squashfs/zlib_wrapper.c b/fs/squashfs/zlib_wrapper.c
index 4dd70e0..70c7571 100644
--- a/fs/squashfs/zlib_wrapper.c
+++ b/fs/squashfs/zlib_wrapper.c
@@ -31,6 +31,7 @@ 
 #include "squashfs_fs_i.h"
 #include "squashfs.h"
 #include "decompressor.h"
+#include "backend.h"
 
 static void *zlib_init(struct squashfs_sb_info *dummy)
 {
@@ -61,13 +62,18 @@  static void zlib_free(void *strm)
 }
 
 
-static int zlib_uncompress(struct squashfs_sb_info *msblk, void **buffer,
-	struct buffer_head **bh, int b, int offset, int length, int srclength,
-	int pages)
+static int zlib_uncompress(struct super_block *sb, void **buffer,
+	int length, int pages)
 {
+	struct squashfs_sb_info *msblk = sb->s_fs_info;
 	int zlib_err = 0, zlib_init = 0;
-	int avail, bytes, k = 0, page = 0;
+	int bytes, page = 0;
 	z_stream *stream = msblk->stream;
+	/* Copying is required by the backend interface for now. */
+	void *input = kmalloc (PAGE_CACHE_SIZE, GFP_KERNEL);
+
+	if (!input)
+		return -ENOMEM;
 
 	mutex_lock(&msblk->read_data_mutex);
 
@@ -76,22 +82,13 @@  static int zlib_uncompress(struct squashfs_sb_info *msblk, void **buffer,
 
 	bytes = length;
 	do {
-		if (stream->avail_in == 0 && k < b) {
-			avail = min(bytes, msblk->devblksize - offset);
-			bytes -= avail;
-			wait_on_buffer(bh[k]);
-			if (!buffer_uptodate(bh[k]))
+		if (stream->avail_in == 0) {
+			stream->avail_in = msblk->backend->read(sb, input, PAGE_CACHE_SIZE);
+			if (!stream->avail_in) {
+				ERROR("unexpected end of input\n");
 				goto release_mutex;
-
-			if (avail == 0) {
-				offset = 0;
-				put_bh(bh[k++]);
-				continue;
 			}
-
-			stream->next_in = bh[k]->b_data + offset;
-			stream->avail_in = avail;
-			offset = 0;
+			stream->next_in = input;
 		}
 
 		if (stream->avail_out == 0 && page < pages) {
@@ -103,8 +100,7 @@  static int zlib_uncompress(struct squashfs_sb_info *msblk, void **buffer,
 			zlib_err = zlib_inflateInit(stream);
 			if (zlib_err != Z_OK) {
 				ERROR("zlib_inflateInit returned unexpected "
-					"result 0x%x, srclength %d\n",
-					zlib_err, srclength);
+					"result 0x%x\n", zlib_err);
 				goto release_mutex;
 			}
 			zlib_init = 1;
@@ -112,8 +108,6 @@  static int zlib_uncompress(struct squashfs_sb_info *msblk, void **buffer,
 
 		zlib_err = zlib_inflate(stream, Z_SYNC_FLUSH);
 
-		if (stream->avail_in == 0 && k < b)
-			put_bh(bh[k++]);
 	} while (zlib_err == Z_OK);
 
 	if (zlib_err != Z_STREAM_END) {
@@ -128,14 +122,12 @@  static int zlib_uncompress(struct squashfs_sb_info *msblk, void **buffer,
 	}
 
 	mutex_unlock(&msblk->read_data_mutex);
+	kfree(input);
 	return stream->total_out;
 
 release_mutex:
 	mutex_unlock(&msblk->read_data_mutex);
-
-	for (; k < b; k++)
-		put_bh(bh[k]);
-
+	kfree(input);
 	return -EIO;
 }
 
-- 
1.5.6.5


From 9cb2a2373ab4b7a5519c5a2ef816894e28778196 Mon Sep 17 00:00:00 2001
From: Ferenc Wagner <wferi@niif.hu>
Date: Tue, 23 Mar 2010 02:21:27 +0100
Subject: [PATCH] squashfs: add mtd backend

---
 fs/squashfs/Kconfig    |    2 +-
 fs/squashfs/Makefile   |    1 +
 fs/squashfs/backend.c  |    3 +
 fs/squashfs/mtd.c      |  167 ++++++++++++++++++++++++++++++++++++++++++++++++
 fs/squashfs/squashfs.h |    3 +
 5 files changed, 175 insertions(+), 1 deletions(-)
 create mode 100644 fs/squashfs/mtd.c

diff --git a/fs/squashfs/Kconfig b/fs/squashfs/Kconfig
index 7ec5d7e..6849e70 100644
--- a/fs/squashfs/Kconfig
+++ b/fs/squashfs/Kconfig
@@ -1,6 +1,6 @@ 
 config SQUASHFS
 	tristate "SquashFS 4.0 - Squashed file system support"
-	depends on BLOCK
+	depends on BLOCK || MTD
 	select ZLIB_INFLATE
 	help
 	  Saying Y here includes support for SquashFS 4.0 (a Compressed
diff --git a/fs/squashfs/Makefile b/fs/squashfs/Makefile
index 0f02891..b6dd60e 100644
--- a/fs/squashfs/Makefile
+++ b/fs/squashfs/Makefile
@@ -7,3 +7,4 @@  squashfs-y += block.o cache.o dir.o export.o file.o fragment.o id.o inode.o
 squashfs-y += namei.o super.o symlink.o zlib_wrapper.o decompressor.o backend.o
 squashfs-$(CONFIG_SQUASHFS_LZMA) += lzma_wrapper.o
 squashfs-$(CONFIG_BLOCK) += bdev.o
+squashfs-$(CONFIG_MTD) += mtd.o
diff --git a/fs/squashfs/backend.c b/fs/squashfs/backend.c
index 49fa6fa..56037e3 100644
--- a/fs/squashfs/backend.c
+++ b/fs/squashfs/backend.c
@@ -8,6 +8,9 @@  const struct squashfs_backend *backends[] = {
 #ifdef CONFIG_BLOCK
 	&squashfs_bdev_ops,
 #endif
+#ifdef CONFIG_MTD
+	&squashfs_mtd_ops,
+#endif
 	NULL
 };
 
diff --git a/fs/squashfs/mtd.c b/fs/squashfs/mtd.c
new file mode 100644
index 0000000..c9ba361
--- /dev/null
+++ b/fs/squashfs/mtd.c
@@ -0,0 +1,167 @@ 
+#include <linux/fs.h>
+#include <linux/mtd/mtd.h>
+#include <linux/mtd/super.h>
+
+#include "squashfs_fs_i.h"
+#include "squashfs.h"
+#include "backend.h"
+
+struct squashfs_mtd {
+	u64 index;
+	int length;
+	int bytes_read;
+};
+
+static int checked_mtd_read(struct super_block *sb, u64 index, int length,
+			    void *buf)
+{
+	struct mtd_info *mi = sb->s_mtd;
+	int ret, retlen;
+
+	TRACE("Entering checked_mtd_read: index=0x%llx, length=%d\n",
+	      index, length);
+	ret = mi->read(mi, index, length, &retlen, buf);
+	if (ret) {
+		if (ret != -EUCLEAN && ret != -EBADMSG) {
+			ERROR("checked_mtd_read(index=0x%llx, length=%d): %d\n",
+			      index, length, ret);
+			return ret;
+		} else
+			WARNING("checked_mtd_read(index=0x%llx, length=%d): "
+				"recoverable error %d\n", index, length, ret);
+	}
+	if (retlen != length) {
+		ERROR("checked_mtd_read(index=0x%llx, length=%d) short read: %d\n",
+		      index, length, retlen);
+		return -EIO;
+	}
+	return 0;
+}
+
+static int mtd_init(struct super_block *sb, u64 index, int length,
+		    int srclength, u64 *next_index, int *compressed)
+{
+	struct squashfs_sb_info *msblk = sb->s_fs_info;
+	struct squashfs_mtd *mtd = msblk->backend_data;
+
+	TRACE("Entering mtd_init: index=0x%llx, length=0x%x, srclength=%d\n",
+		index, length, srclength);
+
+	if (length) { /* datablock */
+		*compressed = SQUASHFS_COMPRESSED_BLOCK(length);
+		length = SQUASHFS_COMPRESSED_SIZE_BLOCK(length);
+		if (next_index)
+			*next_index = index + length;
+		TRACE("Block @ 0x%llx, %scompressed size %d, src size %d\n",
+			index, *compressed ? "" : "un", length, srclength);
+	} else { /* metadata block */
+		u16 metalen;
+		if ((index + 2) > msblk->bytes_used)
+			goto read_failure;
+		if (checked_mtd_read(sb, index, 2, &metalen))
+			goto read_failure;
+		length = le16_to_cpu(metalen);
+		*compressed = SQUASHFS_COMPRESSED(length);
+		length = SQUASHFS_COMPRESSED_SIZE(length);
+		if (next_index)
+			*next_index = index + length + 2;
+		TRACE("Block @ 0x%llx, %scompressed size %d\n", index,
+				*compressed ? "" : "un", length);
+		index += 2;
+	}
+	if (length < 0 || length > srclength ||
+			(index + length) > msblk->bytes_used)
+		goto read_failure;
+
+	mtd->index = index;
+	mtd->length = length;
+	mtd->bytes_read = 0;
+	return length;
+
+read_failure:
+	ERROR("mtd_init failed to read block 0x%llx\n",
+					(unsigned long long) index);
+	return -EIO;
+}
+
+static void mtd_free(struct squashfs_sb_info *msblk)
+{
+	TRACE("mtd_free: no op\n");
+}
+
+static int mtd_read(struct super_block *sb, void *buf, int len)
+{
+	struct squashfs_sb_info *msblk = sb->s_fs_info;
+	struct squashfs_mtd *mtd = msblk->backend_data;
+	int err;
+
+	TRACE("mtd_read: buf=%p, len=%d, length=%d, bytes_read=%d\n",
+	      buf, len, mtd->length, mtd->bytes_read);
+
+	if (mtd->bytes_read == mtd->length) /* EOF */
+		return 0;
+	if (mtd->bytes_read + len > mtd->length)
+		len = mtd->length - mtd->bytes_read;
+	TRACE("mtd_read: copying %d bytes", len);
+	err = checked_mtd_read(sb, mtd->index + mtd->bytes_read, len, buf);
+	if (err)
+		return err;
+	mtd->bytes_read += len;
+	return len;
+}
+
+static int add_mtd_backend(struct super_block *sb)
+{
+	struct squashfs_sb_info *msblk = sb->s_fs_info;
+	struct squashfs_mtd *mtd = kzalloc(sizeof(*mtd), GFP_KERNEL);
+
+	TRACE("add_mtd_backend\n");
+	if (!mtd)
+		return -ENOMEM;
+
+	msblk->backend = &squashfs_mtd_ops;
+	msblk->backend_data = mtd;
+	return 0;
+}
+
+static int fill_mtd_super(struct super_block *sb, void *data, int silent)
+{
+	return squashfs_fill_super(sb, data, silent, add_mtd_backend);
+}
+
+static int mtd_probe(struct file_system_type *fs_type, int flags,
+			const char *dev_name, void *data,
+			struct vfsmount *mnt)
+{
+	return get_sb_mtd(fs_type, flags, dev_name, data,
+				fill_mtd_super, mnt);
+}
+
+static void mtd_kill(struct squashfs_sb_info *msblk)
+{
+	TRACE("mtd_kill: mtd=%p\n", msblk->backend_data);
+	mtd_free(msblk);
+	kfree(msblk->backend_data);
+}
+
+static loff_t mtd_size(const struct super_block *sb)
+{
+	return sb->s_mtd->size;
+}
+
+static const char *mtd_devname(const struct super_block *sb, char *buffer)
+{
+	snprintf(buffer, BDEVNAME_SIZE, "MTD%d", sb->s_mtd->index);
+	return buffer;
+}
+
+const struct squashfs_backend squashfs_mtd_ops = {
+	.init	= mtd_init,
+	.free	= mtd_free,
+	.read	= mtd_read,
+	.probe	= mtd_probe,
+	.killsb = kill_mtd_super,
+	.kill	= mtd_kill,
+	.size	= mtd_size,
+	.devname= mtd_devname
+};
diff --git a/fs/squashfs/squashfs.h b/fs/squashfs/squashfs.h
index 24c4b9e..fc3a57a 100644
--- a/fs/squashfs/squashfs.h
+++ b/fs/squashfs/squashfs.h
@@ -104,3 +104,6 @@  extern const struct squashfs_decompressor squashfs_lzma_comp_ops;
 
 /* bdev.c */
 extern const struct squashfs_backend squashfs_bdev_ops;
+
+/* mtd.c */
+extern const struct squashfs_backend squashfs_mtd_ops;
-- 
1.5.6.5


From 294602c4fed7319aeb5c44fff5bb8649f1f37b6e Mon Sep 17 00:00:00 2001
From: Ferenc Wagner <wferi@niif.hu>
Date: Tue, 23 Mar 2010 10:21:40 +0100
Subject: [PATCH] squashfs: pull out common logic of backend init functions

---
 fs/squashfs/backend.h |    2 +-
 fs/squashfs/bdev.c    |  155 +++++++++----------------------------------------
 fs/squashfs/block.c   |   33 ++++++++++-
 fs/squashfs/mtd.c     |   38 +-----------
 4 files changed, 62 insertions(+), 166 deletions(-)

diff --git a/fs/squashfs/backend.h b/fs/squashfs/backend.h
index 308fe6d..5beba00 100644
--- a/fs/squashfs/backend.h
+++ b/fs/squashfs/backend.h
@@ -4,7 +4,7 @@ 
 #include "squashfs_fs_sb.h"
 
 struct squashfs_backend {
-	int	(*init)(struct super_block *, u64, int, int, u64 *, int *);
+	int	(*init)(struct super_block *, u64, int);
 	void	(*free)(struct squashfs_sb_info *);
 	int	(*read)(struct super_block *, void *, int);
 	int	(*probe)(struct file_system_type *, int, const char *,
diff --git a/fs/squashfs/bdev.c b/fs/squashfs/bdev.c
index 497029b..20065b7 100644
--- a/fs/squashfs/bdev.c
+++ b/fs/squashfs/bdev.c
@@ -16,156 +16,55 @@  struct squashfs_bdev {
 	int b;				/* total number of buffer heads */
 };
 
-/*
- * Read the metadata block length, this is stored in the first two
- * bytes of the metadata block.
- */
-static struct buffer_head *get_block_length(struct super_block *sb,
-			u64 *cur_index, int *offset, int *length)
+static void bdev_free(struct squashfs_sb_info *msblk)
 {
-	struct squashfs_sb_info *msblk = sb->s_fs_info;
 	struct squashfs_bdev *bdev = msblk->backend_data;
-	struct buffer_head *bh;
-
-	TRACE("get_block_length: cur_index=0x%llx, offset=%d\n",
-		(unsigned long long)*cur_index, *offset);
-	bh = sb_bread(sb, *cur_index);
-	if (bh == NULL)
-		return NULL;
 
-	if (bdev->devblksize - *offset == 1) {
-		*length = (unsigned char) bh->b_data[*offset];
-		put_bh(bh);
-		bh = sb_bread(sb, ++(*cur_index));
-		if (bh == NULL)
-			return NULL;
-		*length |= (unsigned char) bh->b_data[0] << 8;
-		*offset = 1;
-	} else {
-		*length = (unsigned char) bh->b_data[*offset] |
-			(unsigned char) bh->b_data[*offset + 1] << 8;
-		*offset += 2;
-	}
-
-	TRACE("get_block_length: length=%d, offset=%d\n", *length, *offset);
-	return bh;
+	TRACE("bdev_free: bh=%p, bh_index=%d, b=%d\n",
+		bdev->bh, bdev->bh_index, bdev->b);
+	while (bdev->bh_index < bdev->b)
+		put_bh(bdev->bh[bdev->bh_index++]);
+	kfree(bdev->bh);
+	bdev->bh = NULL;
 }
 
-/*
- * Read a metadata block or datablock into memory.  Length is non-zero
- * if a datablock is being read (the size is stored elsewhere in the
- * filesystem), otherwise the length is obtained from the first two bytes of
- * the metadata block.  A bit in the length field indicates if the block
- * is stored uncompressed in the filesystem (usually because compression
- * generated a larger block - this does occasionally happen with zlib).
- */
-static int bdev_init(struct super_block *sb, u64 index, int length,
-			int srclength, u64 *next_index, int *compressed)
+static int bdev_init(struct super_block *sb, u64 index, int length)
 {
 	struct squashfs_sb_info *msblk = sb->s_fs_info;
 	struct squashfs_bdev *bdev = msblk->backend_data;
-	struct buffer_head **bh;
 	u64 cur_index;			/* sector of next byte */
 	int bytes;			/* number of bytes handled */
-	int b;				/* for counting buffer heads */
 
-	TRACE("Entering bdev_init: index=0x%llx, length=0x%x, srclength=%d\n",
-		index, length, srclength);
-	bh = kcalloc((max_t(int, msblk->block_size, SQUASHFS_METADATA_SIZE)
-		      >> bdev->devblksize_log2) + 1,
-		     sizeof(*bh), GFP_KERNEL);
-	if (bh == NULL)
+	TRACE("Entering bdev_init: index=0x%llx, length=0x%x\n",
+		index, length);
+	bdev->bh = kcalloc((max_t(int, msblk->block_size, SQUASHFS_METADATA_SIZE)
+			    >> bdev->devblksize_log2) + 1,
+			   sizeof(*bdev->bh), GFP_KERNEL);
+	if (bdev->bh == NULL)
 		return -ENOMEM;
 
 	bdev->offset = index & ((1 << bdev->devblksize_log2) - 1);
 	cur_index = index >> bdev->devblksize_log2;
-	if (length) { /* FIXME: this logic and the comment above should be pulled out! */
-		/*
-		 * Datablock.
-		 */
-		bytes = -bdev->offset;
-		*compressed = SQUASHFS_COMPRESSED_BLOCK(length);
-		length = SQUASHFS_COMPRESSED_SIZE_BLOCK(length);
-		if (next_index)
-			*next_index = index + length;
-
-		TRACE("Block @ 0x%llx, %scompressed size %d, src size %d\n",
-			index, *compressed ? "" : "un", length, srclength);
-
-		if (length < 0 || length > srclength ||
-				(index + length) > msblk->bytes_used)
-			goto read_failure;
-
-		for (b = 0; bytes < length; b++, cur_index++) {
-			bh[b] = sb_getblk(sb, cur_index);
-			if (bh[b] == NULL)
-				goto block_release;
-			bytes += bdev->devblksize;
-		}
-		ll_rw_block(READ, b, bh);
-	} else {
-		/*
-		 * Metadata block.
-		 */
-		if ((index + 2) > msblk->bytes_used)
-			goto read_failure;
-
-		bh[0] = get_block_length(sb, &cur_index, &bdev->offset, &length);
-		if (bh[0] == NULL)
-			goto read_failure;
-		b = 1;
-
-		bytes = bdev->devblksize - bdev->offset;
-		*compressed = SQUASHFS_COMPRESSED(length);
-		length = SQUASHFS_COMPRESSED_SIZE(length);
-		if (next_index)
-			*next_index = index + length + 2;
 
-		TRACE("Block @ 0x%llx, %scompressed size %d\n", index,
-				*compressed ? "" : "un", length);
-
-		if (length < 0 || length > srclength ||
-					(index + length) > msblk->bytes_used)
-			goto block_release;
+	bdev->bh_index = 0;
+	bdev->length = length;
+	bdev->bytes_read = 0;
 
-		for (; bytes < length; b++) {
-			bh[b] = sb_getblk(sb, ++cur_index);
-			if (bh[b] == NULL)
-				goto block_release;
-			bytes += bdev->devblksize;
+	bytes = -bdev->offset;
+	for (bdev->b = 0; bytes < length; bdev->b++, cur_index++) {
+		if ((bdev->bh[bdev->b] = sb_getblk(sb, cur_index)) == NULL) {
+			ERROR("bdev_init failed to read block 0x%llx\n",
+			      (unsigned long long) index);
+			bdev_free(msblk);
+			return -EIO;
 		}
-		ll_rw_block(READ, b - 1, bh + 1);
+		bytes += bdev->devblksize;
 	}
+	ll_rw_block(READ, bdev->b, bdev->bh);
 
-	bdev->bh = bh;
-	bdev->length = length;
-	bdev->bytes_read = 0;
-	bdev->b = b;
-	bdev->bh_index = 0;
 	TRACE("bdev_init: allocated %d buffer heads at %p, returning length=%d",
-	      b, bh, length);
+	      bdev->b, bdev->bh, length);
 	return length;
-
-block_release:
-	while (b--)
-		put_bh(bh[b]);
-read_failure:
-	ERROR("bdev_init failed to read block 0x%llx\n",
-					(unsigned long long) index);
-	kfree(bh);
-	return -EIO;
-}
-
-static void bdev_free(struct squashfs_sb_info *msblk)
-{
-	struct squashfs_bdev *bdev = msblk->backend_data;
-
-	TRACE("bdev_free: bh=%p, bh_index=%d, b=%d\n",
-		bdev->bh, bdev->bh_index, bdev->b);
-	while (bdev->bh_index < bdev->b)
-		put_bh(bdev->bh[bdev->bh_index++]);
-	kfree(bdev->bh);
-	bdev->bh = NULL;
 }
 
 static int bdev_read(struct super_block *sb, void *buf, int len)
diff --git a/fs/squashfs/block.c b/fs/squashfs/block.c
index 1c85063..eb33770 100644
--- a/fs/squashfs/block.c
+++ b/fs/squashfs/block.c
@@ -53,8 +53,37 @@  int squashfs_read_data(struct super_block *sb, void **buffer, u64 index,
 	int compressed;
 	struct squashfs_sb_info *msblk = sb->s_fs_info;
 
-	length = msblk->backend->init(sb, index, length, srclength,
-					next_index, &compressed);
+	if (length) { /* Data block */
+		compressed = SQUASHFS_COMPRESSED_BLOCK(length);
+		length = SQUASHFS_COMPRESSED_SIZE_BLOCK(length);
+		TRACE("Block @ 0x%llx, %scompressed size %d, src size %d\n",
+			index, compressed ? "" : "un", length, srclength);
+	} else { /* Metadata block */
+		int read;
+		u16 metalen;
+
+		length = msblk->backend->init(sb, index, 2);
+		if (length < 0)
+			return length;
+		read = msblk->backend->read(sb, &metalen, 2);
+		if (read != 2)
+			goto read_failure;
+		msblk->backend->free(msblk);
+		length = le16_to_cpu(metalen);
+		compressed = SQUASHFS_COMPRESSED(length);
+		length = SQUASHFS_COMPRESSED_SIZE(length);
+		index += 2;
+		TRACE("Block @ 0x%llx, %scompressed size %d\n", index,
+				compressed ? "" : "un", length);
+	}
+	if (length < 0 || length > srclength ||
+	    (index + length) > msblk->bytes_used)
+		goto read_failure;
+
+	if (next_index)
+		*next_index = index + length;
+
+	length = msblk->backend->init(sb, index, length);
 	if (length < 0)
 		return length;
 
diff --git a/fs/squashfs/mtd.c b/fs/squashfs/mtd.c
index c9ba361..99f2d35 100644
--- a/fs/squashfs/mtd.c
+++ b/fs/squashfs/mtd.c
@@ -38,50 +38,18 @@  static int checked_mtd_read(struct super_block *sb, u64 index, int length,
 	return 0;
 }
 
-static int mtd_init(struct super_block *sb, u64 index, int length,
-		    int srclength, u64 *next_index, int *compressed)
+static int mtd_init(struct super_block *sb, u64 index, int length)
 {
 	struct squashfs_sb_info *msblk = sb->s_fs_info;
 	struct squashfs_mtd *mtd = msblk->backend_data;
 
-	TRACE("Entering mtd_init: index=0x%llx, length=0x%x, srclength=%d\n",
-		index, length, srclength);
-
-	if (length) { /* datablock */
-		*compressed = SQUASHFS_COMPRESSED_BLOCK(length);
-		length = SQUASHFS_COMPRESSED_SIZE_BLOCK(length);
-		if (next_index)
-			*next_index = index + length;
-		TRACE("Block @ 0x%llx, %scompressed size %d, src size %d\n",
-			index, *compressed ? "" : "un", length, srclength);
-	} else { /* metadata block */
-		u16 metalen;
-		if ((index + 2) > msblk->bytes_used)
-			goto read_failure;
-		if (checked_mtd_read(sb, index, 2, &metalen))
-			goto read_failure;
-		length = le16_to_cpu(metalen);
-		*compressed = SQUASHFS_COMPRESSED(length);
-		length = SQUASHFS_COMPRESSED_SIZE(length);
-		if (next_index)
-			*next_index = index + length + 2;
-		TRACE("Block @ 0x%llx, %scompressed size %d\n", index,
-				*compressed ? "" : "un", length);
-		index += 2;
-	}
-	if (length < 0 || length > srclength ||
-			(index + length) > msblk->bytes_used)
-		goto read_failure;
+	TRACE("Entering mtd_init: index=0x%llx, length=0x%x\n",
+		index, length);
 
 	mtd->index = index;
 	mtd->length = length;
 	mtd->bytes_read = 0;
 	return length;
-
-read_failure:
-	ERROR("mtd_init failed to read block 0x%llx\n",
-					(unsigned long long) index);
-	return -EIO;
 }
 
 static void mtd_free(struct squashfs_sb_info *msblk)
-- 
1.5.6.5