/* This Source Code Form is subject to the terms of the Mozilla Public * License, v. 2.0. If a copy of the MPL was not distributed with this * file, You can obtain one at http://mozilla.org/MPL/2.0/. */ /* * Base64 decoding (ascii to binary). */ #include "nssb64.h" #include "nspr.h" #include "secitem.h" #include "secerr.h" /* * XXX We want this basic support to go into NSPR (the PL part). * Until that can happen, the PL interface is going to be kept entirely * internal here -- all static functions and opaque data structures. * When someone can get it moved over into NSPR, that should be done: * - giving everything names that are accepted by the NSPR module owners * (though I tried to choose ones that would work without modification) * - exporting the functions (remove static declarations and add * to nssutil.def as necessary) * - put prototypes into appropriate header file (probably replacing * the entire current lib/libc/include/plbase64.h in NSPR) * along with a typedef for the context structure (which should be * kept opaque -- definition in the source file only, but typedef * ala "typedef struct PLBase64FooStr PLBase64Foo;" in header file) * - modify anything else as necessary to conform to NSPR required style * (I looked but found no formatting guide to follow) * * You will want to move over everything from here down to the comment * which says "XXX End of base64 decoding code to be moved into NSPR", * into a new file in NSPR. */ /* ************************************************************** * XXX Beginning of base64 decoding code to be moved into NSPR. */ /* * This typedef would belong in the NSPR header file (i.e. plbase64.h). */ typedef struct PLBase64DecoderStr PLBase64Decoder; /* * The following implementation of base64 decoding was based on code * found in libmime (specifically, in mimeenc.c). It has been adapted to * use PR types and naming as well as to provide other necessary semantics * (like buffer-in/buffer-out in addition to "streaming" without undue * performance hit of extra copying if you made the buffer versions * use the output_fn). It also incorporates some aspects of the current * NSPR base64 decoding code. As such, you may find similarities to * both of those implementations. I tried to use names that reflected * the original code when possible. For this reason you may find some * inconsistencies -- libmime used lots of "in" and "out" whereas the * NSPR version uses "src" and "dest"; sometimes I changed one to the other * and sometimes I left them when I thought the subroutines were at least * self-consistent. */ PR_BEGIN_EXTERN_C /* * Opaque object used by the decoder to store state. */ struct PLBase64DecoderStr { /* Current token (or portion, if token_size < 4) being decoded. */ unsigned char token[4]; int token_size; /* * Where to write the decoded data (used when streaming, not when * doing all in-memory (buffer) operations). * * Note that this definition is chosen to be compatible with PR_Write. */ PRInt32 (*output_fn)(void *output_arg, const unsigned char *buf, PRInt32 size); void *output_arg; /* * Where the decoded output goes -- either temporarily (in the streaming * case, staged here before it goes to the output function) or what will * be the entire buffered result for users of the buffer version. */ unsigned char *output_buffer; PRUint32 output_buflen; /* the total length of allocated buffer */ PRUint32 output_length; /* the length that is currently populated */ }; PR_END_EXTERN_C /* A constant time range check for unsigned chars. * Returns 255 if a <= x <= b and 0 otherwise. */ static inline unsigned char ct_u8_in_range(unsigned char x, unsigned char a, unsigned char b) { /* Let x, a, b be ints in {0, 1, ... 255}. * The value (a - x - 1) is in {-256, ..., 254}, so the low * 8 bits of * (a - x - 1) >> 8 * are all 1 if a <= x and all 0 if a > x. * * Likewise the low 8 bits of * ((a - x - 1) >> 8) & ((x - c - 1) >> 8) * are all 1 if a <= x <= c and all 0 otherwise. * * The same is true if we perform the shift after the AND * ((a - x - 1) & (x - b - 1)) >> 8. */ return (unsigned char)(((a - x - 1) & (x - b - 1)) >> 8); } /* Convert a base64 code [A-Za-z0-9+/] to its value in {1, 2, ..., 64}. * The use of 1-64 instead of 0-63 is so that the special value of zero can * denote an invalid mapping; that was much easier than trying to fill in the * other values with some value other than zero, and to check for it. * Just remember to SUBTRACT ONE when using the value retrieved. */ static unsigned char pl_base64_codetovaluep1(unsigned char code) { unsigned char mask; unsigned char res = 0; /* The range 'A' to 'Z' is mapped to 1 to 26 */ mask = ct_u8_in_range(code, 'A', 'Z'); res |= mask & (code - 'A' + 1); /* The range 'a' to 'z' is mapped to 27 to 52 */ mask = ct_u8_in_range(code, 'a', 'z'); res |= mask & (code - 'a' + 27); /* The range '0' to '9' is mapped to 53 to 62 */ mask = ct_u8_in_range(code, '0', '9'); res |= mask & (code - '0' + 53); /* The code '+' is mapped to 63 */ mask = ct_u8_in_range(code, '+', '+'); res |= mask & 63; /* The code '/' is mapped to 64 */ mask = ct_u8_in_range(code, '/', '/'); res |= mask & 64; /* All other characters, including '=' are mapped to 0. */ return res; } #define B64_PAD '=' /* * Reads 4; writes 3 (known, or expected, to have no trailing padding). * Returns bytes written; -1 on error (unexpected character). */ static int pl_base64_decode_4to3(const unsigned char *in, unsigned char *out) { int j; PRUint32 num = 0; unsigned char bits; for (j = 0; j < 4; j++) { bits = pl_base64_codetovaluep1(in[j]); if (bits == 0) return -1; num = (num << 6) | (bits - 1); } out[0] = (unsigned char)(num >> 16); out[1] = (unsigned char)((num >> 8) & 0xFF); out[2] = (unsigned char)(num & 0xFF); return 3; } /* * Reads 3; writes 2 (caller already confirmed EOF or trailing padding). * Returns bytes written; -1 on error (unexpected character). */ static int pl_base64_decode_3to2(const unsigned char *in, unsigned char *out) { PRUint32 num = 0; unsigned char bits1, bits2, bits3; bits1 = pl_base64_codetovaluep1(in[0]); bits2 = pl_base64_codetovaluep1(in[1]); bits3 = pl_base64_codetovaluep1(in[2]); if ((bits1 == 0) || (bits2 == 0) || (bits3 == 0)) return -1; num = ((PRUint32)(bits1 - 1)) << 10; num |= ((PRUint32)(bits2 - 1)) << 4; num |= ((PRUint32)(bits3 - 1)) >> 2; out[0] = (unsigned char)(num >> 8); out[1] = (unsigned char)(num & 0xFF); return 2; } /* * Reads 2; writes 1 (caller already confirmed EOF or trailing padding). * Returns bytes written; -1 on error (unexpected character). */ static int pl_base64_decode_2to1(const unsigned char *in, unsigned char *out) { PRUint32 num = 0; unsigned char bits1, bits2; bits1 = pl_base64_codetovaluep1(in[0]); bits2 = pl_base64_codetovaluep1(in[1]); if ((bits1 == 0) || (bits2 == 0)) return -1; num = ((PRUint32)(bits1 - 1)) << 2; num |= ((PRUint32)(bits2 - 1)) >> 4; out[0] = (unsigned char)num; return 1; } /* * Reads 4; writes 0-3. Returns bytes written or -1 on error. * (Writes less than 3 only at (presumed) EOF.) */ static int pl_base64_decode_token(const unsigned char *in, unsigned char *out) { if (in[3] != B64_PAD) return pl_base64_decode_4to3(in, out); if (in[2] == B64_PAD) return pl_base64_decode_2to1(in, out); return pl_base64_decode_3to2(in, out); } static PRStatus pl_base64_decode_buffer(PLBase64Decoder *data, const unsigned char *in, PRUint32 length) { unsigned char *out = data->output_buffer; unsigned char *token = data->token; int i, n = 0; i = data->token_size; data->token_size = 0; while (length > 0) { while (i < 4 && length > 0) { /* * XXX Note that the following simply ignores any unexpected * characters. This is exactly what the original code in * libmime did, and I am leaving it. We certainly want to skip * over whitespace (we must); this does much more than that. * I am not confident changing it, and I don't want to slow * the processing down doing more complicated checking, but * someone else might have different ideas in the future. */ if (pl_base64_codetovaluep1(*in) > 0 || *in == B64_PAD) token[i++] = *in; in++; length--; } if (i < 4) { /* Didn't get enough for a complete token. */ data->token_size = i; break; } i = 0; PR_ASSERT((PRUint32)(out - data->output_buffer + 3) <= data->output_buflen); /* * Assume we are not at the end; the following function only works * for an internal token (no trailing padding characters) but is * faster that way. If it hits an invalid character (padding) it * will return an error; we break out of the loop and try again * calling the routine that will handle a final token. * Note that we intentionally do it this way rather than explicitly * add a check for padding here (because that would just slow down * the normal case) nor do we rely on checking whether we have more * input to process (because that would also slow it down but also * because we want to allow trailing garbage, especially white space * and cannot tell that without read-ahead, also a slow proposition). * Whew. Understand? */ n = pl_base64_decode_4to3(token, out); if (n < 0) break; /* Advance "out" by the number of bytes just written to it. */ out += n; n = 0; } /* * See big comment above, before call to pl_base64_decode_4to3. * Here we check if we error'd out of loop, and allow for the case * that we are processing the last interesting token. If the routine * which should handle padding characters also fails, then we just * have bad input and give up. */ if (n < 0) { n = pl_base64_decode_token(token, out); if (n < 0) return PR_FAILURE; out += n; } /* * As explained above, we can get here with more input remaining, but * it should be all characters we do not care about (i.e. would be * ignored when transferring from "in" to "token" in loop above, * except here we choose to ignore extraneous pad characters, too). * Swallow it, performing that check. If we find more characters that * we would expect to decode, something is wrong. */ while (length > 0) { if (pl_base64_codetovaluep1(*in) > 0) return PR_FAILURE; in++; length--; } /* Record the length of decoded data we have left in output_buffer. */ data->output_length = (PRUint32)(out - data->output_buffer); return PR_SUCCESS; } /* * Flush any remaining buffered characters. Given well-formed input, * this will have nothing to do. If the input was missing the padding * characters at the end, though, there could be 1-3 characters left * behind -- we will tolerate that by adding the padding for them. */ static PRStatus pl_base64_decode_flush(PLBase64Decoder *data) { int count; /* * If no remaining characters, or all are padding (also not well-formed * input, but again, be tolerant), then nothing more to do. (And, that * is considered successful.) */ if (data->token_size == 0 || data->token[0] == B64_PAD) return PR_SUCCESS; /* * Assume we have all the interesting input except for some expected * padding characters. Add them and decode the resulting token. */ while (data->token_size < 4) data->token[data->token_size++] = B64_PAD; data->token_size = 0; /* so a subsequent flush call is a no-op */ count = pl_base64_decode_token(data->token, data->output_buffer + data->output_length); if (count < 0) return PR_FAILURE; /* * If there is an output function, call it with this last bit of data. * Otherwise we are doing all buffered output, and the decoded bytes * are now there, we just need to reflect that in the length. */ if (data->output_fn != NULL) { PRInt32 output_result; PR_ASSERT(data->output_length == 0); output_result = data->output_fn(data->output_arg, data->output_buffer, (PRInt32)count); if (output_result < 0) return PR_FAILURE; } else { data->output_length += count; } return PR_SUCCESS; } /* * The maximum space needed to hold the output of the decoder given * input data of length "size". */ static PRUint32 PL_Base64MaxDecodedLength(PRUint32 size) { return size * 0.75; } /* * A distinct internal creation function for the buffer version to use. * (It does not want to specify an output_fn, and we want the normal * Create function to require that.) If more common initialization * of the decoding context needs to be done, it should be done *here*. */ static PLBase64Decoder * pl_base64_create_decoder(void) { return PR_NEWZAP(PLBase64Decoder); } /* * Function to start a base64 decoding context. * An "output_fn" is required; the "output_arg" parameter to that is optional. */ static PLBase64Decoder * PL_CreateBase64Decoder(PRInt32 (*output_fn)(void *, const unsigned char *, PRInt32), void *output_arg) { PLBase64Decoder *data; if (output_fn == NULL) { PR_SetError(PR_INVALID_ARGUMENT_ERROR, 0); return NULL; } data = pl_base64_create_decoder(); if (data != NULL) { data->output_fn = output_fn; data->output_arg = output_arg; } return data; } /* * Push data through the decoder, causing the output_fn (provided to Create) * to be called with the decoded data. */ static PRStatus PL_UpdateBase64Decoder(PLBase64Decoder *data, const char *buffer, PRUint32 size) { PRUint32 need_length; PRStatus status; /* XXX Should we do argument checking only in debug build? */ if (data == NULL || buffer == NULL || size == 0) { PR_SetError(PR_INVALID_ARGUMENT_ERROR, 0); return PR_FAILURE; } /* * How much space could this update need for decoding? */ need_length = PL_Base64MaxDecodedLength(size + data->token_size); /* * Make sure we have at least that much. If not, (re-)allocate. */ if (need_length > data->output_buflen) { unsigned char *output_buffer = data->output_buffer; if (output_buffer != NULL) output_buffer = (unsigned char *)PR_Realloc(output_buffer, need_length); else output_buffer = (unsigned char *)PR_Malloc(need_length); if (output_buffer == NULL) return PR_FAILURE; data->output_buffer = output_buffer; data->output_buflen = need_length; } /* There should not have been any leftover output data in the buffer. */ PR_ASSERT(data->output_length == 0); data->output_length = 0; status = pl_base64_decode_buffer(data, (const unsigned char *)buffer, size); /* Now that we have some decoded data, write it. */ if (status == PR_SUCCESS && data->output_length > 0) { PRInt32 output_result; PR_ASSERT(data->output_fn != NULL); output_result = data->output_fn(data->output_arg, data->output_buffer, (PRInt32)data->output_length); if (output_result < 0) status = PR_FAILURE; } data->output_length = 0; return status; } /* * When you're done decoding, call this to free the data. If "abort_p" * is false, then calling this may cause the output_fn to be called * one last time (as the last buffered data is flushed out). */ static PRStatus PL_DestroyBase64Decoder(PLBase64Decoder *data, PRBool abort_p) { PRStatus status = PR_SUCCESS; /* XXX Should we do argument checking only in debug build? */ if (data == NULL) { PR_SetError(PR_INVALID_ARGUMENT_ERROR, 0); return PR_FAILURE; } /* Flush out the last few buffered characters. */ if (!abort_p) status = pl_base64_decode_flush(data); if (data->output_buffer != NULL) PR_Free(data->output_buffer); PR_Free(data); return status; } /* * Perform base64 decoding from an input buffer to an output buffer. * The output buffer can be provided (as "dest"); you can also pass in * a NULL and this function will allocate a buffer large enough for you, * and return it. If you do provide the output buffer, you must also * provide the maximum length of that buffer (as "maxdestlen"). * The actual decoded length of output will be returned to you in * "output_destlen". * * Return value is NULL on error, the output buffer (allocated or provided) * otherwise. */ static unsigned char * PL_Base64DecodeBuffer(const char *src, PRUint32 srclen, unsigned char *dest, PRUint32 maxdestlen, PRUint32 *output_destlen) { PRUint32 need_length; unsigned char *output_buffer = NULL; PLBase64Decoder *data = NULL; PRStatus status; PR_ASSERT(srclen > 0); if (srclen == 0) { PR_SetError(PR_INVALID_ARGUMENT_ERROR, 0); return NULL; } /* * How much space could we possibly need for decoding this input? */ need_length = PL_Base64MaxDecodedLength(srclen); /* * Make sure we have at least that much, if output buffer provided. * If no output buffer provided, then we allocate that much. */ if (dest != NULL) { PR_ASSERT(maxdestlen >= need_length); if (maxdestlen < need_length) { PR_SetError(PR_BUFFER_OVERFLOW_ERROR, 0); goto loser; } output_buffer = dest; } else { output_buffer = (unsigned char *)PR_Malloc(need_length); if (output_buffer == NULL) goto loser; maxdestlen = need_length; } data = pl_base64_create_decoder(); if (data == NULL) goto loser; data->output_buflen = maxdestlen; data->output_buffer = output_buffer; status = pl_base64_decode_buffer(data, (const unsigned char *)src, srclen); /* * We do not wait for Destroy to flush, because Destroy will also * get rid of our decoder context, which we need to look at first! */ if (status == PR_SUCCESS) status = pl_base64_decode_flush(data); /* Must clear this or Destroy will free it. */ data->output_buffer = NULL; if (status == PR_SUCCESS) { *output_destlen = data->output_length; status = PL_DestroyBase64Decoder(data, PR_FALSE); data = NULL; if (status == PR_FAILURE) goto loser; return output_buffer; } loser: if (dest == NULL && output_buffer != NULL) PR_Free(output_buffer); if (data != NULL) (void)PL_DestroyBase64Decoder(data, PR_TRUE); return NULL; } /* * XXX End of base64 decoding code to be moved into NSPR. ******************************************************** */ /* * This is the beginning of the NSS cover functions. These will * provide the interface we want to expose as NSS-ish. For example, * they will operate on our Items, do any special handling or checking * we want to do, etc. */ PR_BEGIN_EXTERN_C /* * A boring cover structure for now. Perhaps someday it will include * some more interesting fields. */ struct NSSBase64DecoderStr { PLBase64Decoder *pl_data; }; PR_END_EXTERN_C /* * Function to start a base64 decoding context. */ NSSBase64Decoder * NSSBase64Decoder_Create(PRInt32 (*output_fn)(void *, const unsigned char *, PRInt32), void *output_arg) { PLBase64Decoder *pl_data; NSSBase64Decoder *nss_data; nss_data = PORT_ZNew(NSSBase64Decoder); if (nss_data == NULL) return NULL; pl_data = PL_CreateBase64Decoder(output_fn, output_arg); if (pl_data == NULL) { PORT_Free(nss_data); return NULL; } nss_data->pl_data = pl_data; return nss_data; } /* * Push data through the decoder, causing the output_fn (provided to Create) * to be called with the decoded data. */ SECStatus NSSBase64Decoder_Update(NSSBase64Decoder *data, const char *buffer, PRUint32 size) { PRStatus pr_status; /* XXX Should we do argument checking only in debug build? */ if (data == NULL) { PORT_SetError(SEC_ERROR_INVALID_ARGS); return SECFailure; } pr_status = PL_UpdateBase64Decoder(data->pl_data, buffer, size); if (pr_status == PR_FAILURE) return SECFailure; return SECSuccess; } /* * When you're done decoding, call this to free the data. If "abort_p" * is false, then calling this may cause the output_fn to be called * one last time (as the last buffered data is flushed out). */ SECStatus NSSBase64Decoder_Destroy(NSSBase64Decoder *data, PRBool abort_p) { PRStatus pr_status; /* XXX Should we do argument checking only in debug build? */ if (data == NULL) { PORT_SetError(SEC_ERROR_INVALID_ARGS); return SECFailure; } pr_status = PL_DestroyBase64Decoder(data->pl_data, abort_p); PORT_Free(data); if (pr_status == PR_FAILURE) return SECFailure; return SECSuccess; } /* * Perform base64 decoding from an ascii string "inStr" to an Item. * The length of the input must be provided as "inLen". The Item * may be provided (as "outItemOpt"); you can also pass in a NULL * and the Item will be allocated for you. * * In any case, the data within the Item will be allocated for you. * All allocation will happen out of the passed-in "arenaOpt", if non-NULL. * If "arenaOpt" is NULL, standard allocation (heap) will be used and * you will want to free the result via SECITEM_FreeItem. * * Return value is NULL on error, the Item (allocated or provided) otherwise. */ SECItem * NSSBase64_DecodeBuffer(PLArenaPool *arenaOpt, SECItem *outItemOpt, const char *inStr, unsigned int inLen) { SECItem *out_item = NULL; PRUint32 max_out_len = 0; void *mark = NULL; unsigned char *dummy = NULL; if ((outItemOpt != NULL && outItemOpt->data != NULL) || inLen == 0) { PORT_SetError(SEC_ERROR_INVALID_ARGS); return NULL; } if (arenaOpt != NULL) mark = PORT_ArenaMark(arenaOpt); max_out_len = PL_Base64MaxDecodedLength(inLen); if (max_out_len == 0) { goto loser; } out_item = SECITEM_AllocItem(arenaOpt, outItemOpt, max_out_len); if (out_item == NULL) { goto loser; } dummy = PL_Base64DecodeBuffer(inStr, inLen, out_item->data, max_out_len, &out_item->len); if (dummy == NULL) { goto loser; } if (arenaOpt != NULL) { PORT_ArenaUnmark(arenaOpt, mark); } return out_item; loser: if (arenaOpt != NULL) { PORT_ArenaRelease(arenaOpt, mark); if (outItemOpt != NULL) { outItemOpt->data = NULL; outItemOpt->len = 0; } } else if (dummy == NULL) { SECITEM_FreeItem(out_item, (PRBool)(outItemOpt == NULL)); } return NULL; } /* * XXX Everything below is deprecated. If you add new stuff, put it * *above*, not below. */ /* * XXX The following "ATOB" functions are provided for backward compatibility * with current code. They should be considered strongly deprecated. * When we can convert all our code over to using the new NSSBase64Decoder_ * functions defined above, we should get rid of these altogether. (Remove * protoypes from base64.h as well -- actually, remove that file completely). * If someone thinks either of these functions provides such a very useful * interface (though, as shown, the same functionality can already be * obtained by calling NSSBase64_DecodeBuffer directly), fine -- but then * that API should be provided with a nice new NSSFoo name and using * appropriate types, etc. */ #include "base64.h" /* ** Return an PORT_Alloc'd string which is the base64 decoded version ** of the input string; set *lenp to the length of the returned data. */ unsigned char * ATOB_AsciiToData(const char *string, unsigned int *lenp) { SECItem binary_item, *dummy; binary_item.data = NULL; binary_item.len = 0; dummy = NSSBase64_DecodeBuffer(NULL, &binary_item, string, (PRUint32)PORT_Strlen(string)); if (dummy == NULL) return NULL; PORT_Assert(dummy == &binary_item); *lenp = dummy->len; return dummy->data; } /* ** Convert from ascii to binary encoding of an item. */ SECStatus ATOB_ConvertAsciiToItem(SECItem *binary_item, const char *ascii) { SECItem *dummy; if (binary_item == NULL) { PORT_SetError(SEC_ERROR_INVALID_ARGS); return SECFailure; } /* * XXX Would prefer to assert here if data is non-null (actually, * don't need to, just let NSSBase64_DecodeBuffer do it), so as to * to catch unintended memory leaks, but callers are not clean in * this respect so we need to explicitly clear here to avoid the * assert in NSSBase64_DecodeBuffer. */ binary_item->data = NULL; binary_item->len = 0; dummy = NSSBase64_DecodeBuffer(NULL, binary_item, ascii, (PRUint32)PORT_Strlen(ascii)); if (dummy == NULL) return SECFailure; return SECSuccess; }