decompressors: add boot-time XZ support

This implements the API defined in <linux/decompress/generic.h> which is
used for kernel, initramfs, and initrd decompression.  This patch together
with the first patch is enough for XZ-compressed initramfs and initrd;
XZ-compressed kernel will need arch-specific changes.

The buffering requirements described in decompress_unxz.c are stricter
than with gzip, so the relevant changes should be done to the
arch-specific code when adding support for XZ-compressed kernel.
Similarly, the heap size in arch-specific pre-boot code may need to be
increased (30 KiB is enough).

The XZ decompressor needs memmove(), memeq() (memcmp() == 0), and
memzero() (memset(ptr, 0, size)), which aren't available in all
arch-specific pre-boot environments.  I'm including simple versions in
decompress_unxz.c, but a cleaner solution would naturally be nicer.

Signed-off-by: Lasse Collin <lasse.collin@tukaani.org>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Cc: Alain Knaff <alain@knaff.lu>
Cc: Albin Tonnerre <albin.tonnerre@free-electrons.com>
Cc: Phillip Lougher <phillip@lougher.demon.co.uk>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
diff --git a/include/linux/decompress/unxz.h b/include/linux/decompress/unxz.h
new file mode 100644
index 0000000..41728fc
--- /dev/null
+++ b/include/linux/decompress/unxz.h
@@ -0,0 +1,19 @@
+/*
+ * Wrapper for decompressing XZ-compressed kernel, initramfs, and initrd
+ *
+ * Author: Lasse Collin <lasse.collin@tukaani.org>
+ *
+ * This file has been put into the public domain.
+ * You can do whatever you want with this file.
+ */
+
+#ifndef DECOMPRESS_UNXZ_H
+#define DECOMPRESS_UNXZ_H
+
+int unxz(unsigned char *in, int in_size,
+	 int (*fill)(void *dest, unsigned int size),
+	 int (*flush)(void *src, unsigned int size),
+	 unsigned char *out, int *in_used,
+	 void (*error)(char *x));
+
+#endif
diff --git a/init/Kconfig b/init/Kconfig
index 8dfd094..ea176e8 100644
--- a/init/Kconfig
+++ b/init/Kconfig
@@ -130,13 +130,16 @@
 config HAVE_KERNEL_LZMA
 	bool
 
+config HAVE_KERNEL_XZ
+	bool
+
 config HAVE_KERNEL_LZO
 	bool
 
 choice
 	prompt "Kernel compression mode"
 	default KERNEL_GZIP
-	depends on HAVE_KERNEL_GZIP || HAVE_KERNEL_BZIP2 || HAVE_KERNEL_LZMA || HAVE_KERNEL_LZO
+	depends on HAVE_KERNEL_GZIP || HAVE_KERNEL_BZIP2 || HAVE_KERNEL_LZMA || HAVE_KERNEL_XZ || HAVE_KERNEL_LZO
 	help
 	  The linux kernel is a kind of self-extracting executable.
 	  Several compression algorithms are available, which differ
@@ -181,6 +184,21 @@
 	  two. Compression is slowest.	The kernel size is about 33%
 	  smaller with LZMA in comparison to gzip.
 
+config KERNEL_XZ
+	bool "XZ"
+	depends on HAVE_KERNEL_XZ
+	help
+	  XZ uses the LZMA2 algorithm and instruction set specific
+	  BCJ filters which can improve compression ratio of executable
+	  code. The size of the kernel is about 30% smaller with XZ in
+	  comparison to gzip. On architectures for which there is a BCJ
+	  filter (i386, x86_64, ARM, IA-64, PowerPC, and SPARC), XZ
+	  will create a few percent smaller kernel than plain LZMA.
+
+	  The speed is about the same as with LZMA: The decompression
+	  speed of XZ is better than that of bzip2 but worse than gzip
+	  and LZO. Compression is slow.
+
 config KERNEL_LZO
 	bool "LZO"
 	depends on HAVE_KERNEL_LZO
diff --git a/lib/Kconfig b/lib/Kconfig
index 2b8f854..0ee67e0 100644
--- a/lib/Kconfig
+++ b/lib/Kconfig
@@ -122,6 +122,10 @@
 config DECOMPRESS_LZMA
 	tristate
 
+config DECOMPRESS_XZ
+	select XZ_DEC
+	tristate
+
 config DECOMPRESS_LZO
 	select LZO_DECOMPRESS
 	tristate
diff --git a/lib/Makefile b/lib/Makefile
index 4df2d02..cbb774f 100644
--- a/lib/Makefile
+++ b/lib/Makefile
@@ -75,6 +75,7 @@
 lib-$(CONFIG_DECOMPRESS_GZIP) += decompress_inflate.o
 lib-$(CONFIG_DECOMPRESS_BZIP2) += decompress_bunzip2.o
 lib-$(CONFIG_DECOMPRESS_LZMA) += decompress_unlzma.o
+lib-$(CONFIG_DECOMPRESS_XZ) += decompress_unxz.o
 lib-$(CONFIG_DECOMPRESS_LZO) += decompress_unlzo.o
 
 obj-$(CONFIG_TEXTSEARCH) += textsearch.o
diff --git a/lib/decompress.c b/lib/decompress.c
index a760681..3d766b7f 100644
--- a/lib/decompress.c
+++ b/lib/decompress.c
@@ -8,6 +8,7 @@
 
 #include <linux/decompress/bunzip2.h>
 #include <linux/decompress/unlzma.h>
+#include <linux/decompress/unxz.h>
 #include <linux/decompress/inflate.h>
 #include <linux/decompress/unlzo.h>
 
@@ -23,6 +24,9 @@
 #ifndef CONFIG_DECOMPRESS_LZMA
 # define unlzma NULL
 #endif
+#ifndef CONFIG_DECOMPRESS_XZ
+# define unxz NULL
+#endif
 #ifndef CONFIG_DECOMPRESS_LZO
 # define unlzo NULL
 #endif
@@ -36,6 +40,7 @@
 	{ {037, 0236}, "gzip", gunzip },
 	{ {0x42, 0x5a}, "bzip2", bunzip2 },
 	{ {0x5d, 0x00}, "lzma", unlzma },
+	{ {0xfd, 0x37}, "xz", unxz },
 	{ {0x89, 0x4c}, "lzo", unlzo },
 	{ {0, 0}, NULL, NULL }
 };
diff --git a/lib/decompress_unxz.c b/lib/decompress_unxz.c
new file mode 100644
index 0000000..cecd23d
--- /dev/null
+++ b/lib/decompress_unxz.c
@@ -0,0 +1,397 @@
+/*
+ * Wrapper for decompressing XZ-compressed kernel, initramfs, and initrd
+ *
+ * Author: Lasse Collin <lasse.collin@tukaani.org>
+ *
+ * This file has been put into the public domain.
+ * You can do whatever you want with this file.
+ */
+
+/*
+ * Important notes about in-place decompression
+ *
+ * At least on x86, the kernel is decompressed in place: the compressed data
+ * is placed to the end of the output buffer, and the decompressor overwrites
+ * most of the compressed data. There must be enough safety margin to
+ * guarantee that the write position is always behind the read position.
+ *
+ * The safety margin for XZ with LZMA2 or BCJ+LZMA2 is calculated below.
+ * Note that the margin with XZ is bigger than with Deflate (gzip)!
+ *
+ * The worst case for in-place decompression is that the beginning of
+ * the file is compressed extremely well, and the rest of the file is
+ * uncompressible. Thus, we must look for worst-case expansion when the
+ * compressor is encoding uncompressible data.
+ *
+ * The structure of the .xz file in case of a compresed kernel is as follows.
+ * Sizes (as bytes) of the fields are in parenthesis.
+ *
+ *    Stream Header (12)
+ *    Block Header:
+ *      Block Header (8-12)
+ *      Compressed Data (N)
+ *      Block Padding (0-3)
+ *      CRC32 (4)
+ *    Index (8-20)
+ *    Stream Footer (12)
+ *
+ * Normally there is exactly one Block, but let's assume that there are
+ * 2-4 Blocks just in case. Because Stream Header and also Block Header
+ * of the first Block don't make the decompressor produce any uncompressed
+ * data, we can ignore them from our calculations. Block Headers of possible
+ * additional Blocks have to be taken into account still. With these
+ * assumptions, it is safe to assume that the total header overhead is
+ * less than 128 bytes.
+ *
+ * Compressed Data contains LZMA2 or BCJ+LZMA2 encoded data. Since BCJ
+ * doesn't change the size of the data, it is enough to calculate the
+ * safety margin for LZMA2.
+ *
+ * LZMA2 stores the data in chunks. Each chunk has a header whose size is
+ * a maximum of 6 bytes, but to get round 2^n numbers, let's assume that
+ * the maximum chunk header size is 8 bytes. After the chunk header, there
+ * may be up to 64 KiB of actual payload in the chunk. Often the payload is
+ * quite a bit smaller though; to be safe, let's assume that an average
+ * chunk has only 32 KiB of payload.
+ *
+ * The maximum uncompressed size of the payload is 2 MiB. The minimum
+ * uncompressed size of the payload is in practice never less than the
+ * payload size itself. The LZMA2 format would allow uncompressed size
+ * to be less than the payload size, but no sane compressor creates such
+ * files. LZMA2 supports storing uncompressible data in uncompressed form,
+ * so there's never a need to create payloads whose uncompressed size is
+ * smaller than the compressed size.
+ *
+ * The assumption, that the uncompressed size of the payload is never
+ * smaller than the payload itself, is valid only when talking about
+ * the payload as a whole. It is possible that the payload has parts where
+ * the decompressor consumes more input than it produces output. Calculating
+ * the worst case for this would be tricky. Instead of trying to do that,
+ * let's simply make sure that the decompressor never overwrites any bytes
+ * of the payload which it is currently reading.
+ *
+ * Now we have enough information to calculate the safety margin. We need
+ *   - 128 bytes for the .xz file format headers;
+ *   - 8 bytes per every 32 KiB of uncompressed size (one LZMA2 chunk header
+ *     per chunk, each chunk having average payload size of 32 KiB); and
+ *   - 64 KiB (biggest possible LZMA2 chunk payload size) to make sure that
+ *     the decompressor never overwrites anything from the LZMA2 chunk
+ *     payload it is currently reading.
+ *
+ * We get the following formula:
+ *
+ *    safety_margin = 128 + uncompressed_size * 8 / 32768 + 65536
+ *                  = 128 + (uncompressed_size >> 12) + 65536
+ *
+ * For comparision, according to arch/x86/boot/compressed/misc.c, the
+ * equivalent formula for Deflate is this:
+ *
+ *    safety_margin = 18 + (uncompressed_size >> 12) + 32768
+ *
+ * Thus, when updating Deflate-only in-place kernel decompressor to
+ * support XZ, the fixed overhead has to be increased from 18+32768 bytes
+ * to 128+65536 bytes.
+ */
+
+/*
+ * STATIC is defined to "static" if we are being built for kernel
+ * decompression (pre-boot code). <linux/decompress/mm.h> will define
+ * STATIC to empty if it wasn't already defined. Since we will need to
+ * know later if we are being used for kernel decompression, we define
+ * XZ_PREBOOT here.
+ */
+#ifdef STATIC
+#	define XZ_PREBOOT
+#endif
+#ifdef __KERNEL__
+#	include <linux/decompress/mm.h>
+#endif
+#define XZ_EXTERN STATIC
+
+#ifndef XZ_PREBOOT
+#	include <linux/slab.h>
+#	include <linux/xz.h>
+#else
+/*
+ * Use the internal CRC32 code instead of kernel's CRC32 module, which
+ * is not available in early phase of booting.
+ */
+#define XZ_INTERNAL_CRC32 1
+
+/*
+ * For boot time use, we enable only the BCJ filter of the current
+ * architecture or none if no BCJ filter is available for the architecture.
+ */
+#ifdef CONFIG_X86
+#	define XZ_DEC_X86
+#endif
+#ifdef CONFIG_PPC
+#	define XZ_DEC_POWERPC
+#endif
+#ifdef CONFIG_ARM
+#	define XZ_DEC_ARM
+#endif
+#ifdef CONFIG_IA64
+#	define XZ_DEC_IA64
+#endif
+#ifdef CONFIG_SPARC
+#	define XZ_DEC_SPARC
+#endif
+
+/*
+ * This will get the basic headers so that memeq() and others
+ * can be defined.
+ */
+#include "xz/xz_private.h"
+
+/*
+ * Replace the normal allocation functions with the versions from
+ * <linux/decompress/mm.h>. vfree() needs to support vfree(NULL)
+ * when XZ_DYNALLOC is used, but the pre-boot free() doesn't support it.
+ * Workaround it here because the other decompressors don't need it.
+ */
+#undef kmalloc
+#undef kfree
+#undef vmalloc
+#undef vfree
+#define kmalloc(size, flags) malloc(size)
+#define kfree(ptr) free(ptr)
+#define vmalloc(size) malloc(size)
+#define vfree(ptr) do { if (ptr != NULL) free(ptr); } while (0)
+
+/*
+ * FIXME: Not all basic memory functions are provided in architecture-specific
+ * files (yet). We define our own versions here for now, but this should be
+ * only a temporary solution.
+ *
+ * memeq and memzero are not used much and any remotely sane implementation
+ * is fast enough. memcpy/memmove speed matters in multi-call mode, but
+ * the kernel image is decompressed in single-call mode, in which only
+ * memcpy speed can matter and only if there is a lot of uncompressible data
+ * (LZMA2 stores uncompressible chunks in uncompressed form). Thus, the
+ * functions below should just be kept small; it's probably not worth
+ * optimizing for speed.
+ */
+
+#ifndef memeq
+static bool memeq(const void *a, const void *b, size_t size)
+{
+	const uint8_t *x = a;
+	const uint8_t *y = b;
+	size_t i;
+
+	for (i = 0; i < size; ++i)
+		if (x[i] != y[i])
+			return false;
+
+	return true;
+}
+#endif
+
+#ifndef memzero
+static void memzero(void *buf, size_t size)
+{
+	uint8_t *b = buf;
+	uint8_t *e = b + size;
+
+	while (b != e)
+		*b++ = '\0';
+}
+#endif
+
+#ifndef memmove
+/* Not static to avoid a conflict with the prototype in the Linux headers. */
+void *memmove(void *dest, const void *src, size_t size)
+{
+	uint8_t *d = dest;
+	const uint8_t *s = src;
+	size_t i;
+
+	if (d < s) {
+		for (i = 0; i < size; ++i)
+			d[i] = s[i];
+	} else if (d > s) {
+		i = size;
+		while (i-- > 0)
+			d[i] = s[i];
+	}
+
+	return dest;
+}
+#endif
+
+/*
+ * Since we need memmove anyway, would use it as memcpy too.
+ * Commented out for now to avoid breaking things.
+ */
+/*
+#ifndef memcpy
+#	define memcpy memmove
+#endif
+*/
+
+#include "xz/xz_crc32.c"
+#include "xz/xz_dec_stream.c"
+#include "xz/xz_dec_lzma2.c"
+#include "xz/xz_dec_bcj.c"
+
+#endif /* XZ_PREBOOT */
+
+/* Size of the input and output buffers in multi-call mode */
+#define XZ_IOBUF_SIZE 4096
+
+/*
+ * This function implements the API defined in <linux/decompress/generic.h>.
+ *
+ * This wrapper will automatically choose single-call or multi-call mode
+ * of the native XZ decoder API. The single-call mode can be used only when
+ * both input and output buffers are available as a single chunk, i.e. when
+ * fill() and flush() won't be used.
+ */
+STATIC int INIT unxz(unsigned char *in, int in_size,
+		     int (*fill)(void *dest, unsigned int size),
+		     int (*flush)(void *src, unsigned int size),
+		     unsigned char *out, int *in_used,
+		     void (*error)(char *x))
+{
+	struct xz_buf b;
+	struct xz_dec *s;
+	enum xz_ret ret;
+	bool must_free_in = false;
+
+#if XZ_INTERNAL_CRC32
+	xz_crc32_init();
+#endif
+
+	if (in_used != NULL)
+		*in_used = 0;
+
+	if (fill == NULL && flush == NULL)
+		s = xz_dec_init(XZ_SINGLE, 0);
+	else
+		s = xz_dec_init(XZ_DYNALLOC, (uint32_t)-1);
+
+	if (s == NULL)
+		goto error_alloc_state;
+
+	if (flush == NULL) {
+		b.out = out;
+		b.out_size = (size_t)-1;
+	} else {
+		b.out_size = XZ_IOBUF_SIZE;
+		b.out = malloc(XZ_IOBUF_SIZE);
+		if (b.out == NULL)
+			goto error_alloc_out;
+	}
+
+	if (in == NULL) {
+		must_free_in = true;
+		in = malloc(XZ_IOBUF_SIZE);
+		if (in == NULL)
+			goto error_alloc_in;
+	}
+
+	b.in = in;
+	b.in_pos = 0;
+	b.in_size = in_size;
+	b.out_pos = 0;
+
+	if (fill == NULL && flush == NULL) {
+		ret = xz_dec_run(s, &b);
+	} else {
+		do {
+			if (b.in_pos == b.in_size && fill != NULL) {
+				if (in_used != NULL)
+					*in_used += b.in_pos;
+
+				b.in_pos = 0;
+
+				in_size = fill(in, XZ_IOBUF_SIZE);
+				if (in_size < 0) {
+					/*
+					 * This isn't an optimal error code
+					 * but it probably isn't worth making
+					 * a new one either.
+					 */
+					ret = XZ_BUF_ERROR;
+					break;
+				}
+
+				b.in_size = in_size;
+			}
+
+			ret = xz_dec_run(s, &b);
+
+			if (flush != NULL && (b.out_pos == b.out_size
+					|| (ret != XZ_OK && b.out_pos > 0))) {
+				/*
+				 * Setting ret here may hide an error
+				 * returned by xz_dec_run(), but probably
+				 * it's not too bad.
+				 */
+				if (flush(b.out, b.out_pos) != (int)b.out_pos)
+					ret = XZ_BUF_ERROR;
+
+				b.out_pos = 0;
+			}
+		} while (ret == XZ_OK);
+
+		if (must_free_in)
+			free(in);
+
+		if (flush != NULL)
+			free(b.out);
+	}
+
+	if (in_used != NULL)
+		*in_used += b.in_pos;
+
+	xz_dec_end(s);
+
+	switch (ret) {
+	case XZ_STREAM_END:
+		return 0;
+
+	case XZ_MEM_ERROR:
+		/* This can occur only in multi-call mode. */
+		error("XZ decompressor ran out of memory");
+		break;
+
+	case XZ_FORMAT_ERROR:
+		error("Input is not in the XZ format (wrong magic bytes)");
+		break;
+
+	case XZ_OPTIONS_ERROR:
+		error("Input was encoded with settings that are not "
+				"supported by this XZ decoder");
+		break;
+
+	case XZ_DATA_ERROR:
+	case XZ_BUF_ERROR:
+		error("XZ-compressed data is corrupt");
+		break;
+
+	default:
+		error("Bug in the XZ decompressor");
+		break;
+	}
+
+	return -1;
+
+error_alloc_in:
+	if (flush != NULL)
+		free(b.out);
+
+error_alloc_out:
+	xz_dec_end(s);
+
+error_alloc_state:
+	error("XZ decompressor ran out of memory");
+	return -1;
+}
+
+/*
+ * This macro is used by architecture-specific files to decompress
+ * the kernel image.
+ */
+#define decompress unxz
diff --git a/scripts/gen_initramfs_list.sh b/scripts/gen_initramfs_list.sh
index 5958fff..55caecd 100644
--- a/scripts/gen_initramfs_list.sh
+++ b/scripts/gen_initramfs_list.sh
@@ -243,6 +243,8 @@
 		echo "$output_file" | grep -q "\.gz$" && compr="gzip -9 -f"
 		echo "$output_file" | grep -q "\.bz2$" && compr="bzip2 -9 -f"
 		echo "$output_file" | grep -q "\.lzma$" && compr="lzma -9 -f"
+		echo "$output_file" | grep -q "\.xz$" && \
+				compr="xz --check=crc32 --lzma2=dict=1MiB"
 		echo "$output_file" | grep -q "\.lzo$" && compr="lzop -9 -f"
 		echo "$output_file" | grep -q "\.cpio$" && compr="cat"
 		shift
diff --git a/usr/Kconfig b/usr/Kconfig
index c2c7fe2..4780dea 100644
--- a/usr/Kconfig
+++ b/usr/Kconfig
@@ -72,6 +72,15 @@
 	  Support loading of a LZMA encoded initial ramdisk or cpio buffer
 	  If unsure, say N.
 
+config RD_XZ
+	bool "Support initial ramdisks compressed using XZ" if EMBEDDED
+	default !EMBEDDED
+	depends on BLK_DEV_INITRD
+	select DECOMPRESS_XZ
+	help
+	  Support loading of a XZ encoded initial ramdisk or cpio buffer.
+	  If unsure, say N.
+
 config RD_LZO
 	bool "Support initial ramdisks compressed using LZO" if EMBEDDED
 	default !EMBEDDED
@@ -139,6 +148,15 @@
 	  three. Compression is slowest. The initramfs size is about 33%
 	  smaller with LZMA in comparison to gzip.
 
+config INITRAMFS_COMPRESSION_XZ
+	bool "XZ"
+	depends on RD_XZ
+	help
+	  XZ uses the LZMA2 algorithm. The initramfs size is about 30%
+	  smaller with XZ in comparison to gzip. Decompression speed
+	  is better than that of bzip2 but worse than gzip and LZO.
+	  Compression is slow.
+
 config INITRAMFS_COMPRESSION_LZO
 	bool "LZO"
 	depends on RD_LZO
diff --git a/usr/Makefile b/usr/Makefile
index 6faa444..029ffe6 100644
--- a/usr/Makefile
+++ b/usr/Makefile
@@ -15,6 +15,9 @@
 # Lzma
 suffix_$(CONFIG_INITRAMFS_COMPRESSION_LZMA)   = .lzma
 
+# XZ
+suffix_$(CONFIG_INITRAMFS_COMPRESSION_XZ)     = .xz
+
 # Lzo
 suffix_$(CONFIG_INITRAMFS_COMPRESSION_LZO)   = .lzo
 
@@ -50,7 +53,7 @@
 quiet_cmd_initfs = GEN     $@
       cmd_initfs = $(initramfs) -o $@ $(ramfs-args) $(ramfs-input)
 
-targets := initramfs_data.cpio.gz initramfs_data.cpio.bz2 initramfs_data.cpio.lzma initramfs_data.cpio.lzo initramfs_data.cpio
+targets := initramfs_data.cpio.gz initramfs_data.cpio.bz2 initramfs_data.cpio.lzma initramfs_data.cpio.xz initramfs_data.cpio.lzo initramfs_data.cpio
 # do not try to update files included in initramfs
 $(deps_initramfs): ;