1 // SPDX-License-Identifier: GPL-2.0
4 * Important notes about in-place decompression
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.
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!
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.
19 * The structure of the .zst file in case of a compressed kernel is as follows.
20 * Maximum sizes (as bytes) of the fields are in parenthesis.
26 * The frame header and checksum overhead is at most 22 bytes.
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).
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.
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
46 * - 128 KiB (biggest possible zstd block size) to make sure that the
47 * decompressor never overwrites anything from the block it is currently
50 * We get the following formula:
52 * safety_margin = 22 + uncompressed_size * 3 / 131072 + 131072
53 * <= 22 + (uncompressed_size >> 15) + 131072
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 dependency is xxhash, which has no source
62 * When UNZSTD_PREBOOT is defined we declare __decompress(), which is
63 * used for kernel decompression, instead of unzstd().
65 * Define __DISABLE_EXPORTS in preboot environments to prevent symbols
66 * from xxhash and zstd from being exported by the EXPORT_SYMBOL macro.
69 # define UNZSTD_PREBOOT
71 # include "zstd/decompress_sources.h"
74 #include <linux/decompress/mm.h>
75 #include <linux/kernel.h>
76 #include <linux/zstd.h>
78 /* 128MB is the maximum window size supported by zstd. */
79 #define ZSTD_WINDOWSIZE_MAX (1 << ZSTD_WINDOWLOG_MAX)
81 * Size of the input and output buffers in multi-call mode.
82 * Pick a larger size because it isn't used during kernel decompression,
83 * since that is single pass, and we have to allocate a large buffer for
84 * zstd's window anyway. The larger size speeds up initramfs decompression.
86 #define ZSTD_IOBUF_SIZE (1 << 17)
88 static int INIT handle_zstd_error(size_t ret, void (*error)(char *x))
90 const zstd_error_code err = zstd_get_error_code(ret);
92 if (!zstd_is_error(ret))
96 * zstd_get_error_name() cannot be used because error takes a char *
100 case ZSTD_error_memory_allocation:
101 error("ZSTD decompressor ran out of memory");
103 case ZSTD_error_prefix_unknown:
104 error("Input is not in the ZSTD format (wrong magic bytes)");
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");
112 error("ZSTD-compressed data is probably corrupt");
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.
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))
127 const size_t wksp_size = zstd_dctx_workspace_bound();
128 void *wksp = large_malloc(wksp_size);
129 zstd_dctx *dctx = zstd_init_dctx(wksp, wksp_size);
134 error("Out of memory while allocating zstd_dctx");
139 * Find out how large the frame actually is, there may be junk at
140 * the end of the frame that zstd_decompress_dctx() can't handle.
142 ret = zstd_find_frame_compressed_size(in_buf, in_len);
143 err = handle_zstd_error(ret, error);
148 ret = zstd_decompress_dctx(dctx, out_buf, out_len, in_buf, in_len);
149 err = handle_zstd_error(ret, error);
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,
168 void (*error)(char *x))
172 zstd_frame_header header;
173 void *in_allocated = NULL;
174 void *out_allocated = NULL;
177 zstd_dstream *dstream;
182 * ZSTD decompression code won't be happy if the buffer size is so big
183 * that its end address overflows. When the size is not provided, make
184 * it as big as possible without having the end address overflow.
187 out_len = UINTPTR_MAX - (uintptr_t)out_buf;
189 if (fill == NULL && flush == NULL)
191 * We can decompress faster and with less memory when we have a
194 return decompress_single(in_buf, in_len, out_buf, out_len,
198 * If in_buf is not provided, we must be using fill(), so allocate
199 * a large enough buffer. If it is provided, it must be at least
200 * ZSTD_IOBUF_SIZE large.
202 if (in_buf == NULL) {
203 in_allocated = large_malloc(ZSTD_IOBUF_SIZE);
204 if (in_allocated == NULL) {
205 error("Out of memory while allocating input buffer");
209 in_buf = in_allocated;
212 /* Read the first chunk, since we need to decode the frame header. */
214 in_len = fill(in_buf, ZSTD_IOBUF_SIZE);
216 error("ZSTD-compressed data is truncated");
220 /* Set the first non-empty input buffer. */
224 /* Allocate the output buffer if we are using flush(). */
226 out_allocated = large_malloc(ZSTD_IOBUF_SIZE);
227 if (out_allocated == NULL) {
228 error("Out of memory while allocating output buffer");
232 out_buf = out_allocated;
233 out_len = ZSTD_IOBUF_SIZE;
235 /* Set the output buffer. */
241 * We need to know the window size to allocate the zstd_dstream.
242 * Since we are streaming, we need to allocate a buffer for the sliding
243 * window. The window size varies from 1 KB to ZSTD_WINDOWSIZE_MAX
244 * (8 MB), so it is important to use the actual value so as not to
245 * waste memory when it is smaller.
247 ret = zstd_get_frame_header(&header, in.src, in.size);
248 err = handle_zstd_error(ret, error);
252 error("ZSTD-compressed data has an incomplete frame header");
256 if (header.windowSize > ZSTD_WINDOWSIZE_MAX) {
257 error("ZSTD-compressed data has too large a window size");
263 * Allocate the zstd_dstream now that we know how much memory is
266 wksp_size = zstd_dstream_workspace_bound(header.windowSize);
267 wksp = large_malloc(wksp_size);
268 dstream = zstd_init_dstream(header.windowSize, wksp, wksp_size);
269 if (dstream == NULL) {
270 error("Out of memory while allocating ZSTD_DStream");
276 * Decompression loop:
277 * Read more data if necessary (error if no more data can be read).
278 * Call the decompression function, which returns 0 when finished.
279 * Flush any data produced if using flush().
285 * If we need to reload data, either we have fill() and can
286 * try to get more data, or we don't and the input is truncated.
288 if (in.pos == in.size) {
291 in_len = fill ? fill(in_buf, ZSTD_IOBUF_SIZE) : -1;
293 error("ZSTD-compressed data is truncated");
300 /* Returns zero when the frame is complete. */
301 ret = zstd_decompress_stream(dstream, &out, &in);
302 err = handle_zstd_error(ret, error);
305 /* Flush all of the data produced if using flush(). */
306 if (flush != NULL && out.pos > 0) {
307 if (out.pos != flush(out.dst, out.pos)) {
308 error("Failed to flush()");
321 if (in_allocated != NULL)
322 large_free(in_allocated);
323 if (out_allocated != NULL)
324 large_free(out_allocated);
330 #ifndef UNZSTD_PREBOOT
331 STATIC int INIT unzstd(unsigned char *buf, long len,
332 long (*fill)(void*, unsigned long),
333 long (*flush)(void*, unsigned long),
334 unsigned char *out_buf,
336 void (*error)(char *x))
338 return __unzstd(buf, len, fill, flush, out_buf, 0, pos, error);
341 STATIC int INIT __decompress(unsigned char *buf, long len,
342 long (*fill)(void*, unsigned long),
343 long (*flush)(void*, unsigned long),
344 unsigned char *out_buf, long out_len,
346 void (*error)(char *x))
348 return __unzstd(buf, len, fill, flush, out_buf, out_len, pos, error);