7 files changed, 619 insertions, 0 deletions

diff --git a/inc/dnbd3/shared/crc32.h b/inc/dnbd3/shared/crc32.h
new file mode 100644
index 0000000..00b8bdd
--- /dev/null
+++ b/inc/dnbd3/shared/crc32.h
@@ -0,0 +1,9 @@
+#ifndef _CRC32_H_
+#define _CRC32_H_
+
+#include <stdint.h>
+
+uint32_t crc32(uint32_t crc, const uint8_t *buf, size_t len);
+
+#endif
+
diff --git a/inc/dnbd3/shared/fdsignal.h b/inc/dnbd3/shared/fdsignal.h
new file mode 100644
index 0000000..960a2a9
--- /dev/null
+++ b/inc/dnbd3/shared/fdsignal.h
@@ -0,0 +1,57 @@
+#ifndef _FD_SIGNAL_H_
+#define _FD_SIGNAL_H_
+
+#define SIGNAL_OK (0)
+#define SIGNAL_TIMEOUT (-2)
+#define SIGNAL_ERROR (-1)
+
+typedef struct _dnbd3_signal dnbd3_signal_t;
+
+/**
+ * Create a new signal, nonblocking.
+ * @return NULL on error, pointer to dnbd3_signal_t on success.
+ */
+dnbd3_signal_t* signal_new();
+
+/**
+ * Create a new signal, blocking.
+ * @return NULL on error, pointer to dnbd3_signal_t on success.
+ */
+dnbd3_signal_t* signal_newBlocking();
+
+/**
+ * Trigger the given signal, so a wait or clear call will succeed.
+ * @return SIGNAL_OK on success, SIGNAL_ERROR on error
+ */
+int signal_call(const dnbd3_signal_t* const signal);
+
+/**
+ * Wait for given signal, with an optional timeout.
+ * If timeout == 0, just poll once.
+ * If timeout < 0, wait forever.
+ * @return > 0 telling how many times the signal was called,
+ *    SIGNAL_TIMEOUT if the timeout was reached,
+ *    SIGNAL_ERROR if some error occured
+ */
+int signal_wait(const dnbd3_signal_t* const signal, int timeoutMs);
+
+/**
+ * Clears any pending signals on this signal.
+ * @return number of signals that were pending,
+ *    SIGNAL_ERROR if some error occured
+ */
+int signal_clear(const dnbd3_signal_t* const signal);
+
+/**
+ * Close the given signal.
+ */
+void signal_close(const dnbd3_signal_t* const signal);
+
+/**
+ * Get a file descriptor for the given signal that can be
+ * waited on using poll or similar.
+ * @return -1 if the signal is invalid
+ */
+int signal_getWaitFd(const dnbd3_signal_t* const signal);
+
+#endif
diff --git a/inc/dnbd3/shared/log.h b/inc/dnbd3/shared/log.h
new file mode 100644
index 0000000..2a15f1d
--- /dev/null
+++ b/inc/dnbd3/shared/log.h
@@ -0,0 +1,80 @@
+/*
+ * This file is part of the Distributed Network Block Device 3
+ *
+ * Copyright(c) 2011-2012 Simon Rettberg
+ *
+ * This file may be licensed under the terms of the
+ * GNU General Public License Version 2 (the ``GPL'').
+ *
+ * Software distributed under the License is distributed
+ * on an ``AS IS'' basis, WITHOUT WARRANTY OF ANY KIND, either
+ * express or implied. See the GPL for the specific language
+ * governing rights and limitations.
+ *
+ * You should have received a copy of the GPL along with this
+ * program. If not, go to http://www.gnu.org/licenses/gpl.html
+ * or write to the Free Software Foundation, Inc.,
+ * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301, USA.
+ *
+ */
+
+#ifndef LOG_H_
+#define LOG_H_
+
+#include <stdbool.h>
+#include <unistd.h>
+#include <stdio.h>
+
+typedef unsigned int logmask_t;
+#define LOG_ERROR    ((logmask_t)1)  // Fatal error, server will terminate
+#define LOG_WARNING  ((logmask_t)2)  // Major issue, something is broken but keep running
+#define LOG_MINOR    ((logmask_t)4)  // Minor issue, more of a hickup than serious problem
+#define LOG_INFO     ((logmask_t)8)  // Informational message
+#define LOG_DEBUG1  ((logmask_t)16)  // Debug information, use this for non-spammy stuff
+#define LOG_DEBUG2  ((logmask_t)32)  // Use this for debug messages that will show up a lot
+
+/**
+ * Initialize the logging (constructor)
+ */
+void log_init(void);
+
+/**
+ * Check if cansoleMask | fileMask has all of mask set.
+ */
+bool log_hasMask(const logmask_t mask);
+
+void log_setFileMask(logmask_t mask);
+
+void log_setConsoleMask(logmask_t mask);
+
+void log_setConsoleTimestamps(bool on);
+
+/**
+ * Set console output stream
+ * The output stream can be either stdout or stderr
+ *
+ * Note: A call of this function is optional and only required if the output
+ *       stream should be changed from stdout to stderr since the log
+ *       implementation defaults to the output stream stdout
+ */
+int log_setConsoleOutputStream(FILE *outputStream);
+
+/**
+ * Open or reopen the log file. If path is NULL and the
+ * function was called with a path before, the same path
+ * will be used again.
+ */
+bool log_openLogFile(const char *path);
+
+/**
+ * Add a line to the log
+ */
+void logadd(const logmask_t mask, const char *text, ...)
+	__attribute__ ((format (printf, 2, 3)));
+
+/**
+ * Return last size bytes of log.
+ */
+ssize_t log_fetch(char *buffer, int size);
+
+#endif /* LOG_H_ */
diff --git a/inc/dnbd3/shared/protocol.h b/inc/dnbd3/shared/protocol.h
new file mode 100644
index 0000000..1dd47f8
--- /dev/null
+++ b/inc/dnbd3/shared/protocol.h
@@ -0,0 +1,156 @@
+#ifndef _PROTOCOL_H_
+#define _PROTOCOL_H_
+
+#include <dnbd3/types.h>
+#include <dnbd3/shared/serialize.h>
+#include <dnbd3/shared/sockhelper.h>
+
+#include <errno.h>
+#include <sys/types.h>
+#include <sys/socket.h>
+#include <sys/uio.h>
+
+// Client tells server that it is another server
+#define FLAGS8_SERVER (1)
+// Client (which is a proxy) tells server that it has background-replication enabled
+#define FLAGS8_BG_REP (2)
+
+// 2017-10-16: We now support hop-counting, macro to pass hop count conditinally to a function
+#define COND_HOPCOUNT(vers,hopcount) ( (vers) >= 3 ? (hopcount) : 0 )
+
+// 2017-11-02: Macro to set flags in select image message properly if we're a server, as BG_REP depends on global var
+#define SI_SERVER_FLAGS ( (uint8_t)( (_pretendClient ? 0 : FLAGS8_SERVER) | (_backgroundReplication == BGR_FULL ? FLAGS8_BG_REP : 0) ) )
+
+#define REPLY_OK (0)
+#define REPLY_ERRNO (-1)
+#define REPLY_AGAIN (-2)
+#define REPLY_INTR (-3)
+#define REPLY_CLOSED (-4)
+#define REPLY_INCOMPLETE (-5)
+#define REPLY_WRONGMAGIC (-6)
+
+static inline int dnbd3_read_reply(int sock, dnbd3_reply_t *reply, bool wait)
+{
+	ssize_t ret = recv( sock, reply, sizeof(*reply), (wait ? MSG_WAITALL : MSG_DONTWAIT) | MSG_NOSIGNAL );
+	if ( ret == 0 ) return REPLY_CLOSED;
+	if ( ret < 0 ) {
+		if ( errno == EAGAIN || errno == EWOULDBLOCK ) return REPLY_AGAIN;
+		if ( errno == EINTR ) return REPLY_INTR;
+		return REPLY_ERRNO;
+	}
+	if ( !wait && ret != sizeof(*reply) ) ret += recv( sock, ((char*)reply) + ret, sizeof(*reply) - ret, MSG_WAITALL | MSG_NOSIGNAL );
+	if ( ret != sizeof(*reply) ) return REPLY_INCOMPLETE;
+	fixup_reply( *reply );
+	if ( reply->magic != dnbd3_packet_magic ) return REPLY_WRONGMAGIC;
+	return REPLY_OK;
+}
+
+static inline bool dnbd3_get_reply(int sock, dnbd3_reply_t *reply)
+{
+	int ret;
+	do {
+		ret = dnbd3_read_reply( sock, reply, true );
+	} while ( ret == REPLY_INTR );
+	return ret == REPLY_OK;
+}
+
+static inline bool dnbd3_select_image(int sock, const char *name, uint16_t rid, uint8_t flags8)
+{
+	serialized_buffer_t serialized;
+	dnbd3_request_t request;
+	struct iovec iov[2];
+	serializer_reset_write( &serialized );
+	serializer_put_uint16( &serialized, PROTOCOL_VERSION );
+	serializer_put_string( &serialized, name );
+	serializer_put_uint16( &serialized, rid );
+	serializer_put_uint8( &serialized, flags8 );
+	const ssize_t len = serializer_get_written_length( &serialized );
+	request.magic = dnbd3_packet_magic;
+	request.cmd = CMD_SELECT_IMAGE;
+	request.size = (uint32_t)len;
+	request.handle = 0;
+	request.offset = 0;
+	fixup_request( request );
+	iov[0].iov_base = &request;
+	iov[0].iov_len = sizeof(request);
+	iov[1].iov_base = &serialized;
+	iov[1].iov_len = len;
+	ssize_t ret;
+	do {
+		ret = writev( sock, iov, 2 );
+	} while ( ret == -1 && errno == EINTR );
+	return ret == len + (ssize_t)sizeof(request);
+}
+
+static inline bool dnbd3_get_block(int sock, uint64_t offset, uint32_t size, uint64_t handle, uint8_t hopCount)
+{
+	dnbd3_request_t request;
+	request.magic = dnbd3_packet_magic;
+	request.handle = handle;
+	request.cmd = CMD_GET_BLOCK;
+	// When writing before "fixup", we can get away with assigning to offset instead of offset_small if we
+	// do it before assigning to .hops. Faster on 64bit machines (so, on everything)
+	request.offset = offset;
+	request.hops = hopCount;
+	request.size = size;
+	fixup_request( request );
+	return sock_sendAll( sock, &request, sizeof(request), 2 ) == (ssize_t)sizeof(request);
+}
+
+static inline bool dnbd3_get_crc32(int sock, uint32_t *master, void *buffer, size_t *bufferLen)
+{
+	dnbd3_request_t request;
+	dnbd3_reply_t reply;
+	request.magic = dnbd3_packet_magic;
+	request.handle = 0;
+	request.cmd = CMD_GET_CRC32;
+	request.offset = 0;
+	request.size = 0;
+	fixup_request( request );
+	if ( sock_sendAll( sock, &request, sizeof(request), 2 ) != (ssize_t)sizeof(request) ) return false;
+	if ( !dnbd3_get_reply( sock, &reply ) ) return false;
+	if ( reply.size == 0 ) {
+		*bufferLen = 0;
+		return true;
+	}
+	if ( reply.size < 4 ) return false;
+	reply.size -= 4;
+	if ( reply.cmd != CMD_GET_CRC32 || reply.size > *bufferLen ) return false;
+	*bufferLen = reply.size;
+	if ( sock_recv( sock, master, sizeof(uint32_t) ) != (ssize_t)sizeof(uint32_t) ) return false;
+	return sock_recv( sock, buffer, reply.size ) == (ssize_t)reply.size;
+}
+
+/**
+ * Pass a full serialized_buffer_t and a socket fd. Parsed data will be returned in further arguments.
+ * Note that all strings will point into the passed buffer, so there's no need to free them.
+ * This function will also read the header for you, as this message can only occur during connection,
+ * where no unrequested messages could arrive inbetween.
+ */
+static inline bool dnbd3_select_image_reply(serialized_buffer_t *buffer, int sock, uint16_t *protocol_version, char **name, uint16_t *rid,
+		uint64_t *imageSize)
+{
+	errno = 0;
+	dnbd3_reply_t reply;
+	if ( !dnbd3_get_reply( sock, &reply ) ) {
+		return false;
+	}
+	errno = 0;
+	if ( reply.cmd != CMD_SELECT_IMAGE || reply.size < 3 || reply.size > MAX_PAYLOAD ) {
+		return false;
+	}
+	// receive reply payload
+	ssize_t ret = sock_recv( sock, buffer, reply.size );
+	if ( ret != (ssize_t)reply.size ) {
+		return false;
+	}
+	// handle/check reply payload
+	serializer_reset_read( buffer, reply.size );
+	*protocol_version = serializer_get_uint16( buffer );
+	*name = serializer_get_string( buffer );
+	*rid = serializer_get_uint16( buffer );
+	*imageSize = serializer_get_uint64( buffer );
+	return true;
+}
+
+#endif
diff --git a/inc/dnbd3/shared/serialize.h b/inc/dnbd3/shared/serialize.h
new file mode 100644
index 0000000..b808fd0
--- /dev/null
+++ b/inc/dnbd3/shared/serialize.h
@@ -0,0 +1,39 @@
+#ifndef SERIALIZER_H_
+#define SERIALIZER_H_
+
+#include <dnbd3/config.h>
+#include <dnbd3/types.h>
+
+typedef struct {
+	char buffer[MAX_PAYLOAD]; // This MUST be the first member or send_reply() will blow up
+	char *buffer_end;
+	char *buffer_pointer;
+} serialized_buffer_t;
+
+void serializer_reset_read(serialized_buffer_t *buffer, size_t data_len);
+
+void serializer_reset_write(serialized_buffer_t *buffer);
+
+uint32_t serializer_get_written_length(serialized_buffer_t *buffer);
+
+//
+
+uint8_t serializer_get_uint8(serialized_buffer_t *buffer);
+
+uint16_t serializer_get_uint16(serialized_buffer_t *buffer);
+
+uint64_t serializer_get_uint64(serialized_buffer_t *buffer);
+
+char *serializer_get_string(serialized_buffer_t *buffer);
+
+//
+
+void serializer_put_uint8(serialized_buffer_t *buffer, uint8_t value);
+
+void serializer_put_uint16(serialized_buffer_t *buffer, uint16_t value);
+
+void serializer_put_uint64(serialized_buffer_t *buffer, uint64_t value);
+
+void serializer_put_string(serialized_buffer_t *buffer, const char *value);
+
+#endif
diff --git a/inc/dnbd3/shared/sockhelper.h b/inc/dnbd3/shared/sockhelper.h
new file mode 100644
index 0000000..5c7d903
--- /dev/null
+++ b/inc/dnbd3/shared/sockhelper.h
@@ -0,0 +1,120 @@
+#ifndef SOCKHELPER_H_
+#define SOCKHELPER_H_
+
+/*
+ * Helper functions for dealing with sockets. These functions should
+ * abstract from the IP version by using getaddrinfo() and thelike.
+ */
+
+#include <dnbd3/types.h>
+#include <stdint.h>
+#include <sys/socket.h>
+#include <string.h>
+
+typedef struct _poll_list poll_list_t;
+
+/**
+ * Connect to given dnbd3_host_t.
+ * @param addr - address of host to connect to
+ * @param connect_ms - timeout in milliseconds after which the connection attempt fails
+ * @param rw_ms - read/write timeout in milliseconds to apply on successful connect
+ * @return socket file descriptor, or -1 on error
+ */
+int sock_connect(const dnbd3_host_t * const addr, const int connect_ms, const int rw_ms);
+
+/**
+ * Resolve/parse given address and put the result(s) into passed dnbd3_host_t array,
+ * but only up to count entries.
+ * @return Number of items added to array
+ */
+int sock_resolveToDnbd3Host(const char * const address, dnbd3_host_t * const dest, const int count);
+
+bool sock_sockaddrToDnbd3(struct sockaddr* sa, dnbd3_host_t *host);
+
+void sock_setTimeout(const int sockfd, const int milliseconds);
+
+size_t sock_printHost(const dnbd3_host_t * const host, char *output, const size_t len);
+
+size_t sock_printable(const struct sockaddr * const addr, const socklen_t addrLen, char *output, const size_t len);
+
+/**
+ * Create new poll list.
+ */
+poll_list_t* sock_newPollList();
+
+/**
+ * Delete a poll list, closing all sockets first if necessary.
+ */
+void sock_destroyPollList(poll_list_t *list);
+
+/**
+ * Listen on all interfaces/available IP addresses, using the given protocol.
+ * IPv4 and IPv6 are supported.
+ * @param protocol_family PF_INET or PF_INET6
+ * @param port port to listen on
+ * @return true if any listen call was successful
+ */
+bool sock_listenAny(poll_list_t* list, uint16_t port);
+
+/**
+ * Listen on a specific address and port.
+ * @param bind_addr human readable address to bind to for listening
+ * @param port to listen on
+ */
+bool sock_listen(poll_list_t* list, char* bind_addr, uint16_t port);
+
+/**
+ * Asynchroneously connect to multiple hosts.
+ * This can be called multiple times with varying timeouts. Calling it
+ * the first time on an empty list is identical to sock_connect(). On
+ * consecutive calls, more nonblocking sockets in connecting state will
+ * be added to the list, and on each of these calls, all the pending
+ * sockets will be checked for successful connection (or error), respecting
+ * the passed timeout.
+ * host can be NULL to just wait on the sockets already in the list.
+ * If at least one socket completed the connection
+ * within the given timeout, it will be removed from the list and
+ * returned. On error or timeout, -1 is returned. If there are no more sockets
+ * in the list, -2 is returned.
+ */
+int sock_multiConnect(poll_list_t* list, const dnbd3_host_t* host, int connect_ms, int rw_ms);
+
+/**
+ * This is a multi-socket version of accept. Pass in an array of listening sockets.
+ * If any of the sockets has an incoming connection, accept it and return the new socket's fd.
+ * On error, return -1, just like accept().
+ * @param sockets array of listening socket fds
+ * @param socket_count number of sockets in that array
+ * @return fd of new client socket, -1 on error
+ */
+int sock_accept(poll_list_t *list, struct sockaddr_storage *addr, socklen_t *length_ptr);
+
+void sock_set_nonblock(int sock);
+
+void sock_set_block(int sock);
+
+/**
+ * Add given socket to array. Take an existing empty slot ( == -1) if available,
+ * append to end otherwise. Updates socket count variable passed by reference.
+ *
+ * @param poll_list_t list the poll list to add the socket to
+ * @param sock socket fd to add
+ * @param wantRead whether to set the EPOLLIN flag
+ * @param wantWrite whether to set the EPOLLOUT flag
+ * @return true on success, false iff the array is already full or socket is < 0
+ */
+bool sock_append(poll_list_t *list, const int sock, bool wantRead, bool wantWrite);
+
+/**
+ * Send the whole buffer, calling write() multiple times if neccessary.
+ * Give up after calling write() maxtries times.
+ * Set maxtries < 0 to try infinitely.
+ */
+ssize_t sock_sendAll(const int sock, const void *buffer, const size_t len, int maxtries);
+
+/**
+ * Send given buffer, repeatedly calling recv on partial send or EINTR.
+ */
+ssize_t sock_recv(const int sock, void *buffer, const size_t len);
+
+#endif /* SOCKHELPER_H_ */
diff --git a/inc/dnbd3/shared/timing.h b/inc/dnbd3/shared/timing.h
new file mode 100644
index 0000000..2530416
--- /dev/null
+++ b/inc/dnbd3/shared/timing.h
@@ -0,0 +1,158 @@
+#ifndef _D_TIMING_H
+#define _D_TIMING_H
+
+#include <time.h>
+#include <stdint.h>
+#include <stdbool.h>
+
+#ifdef CLOCK_MONOTONIC_RAW
+#define BEST_CLOCK_SOURCE CLOCK_MONOTONIC_RAW
+#else
+#define BEST_CLOCK_SOURCE CLOCK_MONOTONIC
+#endif
+
+typedef struct timespec ticks;
+
+extern struct timespec basetime;
+
+/**
+ * Assign src to dst while adding secs seconds.
+ */
+#define timing_set(dst,src,secs) do { (dst)->tv_sec = (src)->tv_sec + (secs); (dst)->tv_nsec = (src)->tv_nsec; } while (0)
+
+/**
+ * Define variable now, initialize to timing_get.
+ */
+#define declare_now ticks now; timing_get( &now )
+
+/**
+ * Call this once to calibrate on startup.
+ * Although overflows of CLOCK_MONOTONIC(_RAW) should
+ * by definition never happen, we still have a fixed size
+ * int that could at some point. By forcing the counter
+ * to start at 0 on startup the point of overflow
+ * will be very far in the future (decades for 32bit time_t,
+ * end of universe for 64bit).
+ */
+void timing_setBase();
+
+/**
+ * Internal, do not use. Moved to another function
+ * to prevent inlining of error handling code, which
+ * should be very unlikely to ever trigger.
+ */
+_Noreturn void timing_abort();
+
+/**
+ * Get current time. Shortcut for clock_gettime with error check.
+ */
+static inline void timing_get(ticks* retval)
+{
+	if ( clock_gettime( BEST_CLOCK_SOURCE, retval ) == -1 ) timing_abort();
+	retval->tv_sec -= basetime.tv_sec;
+}
+
+/**
+ * Get a ticks instance somewhere in the future.
+ * Useful for timeouts.
+ */
+static inline void timing_gets(ticks* retval, int32_t addSeconds)
+{
+	timing_get( retval );
+	retval->tv_sec += addSeconds;
+}
+
+static inline void timing_addSeconds(ticks* retval, ticks* base, int32_t addSeconds)
+{
+	retval->tv_sec = base->tv_sec + addSeconds;
+	retval->tv_nsec = base->tv_nsec;
+}
+
+/**
+ * Check whether given timeout is reached.
+ * Might trigger up to one second early.
+ */
+static inline bool timing_reached(const ticks* timeout, const ticks* now)
+{
+	return now->tv_sec >= timeout->tv_sec;
+}
+#define timing_1le2(one,two) timing_reached(one,two)
+
+/**
+ * Precise check whether given timeout has been reached.
+ */
+static inline bool timing_reachedPrecise(const ticks* timeout, const ticks* now)
+{
+	return now->tv_sec > timeout->tv_sec
+			|| (now->tv_sec == timeout->tv_sec && now->tv_nsec > timeout->tv_nsec);
+}
+
+/**
+ * Shortcut for above. Useful if not used in loop.
+ * Might trigger up to one second early.
+ */
+static inline bool timing_isReached(const ticks* timeout)
+{
+	ticks now;
+	timing_get( &now );
+	return timing_reached( timeout, &now );
+}
+/**
+ * Shortcut for above. Useful if not used in loop.
+ */
+static inline bool timing_isReachedPrecise(const ticks* timeout)
+{
+	ticks now;
+	timing_get( &now );
+	return timing_reachedPrecise( timeout, &now );
+}
+
+
+/**
+ * Get difference between two ticks, rounded down to seconds.
+ * Make sure you pass the arguments in the proper order. If
+ * end is before start, 0 will always be returned.
+ */
+static inline uint32_t timing_diff(const ticks *start, const ticks *end)
+{
+	if ( end->tv_sec <= start->tv_sec ) return 0;
+	return (uint32_t)( ( end->tv_sec - start->tv_sec )
+			+ ( start->tv_nsec > end->tv_nsec ? -1 : 0 ) );
+}
+
+/**
+ * Get difference between two ticks, rounded down to milliseconds.
+ * Same as above; passing arguments in reverse will always return 0.
+ */
+static inline uint64_t timing_diffMs(const ticks *start, const ticks *end)
+{
+	if ( end->tv_sec < start->tv_sec ) return 0;
+	uint64_t diff = (uint64_t)( end->tv_sec - start->tv_sec ) * 1000;
+	if ( start->tv_nsec >= end->tv_nsec ) {
+		if ( diff == 0 ) return 0;
+		diff -= (start->tv_nsec - end->tv_nsec) / 1000000;
+	} else {
+		diff += (end->tv_nsec - start->tv_nsec) / 1000000;
+	}
+	return diff;
+}
+
+/**
+ * Get difference between two ticks, rounded down to microseconds.
+ * Same as above; passing arguments in reverse will always return 0.
+ */
+static inline uint64_t timing_diffUs(const ticks *start, const ticks *end)
+{
+	if ( end->tv_sec < start->tv_sec ) return 0;
+	uint64_t diff = (uint64_t)( end->tv_sec - start->tv_sec ) * 1000000;
+	if ( start->tv_nsec >= end->tv_nsec ) {
+		if ( diff == 0 ) return 0;
+		diff -= ( start->tv_nsec - end->tv_nsec ) / 1000;
+	} else {
+		diff += ( end->tv_nsec - start->tv_nsec ) / 1000;
+	}
+	return diff;
+}
+
+
+#endif