1 // SPDX-License-Identifier: GPL-2.0 1 // SPDX-License-Identifier: GPL-2.0
2 2
3 /* 3 /*
4 * Important notes about in-place decompressio 4 * Important notes about in-place decompression
5 * 5 *
6 * At least on x86, the kernel is decompressed 6 * At least on x86, the kernel is decompressed in place: the compressed data
7 * is placed to the end of the output buffer, 7 * is placed to the end of the output buffer, and the decompressor overwrites
8 * most of the compressed data. There must be 8 * most of the compressed data. There must be enough safety margin to
9 * guarantee that the write position is always 9 * guarantee that the write position is always behind the read position.
10 * 10 *
11 * The safety margin for ZSTD with a 128 KB bl 11 * The safety margin for ZSTD with a 128 KB block size is calculated below.
12 * Note that the margin with ZSTD is bigger th 12 * Note that the margin with ZSTD is bigger than with GZIP or XZ!
13 * 13 *
14 * The worst case for in-place decompression i 14 * The worst case for in-place decompression is that the beginning of
15 * the file is compressed extremely well, and 15 * the file is compressed extremely well, and the rest of the file is
16 * uncompressible. Thus, we must look for wors 16 * uncompressible. Thus, we must look for worst-case expansion when the
17 * compressor is encoding uncompressible data. 17 * compressor is encoding uncompressible data.
18 * 18 *
19 * The structure of the .zst file in case of a !! 19 * The structure of the .zst file in case of a compresed kernel is as follows.
20 * Maximum sizes (as bytes) of the fields are 20 * Maximum sizes (as bytes) of the fields are in parenthesis.
21 * 21 *
22 * Frame Header: (18) 22 * Frame Header: (18)
23 * Blocks: (N) 23 * Blocks: (N)
24 * Checksum: (4) 24 * Checksum: (4)
25 * 25 *
26 * The frame header and checksum overhead is a 26 * The frame header and checksum overhead is at most 22 bytes.
27 * 27 *
28 * ZSTD stores the data in blocks. Each block 28 * ZSTD stores the data in blocks. Each block has a header whose size is
29 * a 3 bytes. After the block header, there is 29 * a 3 bytes. After the block header, there is up to 128 KB of payload.
30 * The maximum uncompressed size of the payloa 30 * The maximum uncompressed size of the payload is 128 KB. The minimum
31 * uncompressed size of the payload is never l 31 * uncompressed size of the payload is never less than the payload size
32 * (excluding the block header). 32 * (excluding the block header).
33 * 33 *
34 * The assumption, that the uncompressed size 34 * The assumption, that the uncompressed size of the payload is never
35 * smaller than the payload itself, is valid o 35 * smaller than the payload itself, is valid only when talking about
36 * the payload as a whole. It is possible that 36 * the payload as a whole. It is possible that the payload has parts where
37 * the decompressor consumes more input than i 37 * the decompressor consumes more input than it produces output. Calculating
38 * the worst case for this would be tricky. In 38 * the worst case for this would be tricky. Instead of trying to do that,
39 * let's simply make sure that the decompresso 39 * let's simply make sure that the decompressor never overwrites any bytes
40 * of the payload which it is currently readin 40 * of the payload which it is currently reading.
41 * 41 *
42 * Now we have enough information to calculate 42 * Now we have enough information to calculate the safety margin. We need
43 * - 22 bytes for the .zst file format heade 43 * - 22 bytes for the .zst file format headers;
44 * - 3 bytes per every 128 KiB of uncompress 44 * - 3 bytes per every 128 KiB of uncompressed size (one block header per
45 * block); and 45 * block); and
46 * - 128 KiB (biggest possible zstd block si 46 * - 128 KiB (biggest possible zstd block size) to make sure that the
47 * decompressor never overwrites anything 47 * decompressor never overwrites anything from the block it is currently
48 * reading. 48 * reading.
49 * 49 *
50 * We get the following formula: 50 * We get the following formula:
51 * 51 *
52 * safety_margin = 22 + uncompressed_size * 52 * safety_margin = 22 + uncompressed_size * 3 / 131072 + 131072
53 * <= 22 + (uncompressed_size 53 * <= 22 + (uncompressed_size >> 15) + 131072
54 */ 54 */
55 55
56 /* 56 /*
57 * Preboot environments #include "path/to/deco 57 * Preboot environments #include "path/to/decompress_unzstd.c".
58 * All of the source files we depend on must b 58 * All of the source files we depend on must be #included.
59 * zstd's only source dependency is xxhash, wh !! 59 * zstd's only source dependeny is xxhash, which has no source
60 * dependencies. 60 * dependencies.
61 * 61 *
62 * When UNZSTD_PREBOOT is defined we declare _ 62 * When UNZSTD_PREBOOT is defined we declare __decompress(), which is
63 * used for kernel decompression, instead of u 63 * used for kernel decompression, instead of unzstd().
64 * 64 *
65 * Define __DISABLE_EXPORTS in preboot environ 65 * Define __DISABLE_EXPORTS in preboot environments to prevent symbols
66 * from xxhash and zstd from being exported by 66 * from xxhash and zstd from being exported by the EXPORT_SYMBOL macro.
67 */ 67 */
68 #ifdef STATIC 68 #ifdef STATIC
69 # define UNZSTD_PREBOOT 69 # define UNZSTD_PREBOOT
70 # include "xxhash.c" 70 # include "xxhash.c"
71 # include "zstd/decompress_sources.h" !! 71 # include "zstd/entropy_common.c"
72 #else !! 72 # include "zstd/fse_decompress.c"
73 #include <linux/decompress/unzstd.h> !! 73 # include "zstd/huf_decompress.c"
>> 74 # include "zstd/zstd_common.c"
>> 75 # include "zstd/decompress.c"
74 #endif 76 #endif
75 77
76 #include <linux/decompress/mm.h> 78 #include <linux/decompress/mm.h>
77 #include <linux/kernel.h> 79 #include <linux/kernel.h>
78 #include <linux/zstd.h> 80 #include <linux/zstd.h>
79 81
80 /* 128MB is the maximum window size supported 82 /* 128MB is the maximum window size supported by zstd. */
81 #define ZSTD_WINDOWSIZE_MAX (1 << ZSTD_WIN 83 #define ZSTD_WINDOWSIZE_MAX (1 << ZSTD_WINDOWLOG_MAX)
82 /* 84 /*
83 * Size of the input and output buffers in mul 85 * Size of the input and output buffers in multi-call mode.
84 * Pick a larger size because it isn't used du 86 * Pick a larger size because it isn't used during kernel decompression,
85 * since that is single pass, and we have to a 87 * since that is single pass, and we have to allocate a large buffer for
86 * zstd's window anyway. The larger size speed 88 * zstd's window anyway. The larger size speeds up initramfs decompression.
87 */ 89 */
88 #define ZSTD_IOBUF_SIZE (1 << 17) 90 #define ZSTD_IOBUF_SIZE (1 << 17)
89 91
90 static int INIT handle_zstd_error(size_t ret, 92 static int INIT handle_zstd_error(size_t ret, void (*error)(char *x))
91 { 93 {
92 const zstd_error_code err = zstd_get_e !! 94 const int err = ZSTD_getErrorCode(ret);
93 95
94 if (!zstd_is_error(ret)) !! 96 if (!ZSTD_isError(ret))
95 return 0; 97 return 0;
96 98
97 /* <<
98 * zstd_get_error_name() cannot be use <<
99 * not a const char * <<
100 */ <<
101 switch (err) { 99 switch (err) {
102 case ZSTD_error_memory_allocation: 100 case ZSTD_error_memory_allocation:
103 error("ZSTD decompressor ran o 101 error("ZSTD decompressor ran out of memory");
104 break; 102 break;
105 case ZSTD_error_prefix_unknown: 103 case ZSTD_error_prefix_unknown:
106 error("Input is not in the ZST 104 error("Input is not in the ZSTD format (wrong magic bytes)");
107 break; 105 break;
108 case ZSTD_error_dstSize_tooSmall: 106 case ZSTD_error_dstSize_tooSmall:
109 case ZSTD_error_corruption_detected: 107 case ZSTD_error_corruption_detected:
110 case ZSTD_error_checksum_wrong: 108 case ZSTD_error_checksum_wrong:
111 error("ZSTD-compressed data is 109 error("ZSTD-compressed data is corrupt");
112 break; 110 break;
113 default: 111 default:
114 error("ZSTD-compressed data is 112 error("ZSTD-compressed data is probably corrupt");
115 break; 113 break;
116 } 114 }
117 return -1; 115 return -1;
118 } 116 }
119 117
120 /* 118 /*
121 * Handle the case where we have the entire in 119 * Handle the case where we have the entire input and output in one segment.
122 * We can allocate less memory (no circular bu 120 * We can allocate less memory (no circular buffer for the sliding window),
123 * and avoid some memcpy() calls. 121 * and avoid some memcpy() calls.
124 */ 122 */
125 static int INIT decompress_single(const u8 *in 123 static int INIT decompress_single(const u8 *in_buf, long in_len, u8 *out_buf,
126 long out_len 124 long out_len, long *in_pos,
127 void (*error 125 void (*error)(char *x))
128 { 126 {
129 const size_t wksp_size = zstd_dctx_wor !! 127 const size_t wksp_size = ZSTD_DCtxWorkspaceBound();
130 void *wksp = large_malloc(wksp_size); 128 void *wksp = large_malloc(wksp_size);
131 zstd_dctx *dctx = zstd_init_dctx(wksp, !! 129 ZSTD_DCtx *dctx = ZSTD_initDCtx(wksp, wksp_size);
132 int err; 130 int err;
133 size_t ret; 131 size_t ret;
134 132
135 if (dctx == NULL) { 133 if (dctx == NULL) {
136 error("Out of memory while all !! 134 error("Out of memory while allocating ZSTD_DCtx");
137 err = -1; 135 err = -1;
138 goto out; 136 goto out;
139 } 137 }
140 /* 138 /*
141 * Find out how large the frame actual 139 * Find out how large the frame actually is, there may be junk at
142 * the end of the frame that zstd_deco !! 140 * the end of the frame that ZSTD_decompressDCtx() can't handle.
143 */ 141 */
144 ret = zstd_find_frame_compressed_size( !! 142 ret = ZSTD_findFrameCompressedSize(in_buf, in_len);
145 err = handle_zstd_error(ret, error); 143 err = handle_zstd_error(ret, error);
146 if (err) 144 if (err)
147 goto out; 145 goto out;
148 in_len = (long)ret; 146 in_len = (long)ret;
149 147
150 ret = zstd_decompress_dctx(dctx, out_b !! 148 ret = ZSTD_decompressDCtx(dctx, out_buf, out_len, in_buf, in_len);
151 err = handle_zstd_error(ret, error); 149 err = handle_zstd_error(ret, error);
152 if (err) 150 if (err)
153 goto out; 151 goto out;
154 152
155 if (in_pos != NULL) 153 if (in_pos != NULL)
156 *in_pos = in_len; 154 *in_pos = in_len;
157 155
158 err = 0; 156 err = 0;
159 out: 157 out:
160 if (wksp != NULL) 158 if (wksp != NULL)
161 large_free(wksp); 159 large_free(wksp);
162 return err; 160 return err;
163 } 161 }
164 162
165 static int INIT __unzstd(unsigned char *in_buf 163 static int INIT __unzstd(unsigned char *in_buf, long in_len,
166 long (*fill)(void*, u 164 long (*fill)(void*, unsigned long),
167 long (*flush)(void*, 165 long (*flush)(void*, unsigned long),
168 unsigned char *out_bu 166 unsigned char *out_buf, long out_len,
169 long *in_pos, 167 long *in_pos,
170 void (*error)(char *x 168 void (*error)(char *x))
171 { 169 {
172 zstd_in_buffer in; !! 170 ZSTD_inBuffer in;
173 zstd_out_buffer out; !! 171 ZSTD_outBuffer out;
174 zstd_frame_header header; !! 172 ZSTD_frameParams params;
175 void *in_allocated = NULL; 173 void *in_allocated = NULL;
176 void *out_allocated = NULL; 174 void *out_allocated = NULL;
177 void *wksp = NULL; 175 void *wksp = NULL;
178 size_t wksp_size; 176 size_t wksp_size;
179 zstd_dstream *dstream; !! 177 ZSTD_DStream *dstream;
180 int err; 178 int err;
181 size_t ret; 179 size_t ret;
182 180
183 /* 181 /*
184 * ZSTD decompression code won't be ha 182 * ZSTD decompression code won't be happy if the buffer size is so big
185 * that its end address overflows. Whe 183 * that its end address overflows. When the size is not provided, make
186 * it as big as possible without havin 184 * it as big as possible without having the end address overflow.
187 */ 185 */
188 if (out_len == 0) 186 if (out_len == 0)
189 out_len = UINTPTR_MAX - (uintp 187 out_len = UINTPTR_MAX - (uintptr_t)out_buf;
190 188
191 if (fill == NULL && flush == NULL) 189 if (fill == NULL && flush == NULL)
192 /* 190 /*
193 * We can decompress faster an 191 * We can decompress faster and with less memory when we have a
194 * single chunk. 192 * single chunk.
195 */ 193 */
196 return decompress_single(in_bu 194 return decompress_single(in_buf, in_len, out_buf, out_len,
197 in_po 195 in_pos, error);
198 196
199 /* 197 /*
200 * If in_buf is not provided, we must 198 * If in_buf is not provided, we must be using fill(), so allocate
201 * a large enough buffer. If it is pro 199 * a large enough buffer. If it is provided, it must be at least
202 * ZSTD_IOBUF_SIZE large. 200 * ZSTD_IOBUF_SIZE large.
203 */ 201 */
204 if (in_buf == NULL) { 202 if (in_buf == NULL) {
205 in_allocated = large_malloc(ZS 203 in_allocated = large_malloc(ZSTD_IOBUF_SIZE);
206 if (in_allocated == NULL) { 204 if (in_allocated == NULL) {
207 error("Out of memory w 205 error("Out of memory while allocating input buffer");
208 err = -1; 206 err = -1;
209 goto out; 207 goto out;
210 } 208 }
211 in_buf = in_allocated; 209 in_buf = in_allocated;
212 in_len = 0; 210 in_len = 0;
213 } 211 }
214 /* Read the first chunk, since we need 212 /* Read the first chunk, since we need to decode the frame header. */
215 if (fill != NULL) 213 if (fill != NULL)
216 in_len = fill(in_buf, ZSTD_IOB 214 in_len = fill(in_buf, ZSTD_IOBUF_SIZE);
217 if (in_len < 0) { 215 if (in_len < 0) {
218 error("ZSTD-compressed data is 216 error("ZSTD-compressed data is truncated");
219 err = -1; 217 err = -1;
220 goto out; 218 goto out;
221 } 219 }
222 /* Set the first non-empty input buffe 220 /* Set the first non-empty input buffer. */
223 in.src = in_buf; 221 in.src = in_buf;
224 in.pos = 0; 222 in.pos = 0;
225 in.size = in_len; 223 in.size = in_len;
226 /* Allocate the output buffer if we ar 224 /* Allocate the output buffer if we are using flush(). */
227 if (flush != NULL) { 225 if (flush != NULL) {
228 out_allocated = large_malloc(Z 226 out_allocated = large_malloc(ZSTD_IOBUF_SIZE);
229 if (out_allocated == NULL) { 227 if (out_allocated == NULL) {
230 error("Out of memory w 228 error("Out of memory while allocating output buffer");
231 err = -1; 229 err = -1;
232 goto out; 230 goto out;
233 } 231 }
234 out_buf = out_allocated; 232 out_buf = out_allocated;
235 out_len = ZSTD_IOBUF_SIZE; 233 out_len = ZSTD_IOBUF_SIZE;
236 } 234 }
237 /* Set the output buffer. */ 235 /* Set the output buffer. */
238 out.dst = out_buf; 236 out.dst = out_buf;
239 out.pos = 0; 237 out.pos = 0;
240 out.size = out_len; 238 out.size = out_len;
241 239
242 /* 240 /*
243 * We need to know the window size to !! 241 * We need to know the window size to allocate the ZSTD_DStream.
244 * Since we are streaming, we need to 242 * Since we are streaming, we need to allocate a buffer for the sliding
245 * window. The window size varies from 243 * window. The window size varies from 1 KB to ZSTD_WINDOWSIZE_MAX
246 * (8 MB), so it is important to use t 244 * (8 MB), so it is important to use the actual value so as not to
247 * waste memory when it is smaller. 245 * waste memory when it is smaller.
248 */ 246 */
249 ret = zstd_get_frame_header(&header, i !! 247 ret = ZSTD_getFrameParams(¶ms, in.src, in.size);
250 err = handle_zstd_error(ret, error); 248 err = handle_zstd_error(ret, error);
251 if (err) 249 if (err)
252 goto out; 250 goto out;
253 if (ret != 0) { 251 if (ret != 0) {
254 error("ZSTD-compressed data ha 252 error("ZSTD-compressed data has an incomplete frame header");
255 err = -1; 253 err = -1;
256 goto out; 254 goto out;
257 } 255 }
258 if (header.windowSize > ZSTD_WINDOWSIZ !! 256 if (params.windowSize > ZSTD_WINDOWSIZE_MAX) {
259 error("ZSTD-compressed data ha 257 error("ZSTD-compressed data has too large a window size");
260 err = -1; 258 err = -1;
261 goto out; 259 goto out;
262 } 260 }
263 261
264 /* 262 /*
265 * Allocate the zstd_dstream now that !! 263 * Allocate the ZSTD_DStream now that we know how much memory is
266 * required. 264 * required.
267 */ 265 */
268 wksp_size = zstd_dstream_workspace_bou !! 266 wksp_size = ZSTD_DStreamWorkspaceBound(params.windowSize);
269 wksp = large_malloc(wksp_size); 267 wksp = large_malloc(wksp_size);
270 dstream = zstd_init_dstream(header.win !! 268 dstream = ZSTD_initDStream(params.windowSize, wksp, wksp_size);
271 if (dstream == NULL) { 269 if (dstream == NULL) {
272 error("Out of memory while all 270 error("Out of memory while allocating ZSTD_DStream");
273 err = -1; 271 err = -1;
274 goto out; 272 goto out;
275 } 273 }
276 274
277 /* 275 /*
278 * Decompression loop: 276 * Decompression loop:
279 * Read more data if necessary (error 277 * Read more data if necessary (error if no more data can be read).
280 * Call the decompression function, wh 278 * Call the decompression function, which returns 0 when finished.
281 * Flush any data produced if using fl 279 * Flush any data produced if using flush().
282 */ 280 */
283 if (in_pos != NULL) 281 if (in_pos != NULL)
284 *in_pos = 0; 282 *in_pos = 0;
285 do { 283 do {
286 /* 284 /*
287 * If we need to reload data, 285 * If we need to reload data, either we have fill() and can
288 * try to get more data, or we 286 * try to get more data, or we don't and the input is truncated.
289 */ 287 */
290 if (in.pos == in.size) { 288 if (in.pos == in.size) {
291 if (in_pos != NULL) 289 if (in_pos != NULL)
292 *in_pos += in. 290 *in_pos += in.pos;
293 in_len = fill ? fill(i 291 in_len = fill ? fill(in_buf, ZSTD_IOBUF_SIZE) : -1;
294 if (in_len < 0) { 292 if (in_len < 0) {
295 error("ZSTD-co 293 error("ZSTD-compressed data is truncated");
296 err = -1; 294 err = -1;
297 goto out; 295 goto out;
298 } 296 }
299 in.pos = 0; 297 in.pos = 0;
300 in.size = in_len; 298 in.size = in_len;
301 } 299 }
302 /* Returns zero when the frame 300 /* Returns zero when the frame is complete. */
303 ret = zstd_decompress_stream(d !! 301 ret = ZSTD_decompressStream(dstream, &out, &in);
304 err = handle_zstd_error(ret, e 302 err = handle_zstd_error(ret, error);
305 if (err) 303 if (err)
306 goto out; 304 goto out;
307 /* Flush all of the data produ 305 /* Flush all of the data produced if using flush(). */
308 if (flush != NULL && out.pos > 306 if (flush != NULL && out.pos > 0) {
309 if (out.pos != flush(o 307 if (out.pos != flush(out.dst, out.pos)) {
310 error("Failed 308 error("Failed to flush()");
311 err = -1; 309 err = -1;
312 goto out; 310 goto out;
313 } 311 }
314 out.pos = 0; 312 out.pos = 0;
315 } 313 }
316 } while (ret != 0); 314 } while (ret != 0);
317 315
318 if (in_pos != NULL) 316 if (in_pos != NULL)
319 *in_pos += in.pos; 317 *in_pos += in.pos;
320 318
321 err = 0; 319 err = 0;
322 out: 320 out:
323 if (in_allocated != NULL) 321 if (in_allocated != NULL)
324 large_free(in_allocated); 322 large_free(in_allocated);
325 if (out_allocated != NULL) 323 if (out_allocated != NULL)
326 large_free(out_allocated); 324 large_free(out_allocated);
327 if (wksp != NULL) 325 if (wksp != NULL)
328 large_free(wksp); 326 large_free(wksp);
329 return err; 327 return err;
330 } 328 }
331 329
332 #ifndef UNZSTD_PREBOOT 330 #ifndef UNZSTD_PREBOOT
333 STATIC int INIT unzstd(unsigned char *buf, lon 331 STATIC int INIT unzstd(unsigned char *buf, long len,
334 long (*fill)(void*, uns 332 long (*fill)(void*, unsigned long),
335 long (*flush)(void*, un 333 long (*flush)(void*, unsigned long),
336 unsigned char *out_buf, 334 unsigned char *out_buf,
337 long *pos, 335 long *pos,
338 void (*error)(char *x)) 336 void (*error)(char *x))
339 { 337 {
340 return __unzstd(buf, len, fill, flush, 338 return __unzstd(buf, len, fill, flush, out_buf, 0, pos, error);
341 } 339 }
342 #else 340 #else
343 STATIC int INIT __decompress(unsigned char *bu 341 STATIC int INIT __decompress(unsigned char *buf, long len,
344 long (*fill)(void 342 long (*fill)(void*, unsigned long),
345 long (*flush)(voi 343 long (*flush)(void*, unsigned long),
346 unsigned char *ou 344 unsigned char *out_buf, long out_len,
347 long *pos, 345 long *pos,
348 void (*error)(cha 346 void (*error)(char *x))
349 { 347 {
350 return __unzstd(buf, len, fill, flush, 348 return __unzstd(buf, len, fill, flush, out_buf, out_len, pos, error);
351 } 349 }
352 #endif 350 #endif
353 351
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.