xref: /freebsd/sys/contrib/openzfs/include/sys/zstd/zstd.h (revision 53b70c86)
1 /*
2  * BSD 3-Clause New License (https://spdx.org/licenses/BSD-3-Clause.html)
3  *
4  * Redistribution and use in source and binary forms, with or without
5  * modification, are permitted provided that the following conditions are met:
6  *
7  * 1. Redistributions of source code must retain the above copyright notice,
8  * this list of conditions and the following disclaimer.
9  *
10  * 2. Redistributions in binary form must reproduce the above copyright notice,
11  * this list of conditions and the following disclaimer in the documentation
12  * and/or other materials provided with the distribution.
13  *
14  * 3. Neither the name of the copyright holder nor the names of its
15  * contributors may be used to endorse or promote products derived from this
16  * software without specific prior written permission.
17  *
18  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
19  * AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
20  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
21  * ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS BE
22  * LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
23  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
24  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
25  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
26  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
27  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
28  * POSSIBILITY OF SUCH DAMAGE.
29  */
30 
31 /*
32  * Copyright (c) 2016-2018, Klara Inc.
33  * Copyright (c) 2016-2018, Allan Jude
34  * Copyright (c) 2018-2020, Sebastian Gottschall
35  * Copyright (c) 2019-2020, Michael Niewöhner
36  * Copyright (c) 2020, The FreeBSD Foundation [1]
37  *
38  * [1] Portions of this software were developed by Allan Jude
39  *     under sponsorship from the FreeBSD Foundation.
40  */
41 
42 #ifndef	_ZFS_ZSTD_H
43 #define	_ZFS_ZSTD_H
44 
45 #ifdef	__cplusplus
46 extern "C" {
47 #endif
48 
49 /*
50  * ZSTD block header
51  * NOTE: all fields in this header are in big endian order.
52  */
53 typedef struct zfs_zstd_header {
54 	/* Compressed size of data */
55 	uint32_t c_len;
56 
57 	/*
58 	 * Version and compression level
59 	 * We used to use a union to reference compression level
60 	 * and version easily, but as it turns out, relying on the
61 	 * ordering of bitfields is not remotely portable.
62 	 * So now we have get/set functions in zfs_zstd.c for
63 	 * manipulating this in just the right way forever.
64 	 */
65 	uint32_t raw_version_level;
66 	char data[];
67 } zfs_zstdhdr_t;
68 
69 /*
70  * Simple struct to pass the data from raw_version_level around.
71  */
72 typedef struct zfs_zstd_meta {
73 	uint8_t level;
74 	uint32_t version;
75 } zfs_zstdmeta_t;
76 
77 /*
78  * kstat helper macros
79  */
80 #define	ZSTDSTAT(stat)		(zstd_stats.stat.value.ui64)
81 #define	ZSTDSTAT_ADD(stat, val) \
82 	atomic_add_64(&zstd_stats.stat.value.ui64, (val))
83 #define	ZSTDSTAT_SUB(stat, val) \
84 	atomic_sub_64(&zstd_stats.stat.value.ui64, (val))
85 #define	ZSTDSTAT_BUMP(stat)	ZSTDSTAT_ADD(stat, 1)
86 
87 /* (de)init for user space / kernel emulation */
88 int zstd_init(void);
89 void zstd_fini(void);
90 
91 size_t zfs_zstd_compress(void *s_start, void *d_start, size_t s_len,
92     size_t d_len, int level);
93 int zfs_zstd_get_level(void *s_start, size_t s_len, uint8_t *level);
94 int zfs_zstd_decompress_level(void *s_start, void *d_start, size_t s_len,
95     size_t d_len, uint8_t *level);
96 int zfs_zstd_decompress(void *s_start, void *d_start, size_t s_len,
97     size_t d_len, int n);
98 void zfs_zstd_cache_reap_now(void);
99 
100 /*
101  * So, the reason we have all these complicated set/get functions is that
102  * originally, in the zstd "header" we wrote out to disk, we used a 32-bit
103  * bitfield to store the "level" (8 bits) and "version" (24 bits).
104  *
105  * Unfortunately, bitfields make few promises about how they're arranged in
106  * memory...
107  *
108  * By way of example, if we were using version 1.4.5 and level 3, it'd be
109  * level = 0x03, version = 10405/0x0028A5, which gets broken into Vhigh = 0x00,
110  * Vmid = 0x28, Vlow = 0xA5. We include these positions below to help follow
111  * which data winds up where.
112  *
113  * As a consequence, we wound up with little endian platforms with a layout
114  * like this in memory:
115  *
116  *      0       8      16      24      32
117  *      +-------+-------+-------+-------+
118  *      | Vlow  | Vmid  | Vhigh | level |
119  *      +-------+-------+-------+-------+
120  *        =A5     =28     =00     =03
121  *
122  * ...and then, after being run through BE_32(), serializing this out to
123  * disk:
124  *
125  *      0       8      16      24      32
126  *      +-------+-------+-------+-------+
127  *      | level | Vhigh | Vmid  | Vlow  |
128  *      +-------+-------+-------+-------+
129  *        =03     =00     =28     =A5
130  *
131  * while on big-endian systems, since BE_32() is a noop there, both in
132  * memory and on disk, we wind up with:
133  *
134  *      0       8      16      24      32
135  *      +-------+-------+-------+-------+
136  *      | Vhigh | Vmid  | Vlow  | level |
137  *      +-------+-------+-------+-------+
138  *        =00     =28     =A5     =03
139  *
140  * (Vhigh is always 0 until version exceeds 6.55.35. Vmid and Vlow are the
141  * other two bytes of the "version" data.)
142  *
143  * So now we use the BF32_SET macros to get consistent behavior (the
144  * ondisk LE encoding, since x86 currently rules the world) across
145  * platforms, but the "get" behavior requires that we check each of the
146  * bytes in the aforementioned former-bitfield for 0x00, and from there,
147  * we can know which possible layout we're dealing with. (Only the two
148  * that have been observed in the wild are illustrated above, but handlers
149  * for all 4 positions of 0x00 are implemented.
150  */
151 
152 static inline void
153 zfs_get_hdrmeta(const zfs_zstdhdr_t *blob, zfs_zstdmeta_t *res)
154 {
155 	uint32_t raw = blob->raw_version_level;
156 	uint8_t findme = 0xff;
157 	int shift;
158 	for (shift = 0; shift < 4; shift++) {
159 		findme = BF32_GET(raw, 8*shift, 8);
160 		if (findme == 0)
161 			break;
162 	}
163 	switch (shift) {
164 	case 0:
165 		res->level = BF32_GET(raw, 24, 8);
166 		res->version = BSWAP_32(raw);
167 		res->version = BF32_GET(res->version, 8, 24);
168 		break;
169 	case 1:
170 		res->level = BF32_GET(raw, 0, 8);
171 		res->version = BSWAP_32(raw);
172 		res->version = BF32_GET(res->version, 0, 24);
173 		break;
174 	case 2:
175 		res->level = BF32_GET(raw, 24, 8);
176 		res->version = BF32_GET(raw, 0, 24);
177 		break;
178 	case 3:
179 		res->level = BF32_GET(raw, 0, 8);
180 		res->version = BF32_GET(raw, 8, 24);
181 		break;
182 	default:
183 		res->level = 0;
184 		res->version = 0;
185 		break;
186 	}
187 }
188 
189 static inline uint8_t
190 zfs_get_hdrlevel(const zfs_zstdhdr_t *blob)
191 {
192 	uint8_t level = 0;
193 	zfs_zstdmeta_t res;
194 	zfs_get_hdrmeta(blob, &res);
195 	level = res.level;
196 	return (level);
197 }
198 
199 static inline uint32_t
200 zfs_get_hdrversion(const zfs_zstdhdr_t *blob)
201 {
202 	uint32_t version = 0;
203 	zfs_zstdmeta_t res;
204 	zfs_get_hdrmeta(blob, &res);
205 	version = res.version;
206 	return (version);
207 
208 }
209 
210 static inline void
211 zfs_set_hdrversion(zfs_zstdhdr_t *blob, uint32_t version)
212 {
213 	BF32_SET(blob->raw_version_level, 0, 24, version);
214 }
215 
216 static inline void
217 zfs_set_hdrlevel(zfs_zstdhdr_t *blob, uint8_t level)
218 {
219 	BF32_SET(blob->raw_version_level, 24, 8, level);
220 }
221 
222 
223 #ifdef	__cplusplus
224 }
225 #endif
226 
227 #endif /* _ZFS_ZSTD_H */
228