/** * lsyncd.c Live (Mirror) Syncing Demon * * License: GPLv2 (see COPYING) or any later version * * Authors: Axel Kittenberger * Eugene Sanivsky */ #include "config.h" #include #include #include #ifdef HAVE_SYS_INOTIFY_H # include #else # include "inotify-nosys.h" #endif #include #include #include #include #include #include #include #include #include #include #include #define INOTIFY_BUF_LEN (512 * (sizeof(struct inotify_event) + 16)) #define LOG_DEBUG 1 #define LOG_NORMAL 2 #define LOG_ERROR 3 #define VER_MAJOR 1 #define VER_MINOR 0 int loglevel = LOG_NORMAL; /** * Option: if true rsync will not be actually called. */ int flag_dryrun = 0; /** * Option: if true, do not detach and log to stdout/stderr. */ int flag_nodaemon = 0; /** * Option: Source dir to watch */ char * option_source = NULL; /** * Option: Target to rsync to. */ char * option_target = NULL; /** * Option: rsync binary to call. */ char * rsync_binary = "/usr/bin/rsync"; /** * Option: the exclude-file to pass to rsync. */ char * exclude_file = NULL; /** * The maximum character length paths may take */ #define MAX_PATH (8*1024) /** * Structure to store the directory watches of the deamon. */ struct dir_watch { /** * The watch descriptor returned by kernel. */ int wd; /** * The name of the directory. * In case of the root dir to be watched, it is a full path * and parent == NULL. Otherwise its just the name of the * directory and parent points to the parent directory thats * also watched. */ char * dirname; /** * Call this directory that way on destiation. * if NULL call it like dirname. */ char * destname; /** * Points to the index of the parent. * -1 if no parent */ int parent; }; /** * Structure to store strings for the diversve Inotfy masked events. * Actually used for compfortable debugging only. */ struct inotify_mask_text { int mask; char const * text; }; /** * A constant that assigns every inotify mask a printable string. * Used for debugging. */ struct inotify_mask_text mask_texts[] = { { IN_ACCESS, "ACCESS" }, { IN_ATTRIB, "ATTRIB" }, { IN_CLOSE_WRITE, "CLOSE_WRITE" }, { IN_CLOSE_NOWRITE, "CLOSE_NOWRITE" }, { IN_CREATE, "CREATE" }, { IN_DELETE, "DELETE" }, { IN_DELETE_SELF, "DELETE_SELF" }, { IN_MODIFY, "MODIFY" }, { IN_MOVE_SELF, "MOVE_SELF" }, { IN_MOVED_FROM, "MOVED_FROM" }, { IN_MOVED_TO, "MOVED_TO" }, { IN_OPEN, "OPEN" }, { 0, "" }, }; /** * Holds an allocated array of all directories watched. */ struct dir_watch *dir_watches; /** * The allocated size of dir_watches; */ int dir_watch_size = 0; /** * The largest dir_watch number used; */ int dir_watch_num = 0; /** * lsyncd will log into this file/stream. */ char * logfile = "/var/log/lsyncd"; /** * The inotify instance. */ int inotf; /** * Array of strings of directory names to include. * This is limited to MAX_EXCLUDES. * It's not worth to code a dynamic size handling... */ #define MAX_EXCLUDES 256 char * exclude_dirs[MAX_EXCLUDES] = {NULL, }; int exclude_dir_n = 0; volatile sig_atomic_t keep_going = 1; void catch_alarm(int sig) { keep_going = 0; } /** * Prints a message to the log stream, preceding a timestamp. * Otherwise it behaves like printf(); */ void printlogf(int level, const char *fmt, ...) { va_list ap; char * ct; time_t mtime; FILE * flog; if (level < loglevel) { return; } if (!flag_nodaemon) { flog = fopen(logfile, "a"); if (flog == NULL) { printf("cannot open logfile [%s]!\n", logfile); exit(-1); } } else { flog = stdout; } va_start(ap, fmt); time(&mtime); ct = ctime(&mtime); ct[strlen(ct) - 1] = 0; // cut trailing \n fprintf(flog, "%s: ", ct); switch (level) { case LOG_DEBUG : break; case LOG_NORMAL : break; case LOG_ERROR : fprintf(flog, "ERROR :"); break; } vfprintf(flog, fmt, ap); fprintf(flog, "\n"); va_end(ap); if (!flag_nodaemon) { fclose(flog); } } /** * "secured" malloc, meaning the deamon shall kill itself * in case of out of memory. * * On linux systems, which is actually the only system this * deamon will run at, due to the use of inotify, this is * an "academic" cleaness only, linux will never return out * memory, but kill a process to ensure memory will be * available. */ void *s_malloc(size_t size) { void *r = malloc(size); if (r == NULL) { printlogf(LOG_ERROR, "Out of memory!"); exit(-1); } return r; } /** * "secured" calloc. */ void *s_calloc(size_t nmemb, size_t size) { void *r = calloc(nmemb, size); if (r == NULL) { printlogf(LOG_ERROR, "Out of memory!"); exit(-1); } return r; } /** * "secured" realloc. */ void *s_realloc(void *ptr, size_t size) { void *r = realloc(ptr, size); if (r == NULL) { printlogf(LOG_ERROR, "Out of memory!"); exit(-1); } return r; } /** * Calls rsync to sync from src to dest. * Returns after rsync has finished. * * @param src Source string. * @param dest Destination string, * @param recursive If true -r will be handled on, -d (single directory) otherwise */ bool rsync(char const * src, const char * dest, bool recursive) { pid_t pid; int status; char const * opts = recursive ? "-ltr" : "-ltd"; if (exclude_file) { printlogf(LOG_DEBUG, "exec %s(%s,%s,%s,%s,%s,%s)", rsync_binary, "--delete", opts, "--exclude-from", exclude_file, src, dest); } else { printlogf(LOG_DEBUG, "exec %s(%s,%s,%s,%s)", rsync_binary, "--delete", opts, src, dest); } if (flag_dryrun) { return true; } pid = fork(); if (pid == 0) { if (!flag_nodaemon) { freopen(logfile, "a", stdout); freopen(logfile, "a", stderr); } if (exclude_file) { execl(rsync_binary, rsync_binary, "--delete", opts, "--exclude-from", exclude_file, src, dest, NULL); } else { execl(rsync_binary, rsync_binary, "--delete", opts, src, dest, NULL); } printlogf(LOG_ERROR, "Failed executing [%s]", rsync_binary); exit(-1); } waitpid(pid, &status, 0); assert(WIFEXITED(status)); if (WEXITSTATUS(status)){ printlogf(LOG_ERROR, "Forked rsync process returned non-zero return code"); //TODO: dispute, to we really want to terminate in this case? //exit(-1); } printlogf(LOG_DEBUG, "Rsync of [%s] -> [%s] finished", src, dest); return true; } /** * Adds a directory to watch * * @param pathname the absolute path of the directory to watch. * @param dirname the name of the directory only (yes this is a bit redudant, but oh well) * @param destname if not NULL call this dir that way on destionation. * @param parent if not -1 the index to the parent directory that is already watched * * @return index to dir_watches of the new dir, -1 on error. */ int add_watch(char const * pathname, char const * dirname, char const * destname, int parent) { int wd; char * nn; int newdw; wd = inotify_add_watch(inotf, pathname, IN_ATTRIB | IN_CLOSE_WRITE | IN_CREATE | IN_DELETE | IN_DELETE_SELF | IN_MOVED_FROM | IN_MOVED_TO | IN_DONT_FOLLOW | IN_ONLYDIR); if (wd == -1) { printlogf(LOG_ERROR, "Cannot add watch %s (%d:%s)", pathname, errno, strerror(errno)); return -1; } // look if an unused slot can be found. for (newdw = 0; newdw < dir_watch_num; newdw++) { if (dir_watches[newdw].wd < 0) { break; } } if (newdw == dir_watch_num) { if (dir_watch_num + 1 >= dir_watch_size) { dir_watch_size *= 2; dir_watches = s_realloc(dir_watches, dir_watch_size * sizeof(struct dir_watch)); } dir_watch_num++; } dir_watches[newdw].wd = wd; dir_watches[newdw].parent = parent; nn = s_malloc(strlen(dirname) + 1); strcpy(nn, dirname); dir_watches[newdw].dirname = nn; if (destname) { nn = s_malloc(strlen(destname) + 1); strcpy(nn, destname); dir_watches[newdw].destname = nn; } else { dir_watches[newdw].destname = NULL; } return newdw; } /** * Builds the abolute path name of a given directory beeing watched. * * @param pathname destination buffer to store the result to. * @param pathsize max size of this buffer * @param watch the index in dir_watches to the directory. * @param name if not null, this beeing the watch dir itself is appended. * @param prefix if not NULL it is added at the bein of pathname */ bool buildpath(char *pathname, int pathsize, int watch, char const *name, char const * prefix) { int j, k, p, ps; pathname[0] = 0; if (prefix) { strcat(pathname, prefix); } // count how big the parent stack is for (p = watch, ps = 0; p != -1; p = dir_watches[p].parent, ps++) { } // now add the parent paths from back to front for (j = ps; j > 0; j--) { char * name; // go j steps behind stack for (p = watch, k = 0; k + 1 < j; p = dir_watches[p].parent, k++) { } name = (prefix && dir_watches[p].destname) ? dir_watches[p].destname : dir_watches[p].dirname; if (strlen(pathname) + strlen(name) + 2 >= pathsize) { printlogf(LOG_ERROR, "path too long %s/...", name); return false; } strcat(pathname, name); strcat(pathname, "/"); } if (name) { if (strlen(pathname) + strlen(name) + 2 >= pathsize) { printlogf(LOG_ERROR, "path too long %s//%s", pathname, name); return false; } strcat(pathname, name); } return true; } /** * Adds a dir to watch. * * @param dirname The name or absolute path of the directory to watch. * @param destname If not NULL call this dir that way on sync destination. * @param recursive If true, will also watch all sub-directories. * @param parent If not -1, the index in dir_watches to the parent directory already watched. * Must have absolute path if parent == -1. */ bool add_dirwatch(char const * dirname, char const * destname, bool recursive, int parent) { DIR *d; struct dirent *de; int dw, i; char pathname[MAX_PATH]; printlogf(LOG_DEBUG, "add_dirwatch(%s, %s, %d, p->dirname:%s)", dirname, destname, recursive, parent >= 0 ? dir_watches[parent].dirname : "NULL"); if (!buildpath(pathname, sizeof(pathname), parent, dirname, NULL)) { return false; } for (i = 0; i < exclude_dir_n; i++) { if (!strcmp(dirname, exclude_dirs[i])) { return true; } } dw = add_watch(pathname, dirname, destname, parent); if (dw == -1) { return false; } if (strlen(pathname) + strlen(dirname) + 2 > sizeof(pathname)) { printlogf(LOG_ERROR, "pathname too long %s//%s", pathname, dirname); return false; } d = opendir(pathname); if (d == NULL) { printlogf(LOG_ERROR, "cannot open dir %s.", dirname); return false; } while (keep_going) { de = readdir(d); if (de == NULL) { break; } if (de->d_type == DT_DIR && strcmp(de->d_name, "..") && strcmp(de->d_name, ".")) { add_dirwatch(de->d_name, NULL, true, dw); } } closedir(d); return true; } /** * Removes a watched dir, including recursevily all subdirs. * * @param name Optionally. If not NULL, the directory name to remove which is a child of parent. * @param parent The index to the parent directory of the directory 'name' to remove, * or to be removed itself if name == NULL. */ bool remove_dirwatch(const char * name, int parent) { int i; int dw; if (name) { // look for the child with the name for (i = 0; i < dir_watch_num; i++) { if (dir_watches[i].wd >= 0 && dir_watches[i].parent == parent && !strcmp(name, dir_watches[i].dirname)) { dw = i; break; } } if (i >= dir_watch_num) { printlogf(LOG_ERROR, "Cannot find entry for %s:/:%s :-(", dir_watches[parent].dirname, name); return false; } } else { dw = parent; } for (i = 0; i < dir_watch_num; i++) { if (dir_watches[i].wd >= 0 && dir_watches[i].parent == dw) { remove_dirwatch(NULL, i); } } inotify_rm_watch(inotf, dir_watches[dw].wd); dir_watches[dw].wd = -1; free(dir_watches[dw].dirname); dir_watches[dw].dirname = NULL; if (dir_watches[dw].destname) { free(dir_watches[dw].destname); dir_watches[dw].destname = NULL; } return true; } /** * Handles an inotify event. * * @param event The event to handle */ bool handle_event(struct inotify_event *event) { char masktext[255] = {0,}; char pathname[MAX_PATH]; char destname[MAX_PATH]; int mask = event->mask; int i; struct inotify_mask_text *p; if (IN_IGNORED & event->mask) { return true; } for (i = 0; i < exclude_dir_n; i++) { if (!strcmp(event->name, exclude_dirs[i])) { return true; } } for (p = mask_texts; p->mask; p++) { if (mask & p->mask) { if (strlen(masktext) + strlen(p->text) + 3 >= sizeof(masktext)) { printlogf(LOG_ERROR, "bufferoverflow in handle_event"); return false; } if (*masktext) { strcat(masktext, ", "); } strcat(masktext, p->text); } } for (i = 0; i < dir_watch_num; i++) { if (dir_watches[i].wd == event->wd) { break; } } if (i >= dir_watch_num) { printlogf(LOG_ERROR, "received unkown inotify event :-(%d)", event->mask); return false; } if (((IN_CREATE | IN_MOVED_TO) & event->mask) && (IN_ISDIR & event->mask)) { add_dirwatch(event->name, NULL, false, i); } if (((IN_DELETE | IN_MOVED_FROM) & event->mask) && (IN_ISDIR & event->mask)) { remove_dirwatch(event->name, i); } if (!buildpath(pathname, sizeof(pathname), i, NULL, NULL)) { return false; } if (!buildpath(destname, sizeof(destname), i, NULL, option_target)) { return false; } // call rsync to propagate changes in the directory if ((IN_CREATE | IN_CLOSE_WRITE | IN_DELETE | IN_MOVED_TO | IN_MOVED_FROM) & event->mask) { printlogf(LOG_NORMAL, "%s of %s in %s --> %s", masktext, event->name, pathname, destname); rsync(pathname, destname, false); } return 0; } /** * The control loop waiting for inotify events. */ bool master_loop() { char buf[INOTIFY_BUF_LEN]; int len, i = 0; while (keep_going) { len = read (inotf, buf, INOTIFY_BUF_LEN); if (len < 0) { printlogf(LOG_ERROR, "failed to read from inotify (%d:%s)", errno, strerror(errno)); return false; } if (len == 0) { printlogf(LOG_ERROR, "eof?"); return false; } i = 0; while (i < len) { struct inotify_event *event = (struct inotify_event *) &buf[i]; handle_event(event); i += sizeof(struct inotify_event) + event->len; } } return true; } /** * Scans all dirs in /home, looking if a www subdir exists. * Syncs this dir immediately, and adds watches to it. */ bool scan_homes() { DIR *d; DIR *d2; char path[MAX_PATH]; char destpath[MAX_PATH]; struct dirent *de; d = opendir("/home"); if (d == NULL) { printlogf(LOG_ERROR, "Cannot open /home"); return false; } while (keep_going) { de = readdir(d); if (de == NULL) { break; } if (de->d_type == DT_DIR && strcmp(de->d_name, "..") && strcmp(de->d_name, ".")) { snprintf(path, sizeof(path), "/home/%s/www/", de->d_name); d2 = opendir(path); if (d2 == NULL) { //has no www dir or is not readable printlogf(LOG_NORMAL, "skipping %s. it has no readable www directory.", de->d_name); continue; } closedir(d2); printlogf(LOG_NORMAL, "watching %s's www directory (%s)", de->d_name, path); add_dirwatch(path, de->d_name, true, -1); snprintf(destpath, sizeof(destpath), "%s/%s/", option_target, de->d_name); rsync(path, destpath, true); } } closedir(d); return true; } /** * Prints the help text and exits 0. * * @param arg0 argv[0] to show what lsyncd was called with. */ void print_help(char *arg0) { printf("\n"); printf("USAGE: %s [OPTION]... SOURCE TARGET\n", arg0); printf("\n"); printf("SOURCE: a directory to watch and rsync.\n"); printf(" specify special \"%%userwww\" to scan all users in /home and watch their www directories. \n"); printf("\n"); printf("TARGET: can be any name accepted by rsync. e.g. \"foohost::barmodule/\"\n"); printf("\n"); printf("OPTIONS:\n"); printf(" --debug Log debug messages\n"); printf(" --dryrun Do not call rsync, run dry only\n"); printf(" --exclude-from FILE Exclude file handlet to rsync (DEFAULT: None)\n"); printf(" --help Print this help text and exit.\n"); printf(" --logfile FILE Put log here (DEFAULT: %s)\n", logfile); printf(" --no-daemon Do not detach, log to stdout/stderr\n"); printf(" --rsync-binary FILE Call this binary to sync (DEFAULT: %s)\n", rsync_binary); printf(" --scarce Only log errors\n"); printf(" --version Print version an exit.\n"); printf("\n"); printf("Take care that lsyncd is allowed to write to the logfile specified.\n"); printf("\n"); printf("EXCLUDE FILE: \n"); printf(" The exclude file may have either filebased general masks like \"*.php\" without directory specifications,\n"); printf(" or exclude complete directories like \"Data/\". lsyncd will recognize directory excludes by the trailing '/'\n"); printf(" and will not add watches of directories of exactly such name including sub-directories of them.\n"); printf(" Please do not try to use more sophisticated exclude masks like \"Data/*.dat\" or \"Da*a/\", \"Data/Volatile/\" etc.\n"); printf(" This will not work like you would expect it to.\n"); printf("\n"); printf("LICENSE\n"); printf(" GPLv2 or any later version. See COPYING\n"); printf("\n"); exit(0); } /** * Parses the command line options. */ bool parse_options(int argc, char **argv) { static struct option long_options[] = { {"debug", 0, &loglevel, 1}, {"dryrun", 0, &flag_dryrun, 1}, {"exclude-from", 1, NULL, 0}, {"help", 0, NULL, 0}, {"logfile", 1, NULL, 0}, {"no-daemon", 0, &flag_nodaemon, 1}, {"rsync-binary", 1, NULL, 0}, {"scarce", 0, &loglevel, 3}, {"version", 0, NULL, 0}, {0, 0, 0, 0} }; int c; while (1) { int oi = 0; c = getopt_long_only(argc, argv, "", long_options, &oi); if (c == -1) { break; } if (c == '?') { return false; } if (c == 0) { // longoption if (!strcmp("help", long_options[oi].name)) { print_help(argv[0]); } if (!strcmp("version", long_options[oi].name)) { printf("Version: %d.%d\n", VER_MAJOR, VER_MINOR); exit(0); } if (!strcmp("logfile", long_options[oi].name)) { logfile = s_malloc(strlen(optarg) + 1); strcpy(logfile, optarg); } if (!strcmp("exclude-from", long_options[oi].name)) { exclude_file = s_malloc(strlen(optarg) + 1); strcpy(exclude_file, optarg); } } } if (optind + 2 != argc) { printf("Error: please specify SOURCE and TARGET (see --help)\n"); exit(-1); } option_source = argv[optind]; option_target = argv[optind + 1]; printf("syncing %s -> %s\n", option_source, option_target); return true; } /** * Parses the exclude file looking for directory masks to not watch. */ bool parse_exclude_file() { FILE * ef; char line[MAX_PATH]; int sl; ef = fopen(exclude_file, "r"); if (ef == NULL) { printlogf(LOG_ERROR, "Meh, cannot open exclude file '%s'\n", exclude_file); exit(-1); } while (1) { if (!fgets(line, sizeof(line), ef)) { if (feof(ef)) { fclose(ef); return true; } printlogf(LOG_ERROR, "Reading file '%s' (%d=%s)\n", exclude_file, errno, strerror(errno)); exit(-1); } sl = strlen(line); if (sl == 0) { continue; } if (line[sl - 1] == '\n') { line[sl - 1] = 0; sl--; } if (sl == 0) { continue; } if (line[sl - 1] == '/') { if (exclude_dir_n + 1 >= MAX_EXCLUDES) { printlogf(LOG_ERROR, "Too many directory excludes, can only have %d at the most", MAX_EXCLUDES); exit(-1); } line[sl - 1] = 0; sl--; if (sl == 0) { continue; } printlogf(LOG_NORMAL, "Excluding directories of the name '%s'", line); exclude_dirs[exclude_dir_n] = s_malloc(strlen(line) + 1); strcpy(exclude_dirs[exclude_dir_n], line); exclude_dir_n++; } } return true; } /** * main */ int main(int argc, char **argv) { if (!parse_options(argc, argv)) { return -1; } if (exclude_file) { parse_exclude_file(); } inotf = inotify_init(); if (inotf == -1) { printlogf(LOG_ERROR, "Cannot create inotify instance! (%d:%s)", errno, strerror(errno)); return -1; } if (!flag_nodaemon) { // this will make this process child of init, close stdin/stdout/stderr and chdir to / daemon(0, 0); } printlogf(LOG_NORMAL, "Starting up"); dir_watch_size = 2; dir_watches = s_calloc(dir_watch_size, sizeof(struct dir_watch)); if (!strcmp(option_source, "%userwww")) { printlogf(LOG_NORMAL, "do userwww"); scan_homes(); } else { printlogf(LOG_NORMAL, "watching %s", option_source); add_dirwatch(option_source, "", true, -1); rsync(option_source, option_target, true); } printlogf(LOG_NORMAL, "--- Entering normal operation with [%d] monitored directories ---", dir_watch_num); signal(SIGTERM, catch_alarm); master_loop(); return 0; }