diff options
| -rw-r--r-- | Makefile | 2 | ||||
| -rw-r--r-- | src/filter.h | 2 | ||||
| -rw-r--r-- | src/message.c | 554 | ||||
| -rw-r--r-- | src/message.h | 132 | ||||
| -rw-r--r-- | src/output.h | 22 | ||||
| -rw-r--r-- | src/ramps.h | 1 | ||||
| -rw-r--r-- | src/util.c | 75 | ||||
| -rw-r--r-- | src/util.h | 13 |
8 files changed, 800 insertions, 1 deletions
@@ -3,7 +3,7 @@ COMMAND = gammad KERNEL = $(shell uname | tr '[A-Z]_' '[a-z]-') -SRC = filter gammad output ramps util +SRC = filter gammad output ramps util message OPTIMISE = -Og -g diff --git a/src/filter.h b/src/filter.h index f625125..9df24e1 100644 --- a/src/filter.h +++ b/src/filter.h @@ -41,6 +41,7 @@ enum lifespan * The filter should be removed now */ LIFESPAN_REMOVE + }; @@ -85,6 +86,7 @@ struct filter * `lifespan == LIFESPAN_REMOVE`. */ void* ramps; + }; diff --git a/src/message.c b/src/message.c new file mode 100644 index 0000000..c0372bd --- /dev/null +++ b/src/message.c @@ -0,0 +1,554 @@ +/** + * gammad -- Cooperative gamma server + * Copyright (C) 2016 Mattias Andrée (maandree@kth.se) + * + * This library is free software: you can redistribute it and/or modify + * it under the terms of the GNU General Public License as published by + * the Free Software Foundation, either version 3 of the License, or + * (at your option) any later version. + * + * This library is distributed in the hope that it will be useful, + * but WITHOUT ANY WARRANTY; without even the implied warranty of + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + * GNU General Public License for more details. + * + * You should have received a copy of the GNU General Public License + * along with this library. If not, see <http://www.gnu.org/licenses/>. + */ +#include "message.h" +#include "util.h" + +#include <sys/socket.h> +#include <errno.h> +#include <stdint.h> +#include <stdlib.h> +#include <string.h> + + + +/** + * Initialise a message slot so that it can + * be used by to read messages + * + * @param this Memory slot in which to store the new message + * @return Non-zero on error, `errno` will be set accordingly + */ +int message_initialise(struct message* this) +{ + this->headers = NULL; + this->header_count = 0; + this->payload = NULL; + this->payload_size = 0; + this->payload_ptr = 0; + this->buffer_size = 128; + this->buffer_ptr = 0; + this->stage = 0; + this->buffer = malloc(this->buffer_size); + if (this->buffer == NULL) + return -1; + return 0; +} + + +/** + * Release all resources in a message, should + * be done even if initialisation fails + * + * @param this The message + */ +void message_destroy(struct message* this) +{ + size_t i; + if (this->headers != NULL) + for (i = 0; i < this->header_count; i++) + free(this->headers[i]); + + free(this->headers); + free(this->payload); + free(this->buffer); +} + + +/** + * Marshal a message for state serialisation + * + * @param this The message + * @param buf Output buffer for the marshalled data, + * `NULL` just measure how large the buffers + * needs to be + * @return The number of marshalled byte + */ +size_t message_marshal(const struct message* this, void* buf) +{ + size_t i, n, off = 0; + char* bs = buf; + + if (bs != NULL) + *(size_t*)(bs + off) = this->header_count; + off += sizeof(size_t); + + if (bs != NULL) + *(size_t*)(bs + off) = this->payload_size; + off += sizeof(size_t); + + if (bs != NULL) + *(size_t*)(bs + off) = this->payload_ptr; + off += sizeof(size_t); + + if (bs != NULL) + *(size_t*)(bs + off) = this->buffer_ptr; + off += sizeof(size_t); + + if (bs != NULL) + *(int*)(bs + off) = this->stage; + off += sizeof(int); + + for (i = 0; i < this->header_count; i++) + { + n = strlen(this->headers[i]) + 1; + if (bs != NULL) + memcpy(bs + off, this->headers[i], n); + off += n; + } + + if (bs != NULL) + memcpy(bs + off, this->payload, this->payload_ptr); + off += this->payload_ptr; + + if (bs != NULL) + memcpy(bs + off, this->buffer, this->buffer_ptr); + off += this->buffer_ptr; + + return off; +} + + +/** + * Unmarshal a message for state deserialisation + * + * @param this Memory slot in which to store the new message + * @param buf In buffer with the marshalled data + * @return The number of unmarshalled bytes, 0 on error + */ +size_t message_unmarshal(struct message* this, const void* buf) +{ + size_t i, n, off = 0, header_count; + const char* bs = buf; + + this->header_count = 0; + + header_count = *(const size_t*)(bs + off); + off += sizeof(size_t); + + this->payload_size = *(const size_t*)(bs + off); + off += sizeof(size_t); + + this->payload_ptr = *(const size_t*)(bs + off); + off += sizeof(size_t); + + this->buffer_size = this->buffer_ptr = *(const size_t*)(bs + off); + off += sizeof(size_t); + + this->stage = *(const int*)(bs + off); + off += sizeof(int); + + /* Make sure that the pointers are NULL so that they are + not freed without being allocated when the message is + destroyed if this function fails. */ + this->headers = NULL; + this->payload = NULL; + this->buffer = NULL; + + /* To 2-power-multiple of 128 bytes. */ + this->buffer_size >>= 7; + if (this->buffer_size == 0) + this->buffer_size = 1; + else + { + this->buffer_size -= 1; + this->buffer_size |= this->buffer_size >> 1; + this->buffer_size |= this->buffer_size >> 2; + this->buffer_size |= this->buffer_size >> 4; + this->buffer_size |= this->buffer_size >> 8; + this->buffer_size |= this->buffer_size >> 16; +#if SIZE_MAX == UINT64_MAX + this->buffer_size |= this->buffer_size >> 32; +#endif + this->buffer_size += 1; + } + this->buffer_size <<= 7; + + /* Allocate header list, payload and read buffer. */ + + if (header_count > 0) + if (!(this->headers = malloc(header_count * sizeof(char*)))) + goto fail; + + if (this->payload_size > 0) + if (!(this->payload = malloc(this->payload_size))) + goto fail; + + if (!(this->buffer = malloc(this->buffer_size))) + goto fail; + + /* Fill the header list, payload and read buffer. */ + + for (i = 0; i < header_count; i++) + { + n = strlen(bs + off) + 1; + this->headers[i] = memdup(bs + off, n); + if (this->headers[i] == NULL) + goto fail; + off += n; + this->header_count++; + } + + memcpy(this->payload, bs + off, this->payload_ptr); + off += this->payload_ptr; + + memcpy(this->buffer, bs + off, this->buffer_ptr); + off += this->buffer_ptr; + + return off; + + fail: + return 0; +} + + +/** + * Extend the header list's allocation + * + * @param this The message + * @param extent The number of additional entries + * @return Zero on success, -1 on error + */ +static int extend_headers(struct message* this, size_t extent) +{ + char** new; + if (!(new = realloc(this->headers, (this->header_count + extent) * sizeof(char*)))) + return -1; + this->headers = new; + return 0; +} + + +/** + * Extend the read buffer by way of doubling + * + * @param this The message + * @return Zero on success, -1 on error + */ +static int extend_buffer(struct message* this) +{ + char* new; + if (!(new = realloc(this->buffer, (this->buffer_size << 1) * sizeof(char)))) + return -1; + this->buffer = new; + this->buffer_size <<= 1; + return 0; +} + + +/** + * Reset the header list and the payload + * + * @param this The message + */ +static void reset_message(struct message* this) +{ + size_t i; + if (this->headers != NULL) + for (i = 0; i < this->header_count; i++) + free(this->headers[i]); + free(this->headers); + this->headers = NULL; + this->header_count = 0; + + free(this->payload); + this->payload = NULL; + this->payload_size = 0; + this->payload_ptr = 0; +} + + +/** + * Read the headers the message and determine, and store, its payload's length + * + * @param this The message + * @return Zero on success, negative on error (malformated message: unrecoverable state) + */ +#if defined(__GNUC__) +__attribute__((pure)) +#endif +static int get_payload_length(struct message* this) +{ + char* header; + size_t i; + + for (i = 0; i < this->header_count; i++) + if (strstr(this->headers[i], "Length: ") == this->headers[i]) + { + /* Store the message length. */ + header = this->headers[i] + strlen("Length: "); + this->payload_size = (size_t)atol(header); + + /* Do not except a length that is not correctly formated. */ + for (; *header; header++) + if ((*header < '0') || ('9' < *header)) + return -2; /* Malformated value, enters unrecoverable state. */ + + /* Stop searching for the ‘Length’ header, we have found and parsed it. */ + break; + } + + return 0; +} + + +/** + * Verify that a header is correctly formatted + * + * @param header The header, must be NUL-terminated + * @param length The length of the header + * @return Zero if valid, negative if invalid (malformated message: unrecoverable state) + */ +#if defined(__GNUC__) +__attribute__((pure)) +#endif +static int validate_header(const char* header, size_t length) +{ + char* p = memchr(header, ':', length * sizeof(char)); + + if (verify_utf8(header, 0) < 0) + /* Either the string is not UTF-8, or your are under an UTF-8 attack, + lets just call this unrecoverable because the client will not correct. */ + return -2; + + if ((p == NULL) || /* Buck you, rawmemchr should not segfault the program. */ + (p[1] != ' ')) /* Also an invalid format. ' ' is mandated after the ':'. */ + return -2; + + return 0; +} + + +/** + * Remove the beginning of the read buffer + * + * @param this The message + * @param length The number of characters to remove + * @param update_ptr Whether to update the buffer pointer + */ +static void unbuffer_beginning(struct message* this, size_t length, int update_ptr) +{ + memmove(this->buffer, this->buffer + length, (this->buffer_ptr - length) * sizeof(char)); + if (update_ptr) + this->buffer_ptr -= length; +} + + +/** + * Remove the header–payload delimiter from the buffer, + * get the payload's size and allocate the payload + * + * @param this The message + * @return The return value follows the rules of `message_read` + */ +static int initialise_payload(struct message* this) +{ + /* Remove the \n (end of empty line) we found from the buffer. */ + unbuffer_beginning(this, 1, 1); + + /* Get the length of the payload. */ + if (get_payload_length(this) < 0) + return -2; /* Malformated value, enters unrecoverable state. */ + + /* Allocate the payload buffer. */ + if (this->payload_size > 0) + if (!(this->payload = malloc(this->payload_size))) + return -1; + + return 0; +} + + +/** + * Create a header from the buffer and store it + * + * @param this The message + * @param length The length of the header, including LF-termination + * @return The return value follows the rules of `message_read` + */ +static int store_header(struct message* this, size_t length) +{ + char* header; + + /* Allocate the header. */ + if (!(header = malloc(length))) /* Last char is a LF, which is substituted with NUL. */ + return -1; + /* Copy the header data into the allocated header, */ + memcpy(header, this->buffer, length * sizeof(char)); + /* and NUL-terminate it. */ + header[length - 1] = '\0'; + + /* Remove the header data from the read buffer. */ + unbuffer_beginning(this, length, 1); + + /* Make sure the the header syntax is correct so that + the program does not need to care about it. */ + if (validate_header(header, length)) + { + free(header); + return -2; + } + + /* Store the header in the header list. */ + this->headers[this->header_count++] = header; + + return 0; +} + + +/** + * Continue reading from the socket into the buffer + * + * @param this The message + * @param fd The file descriptor of the socket + * @return The return value follows the rules of `message_read` + */ +static int continue_read(struct message* this, int fd) +{ + size_t n; + ssize_t got; + int r; + + /* Figure out how much space we have left in the read buffer. */ + n = this->buffer_size - this->buffer_ptr; + + /* If we do not have too much left, */ + if (n < 128) + { + /* grow the buffer, */ + if ((r = extend_buffer(this)) < 0) + return r; + + /* and recalculate how much space we have left. */ + n = this->buffer_size - this->buffer_ptr; + } + + /* Then read from the socket. */ + errno = 0; + got = recv(fd, this->buffer + this->buffer_ptr, n, 0); + this->buffer_ptr += (size_t)(got < 0 ? 0 : got); + if (errno) + return -1; + if (got == 0) + { + errno = ECONNRESET; + return -1; + } + + return 0; +} + + +/** + * Read the next message from a file descriptor + * + * @param this Memory slot in which to store the new message + * @param fd The file descriptor + * @return Non-zero on error or interruption, `errno` will be + * set accordingly. Destroy the message on error, + * be aware that the reading could have been + * interrupted by a signal rather than canonical error. + * If -2 is returned `errno` will not have been set, + * -2 indicates that the message is malformated, + * which is a state that cannot be recovered from. + */ +int message_read(struct message* this, int fd) +{ + size_t header_commit_buffer = 0; + int r; + + /* If we are at stage 2, we are done and it is time to start over. + This is important because the function could have been interrupted. */ + if (this->stage == 2) + { + reset_message(this); + this->stage = 0; + } + + /* Read from file descriptor until we have a full message. */ + for (;;) + { + char* p; + size_t length; + + /* Stage 0: headers. */ + /* Read all headers that we have stored into the read buffer. */ + while ((this->stage == 0) && + ((p = memchr(this->buffer, '\n', this->buffer_ptr * sizeof(char))) != NULL)) + if ((length = (size_t)(p - this->buffer))) + { + /* We have found a header. */ + + /* On every eighth header found with this function call, + we prepare the header list for eight more headers so + that it does not need to be reallocated again and again. */ + if (header_commit_buffer == 0) + if ((r = extend_headers(this, header_commit_buffer = 8)) < 0) + return r; + + /* Create and store header. */ + if ((r = store_header(this, length + 1)) < 0) + return r; + header_commit_buffer -= 1; + } + else + { + /* We have found an empty line, i.e. the end of the headers. */ + + /* Remove the header–payload delimiter from the buffer, + get the payload's size and allocate the payload. */ + if ((r = initialise_payload(this)) < 0) + return r; + + /* Mark end of stage, next stage is getting the payload. */ + this->stage = 1; + } + + + /* Stage 1: payload. */ + if ((this->stage == 1) && (this->payload_size > 0)) + { + /* How much of the payload that has not yet been filled. */ + size_t need = this->payload_size - this->payload_ptr; + /* How much we have of that what is needed. */ + size_t move = this->buffer_ptr < need ? this->buffer_ptr : need; + + /* Copy what we have, and remove it from the the read buffer. */ + memcpy(this->payload + this->payload_ptr, this->buffer, move * sizeof(char)); + unbuffer_beginning(this, move, 1); + + /* Keep track of how much we have read. */ + this->payload_ptr += move; + } + if ((this->stage == 1) && (this->payload_ptr == this->payload_size)) + { + /* If we have filled the payload (or there was no payload), + mark the end of this stage, i.e. that the message is + complete, and return with success. */ + this->stage = 2; + return 0; + } + + + /* If stage 1 was not completed. */ + + /* Continue reading from the socket into the buffer. */ + if ((r = continue_read(this, fd)) < 0) + return r; + } +} + diff --git a/src/message.h b/src/message.h new file mode 100644 index 0000000..163d806 --- /dev/null +++ b/src/message.h @@ -0,0 +1,132 @@ +/** + * gammad -- Cooperative gamma server + * Copyright (C) 2016 Mattias Andrée (maandree@kth.se) + * + * This library is free software: you can redistribute it and/or modify + * it under the terms of the GNU General Public License as published by + * the Free Software Foundation, either version 3 of the License, or + * (at your option) any later version. + * + * This library is distributed in the hope that it will be useful, + * but WITHOUT ANY WARRANTY; without even the implied warranty of + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + * GNU General Public License for more details. + * + * You should have received a copy of the GNU General Public License + * along with this library. If not, see <http://www.gnu.org/licenses/>. + */ +#include <stddef.h> + + + +/** + * Message passed between a server and a client + */ +struct message +{ + /** + * The headers in the message, each element in this list + * as an unparsed header, it consists of both the header + * name and its associated value, joined by ": ". A header + * cannot be `NULL` (unless its memory allocation failed,) + * but `headers` itself is `NULL` if there are no headers. + * The "Length" header should be included in this list. + */ + char** headers; + + /** + * The number of headers in the message + */ + size_t header_count; + + /** + * The payload of the message, `NULL` if none (of zero-length) + */ + char* payload; + + /** + * The size of the payload + */ + size_t payload_size; + + /** + * How much of the payload that has been stored (internal data) + */ + size_t payload_ptr; + + /** + * Internal buffer for the reading function (internal data) + */ + char* buffer; + + /** + * The size allocated to `buffer` (internal data) + */ + size_t buffer_size; + + /** + * The number of bytes used in `buffer` (internal data) + */ + size_t buffer_ptr; + + /** + * 0 while reading headers, 1 while reading payload, and 2 when done (internal data) + */ + int stage; + +}; + + + +/** + * Initialise a message slot so that it can + * be used by to read messages + * + * @param this Memory slot in which to store the new message + * @return Non-zero on error, `errno` will be set accordingly + */ +int message_initialise(struct message* this); + +/** + * Release all resources in a message, should + * be done even if initialisation fails + * + * @param this The message + */ +void message_destroy(struct message* this); + +/** + * Marshal a message for state serialisation + * + * @param this The message + * @param buf Output buffer for the marshalled data, + * `NULL` just measure how large the buffers + * needs to be + * @return The number of marshalled byte + */ +size_t message_marshal(const struct message* this, void* buf); + +/** + * Unmarshal a message for state deserialisation + * + * @param this Memory slot in which to store the new message + * @param buf In buffer with the marshalled data + * @return The number of unmarshalled bytes, 0 on error + */ +size_t message_unmarshal(struct message* this, const void* buf); + +/** + * Read the next message from a file descriptor + * + * @param this Memory slot in which to store the new message + * @param fd The file descriptor + * @return Non-zero on error or interruption, `errno` will be + * set accordingly. Destroy the message on error, + * be aware that the reading could have been + * interrupted by a signal rather than canonical error. + * If -2 is returned `errno` will not have been set, + * -2 indicates that the message is malformated, + * which is a state that cannot be recovered from. + */ +int message_read(struct message* this, int fd); + diff --git a/src/output.h b/src/output.h index 3e47da7..aa2ab8b 100644 --- a/src/output.h +++ b/src/output.h @@ -103,10 +103,32 @@ struct output */ union gamma_ramps saved_ramps; + /** + * The table of all applied filters + */ struct filter* table_filters; + + /** + * `.table_sums[i]` is the resulting + * adjustment made when all filter + * from `.table_filters[0]` up to and + * including `.table_filters[i]` has + * been applied + */ union gamma_ramps* table_sums; + + /** + * The number of elements allocated + * for `.table_filters` and for `.table_sums` + */ size_t table_alloc; + + /** + * The number of elements stored in + * `.table_filters` and in `.table_sums` + */ size_t table_size; + }; diff --git a/src/ramps.h b/src/ramps.h index d01a22e..799ad0a 100644 --- a/src/ramps.h +++ b/src/ramps.h @@ -54,6 +54,7 @@ union gamma_ramps * Ramps with `double` value */ libgamma_gamma_rampsd_t d; + }; @@ -152,3 +152,78 @@ void msleep(int ms) nanosleep(&ts, NULL); } +/** + * Check whether a NUL-terminated string is encoded in UTF-8 + * + * @param string The string + * @param allow_modified_nul Whether Modified UTF-8 is allowed, which allows a two-byte encoding for NUL + * @return Zero if good, -1 on encoding error + */ +int verify_utf8(const char* string, int allow_modified_nul) +{ + static long BYTES_TO_MIN_BITS[] = {0, 0, 8, 12, 17, 22, 37}; + static long BYTES_TO_MAX_BITS[] = {0, 7, 11, 16, 21, 26, 31}; + long bytes = 0, read_bytes = 0, bits = 0, c, character; + + /* min bits max bits + 0....... 0 7 + 110..... 10...... 8 11 + 1110.... 10...... 10...... 12 16 + 11110... 10...... 10...... 10...... 17 21 + 111110.. 10...... 10...... 10...... 10...... 22 26 + 1111110. 10...... 10...... 10...... 10...... 10...... 27 31 + */ + + while ((c = (long)(*string++))) + if (read_bytes == 0) + { + /* First byte of the character. */ + + if ((c & 0x80) == 0x00) + /* Single-byte character. */ + continue; + + if ((c & 0xC0) == 0x80) + /* Single-byte character marked as multibyte, or + a non-first byte in a multibyte character. */ + return -1; + + /* Multibyte character. */ + while ((c & 0x80)) + bytes++, c <<= 1; + read_bytes = 1; + character = c & 0x7F; + if (bytes > 6) + /* 31-bit characters can be encoded with 6-bytes, + and UTF-8 does not cover higher code points. */ + return -1; + } + else + { + /* Not first byte of the character. */ + + if ((c & 0xC0) != 0x80) + /* Beginning of new character before a + multibyte character has ended. */ + return -1; + + character = (character << 6) | (c & 0x7F); + + if (++read_bytes < bytes) + /* Not at last byte yet. */ + continue; + + /* Check that the character is not unnecessarily long. */ + while (character) + character >>= 1, bits++; + bits = ((bits == 0) && (bytes == 2) && allow_modified_nul) ? 8 : bits; + if ((bits < BYTES_TO_MIN_BITS[bytes]) || (BYTES_TO_MAX_BITS[bytes] < bits)) + return -1; + + read_bytes = bytes = bits = 0; + } + + /* Make sure we did not stop at the middle of a multibyte character. */ + return read_bytes == 0 ? 0 : -1; +} + @@ -63,3 +63,16 @@ int dup2atleast(int fd, int atleast); */ void msleep(int ms); + +/** + * Check whether a NUL-terminated string is encoded in UTF-8 + * + * @param string The string + * @param allow_modified_nul Whether Modified UTF-8 is allowed, which allows a two-byte encoding for NUL + * @return Zero if good, -1 on encoding error + */ +#if defined(__GNUC__) +__attribute__((pure)) +#endif +int verify_utf8(const char* string, int allow_modified_nul); + |