Laurence Lundblade | 501d3f8 | 2022-04-08 14:52:55 -0600 | [diff] [blame^] | 1 | // |
| 2 | // ub-example.c |
| 3 | // QCBOR |
| 4 | // |
| 5 | // Created by Laurence Lundblade on 4/8/22. |
| 6 | // Copyright © 2022 Laurence Lundblade. All rights reserved. |
| 7 | // |
| 8 | |
| 9 | #include "ub-example.h" |
| 10 | |
| 11 | #include "UsefulBuf.h" |
| 12 | |
| 13 | |
| 14 | /* |
| 15 | A large number of the security issues with C code come from mistakes |
| 16 | made with a pointer and length for a buffer or some binary data. |
| 17 | UsefulBuf adopts a convention that a pointer and length *always* |
| 18 | go together to migitigate this. With UsefulBuf there are never |
| 19 | pointers without lengths so you always know how big the buffer |
| 20 | or the data is. |
| 21 | |
| 22 | C99 allows passing structures so a structure is used. Compilers |
| 23 | are smart these days so the object code produced is no different |
| 24 | than passing two separate parameters. Passing structures also |
| 25 | makes the interfaces prettier. Assignments of structures also |
| 26 | can make code prettier. |
| 27 | |
| 28 | There are a bunch of (tested!) functions to manipulate UsefulBuf's so |
| 29 | code using it may have no pointer manipulation at all! |
| 30 | |
| 31 | In this example the buffers that are filled in with data |
| 32 | are const and the ones that are to-be-filled in are not |
| 33 | const. Keeping const distinct from non-const is helpful |
| 34 | when reading the code and helps avoid some coding mistakes. |
| 35 | See this: |
| 36 | https://stackoverflow.com/questions/117293/use-of-const-for-function-parameters |
| 37 | |
| 38 | This contrived example copies data from input to output |
| 39 | expanding bytes with the value 'x' to 'xx'. |
| 40 | |
| 41 | Input -- This is the pointer and length of the input, the |
| 42 | bytes to copy. Note that UsefulBufC.ptr is a const void * |
| 43 | indicates that input data won't be changed by this function. |
| 44 | There is a "C" in UsefulBufC to indicate the value is const. |
| 45 | The length here is the length of the valid input data. Note |
| 46 | also that the parameter Input is const, so this is fully |
| 47 | const and clearly an [in] parameter. |
| 48 | |
| 49 | Output -- This is a pointer and length of |
| 50 | the memory to used to store the output. The correct length |
| 51 | here is critical for code security. Note that UsefulBuf.ptr |
| 52 | is void *, it is not const indicating data can be written to |
| 53 | it. Note that the parameter itself *is* const indicating |
| 54 | that the code below will not point this to some other buffer |
| 55 | or change the length and clearly marked as an [in] parameter. |
| 56 | |
| 57 | Output -- This is the interesting and unusual one. To stay |
| 58 | consistent with always paring and a length and for |
| 59 | a pointer to valid data to always be const, this is returned as |
| 60 | a UsefulBufC. Note that the parameter is a pointer to a |
| 61 | UsefulBufC, a *place* to return a UsefulBufC. |
| 62 | |
| 63 | In this case and most cases the pointer in Output->ptr |
| 64 | will be the same as OutputBuffer.ptr. This may seem |
| 65 | redundant, but there's a few reasons for it. First, |
| 66 | is the goal of always pairing a pointer and a length. |
| 67 | Second is being more strict with constness. Third |
| 68 | is the code hygene and clarity of having |
| 69 | variables for to-be-filled buffers be distinct from those |
| 70 | containing valid data. Fourth, there are no [in,out] |
| 71 | parameters, only [in] parameters and [out] parameters |
| 72 | (the to-be-filled-in buffer is considered an [in] |
| 73 | parameter). |
| 74 | |
| 75 | Note that the compiler will be smart about all |
| 76 | this and should generate pretty much the same code |
| 77 | as for a traditional interface with the |
| 78 | length parameter. On x86 with gcc-11 and no stack guards, |
| 79 | the UB code is 81 bytes and the traditional code is 77 bytes. |
| 80 | |
| 81 | This supports computing of the would-be output |
| 82 | without actually doing any outputing by making |
| 83 | the OutputBuffer have a NULL pointer and a very |
| 84 | large length, e.g., {NULL, SIZE_MAX}. |
| 85 | |
| 86 | */ |
| 87 | int |
| 88 | ExpandUB(const UsefulBufC Input, |
| 89 | const UsefulBuf OutputBuffer, |
| 90 | UsefulBufC *Output) |
| 91 | { |
| 92 | size_t nInputPosition; |
| 93 | size_t nOutputPosition; |
| 94 | |
| 95 | nOutputPosition = 0; |
| 96 | |
| 97 | /* Loop over all the bytes in Input */ |
| 98 | for(nInputPosition = 0; nInputPosition < Input.len; nInputPosition++) { |
| 99 | const uint8_t nInputByte = ((uint8_t*)Input.ptr)[nInputPosition]; |
| 100 | |
| 101 | /* Copy every byte */ |
| 102 | if(OutputBuffer.ptr != NULL) { |
| 103 | ((uint8_t *)OutputBuffer.ptr)[nOutputPosition] = nInputByte; |
| 104 | } |
| 105 | nOutputPosition++; |
| 106 | if(nOutputPosition >= OutputBuffer.len) { |
| 107 | return -1l; |
| 108 | } |
| 109 | |
| 110 | /* Double output 'x' because that is what this contrived example does */ |
| 111 | if(nInputByte== 'x') { |
| 112 | if(OutputBuffer.ptr != NULL) { |
| 113 | ((uint8_t *)OutputBuffer.ptr)[nOutputPosition] = 'x'; |
| 114 | } |
| 115 | nOutputPosition++; |
| 116 | if(nOutputPosition >= OutputBuffer.len) { |
| 117 | return -1l; |
| 118 | } |
| 119 | } |
| 120 | } |
| 121 | |
| 122 | *Output = (UsefulBufC){OutputBuffer.ptr, nOutputPosition}; |
| 123 | |
| 124 | return 0; /* success */ |
| 125 | } |
| 126 | |
| 127 | |
| 128 | /* This is the more tradional way to implement this. */ |
| 129 | int ExpandTraditional(const uint8_t *pInputPointer, |
| 130 | const size_t uInputLength, |
| 131 | uint8_t *pOutputBuffer, |
| 132 | const size_t uOutputBufferLength, |
| 133 | size_t *puOutputLength) |
| 134 | { |
| 135 | size_t nInputPosition; |
| 136 | size_t nOutputPosition; |
| 137 | |
| 138 | nOutputPosition = 0; |
| 139 | |
| 140 | /* Loop over all the bytes in Input */ |
| 141 | for(nInputPosition = 0; nInputPosition < uInputLength; nInputPosition++) { |
| 142 | const uint8_t nInputByte = ((uint8_t*)pInputPointer)[nInputPosition]; |
| 143 | |
| 144 | /* Copy every byte */ |
| 145 | if(pOutputBuffer != NULL) { |
| 146 | ((uint8_t *)pOutputBuffer)[nOutputPosition] = nInputByte; |
| 147 | } |
| 148 | nOutputPosition++; |
| 149 | if(nOutputPosition >= uOutputBufferLength) { |
| 150 | return -1l; |
| 151 | } |
| 152 | |
| 153 | /* Double output 'x' because that is what this contrived example does */ |
| 154 | if(nInputByte== 'x') { |
| 155 | if(pOutputBuffer != NULL) { |
| 156 | ((uint8_t *)pOutputBuffer)[nOutputPosition] = 'x'; |
| 157 | } |
| 158 | nOutputPosition++; |
| 159 | if(nOutputPosition >= uOutputBufferLength) { |
| 160 | return -1l; |
| 161 | } |
| 162 | } |
| 163 | } |
| 164 | |
| 165 | *puOutputLength = nOutputPosition; |
| 166 | |
| 167 | return 0; /* success */ |
| 168 | } |
| 169 | |
| 170 | |
| 171 | /* |
| 172 | Here's an example of going from a traditional interface |
| 173 | interface to a UsefulBuf interface. |
| 174 | */ |
| 175 | int ExpandTraditionalAdapted(const uint8_t *pInputPointer, |
| 176 | size_t uInputLength, |
| 177 | uint8_t *pOutputBuffer, |
| 178 | size_t uOutputBufferLength, |
| 179 | size_t *puOutputLength) |
| 180 | { |
| 181 | UsefulBufC Input; |
| 182 | UsefulBuf OutputBuffer; |
| 183 | UsefulBufC Output; |
| 184 | int nReturn; |
| 185 | |
| 186 | Input = (UsefulBufC){pInputPointer, uInputLength}; |
| 187 | OutputBuffer = (UsefulBuf){pOutputBuffer, uOutputBufferLength}; |
| 188 | |
| 189 | nReturn = ExpandUB(Input, OutputBuffer, &Output); |
| 190 | |
| 191 | *puOutputLength = Output.len; |
| 192 | |
| 193 | return nReturn; |
| 194 | } |
| 195 | |
| 196 | |
| 197 | /* Here's an example for going from a UsefulBuf interface |
| 198 | to a traditional interface. */ |
| 199 | int |
| 200 | ExpandUBAdapted(const UsefulBufC Input, |
| 201 | const UsefulBuf OutputBuffer, |
| 202 | UsefulBufC *Output) |
| 203 | { |
| 204 | Output->ptr = OutputBuffer.ptr; |
| 205 | |
| 206 | return ExpandTraditional(Input.ptr, Input.len, |
| 207 | OutputBuffer.ptr, OutputBuffer.len, |
| 208 | &(Output->len)); |
| 209 | } |
| 210 | |
| 211 | |
| 212 | |
| 213 | #define INPUT "xyz123xyz" |
| 214 | |
| 215 | int32_t RunUsefulBufExample() |
| 216 | { |
| 217 | /* ------------ UsefulBuf examples ------------- */ |
| 218 | UsefulBufC Input = UsefulBuf_FROM_SZ_LITERAL(INPUT); |
| 219 | |
| 220 | /* This macros makes a 20 byte buffer on the stack. It also makes |
| 221 | * a UsefulBuf on the stack. It sets up the UsefulBuf to point to |
| 222 | * the 20 byte buffer and sets it's length to 20 bytes. This |
| 223 | * is the empty, to-be-filled in memory for the output. It is not |
| 224 | * const. */ |
| 225 | MakeUsefulBufOnStack(OutBuf, sizeof(INPUT) * 2); |
| 226 | |
| 227 | /* This is were the pointer and the length of the completed output |
| 228 | * will be placed. Output.ptr is a pointer to const bytes. */ |
| 229 | UsefulBufC Output; |
| 230 | |
| 231 | ExpandUB(Input, OutBuf, &Output); |
| 232 | |
| 233 | ExpandUBAdapted(Input, OutBuf, &Output); |
| 234 | |
| 235 | |
| 236 | |
| 237 | /* ------ Get Size example -------- */ |
| 238 | ExpandUB(Input, (UsefulBuf){NULL, SIZE_MAX}, &Output); |
| 239 | |
| 240 | /* Size is in Output.len */ |
| 241 | |
| 242 | |
| 243 | |
| 244 | /* ---------- Traditional examples (for comparison) --------- */ |
| 245 | uint8_t puBuffer[sizeof(INPUT) * 2]; |
| 246 | size_t uOutputSize; |
| 247 | |
| 248 | ExpandTraditional((const uint8_t *)INPUT, sizeof(INPUT), |
| 249 | puBuffer, sizeof(puBuffer), |
| 250 | &uOutputSize); |
| 251 | |
| 252 | |
| 253 | ExpandTraditionalAdapted((const uint8_t *)INPUT, sizeof(INPUT), |
| 254 | puBuffer, sizeof(puBuffer), |
| 255 | &uOutputSize); |
| 256 | |
| 257 | return 0; |
| 258 | } |