diff --git a/dm-compress/util.c b/dm-compress/util.c index b7f32b6..ec6a4db 100644 --- a/dm-compress/util.c +++ b/dm-compress/util.c @@ -203,3 +203,236 @@ pblk_endio(struct bio* bio) bio_put(bio); return ret; } + +static inline u32 +cshift (u32 x, uint n) +{ + return (x << n) | (x >> (32 - n)); +} + +#define A ctx->counter[0] +#define B ctx->counter[1] +#define C ctx->counter[2] +#define D ctx->counter[3] +#define X data + +void +md5_init(struct md5* ctx) +{ + ctx->sz[0] = 0; + ctx->sz[1] = 0; + D = 0x10325476; + C = 0x98badcfe; + B = 0xefcdab89; + A = 0x67452301; +} + +#define F(x,y,z) ((x & y) | (~x & z)) +#define G(x,y,z) ((x & z) | (y & ~z)) +#define H(x,y,z) (x ^ y ^ z) +#define I(x,y,z) (y ^ (x | ~z)) + +#define DOIT(a,b,c,d,k,s,i,OP) \ +a = b + cshift(a + OP(b,c,d) + X[k] + (i), s) + +#define DO1(a,b,c,d,k,s,i) DOIT(a,b,c,d,k,s,i,F) +#define DO2(a,b,c,d,k,s,i) DOIT(a,b,c,d,k,s,i,G) +#define DO3(a,b,c,d,k,s,i) DOIT(a,b,c,d,k,s,i,H) +#define DO4(a,b,c,d,k,s,i) DOIT(a,b,c,d,k,s,i,I) + +static inline void +calc(struct md5* ctx, u32* data) +{ + u32 AA, BB, CC, DD; + + AA = A; + BB = B; + CC = C; + DD = D; + + /* Round 1 */ + + DO1(A,B,C,D,0,7,0xd76aa478); + DO1(D,A,B,C,1,12,0xe8c7b756); + DO1(C,D,A,B,2,17,0x242070db); + DO1(B,C,D,A,3,22,0xc1bdceee); + + DO1(A,B,C,D,4,7,0xf57c0faf); + DO1(D,A,B,C,5,12,0x4787c62a); + DO1(C,D,A,B,6,17,0xa8304613); + DO1(B,C,D,A,7,22,0xfd469501); + + DO1(A,B,C,D,8,7,0x698098d8); + DO1(D,A,B,C,9,12,0x8b44f7af); + DO1(C,D,A,B,10,17,0xffff5bb1); + DO1(B,C,D,A,11,22,0x895cd7be); + + DO1(A,B,C,D,12,7,0x6b901122); + DO1(D,A,B,C,13,12,0xfd987193); + DO1(C,D,A,B,14,17,0xa679438e); + DO1(B,C,D,A,15,22,0x49b40821); + + /* Round 2 */ + + DO2(A,B,C,D,1,5,0xf61e2562); + DO2(D,A,B,C,6,9,0xc040b340); + DO2(C,D,A,B,11,14,0x265e5a51); + DO2(B,C,D,A,0,20,0xe9b6c7aa); + + DO2(A,B,C,D,5,5,0xd62f105d); + DO2(D,A,B,C,10,9,0x2441453); + DO2(C,D,A,B,15,14,0xd8a1e681); + DO2(B,C,D,A,4,20,0xe7d3fbc8); + + DO2(A,B,C,D,9,5,0x21e1cde6); + DO2(D,A,B,C,14,9,0xc33707d6); + DO2(C,D,A,B,3,14,0xf4d50d87); + DO2(B,C,D,A,8,20,0x455a14ed); + + DO2(A,B,C,D,13,5,0xa9e3e905); + DO2(D,A,B,C,2,9,0xfcefa3f8); + DO2(C,D,A,B,7,14,0x676f02d9); + DO2(B,C,D,A,12,20,0x8d2a4c8a); + + /* Round 3 */ + + DO3(A,B,C,D,5,4,0xfffa3942); + DO3(D,A,B,C,8,11,0x8771f681); + DO3(C,D,A,B,11,16,0x6d9d6122); + DO3(B,C,D,A,14,23,0xfde5380c); + + DO3(A,B,C,D,1,4,0xa4beea44); + DO3(D,A,B,C,4,11,0x4bdecfa9); + DO3(C,D,A,B,7,16,0xf6bb4b60); + DO3(B,C,D,A,10,23,0xbebfbc70); + + DO3(A,B,C,D,13,4,0x289b7ec6); + DO3(D,A,B,C,0,11,0xeaa127fa); + DO3(C,D,A,B,3,16,0xd4ef3085); + DO3(B,C,D,A,6,23,0x4881d05); + + DO3(A,B,C,D,9,4,0xd9d4d039); + DO3(D,A,B,C,12,11,0xe6db99e5); + DO3(C,D,A,B,15,16,0x1fa27cf8); + DO3(B,C,D,A,2,23,0xc4ac5665); + + /* Round 4 */ + + DO4(A,B,C,D,0,6,0xf4292244); + DO4(D,A,B,C,7,10,0x432aff97); + DO4(C,D,A,B,14,15,0xab9423a7); + DO4(B,C,D,A,5,21,0xfc93a039); + + DO4(A,B,C,D,12,6,0x655b59c3); + DO4(D,A,B,C,3,10,0x8f0ccc92); + DO4(C,D,A,B,10,15,0xffeff47d); + DO4(B,C,D,A,1,21,0x85845dd1); + + DO4(A,B,C,D,8,6,0x6fa87e4f); + DO4(D,A,B,C,15,10,0xfe2ce6e0); + DO4(C,D,A,B,6,15,0xa3014314); + DO4(B,C,D,A,13,21,0x4e0811a1); + + DO4(A,B,C,D,4,6,0xf7537e82); + DO4(D,A,B,C,11,10,0xbd3af235); + DO4(C,D,A,B,2,15,0x2ad7d2bb); + DO4(B,C,D,A,9,21,0xeb86d391); + + A += AA; + B += BB; + C += CC; + D += DD; +} + +/* + * From `Performance analysis of MD5' by Joseph D. Touch + */ +#if !defined(__BYTE_ORDER__) || !defined (__ORDER_BIG_ENDIAN__) +#error __BYTE_ORDER macros not defined +#endif + +#if __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__ +static inline u32 +swap_u32(u32 t) +{ + u32 temp1, temp2; + + temp1 = cshift(t, 16); + temp2 = temp1 >> 8; + temp1 &= 0x00ff00ff; + temp2 &= 0x00ff00ff; + temp1 <<= 8; + return temp1 | temp2; +} +#endif + +struct x32 { + uint a:32; + uint b:32; +}; + +void +md5_update(struct md5* ctx, const void* data, size_t len) +{ + const byte* p = data; + size_t old_sz = ctx->sz[0]; + size_t offset; + + ctx->sz[0] += len * 8; + if (ctx->sz[0] < old_sz) + ++ctx->sz[1]; + offset = (old_sz / 8) % 64; + while(len > 0){ + size_t l = min(len, 64 - offset); + memcpy(ctx->save + offset, p, l); + offset += l; + p += l; + len -= l; + if (offset == 64) { +#if __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__ + int i; + u32 current[16]; + struct x32 *u = (struct x32*)ctx->save; + for (i = 0; i < 8; i++) { + current[2*i+0] = swap_u32(u[i].a); + current[2*i+1] = swap_u32(u[i].b); + } + calc(ctx, current); +#else + calc(ctx, (u32*)ctx->save); +#endif + offset = 0; + } + } +} + +void +md5_final(struct md5* ctx, byte* buf) +{ + byte zeros[72]; + uint offset = (ctx->sz[0] / 8) % 64; + uint dstart = (120 - offset - 1) % 64 + 1; + + *zeros = 0x80; + memset (zeros + 1, 0, sizeof(zeros) - 1); + zeros[dstart+0] = (ctx->sz[0] >> 0) & 0xff; + zeros[dstart+1] = (ctx->sz[0] >> 8) & 0xff; + zeros[dstart+2] = (ctx->sz[0] >> 16) & 0xff; + zeros[dstart+3] = (ctx->sz[0] >> 24) & 0xff; + zeros[dstart+4] = (ctx->sz[1] >> 0) & 0xff; + zeros[dstart+5] = (ctx->sz[1] >> 8) & 0xff; + zeros[dstart+6] = (ctx->sz[1] >> 16) & 0xff; + zeros[dstart+7] = (ctx->sz[1] >> 24) & 0xff; + md5_update(ctx, zeros, dstart + 8); + { + int i; + byte *r = buf; + + for (i = 0; i < 4; ++i) { + r[4*i] = ctx->counter[i] & 0xFF; + r[4*i+1] = (ctx->counter[i] >> 8) & 0xFF; + r[4*i+2] = (ctx->counter[i] >> 16) & 0xFF; + r[4*i+3] = (ctx->counter[i] >> 24) & 0xFF; + } + } +} diff --git a/include/linux/dm-compress.h b/include/linux/dm-compress.h index 7772b51..0b3c46f 100644 --- a/include/linux/dm-compress.h +++ b/include/linux/dm-compress.h @@ -466,6 +466,18 @@ void cbd_free_pages(struct page* pages, size_t len); bool cbd_alloc_pagev(struct page** pagev, size_t len); void cbd_free_pagev(struct page** pagev, size_t len); +/* Debug stuff */ +typedef unsigned char byte; +#define MD5_DIGEST_SIZE 16 +struct md5 { + uint sz[2]; + u32 counter[4]; + byte save[64]; +}; +void md5_init(struct md5* ctx); +void md5_update(struct md5* ctx, const void* data, size_t len); +void md5_final(struct md5* ctx, byte* buf); + int pblk_read_wait(struct cbd_params* params, u64 pblk, u32 count, struct page** pagev); int pblk_read(struct cbd_params* params,