xref: /openbmc/linux/lib/decompress_unzstd.c (revision 8dce88fe)
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 compressed 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 dependency 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/decompress_sources.h"
72 #endif
73 
74 #include <linux/decompress/mm.h>
75 #include <linux/kernel.h>
76 #include <linux/zstd.h>
77 
78 /* 128MB is the maximum window size supported by zstd. */
79 #define ZSTD_WINDOWSIZE_MAX	(1 << ZSTD_WINDOWLOG_MAX)
80 /*
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.
85  */
86 #define ZSTD_IOBUF_SIZE		(1 << 17)
87 
88 static int INIT handle_zstd_error(size_t ret, void (*error)(char *x))
89 {
90 	const zstd_error_code err = zstd_get_error_code(ret);
91 
92 	if (!zstd_is_error(ret))
93 		return 0;
94 
95 	/*
96 	 * zstd_get_error_name() cannot be used because error takes a char *
97 	 * not a const char *
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_dctx_workspace_bound();
128 	void *wksp = large_malloc(wksp_size);
129 	zstd_dctx *dctx = zstd_init_dctx(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_decompress_dctx() can't handle.
141 	 */
142 	ret = zstd_find_frame_compressed_size(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_decompress_dctx(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_in_buffer in;
171 	zstd_out_buffer out;
172 	zstd_frame_header header;
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 	/*
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.
185 	 */
186 	if (out_len == 0)
187 		out_len = UINTPTR_MAX - (uintptr_t)out_buf;
188 
189 	if (fill == NULL && flush == NULL)
190 		/*
191 		 * We can decompress faster and with less memory when we have a
192 		 * single chunk.
193 		 */
194 		return decompress_single(in_buf, in_len, out_buf, out_len,
195 					 in_pos, error);
196 
197 	/*
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.
201 	 */
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");
206 			err = -1;
207 			goto out;
208 		}
209 		in_buf = in_allocated;
210 		in_len = 0;
211 	}
212 	/* Read the first chunk, since we need to decode the frame header. */
213 	if (fill != NULL)
214 		in_len = fill(in_buf, ZSTD_IOBUF_SIZE);
215 	if (in_len < 0) {
216 		error("ZSTD-compressed data is truncated");
217 		err = -1;
218 		goto out;
219 	}
220 	/* Set the first non-empty input buffer. */
221 	in.src = in_buf;
222 	in.pos = 0;
223 	in.size = in_len;
224 	/* Allocate the output buffer if we are using flush(). */
225 	if (flush != NULL) {
226 		out_allocated = large_malloc(ZSTD_IOBUF_SIZE);
227 		if (out_allocated == NULL) {
228 			error("Out of memory while allocating output buffer");
229 			err = -1;
230 			goto out;
231 		}
232 		out_buf = out_allocated;
233 		out_len = ZSTD_IOBUF_SIZE;
234 	}
235 	/* Set the output buffer. */
236 	out.dst = out_buf;
237 	out.pos = 0;
238 	out.size = out_len;
239 
240 	/*
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.
246 	 */
247 	ret = zstd_get_frame_header(&header, in.src, in.size);
248 	err = handle_zstd_error(ret, error);
249 	if (err)
250 		goto out;
251 	if (ret != 0) {
252 		error("ZSTD-compressed data has an incomplete frame header");
253 		err = -1;
254 		goto out;
255 	}
256 	if (header.windowSize > ZSTD_WINDOWSIZE_MAX) {
257 		error("ZSTD-compressed data has too large a window size");
258 		err = -1;
259 		goto out;
260 	}
261 
262 	/*
263 	 * Allocate the zstd_dstream now that we know how much memory is
264 	 * required.
265 	 */
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");
271 		err = -1;
272 		goto out;
273 	}
274 
275 	/*
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().
280 	 */
281 	if (in_pos != NULL)
282 		*in_pos = 0;
283 	do {
284 		/*
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.
287 		 */
288 		if (in.pos == in.size) {
289 			if (in_pos != NULL)
290 				*in_pos += in.pos;
291 			in_len = fill ? fill(in_buf, ZSTD_IOBUF_SIZE) : -1;
292 			if (in_len < 0) {
293 				error("ZSTD-compressed data is truncated");
294 				err = -1;
295 				goto out;
296 			}
297 			in.pos = 0;
298 			in.size = in_len;
299 		}
300 		/* Returns zero when the frame is complete. */
301 		ret = zstd_decompress_stream(dstream, &out, &in);
302 		err = handle_zstd_error(ret, error);
303 		if (err)
304 			goto out;
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()");
309 				err = -1;
310 				goto out;
311 			}
312 			out.pos = 0;
313 		}
314 	} while (ret != 0);
315 
316 	if (in_pos != NULL)
317 		*in_pos += in.pos;
318 
319 	err = 0;
320 out:
321 	if (in_allocated != NULL)
322 		large_free(in_allocated);
323 	if (out_allocated != NULL)
324 		large_free(out_allocated);
325 	if (wksp != NULL)
326 		large_free(wksp);
327 	return err;
328 }
329 
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,
335 		       long *pos,
336 		       void (*error)(char *x))
337 {
338 	return __unzstd(buf, len, fill, flush, out_buf, 0, pos, error);
339 }
340 #else
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,
345 			     long *pos,
346 			     void (*error)(char *x))
347 {
348 	return __unzstd(buf, len, fill, flush, out_buf, out_len, pos, error);
349 }
350 #endif
351