xref: /OK3568_Linux_fs/kernel/lib/decompress_unxz.c (revision 4882a59341e53eb6f0b4789bf948001014eff981)
1 /*
2  * Wrapper for decompressing XZ-compressed kernel, initramfs, and initrd
3  *
4  * Author: Lasse Collin <lasse.collin@tukaani.org>
5  *
6  * This file has been put into the public domain.
7  * You can do whatever you want with this file.
8  */
9 
10 /*
11  * Important notes about in-place decompression
12  *
13  * At least on x86, the kernel is decompressed in place: the compressed data
14  * is placed to the end of the output buffer, and the decompressor overwrites
15  * most of the compressed data. There must be enough safety margin to
16  * guarantee that the write position is always behind the read position.
17  *
18  * The safety margin for XZ with LZMA2 or BCJ+LZMA2 is calculated below.
19  * Note that the margin with XZ is bigger than with Deflate (gzip)!
20  *
21  * The worst case for in-place decompression is that the beginning of
22  * the file is compressed extremely well, and the rest of the file is
23  * uncompressible. Thus, we must look for worst-case expansion when the
24  * compressor is encoding uncompressible data.
25  *
26  * The structure of the .xz file in case of a compresed kernel is as follows.
27  * Sizes (as bytes) of the fields are in parenthesis.
28  *
29  *    Stream Header (12)
30  *    Block Header:
31  *      Block Header (8-12)
32  *      Compressed Data (N)
33  *      Block Padding (0-3)
34  *      CRC32 (4)
35  *    Index (8-20)
36  *    Stream Footer (12)
37  *
38  * Normally there is exactly one Block, but let's assume that there are
39  * 2-4 Blocks just in case. Because Stream Header and also Block Header
40  * of the first Block don't make the decompressor produce any uncompressed
41  * data, we can ignore them from our calculations. Block Headers of possible
42  * additional Blocks have to be taken into account still. With these
43  * assumptions, it is safe to assume that the total header overhead is
44  * less than 128 bytes.
45  *
46  * Compressed Data contains LZMA2 or BCJ+LZMA2 encoded data. Since BCJ
47  * doesn't change the size of the data, it is enough to calculate the
48  * safety margin for LZMA2.
49  *
50  * LZMA2 stores the data in chunks. Each chunk has a header whose size is
51  * a maximum of 6 bytes, but to get round 2^n numbers, let's assume that
52  * the maximum chunk header size is 8 bytes. After the chunk header, there
53  * may be up to 64 KiB of actual payload in the chunk. Often the payload is
54  * quite a bit smaller though; to be safe, let's assume that an average
55  * chunk has only 32 KiB of payload.
56  *
57  * The maximum uncompressed size of the payload is 2 MiB. The minimum
58  * uncompressed size of the payload is in practice never less than the
59  * payload size itself. The LZMA2 format would allow uncompressed size
60  * to be less than the payload size, but no sane compressor creates such
61  * files. LZMA2 supports storing uncompressible data in uncompressed form,
62  * so there's never a need to create payloads whose uncompressed size is
63  * smaller than the compressed size.
64  *
65  * The assumption, that the uncompressed size of the payload is never
66  * smaller than the payload itself, is valid only when talking about
67  * the payload as a whole. It is possible that the payload has parts where
68  * the decompressor consumes more input than it produces output. Calculating
69  * the worst case for this would be tricky. Instead of trying to do that,
70  * let's simply make sure that the decompressor never overwrites any bytes
71  * of the payload which it is currently reading.
72  *
73  * Now we have enough information to calculate the safety margin. We need
74  *   - 128 bytes for the .xz file format headers;
75  *   - 8 bytes per every 32 KiB of uncompressed size (one LZMA2 chunk header
76  *     per chunk, each chunk having average payload size of 32 KiB); and
77  *   - 64 KiB (biggest possible LZMA2 chunk payload size) to make sure that
78  *     the decompressor never overwrites anything from the LZMA2 chunk
79  *     payload it is currently reading.
80  *
81  * We get the following formula:
82  *
83  *    safety_margin = 128 + uncompressed_size * 8 / 32768 + 65536
84  *                  = 128 + (uncompressed_size >> 12) + 65536
85  *
86  * For comparison, according to arch/x86/boot/compressed/misc.c, the
87  * equivalent formula for Deflate is this:
88  *
89  *    safety_margin = 18 + (uncompressed_size >> 12) + 32768
90  *
91  * Thus, when updating Deflate-only in-place kernel decompressor to
92  * support XZ, the fixed overhead has to be increased from 18+32768 bytes
93  * to 128+65536 bytes.
94  */
95 
96 /*
97  * STATIC is defined to "static" if we are being built for kernel
98  * decompression (pre-boot code). <linux/decompress/mm.h> will define
99  * STATIC to empty if it wasn't already defined. Since we will need to
100  * know later if we are being used for kernel decompression, we define
101  * XZ_PREBOOT here.
102  */
103 #ifdef STATIC
104 #	define XZ_PREBOOT
105 #endif
106 #ifdef __KERNEL__
107 #	include <linux/decompress/mm.h>
108 #endif
109 #define XZ_EXTERN STATIC
110 
111 #ifndef XZ_PREBOOT
112 #	include <linux/slab.h>
113 #	include <linux/xz.h>
114 #else
115 /*
116  * Use the internal CRC32 code instead of kernel's CRC32 module, which
117  * is not available in early phase of booting.
118  */
119 #define XZ_INTERNAL_CRC32 1
120 
121 /*
122  * For boot time use, we enable only the BCJ filter of the current
123  * architecture or none if no BCJ filter is available for the architecture.
124  */
125 #ifdef CONFIG_X86
126 #	define XZ_DEC_X86
127 #endif
128 #ifdef CONFIG_PPC
129 #	define XZ_DEC_POWERPC
130 #endif
131 #ifdef CONFIG_ARM
132 #	ifdef CONFIG_THUMB2_KERNEL
133 #		define XZ_DEC_ARMTHUMB
134 #	else
135 #		define XZ_DEC_ARM
136 #	endif
137 #endif
138 #ifdef CONFIG_IA64
139 #	define XZ_DEC_IA64
140 #endif
141 #ifdef CONFIG_SPARC
142 #	define XZ_DEC_SPARC
143 #endif
144 
145 /*
146  * This will get the basic headers so that memeq() and others
147  * can be defined.
148  */
149 #include "xz/xz_private.h"
150 
151 /*
152  * Replace the normal allocation functions with the versions from
153  * <linux/decompress/mm.h>. vfree() needs to support vfree(NULL)
154  * when XZ_DYNALLOC is used, but the pre-boot free() doesn't support it.
155  * Workaround it here because the other decompressors don't need it.
156  */
157 #undef kmalloc
158 #undef kfree
159 #undef vmalloc
160 #undef vfree
161 #define kmalloc(size, flags) malloc(size)
162 #define kfree(ptr) free(ptr)
163 #define vmalloc(size) malloc(size)
164 #define vfree(ptr) do { if (ptr != NULL) free(ptr); } while (0)
165 
166 /*
167  * FIXME: Not all basic memory functions are provided in architecture-specific
168  * files (yet). We define our own versions here for now, but this should be
169  * only a temporary solution.
170  *
171  * memeq and memzero are not used much and any remotely sane implementation
172  * is fast enough. memcpy/memmove speed matters in multi-call mode, but
173  * the kernel image is decompressed in single-call mode, in which only
174  * memmove speed can matter and only if there is a lot of uncompressible data
175  * (LZMA2 stores uncompressible chunks in uncompressed form). Thus, the
176  * functions below should just be kept small; it's probably not worth
177  * optimizing for speed.
178  */
179 
180 #ifndef memeq
memeq(const void * a,const void * b,size_t size)181 static bool memeq(const void *a, const void *b, size_t size)
182 {
183 	const uint8_t *x = a;
184 	const uint8_t *y = b;
185 	size_t i;
186 
187 	for (i = 0; i < size; ++i)
188 		if (x[i] != y[i])
189 			return false;
190 
191 	return true;
192 }
193 #endif
194 
195 #ifndef memzero
memzero(void * buf,size_t size)196 static void memzero(void *buf, size_t size)
197 {
198 	uint8_t *b = buf;
199 	uint8_t *e = b + size;
200 
201 	while (b != e)
202 		*b++ = '\0';
203 }
204 #endif
205 
206 #ifndef memmove
207 /* Not static to avoid a conflict with the prototype in the Linux headers. */
memmove(void * dest,const void * src,size_t size)208 void *memmove(void *dest, const void *src, size_t size)
209 {
210 	uint8_t *d = dest;
211 	const uint8_t *s = src;
212 	size_t i;
213 
214 	if (d < s) {
215 		for (i = 0; i < size; ++i)
216 			d[i] = s[i];
217 	} else if (d > s) {
218 		i = size;
219 		while (i-- > 0)
220 			d[i] = s[i];
221 	}
222 
223 	return dest;
224 }
225 #endif
226 
227 /*
228  * Since we need memmove anyway, would use it as memcpy too.
229  * Commented out for now to avoid breaking things.
230  */
231 /*
232 #ifndef memcpy
233 #	define memcpy memmove
234 #endif
235 */
236 
237 #include "xz/xz_crc32.c"
238 #include "xz/xz_dec_stream.c"
239 #include "xz/xz_dec_lzma2.c"
240 #include "xz/xz_dec_bcj.c"
241 
242 #endif /* XZ_PREBOOT */
243 
244 /* Size of the input and output buffers in multi-call mode */
245 #define XZ_IOBUF_SIZE 4096
246 
247 /*
248  * This function implements the API defined in <linux/decompress/generic.h>.
249  *
250  * This wrapper will automatically choose single-call or multi-call mode
251  * of the native XZ decoder API. The single-call mode can be used only when
252  * both input and output buffers are available as a single chunk, i.e. when
253  * fill() and flush() won't be used.
254  */
unxz(unsigned char * in,long in_size,long (* fill)(void * dest,unsigned long size),long (* flush)(void * src,unsigned long size),unsigned char * out,long * in_used,void (* error)(char * x))255 STATIC int INIT unxz(unsigned char *in, long in_size,
256 		     long (*fill)(void *dest, unsigned long size),
257 		     long (*flush)(void *src, unsigned long size),
258 		     unsigned char *out, long *in_used,
259 		     void (*error)(char *x))
260 {
261 	struct xz_buf b;
262 	struct xz_dec *s;
263 	enum xz_ret ret;
264 	bool must_free_in = false;
265 
266 #if XZ_INTERNAL_CRC32
267 	xz_crc32_init();
268 #endif
269 
270 	if (in_used != NULL)
271 		*in_used = 0;
272 
273 	if (fill == NULL && flush == NULL)
274 		s = xz_dec_init(XZ_SINGLE, 0);
275 	else
276 		s = xz_dec_init(XZ_DYNALLOC, (uint32_t)-1);
277 
278 	if (s == NULL)
279 		goto error_alloc_state;
280 
281 	if (flush == NULL) {
282 		b.out = out;
283 		b.out_size = (size_t)-1;
284 	} else {
285 		b.out_size = XZ_IOBUF_SIZE;
286 		b.out = malloc(XZ_IOBUF_SIZE);
287 		if (b.out == NULL)
288 			goto error_alloc_out;
289 	}
290 
291 	if (in == NULL) {
292 		must_free_in = true;
293 		in = malloc(XZ_IOBUF_SIZE);
294 		if (in == NULL)
295 			goto error_alloc_in;
296 	}
297 
298 	b.in = in;
299 	b.in_pos = 0;
300 	b.in_size = in_size;
301 	b.out_pos = 0;
302 
303 	if (fill == NULL && flush == NULL) {
304 		ret = xz_dec_run(s, &b);
305 	} else {
306 		do {
307 			if (b.in_pos == b.in_size && fill != NULL) {
308 				if (in_used != NULL)
309 					*in_used += b.in_pos;
310 
311 				b.in_pos = 0;
312 
313 				in_size = fill(in, XZ_IOBUF_SIZE);
314 				if (in_size < 0) {
315 					/*
316 					 * This isn't an optimal error code
317 					 * but it probably isn't worth making
318 					 * a new one either.
319 					 */
320 					ret = XZ_BUF_ERROR;
321 					break;
322 				}
323 
324 				b.in_size = in_size;
325 			}
326 
327 			ret = xz_dec_run(s, &b);
328 
329 			if (flush != NULL && (b.out_pos == b.out_size
330 					|| (ret != XZ_OK && b.out_pos > 0))) {
331 				/*
332 				 * Setting ret here may hide an error
333 				 * returned by xz_dec_run(), but probably
334 				 * it's not too bad.
335 				 */
336 				if (flush(b.out, b.out_pos) != (long)b.out_pos)
337 					ret = XZ_BUF_ERROR;
338 
339 				b.out_pos = 0;
340 			}
341 		} while (ret == XZ_OK);
342 
343 		if (must_free_in)
344 			free(in);
345 
346 		if (flush != NULL)
347 			free(b.out);
348 	}
349 
350 	if (in_used != NULL)
351 		*in_used += b.in_pos;
352 
353 	xz_dec_end(s);
354 
355 	switch (ret) {
356 	case XZ_STREAM_END:
357 		return 0;
358 
359 	case XZ_MEM_ERROR:
360 		/* This can occur only in multi-call mode. */
361 		error("XZ decompressor ran out of memory");
362 		break;
363 
364 	case XZ_FORMAT_ERROR:
365 		error("Input is not in the XZ format (wrong magic bytes)");
366 		break;
367 
368 	case XZ_OPTIONS_ERROR:
369 		error("Input was encoded with settings that are not "
370 				"supported by this XZ decoder");
371 		break;
372 
373 	case XZ_DATA_ERROR:
374 	case XZ_BUF_ERROR:
375 		error("XZ-compressed data is corrupt");
376 		break;
377 
378 	default:
379 		error("Bug in the XZ decompressor");
380 		break;
381 	}
382 
383 	return -1;
384 
385 error_alloc_in:
386 	if (flush != NULL)
387 		free(b.out);
388 
389 error_alloc_out:
390 	xz_dec_end(s);
391 
392 error_alloc_state:
393 	error("XZ decompressor ran out of memory");
394 	return -1;
395 }
396 
397 /*
398  * This macro is used by architecture-specific files to decompress
399  * the kernel image.
400  */
401 #ifdef XZ_PREBOOT
__decompress(unsigned char * buf,long len,long (* fill)(void *,unsigned long),long (* flush)(void *,unsigned long),unsigned char * out_buf,long olen,long * pos,void (* error)(char * x))402 STATIC int INIT __decompress(unsigned char *buf, long len,
403 			   long (*fill)(void*, unsigned long),
404 			   long (*flush)(void*, unsigned long),
405 			   unsigned char *out_buf, long olen,
406 			   long *pos,
407 			   void (*error)(char *x))
408 {
409 	return unxz(buf, len, fill, flush, out_buf, pos, error);
410 }
411 #endif
412