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