From: Namjae Jeon Date: Wed, 10 Jun 2026 09:44:55 +0000 (+0900) Subject: smb: move LZ77 compression into common code X-Git-Url: http://git.ipfire.org/gitweb.cgi?a=commitdiff_plain;h=0121b154147f5affd1039ca4f8d9a0fc194142f6;p=thirdparty%2Fkernel%2Flinux.git smb: move LZ77 compression into common code Move the LZ77 codec in cifs.ko to smb/common/ so both the SMB client and ksmbd can use it. Signed-off-by: Namjae Jeon Signed-off-by: Steve French --- diff --git a/fs/smb/client/Makefile b/fs/smb/client/Makefile index 6e83b5204699c..fc6b9d35c9624 100644 --- a/fs/smb/client/Makefile +++ b/fs/smb/client/Makefile @@ -42,7 +42,7 @@ cifs-$(CONFIG_CIFS_ALLOW_INSECURE_LEGACY) += \ smb1session.o \ smb1transport.o -cifs-$(CONFIG_CIFS_COMPRESSION) += compress.o compress/lz77.o +cifs-$(CONFIG_CIFS_COMPRESSION) += compress.o ifneq ($(CONFIG_CIFS_ALLOW_INSECURE_LEGACY),) # diff --git a/fs/smb/client/compress.c b/fs/smb/client/compress.c index be9023f841e69..8f08609707414 100644 --- a/fs/smb/client/compress.c +++ b/fs/smb/client/compress.c @@ -22,7 +22,7 @@ #include "cifsproto.h" #include "smb2proto.h" -#include "compress/lz77.h" +#include "../common/compress/lz77.h" #include "compress.h" /* @@ -44,6 +44,11 @@ struct bucket { unsigned int count; }; +static inline size_t pow4(size_t n) +{ + return n * n * n * n; +} + /* * has_low_entropy() - Compute Shannon entropy of the sampled data. * @bkt: Bytes counts of the sample. @@ -65,7 +70,6 @@ static bool has_low_entropy(struct bucket *bkt, size_t slen) const size_t threshold = 65, max_entropy = 8 * ilog2(16); size_t i, p, p2, len, sum = 0; -#define pow4(n) (n * n * n * n) len = ilog2(pow4(slen)); for (i = 0; i < 256 && bkt[i].count > 0; i++) { @@ -329,14 +333,14 @@ int smb_compress(struct TCP_Server_Info *server, struct smb_rqst *rq, compress_s goto err_free; } - dlen = lz77_compressed_alloc_size(slen); + dlen = smb_lz77_compressed_alloc_size(slen); dst = kvzalloc(dlen, GFP_KERNEL); if (!dst) { ret = -ENOMEM; goto err_free; } - ret = lz77_compress(src, slen, dst, &dlen); + ret = smb_lz77_compress(src, slen, dst, &dlen); if (!ret) { struct smb2_compression_hdr hdr = { 0 }; struct smb_rqst comp_rq = { .rq_nvec = 3, }; diff --git a/fs/smb/client/compress.h b/fs/smb/client/compress.h index 2679baca129b7..e08e6d339d21e 100644 --- a/fs/smb/client/compress.h +++ b/fs/smb/client/compress.h @@ -18,6 +18,7 @@ #include #include #include "../common/smb2pdu.h" +#include "../common/compress/compress.h" #include "cifsglob.h" /* sizeof(smb2_compression_hdr) - sizeof(OriginalPayloadSize) */ @@ -34,29 +35,6 @@ int smb_compress(struct TCP_Server_Info *server, struct smb_rqst *rq, compress_send_fn send_fn); bool should_compress(const struct cifs_tcon *tcon, const struct smb_rqst *rq); -/* - * smb_compress_alg_valid() - Validate a compression algorithm. - * @alg: Compression algorithm to check. - * @valid_none: Conditional check whether NONE algorithm should be - * considered valid or not. - * - * If @alg is SMB3_COMPRESS_NONE, this function returns @valid_none. - * - * Note that 'NONE' (0) compressor type is considered invalid in protocol - * negotiation, as it's never requested to/returned from the server. - * - * Return: true if @alg is valid/supported, false otherwise. - */ -static __always_inline int smb_compress_alg_valid(__le16 alg, bool valid_none) -{ - if (alg == SMB3_COMPRESS_NONE) - return valid_none; - - if (alg == SMB3_COMPRESS_LZ77 || alg == SMB3_COMPRESS_PATTERN) - return true; - - return false; -} #else /* !CONFIG_CIFS_COMPRESSION */ static inline int smb_compress(void *unused1, void *unused2, void *unused3) { @@ -68,9 +46,5 @@ static inline bool should_compress(void *unused1, void *unused2) return false; } -static inline int smb_compress_alg_valid(__le16 unused1, bool unused2) -{ - return -EOPNOTSUPP; -} #endif /* !CONFIG_CIFS_COMPRESSION */ #endif /* _SMB_COMPRESS_H */ diff --git a/fs/smb/common/Makefile b/fs/smb/common/Makefile index 9e0730a385fb1..bd188d36fb6b5 100644 --- a/fs/smb/common/Makefile +++ b/fs/smb/common/Makefile @@ -4,3 +4,6 @@ # obj-$(CONFIG_SMBFS) += cifs_md4.o +obj-$(CONFIG_SMBFS) += smb_compress.o + +smb_compress-y := compress/lz77.o diff --git a/fs/smb/common/compress/compress.h b/fs/smb/common/compress/compress.h new file mode 100644 index 0000000000000..b504cd38b128b --- /dev/null +++ b/fs/smb/common/compress/compress.h @@ -0,0 +1,22 @@ +/* SPDX-License-Identifier: GPL-2.0-only */ +/* + * Copyright (C) 2026 Namjae Jeon + */ +#ifndef _COMMON_SMB_COMPRESS_H +#define _COMMON_SMB_COMPRESS_H + +#include "../smb2pdu.h" + +/* + * SMB3_COMPRESS_NONE is valid only in chained payload headers. It is never + * negotiated as a compression algorithm. + */ +static __always_inline bool smb_compress_alg_valid(__le16 alg, bool valid_none) +{ + if (alg == SMB3_COMPRESS_NONE) + return valid_none; + + return alg == SMB3_COMPRESS_LZ77 || alg == SMB3_COMPRESS_PATTERN; +} + +#endif /* _COMMON_SMB_COMPRESS_H */ diff --git a/fs/smb/client/compress/lz77.c b/fs/smb/common/compress/lz77.c similarity index 75% rename from fs/smb/client/compress/lz77.c rename to fs/smb/common/compress/lz77.c index 7365d0f97396c..9216d973d8769 100644 --- a/fs/smb/client/compress/lz77.c +++ b/fs/smb/common/compress/lz77.c @@ -1,8 +1,10 @@ // SPDX-License-Identifier: GPL-2.0-only /* * Copyright (C) 2024-2026, SUSE LLC + * Copyright (C) 2026 Namjae Jeon * * Authors: Enzo Matsumiya + * Namjae Jeon * * Implementation of the LZ77 "plain" compression algorithm, as per MS-XCA spec. */ @@ -10,6 +12,8 @@ #include #include #include +#include +#include #include "lz77.h" @@ -32,7 +36,7 @@ */ #define LZ77_MATCH_MAX_DIST SZ_8K #define LZ77_HASH_LOG 15 -#define LZ77_HASH_SIZE (1 << LZ77_HASH_LOG) +#define LZ77_HASH_SIZE BIT(LZ77_HASH_LOG) #define LZ77_RSTEP_SIZE sizeof(u32) #define LZ77_MSTEP_SIZE sizeof(u64) #define LZ77_SKIP_TRIGGER 4 @@ -215,7 +219,8 @@ static __always_inline u32 lz77_hash(const u32 v) return ((v ^ 0x9E3779B9) * 0x85EBCA6B) >> (32 - LZ77_HASH_LOG); } -noinline int lz77_compress(const void *src, const u32 slen, void *dst, u32 *dlen) +noinline int smb_lz77_compress(const void *src, const u32 slen, + void *dst, u32 *dlen) { const void *srcp, *rlim, *end, *anchor; u32 *htable, hash, flag_count = 0; @@ -223,10 +228,11 @@ noinline int lz77_compress(const void *src, const u32 slen, void *dst, u32 *dlen long flag = 0; /* This is probably a bug, so throw a warning. */ - if (WARN_ON_ONCE(*dlen < lz77_compressed_alloc_size(slen))) + if (WARN_ON_ONCE(*dlen < smb_lz77_compressed_alloc_size(slen))) return -EINVAL; - srcp = anchor = src; + srcp = src; + anchor = src; end = srcp + slen; /* absolute end */ rlim = end - LZ77_MSTEP_SIZE; /* read limit (for lz77_match_len()) */ dstp = dst; @@ -251,7 +257,8 @@ noinline int lz77_compress(const void *src, const u32 slen, void *dst, u32 *dlen /* * Main loop. * - * @dlen is >= lz77_compressed_alloc_size(), so run without bound-checking @dstp. + * @dlen is >= smb_lz77_compressed_alloc_size(), so run without + * bound-checking @dstp. * * This code was crafted in a way to best utilise fetch-decode-execute CPU flow. * Any attempt to optimize it, or even organize it, can lead to huge performance loss. @@ -333,3 +340,122 @@ out: return -EMSGSIZE; } +EXPORT_SYMBOL_GPL(smb_lz77_compress); + +static int lz77_decode_match_len(const u8 **src, const u8 *end, u16 token, + u8 *nibble, bool *have_nibble, u32 *len) +{ + u8 extra; + + *len = (token & 0x7) + 3; + if ((token & 0x7) != 0x7) + return 0; + + if (!*have_nibble) { + if (*src >= end) + return -EINVAL; + *nibble = *(*src)++; + extra = *nibble & 0xf; + *have_nibble = true; + } else { + extra = *nibble >> 4; + *have_nibble = false; + } + + *len += extra; + if (extra == 0xf) { + u8 b; + + if (*src >= end) + return -EINVAL; + b = *(*src)++; + if (b != 0xff) { + *len += b; + } else { + u16 w; + + if (end - *src < 2) + return -EINVAL; + w = get_unaligned_le16(*src); + *src += 2; + if (w) { + *len = w + 3; + } else { + u32 long_len; + + if (end - *src < 4) + return -EINVAL; + long_len = get_unaligned_le32(*src); + *src += 4; + if (check_add_overflow(long_len, 3, len)) + return -EINVAL; + } + } + } + + return 0; +} + +int smb_lz77_decompress(const void *src, const u32 slen, void *dst, + const u32 dlen) +{ + const u8 *sp = src, *send = sp + slen; + u8 *dp = dst, *dend = dp + dlen; + u32 flags = 0; + int flag_count = 0; + u8 nibble = 0; + bool have_nibble = false; + + while (dp < dend) { + u32 len, dist; + u16 token; + + if (!flag_count) { + if (send - sp < 4) + return -EINVAL; + flags = get_unaligned_le32(sp); + sp += 4; + flag_count = 32; + } + + if (!(flags & 0x80000000)) { + if (sp >= send) + return -EINVAL; + *dp++ = *sp++; + flags <<= 1; + flag_count--; + continue; + } + + flags <<= 1; + flag_count--; + + if (send - sp < 2) + return -EINVAL; + + token = get_unaligned_le16(sp); + sp += 2; + + dist = (token >> 3) + 1; + if (dist > dp - (u8 *)dst) + return -EINVAL; + + if (lz77_decode_match_len(&sp, send, token, &nibble, + &have_nibble, &len)) + return -EINVAL; + + if (len > dend - dp) + return -EINVAL; + + while (len--) { + *dp = *(dp - dist); + dp++; + } + } + + return 0; +} +EXPORT_SYMBOL_GPL(smb_lz77_decompress); + +MODULE_LICENSE("GPL"); +MODULE_DESCRIPTION("SMB plain LZ77 compression"); diff --git a/fs/smb/client/compress/lz77.h b/fs/smb/common/compress/lz77.h similarity index 65% rename from fs/smb/client/compress/lz77.h rename to fs/smb/common/compress/lz77.h index 4e570846aefa2..e032c0f1b48dd 100644 --- a/fs/smb/client/compress/lz77.h +++ b/fs/smb/common/compress/lz77.h @@ -12,7 +12,7 @@ #include /** - * lz77_compressed_alloc_size() - Compute compressed buffer size. + * smb_lz77_compressed_alloc_size() - Compute compressed buffer size. * @size: uncompressed (src) size * * Compute allocation size for the compressed buffer based on uncompressed size. @@ -26,18 +26,21 @@ * Worst case scenario is an all-literal compression, which means: * metadata bytes = 4 + ((@size / 32) * 4) + 4, or, simplified, (@size >> 3) + 8 * - * The worst case scenario rarely happens, but such overprovisioning also allows lz77_compress() - * main loop to run without ever bound checking dst, which is a huge perf improvement, while also - * being safe when compression goes bad. + * The worst case scenario rarely happens, but such overprovisioning also + * allows smb_lz77_compress() main loop to run without ever bound checking dst, + * which is a huge perf improvement, while also being safe when compression goes + * bad. * * Return: required (*) allocation size for compressed buffer. * - * (*) checked once in the beginning of lz77_compress() + * (*) checked once in the beginning of smb_lz77_compress() */ -static __always_inline u32 lz77_compressed_alloc_size(const u32 size) +static __always_inline u32 smb_lz77_compressed_alloc_size(const u32 size) { return size + (size >> 3) + 8; } -int lz77_compress(const void *src, const u32 slen, void *dst, u32 *dlen); +int smb_lz77_compress(const void *src, const u32 slen, void *dst, u32 *dlen); +int smb_lz77_decompress(const void *src, const u32 slen, void *dst, + const u32 dlen); #endif /* _SMB_COMPRESS_LZ77_H */