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 | * incompressible. Thus, we must look for worst-case expansion when the |
24 | * compressor is encoding incompressible data. |
25 | * |
26 | * The structure of the .xz file in case of a compressed 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 incompressible 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 | #else |
106 | #include <linux/decompress/unxz.h> |
107 | #endif |
108 | #ifdef __KERNEL__ |
109 | # include <linux/decompress/mm.h> |
110 | #endif |
111 | #define XZ_EXTERN STATIC |
112 | |
113 | #ifndef XZ_PREBOOT |
114 | # include <linux/slab.h> |
115 | # include <linux/xz.h> |
116 | #else |
117 | /* |
118 | * Use the internal CRC32 code instead of kernel's CRC32 module, which |
119 | * is not available in early phase of booting. |
120 | */ |
121 | #define XZ_INTERNAL_CRC32 1 |
122 | |
123 | /* |
124 | * For boot time use, we enable only the BCJ filter of the current |
125 | * architecture or none if no BCJ filter is available for the architecture. |
126 | */ |
127 | #ifdef CONFIG_X86 |
128 | # define XZ_DEC_X86 |
129 | #endif |
130 | #ifdef CONFIG_PPC |
131 | # define XZ_DEC_POWERPC |
132 | #endif |
133 | #ifdef CONFIG_ARM |
134 | # define XZ_DEC_ARM |
135 | #endif |
136 | #ifdef CONFIG_SPARC |
137 | # define XZ_DEC_SPARC |
138 | #endif |
139 | |
140 | /* |
141 | * This will get the basic headers so that memeq() and others |
142 | * can be defined. |
143 | */ |
144 | #include "xz/xz_private.h" |
145 | |
146 | /* |
147 | * Replace the normal allocation functions with the versions from |
148 | * <linux/decompress/mm.h>. vfree() needs to support vfree(NULL) |
149 | * when XZ_DYNALLOC is used, but the pre-boot free() doesn't support it. |
150 | * Workaround it here because the other decompressors don't need it. |
151 | */ |
152 | #undef kmalloc |
153 | #undef kfree |
154 | #undef vmalloc |
155 | #undef vfree |
156 | #define kmalloc(size, flags) malloc(size) |
157 | #define kfree(ptr) free(ptr) |
158 | #define vmalloc(size) malloc(size) |
159 | #define vfree(ptr) do { if (ptr != NULL) free(ptr); } while (0) |
160 | |
161 | /* |
162 | * FIXME: Not all basic memory functions are provided in architecture-specific |
163 | * files (yet). We define our own versions here for now, but this should be |
164 | * only a temporary solution. |
165 | * |
166 | * memeq and memzero are not used much and any remotely sane implementation |
167 | * is fast enough. memcpy/memmove speed matters in multi-call mode, but |
168 | * the kernel image is decompressed in single-call mode, in which only |
169 | * memmove speed can matter and only if there is a lot of incompressible data |
170 | * (LZMA2 stores incompressible chunks in uncompressed form). Thus, the |
171 | * functions below should just be kept small; it's probably not worth |
172 | * optimizing for speed. |
173 | */ |
174 | |
175 | #ifndef memeq |
176 | static bool memeq(const void *a, const void *b, size_t size) |
177 | { |
178 | const uint8_t *x = a; |
179 | const uint8_t *y = b; |
180 | size_t i; |
181 | |
182 | for (i = 0; i < size; ++i) |
183 | if (x[i] != y[i]) |
184 | return false; |
185 | |
186 | return true; |
187 | } |
188 | #endif |
189 | |
190 | #ifndef memzero |
191 | static void memzero(void *buf, size_t size) |
192 | { |
193 | uint8_t *b = buf; |
194 | uint8_t *e = b + size; |
195 | |
196 | while (b != e) |
197 | *b++ = '\0'; |
198 | } |
199 | #endif |
200 | |
201 | #ifndef memmove |
202 | /* Not static to avoid a conflict with the prototype in the Linux headers. */ |
203 | void *memmove(void *dest, const void *src, size_t size) |
204 | { |
205 | uint8_t *d = dest; |
206 | const uint8_t *s = src; |
207 | size_t i; |
208 | |
209 | if (d < s) { |
210 | for (i = 0; i < size; ++i) |
211 | d[i] = s[i]; |
212 | } else if (d > s) { |
213 | i = size; |
214 | while (i-- > 0) |
215 | d[i] = s[i]; |
216 | } |
217 | |
218 | return dest; |
219 | } |
220 | #endif |
221 | |
222 | /* |
223 | * Since we need memmove anyway, would use it as memcpy too. |
224 | * Commented out for now to avoid breaking things. |
225 | */ |
226 | /* |
227 | #ifndef memcpy |
228 | # define memcpy memmove |
229 | #endif |
230 | */ |
231 | |
232 | #include "xz/xz_crc32.c" |
233 | #include "xz/xz_dec_stream.c" |
234 | #include "xz/xz_dec_lzma2.c" |
235 | #include "xz/xz_dec_bcj.c" |
236 | |
237 | #endif /* XZ_PREBOOT */ |
238 | |
239 | /* Size of the input and output buffers in multi-call mode */ |
240 | #define XZ_IOBUF_SIZE 4096 |
241 | |
242 | /* |
243 | * This function implements the API defined in <linux/decompress/generic.h>. |
244 | * |
245 | * This wrapper will automatically choose single-call or multi-call mode |
246 | * of the native XZ decoder API. The single-call mode can be used only when |
247 | * both input and output buffers are available as a single chunk, i.e. when |
248 | * fill() and flush() won't be used. |
249 | */ |
250 | STATIC int INIT unxz(unsigned char *in, long in_size, |
251 | long (*fill)(void *dest, unsigned long size), |
252 | long (*flush)(void *src, unsigned long size), |
253 | unsigned char *out, long *in_used, |
254 | void (*error)(char *x)) |
255 | { |
256 | struct xz_buf b; |
257 | struct xz_dec *s; |
258 | enum xz_ret ret; |
259 | bool must_free_in = false; |
260 | |
261 | #if XZ_INTERNAL_CRC32 |
262 | xz_crc32_init(); |
263 | #endif |
264 | |
265 | if (in_used != NULL) |
266 | *in_used = 0; |
267 | |
268 | if (fill == NULL && flush == NULL) |
269 | s = xz_dec_init(mode: XZ_SINGLE, dict_max: 0); |
270 | else |
271 | s = xz_dec_init(mode: XZ_DYNALLOC, dict_max: (uint32_t)-1); |
272 | |
273 | if (s == NULL) |
274 | goto error_alloc_state; |
275 | |
276 | if (flush == NULL) { |
277 | b.out = out; |
278 | b.out_size = (size_t)-1; |
279 | } else { |
280 | b.out_size = XZ_IOBUF_SIZE; |
281 | b.out = malloc(XZ_IOBUF_SIZE); |
282 | if (b.out == NULL) |
283 | goto error_alloc_out; |
284 | } |
285 | |
286 | if (in == NULL) { |
287 | must_free_in = true; |
288 | in = malloc(XZ_IOBUF_SIZE); |
289 | if (in == NULL) |
290 | goto error_alloc_in; |
291 | } |
292 | |
293 | b.in = in; |
294 | b.in_pos = 0; |
295 | b.in_size = in_size; |
296 | b.out_pos = 0; |
297 | |
298 | if (fill == NULL && flush == NULL) { |
299 | ret = xz_dec_run(s, b: &b); |
300 | } else { |
301 | do { |
302 | if (b.in_pos == b.in_size && fill != NULL) { |
303 | if (in_used != NULL) |
304 | *in_used += b.in_pos; |
305 | |
306 | b.in_pos = 0; |
307 | |
308 | in_size = fill(in, XZ_IOBUF_SIZE); |
309 | if (in_size < 0) { |
310 | /* |
311 | * This isn't an optimal error code |
312 | * but it probably isn't worth making |
313 | * a new one either. |
314 | */ |
315 | ret = XZ_BUF_ERROR; |
316 | break; |
317 | } |
318 | |
319 | b.in_size = in_size; |
320 | } |
321 | |
322 | ret = xz_dec_run(s, b: &b); |
323 | |
324 | if (flush != NULL && (b.out_pos == b.out_size |
325 | || (ret != XZ_OK && b.out_pos > 0))) { |
326 | /* |
327 | * Setting ret here may hide an error |
328 | * returned by xz_dec_run(), but probably |
329 | * it's not too bad. |
330 | */ |
331 | if (flush(b.out, b.out_pos) != (long)b.out_pos) |
332 | ret = XZ_BUF_ERROR; |
333 | |
334 | b.out_pos = 0; |
335 | } |
336 | } while (ret == XZ_OK); |
337 | |
338 | if (must_free_in) |
339 | free(in); |
340 | |
341 | if (flush != NULL) |
342 | free(b.out); |
343 | } |
344 | |
345 | if (in_used != NULL) |
346 | *in_used += b.in_pos; |
347 | |
348 | xz_dec_end(s); |
349 | |
350 | switch (ret) { |
351 | case XZ_STREAM_END: |
352 | return 0; |
353 | |
354 | case XZ_MEM_ERROR: |
355 | /* This can occur only in multi-call mode. */ |
356 | error("XZ decompressor ran out of memory" ); |
357 | break; |
358 | |
359 | case XZ_FORMAT_ERROR: |
360 | error("Input is not in the XZ format (wrong magic bytes)" ); |
361 | break; |
362 | |
363 | case XZ_OPTIONS_ERROR: |
364 | error("Input was encoded with settings that are not " |
365 | "supported by this XZ decoder" ); |
366 | break; |
367 | |
368 | case XZ_DATA_ERROR: |
369 | case XZ_BUF_ERROR: |
370 | error("XZ-compressed data is corrupt" ); |
371 | break; |
372 | |
373 | default: |
374 | error("Bug in the XZ decompressor" ); |
375 | break; |
376 | } |
377 | |
378 | return -1; |
379 | |
380 | error_alloc_in: |
381 | if (flush != NULL) |
382 | free(b.out); |
383 | |
384 | error_alloc_out: |
385 | xz_dec_end(s); |
386 | |
387 | error_alloc_state: |
388 | error("XZ decompressor ran out of memory" ); |
389 | return -1; |
390 | } |
391 | |
392 | /* |
393 | * This macro is used by architecture-specific files to decompress |
394 | * the kernel image. |
395 | */ |
396 | #ifdef XZ_PREBOOT |
397 | STATIC int INIT __decompress(unsigned char *buf, long len, |
398 | long (*fill)(void*, unsigned long), |
399 | long (*flush)(void*, unsigned long), |
400 | unsigned char *out_buf, long olen, |
401 | long *pos, |
402 | void (*error)(char *x)) |
403 | { |
404 | return unxz(buf, len, fill, flush, out_buf, pos, error); |
405 | } |
406 | #endif |
407 | |