diff --git a/exclude.c b/exclude.c index 7989fb3..13c4253 100644 --- a/exclude.c +++ b/exclude.c @@ -24,18 +24,26 @@ extern int am_server; extern int am_sender; +extern int am_generator; extern int eol_nulls; extern int io_error; +extern int xfer_dirs; +extern int recurse; extern int local_server; extern int prune_empty_dirs; extern int ignore_perishable; +extern int old_style_args; +extern int relative_paths; extern int delete_mode; extern int delete_excluded; extern int cvs_exclude; extern int sanitize_paths; extern int protocol_version; +extern int read_batch; +extern int list_only; extern int module_id; +extern char *filesfrom_host; extern char curr_dir[MAXPATHLEN]; extern unsigned int curr_dir_len; extern unsigned int module_dirlen; @@ -43,8 +51,10 @@ extern unsigned int module_dirlen; filter_rule_list filter_list = { .debug_type = "" }; filter_rule_list cvs_filter_list = { .debug_type = " [global CVS]" }; filter_rule_list daemon_filter_list = { .debug_type = " [daemon]" }; +filter_rule_list implied_filter_list = { .debug_type = " [implied]" }; int saw_xattr_filter = 0; +int trust_sender_filter = 0; /* Need room enough for ":MODS " prefix plus some room to grow. */ #define MAX_RULE_PREFIX (16) @@ -293,6 +303,233 @@ static void add_rule(filter_rule_list *listp, const char *pat, unsigned int pat_ } } +/* If the wildcards failed, the remote shell might give us a file matching the literal + * wildcards. Since "*" & "?" already match themselves, this just needs to deal with + * failed "[foo]" idioms. + */ +static void maybe_add_literal_brackets_rule(filter_rule const *based_on, int arg_len) +{ + filter_rule *rule; + const char *arg = based_on->pattern, *cp; + char *p; + int cnt = 0; + + if (arg_len < 0) + arg_len = strlen(arg); + + for (cp = arg; *cp; cp++) { + if (*cp == '\\' && cp[1]) { + cp++; + } else if (*cp == '[') + cnt++; + } + if (!cnt) + return; + + rule = new0(filter_rule); + rule->rflags = based_on->rflags; + rule->u.slash_cnt = based_on->u.slash_cnt; + p = rule->pattern = new_array(char, arg_len + cnt + 1); + for (cp = arg; *cp; ) { + if (*cp == '\\' && cp[1]) { + *p++ = *cp++; + } else if (*cp == '[') + *p++ = '\\'; + *p++ = *cp++; + } + *p++ = '\0'; + + rule->next = implied_filter_list.head; + implied_filter_list.head = rule; + if (DEBUG_GTE(FILTER, 3)) { + rprintf(FINFO, "[%s] add_implied_include(%s%s)\n", who_am_i(), rule->pattern, + rule->rflags & FILTRULE_DIRECTORY ? "/" : ""); + } +} + +static char *partial_string_buf = NULL; +static int partial_string_len = 0; +void implied_include_partial_string(const char *s_start, const char *s_end) +{ + partial_string_len = s_end - s_start; + if (partial_string_len <= 0 || partial_string_len >= MAXPATHLEN) { /* too-large should be impossible... */ + partial_string_len = 0; + return; + } + if (!partial_string_buf) + partial_string_buf = new_array(char, MAXPATHLEN); + memcpy(partial_string_buf, s_start, partial_string_len); +} + +void free_implied_include_partial_string() +{ + if (partial_string_buf) { + free(partial_string_buf); + partial_string_buf = NULL; + } + partial_string_len = 0; /* paranoia */ +} + +/* Each arg the client sends to the remote sender turns into an implied include + * that the receiver uses to validate the file list from the sender. */ +void add_implied_include(const char *arg, int skip_daemon_module) +{ + filter_rule *rule; + int arg_len, saw_wild = 0, saw_live_open_brkt = 0, backslash_cnt = 0; + int slash_cnt = 1; /* We know we're adding a leading slash. */ + const char *cp; + char *p; + if (am_server || old_style_args || list_only || read_batch || filesfrom_host != NULL) + return; + if (partial_string_len) { + arg_len = strlen(arg); + if (partial_string_len + arg_len >= MAXPATHLEN) { + partial_string_len = 0; + return; /* Should be impossible... */ + } + memcpy(partial_string_buf + partial_string_len, arg, arg_len + 1); + partial_string_len = 0; + arg = partial_string_buf; + } + if (skip_daemon_module) { + if ((cp = strchr(arg, '/')) != NULL) + arg = cp + 1; + else + arg = ""; + } + if (relative_paths) { + if ((cp = strstr(arg, "/./")) != NULL) + arg = cp + 3; + } else if ((cp = strrchr(arg, '/')) != NULL) { + arg = cp + 1; + } + if (*arg == '.' && arg[1] == '\0') + arg++; + arg_len = strlen(arg); + if (arg_len) { + if (strpbrk(arg, "*[?")) { + /* We need to add room to escape backslashes if wildcard chars are present. */ + for (cp = arg; (cp = strchr(cp, '\\')) != NULL; cp++) + arg_len++; + saw_wild = 1; + } + arg_len++; /* Leave room for the prefixed slash */ + rule = new0(filter_rule); + if (!implied_filter_list.head) + implied_filter_list.head = implied_filter_list.tail = rule; + else { + rule->next = implied_filter_list.head; + implied_filter_list.head = rule; + } + rule->rflags = FILTRULE_INCLUDE + (saw_wild ? FILTRULE_WILD : 0); + p = rule->pattern = new_array(char, arg_len + 1); + *p++ = '/'; + for (cp = arg; *cp; ) { + switch (*cp) { + case '\\': + if (cp[1] == ']') { + if (!saw_wild) + cp++; /* A \] in a non-wild filter causes a problem, so drop the \ . */ + } else if (!strchr("*[?", cp[1])) { + backslash_cnt++; + if (saw_wild) + *p++ = '\\'; + } + *p++ = *cp++; + break; + case '/': + if (p[-1] == '/') { /* This is safe because of the initial slash. */ + cp++; + break; + } + if (relative_paths) { + filter_rule const *ent; + int found = 0; + *p = '\0'; + for (ent = implied_filter_list.head; ent; ent = ent->next) { + if (ent != rule && strcmp(ent->pattern, rule->pattern) == 0) { + found = 1; + break; + } + } + if (!found) { + filter_rule *R_rule = new0(filter_rule); + R_rule->rflags = FILTRULE_INCLUDE | FILTRULE_DIRECTORY; + /* Check if our sub-path has wildcards or escaped backslashes */ + if (saw_wild && strpbrk(rule->pattern, "*[?\\")) + R_rule->rflags |= FILTRULE_WILD; + R_rule->pattern = strdup(rule->pattern); + R_rule->u.slash_cnt = slash_cnt; + R_rule->next = implied_filter_list.head; + implied_filter_list.head = R_rule; + if (DEBUG_GTE(FILTER, 3)) { + rprintf(FINFO, "[%s] add_implied_include(%s/)\n", + who_am_i(), R_rule->pattern); + } + if (saw_live_open_brkt) + maybe_add_literal_brackets_rule(R_rule, -1); + } + } + slash_cnt++; + *p++ = *cp++; + break; + case '[': + saw_live_open_brkt = 1; + *p++ = *cp++; + break; + default: + *p++ = *cp++; + break; + } + } + *p = '\0'; + rule->u.slash_cnt = slash_cnt; + arg = rule->pattern; + arg_len = p - arg; /* We recompute it due to backslash weirdness. */ + if (DEBUG_GTE(FILTER, 3)) + rprintf(FINFO, "[%s] add_implied_include(%s)\n", who_am_i(), rule->pattern); + if (saw_live_open_brkt) + maybe_add_literal_brackets_rule(rule, arg_len); + } + + if (recurse || xfer_dirs) { + /* Now create a rule with an added "/" & "**" or "*" at the end */ + rule = new0(filter_rule); + rule->rflags = FILTRULE_INCLUDE | FILTRULE_WILD; + if (recurse) + rule->rflags |= FILTRULE_WILD2; + /* We must leave enough room for / * * \0. */ + if (!saw_wild && backslash_cnt) { + /* We are appending a wildcard, so now the backslashes need to be escaped. */ + p = rule->pattern = new_array(char, arg_len + backslash_cnt + 3 + 1); + for (cp = arg; *cp; ) { + if (*cp == '\\') + *p++ = '\\'; + *p++ = *cp++; + } + } else { + p = rule->pattern = new_array(char, arg_len + 3 + 1); + if (arg_len) { + memcpy(p, arg, arg_len); + p += arg_len; + } + } + if (p[-1] != '/') + *p++ = '/'; + *p++ = '*'; + if (recurse) + *p++ = '*'; + *p = '\0'; + rule->u.slash_cnt = slash_cnt + 1; + rule->next = implied_filter_list.head; + implied_filter_list.head = rule; + if (DEBUG_GTE(FILTER, 3)) + rprintf(FINFO, "[%s] add_implied_include(%s)\n", who_am_i(), rule->pattern); + if (saw_live_open_brkt) + maybe_add_literal_brackets_rule(rule, p - rule->pattern); + } +} + /* This frees any non-inherited items, leaving just inherited items on the list. */ static void pop_filter_list(filter_rule_list *listp) { @@ -709,11 +946,12 @@ static void report_filter_result(enum logcode code, char const *name, filter_rule const *ent, int name_flags, const char *type) { + int log_level = am_sender || am_generator ? 1 : 3; + /* If a trailing slash is present to match only directories, * then it is stripped out by add_rule(). So as a special - * case we add it back in here. */ - - if (DEBUG_GTE(FILTER, 1)) { + * case we add it back in the log output. */ + if (DEBUG_GTE(FILTER, log_level)) { static char *actions[2][2] = { {"show", "hid"}, {"risk", "protect"} }; const char *w = who_am_i(); @@ -721,7 +959,7 @@ static void report_filter_result(enum logcode code, char const *name, : name_flags & NAME_IS_DIR ? "directory" : "file"; rprintf(code, "[%s] %sing %s %s because of pattern %s%s%s\n", - w, actions[*w!='s'][!(ent->rflags & FILTRULE_INCLUDE)], + w, actions[*w=='g'][!(ent->rflags & FILTRULE_INCLUDE)], t, name, ent->pattern, ent->rflags & FILTRULE_DIRECTORY ? "/" : "", type); } @@ -894,6 +1132,7 @@ static filter_rule *parse_rule_tok(const char **rulestr_ptr, } switch (ch) { case ':': + trust_sender_filter = 1; rule->rflags |= FILTRULE_PERDIR_MERGE | FILTRULE_FINISH_SETUP; /* FALL THROUGH */ diff --git a/flist.c b/flist.c index 499440c..630d685 100644 --- a/flist.c +++ b/flist.c @@ -70,6 +70,7 @@ extern int need_unsorted_flist; extern int sender_symlink_iconv; extern int output_needs_newline; extern int sender_keeps_checksum; +extern int trust_sender_filter; extern int unsort_ndx; extern uid_t our_uid; extern struct stats stats; @@ -80,8 +81,7 @@ extern char curr_dir[MAXPATHLEN]; extern struct chmod_mode_struct *chmod_modes; -extern filter_rule_list filter_list; -extern filter_rule_list daemon_filter_list; +extern filter_rule_list filter_list, implied_filter_list, daemon_filter_list; #ifdef ICONV_OPTION extern int filesfrom_convert; @@ -904,6 +904,19 @@ static struct file_struct *recv_file_entry(int f, struct file_list *flist, int x exit_cleanup(RERR_UNSUPPORTED); } + if (*thisname != '.' || thisname[1] != '\0') { + int filt_flags = S_ISDIR(mode) ? NAME_IS_DIR : NAME_IS_FILE; + if (!trust_sender_filter /* a per-dir filter rule means we must trust the sender's filtering */ + && filter_list.head && check_filter(&filter_list, FINFO, thisname, filt_flags) < 0) { + rprintf(FERROR, "ERROR: rejecting excluded file-list name: %s\n", thisname); + exit_cleanup(RERR_PROTOCOL); + } + if (implied_filter_list.head && check_filter(&implied_filter_list, FINFO, thisname, filt_flags) <= 0) { + rprintf(FERROR, "ERROR: rejecting unrequested file-list name: %s\n", thisname); + exit_cleanup(RERR_PROTOCOL); + } + } + if (inc_recurse && S_ISDIR(mode)) { if (one_file_system) { /* Room to save the dir's device for -x */ diff --git a/io.c b/io.c index 59105ba..3aea50f 100644 --- a/io.c +++ b/io.c @@ -374,6 +374,7 @@ static void forward_filesfrom_data(void) free_xbuf(&ff_xb); if (ff_reenable_multiplex >= 0) io_start_multiplex_out(ff_reenable_multiplex); + free_implied_include_partial_string(); } return; } @@ -415,6 +416,7 @@ static void forward_filesfrom_data(void) while (s != eob) { if (*s++ == '\0') { ff_xb.len = s - sob - 1; + add_implied_include(sob, 0); if (iconvbufs(ic_send, &ff_xb, &iobuf.out, flags) < 0) exit_cleanup(RERR_PROTOCOL); /* impossible? */ write_buf(iobuf.out_fd, s-1, 1); /* Send the '\0'. */ @@ -430,6 +432,7 @@ static void forward_filesfrom_data(void) ff_lastchar = '\0'; else { /* Handle a partial string specially, saving any incomplete chars. */ + implied_include_partial_string(sob, s); flags &= ~ICB_INCLUDE_INCOMPLETE; if (iconvbufs(ic_send, &ff_xb, &iobuf.out, flags) < 0) { if (errno == E2BIG) @@ -446,13 +449,17 @@ static void forward_filesfrom_data(void) char *f = ff_xb.buf + ff_xb.pos; char *t = ff_xb.buf; char *eob = f + len; + char *cur = t; /* Eliminate any multi-'\0' runs. */ while (f != eob) { if (!(*t++ = *f++)) { + add_implied_include(cur, 0); + cur = t; while (f != eob && *f == '\0') f++; } } + implied_include_partial_string(cur, t); ff_lastchar = f[-1]; if ((len = t - ff_xb.buf) != 0) { /* This will not circle back to perform_io() because we only get diff --git a/main.c b/main.c index 6113563..abe2ebf 100644 --- a/main.c +++ b/main.c @@ -42,6 +42,7 @@ extern int output_needs_newline; extern int need_messages_from_generator; extern int kluge_around_eof; extern int got_xfer_error; +extern int old_style_args; extern int msgs2stderr; extern int module_id; extern int read_only; @@ -78,6 +79,7 @@ extern BOOL flist_receiving_enabled; extern BOOL shutting_down; extern int backup_dir_len; extern int basis_dir_cnt; +extern int trust_sender_filter; extern struct stats stats; extern char *stdout_format; extern char *logfile_format; @@ -93,7 +95,7 @@ extern char curr_dir[MAXPATHLEN]; extern char backup_dir_buf[MAXPATHLEN]; extern char *basis_dir[MAX_BASIS_DIRS+1]; extern struct file_list *first_flist; -extern filter_rule_list daemon_filter_list; +extern filter_rule_list daemon_filter_list, implied_filter_list; uid_t our_uid; gid_t our_gid; @@ -503,11 +505,7 @@ static pid_t do_cmd(char *cmd, char *machine, char *user, char **remote_argv, in rprintf(FERROR, "internal: args[] overflowed in do_cmd()\n"); exit_cleanup(RERR_SYNTAX); } - if (**remote_argv == '-') { - if (asprintf(args + argc++, "./%s", *remote_argv++) < 0) - out_of_memory("do_cmd"); - } else - args[argc++] = *remote_argv++; + args[argc++] = safe_arg(NULL, *remote_argv++); remote_argc--; } } @@ -534,6 +532,7 @@ static pid_t do_cmd(char *cmd, char *machine, char *user, char **remote_argv, in #ifdef ICONV_CONST setup_iconv(); #endif + trust_sender_filter = 1; } else if (local_server) { /* If the user didn't request --[no-]whole-file, force * it on, but only if we're not batch processing. */ @@ -943,6 +942,7 @@ static int do_recv(int f_in, int f_out, char *local_name) } am_generator = 1; + implied_filter_list.head = implied_filter_list.tail = NULL; flist_receiving_enabled = True; io_end_multiplex_in(MPLX_SWITCHING); @@ -1340,6 +1340,10 @@ static int start_client(int argc, char *argv[]) remote_argc = argc = 1; } + /* A local transfer doesn't unbackslash anything, so leave the args alone. */ + if (local_server) + old_style_args = 2; + if (!rsync_port && remote_argc && !**remote_argv) /* Turn an empty arg into a dot dir. */ *remote_argv = "."; @@ -1358,6 +1362,8 @@ static int start_client(int argc, char *argv[]) char *dummy_host; int dummy_port = rsync_port; int i; + if (filesfrom_fd < 0) + add_implied_include(remote_argv[0], daemon_over_rsh); /* For remote source, any extra source args must have either * the same hostname or an empty hostname. */ for (i = 1; i < remote_argc; i++) { @@ -1381,6 +1387,7 @@ static int start_client(int argc, char *argv[]) if (!rsync_port && !*arg) /* Turn an empty arg into a dot dir. */ arg = "."; remote_argv[i] = arg; + add_implied_include(arg, daemon_over_rsh); } } diff --git a/receiver.c b/receiver.c index d6a48f1..c0aa893 100644 --- a/receiver.c +++ b/receiver.c @@ -577,10 +577,13 @@ int recv_files(int f_in, int f_out, char *local_name) if (DEBUG_GTE(RECV, 1)) rprintf(FINFO, "recv_files(%s)\n", fname); - if (daemon_filter_list.head && (*fname != '.' || fname[1] != '\0') - && check_filter(&daemon_filter_list, FLOG, fname, 0) < 0) { - rprintf(FERROR, "attempt to hack rsync failed.\n"); - exit_cleanup(RERR_PROTOCOL); + if (daemon_filter_list.head && (*fname != '.' || fname[1] != '\0')) { + int filt_flags = S_ISDIR(file->mode) ? NAME_IS_DIR : NAME_IS_FILE; + if (check_filter(&daemon_filter_list, FLOG, fname, filt_flags) < 0) { + rprintf(FERROR, "ERROR: rejecting file transfer request for daemon excluded file: %s\n", + fname); + exit_cleanup(RERR_PROTOCOL); + } } #ifdef SUPPORT_XATTRS diff --git a/options.c b/options.c index 43e8257..aaf8cc9 100644 --- a/options.c +++ b/options.c @@ -99,6 +99,7 @@ int filesfrom_fd = -1; char *filesfrom_host = NULL; int eol_nulls = 0; int protect_args = -1; +int old_style_args = -1; int human_readable = 1; int recurse = 0; int allow_inc_recurse = 1; @@ -277,7 +278,7 @@ static struct output_struct debug_words[COUNT_DEBUG+1] = { DEBUG_WORD(DELTASUM, W_SND|W_REC, "Debug delta-transfer checksumming (levels 1-4)"), DEBUG_WORD(DUP, W_REC, "Debug weeding of duplicate names"), DEBUG_WORD(EXIT, W_CLI|W_SRV, "Debug exit events (levels 1-3)"), - DEBUG_WORD(FILTER, W_SND|W_REC, "Debug filter actions (levels 1-2)"), + DEBUG_WORD(FILTER, W_SND|W_REC, "Debug filter actions (levels 1-3)"), DEBUG_WORD(FLIST, W_SND|W_REC, "Debug file-list operations (levels 1-4)"), DEBUG_WORD(FUZZY, W_REC, "Debug fuzzy scoring (levels 1-2)"), DEBUG_WORD(GENR, W_REC, "Debug generator functions"), @@ -824,7 +825,7 @@ enum {OPT_VERSION = 1000, OPT_DAEMON, OPT_SENDER, OPT_EXCLUDE, OPT_EXCLUDE_FROM, OPT_INCLUDE, OPT_INCLUDE_FROM, OPT_MODIFY_WINDOW, OPT_MIN_SIZE, OPT_CHMOD, OPT_READ_BATCH, OPT_WRITE_BATCH, OPT_ONLY_WRITE_BATCH, OPT_MAX_SIZE, OPT_NO_D, OPT_APPEND, OPT_NO_ICONV, OPT_INFO, OPT_DEBUG, - OPT_USERMAP, OPT_GROUPMAP, OPT_CHOWN, OPT_BWLIMIT, + OPT_USERMAP, OPT_GROUPMAP, OPT_CHOWN, OPT_BWLIMIT, OPT_OLD_ARGS, OPT_SERVER, OPT_REFUSED_BASE = 9000}; static struct poptOption long_options[] = { @@ -1011,6 +1012,8 @@ static struct poptOption long_options[] = { {"files-from", 0, POPT_ARG_STRING, &files_from, 0, 0, 0 }, {"from0", '0', POPT_ARG_VAL, &eol_nulls, 1, 0, 0}, {"no-from0", 0, POPT_ARG_VAL, &eol_nulls, 0, 0, 0}, + {"old-args", 0, POPT_ARG_NONE, 0, OPT_OLD_ARGS, 0, 0}, + {"no-old-args", 0, POPT_ARG_VAL, &old_style_args, 0, 0, 0}, {"protect-args", 's', POPT_ARG_VAL, &protect_args, 1, 0, 0}, {"no-protect-args", 0, POPT_ARG_VAL, &protect_args, 0, 0, 0}, {"no-s", 0, POPT_ARG_VAL, &protect_args, 0, 0, 0}, @@ -1577,6 +1580,13 @@ int parse_arguments(int *argc_p, const char ***argv_p) do_compression++; break; + case OPT_OLD_ARGS: + if (old_style_args <= 0) + old_style_args = 1; + else + old_style_args++; + break; + case 'M': arg = poptGetOptArg(pc); if (*arg != '-') { @@ -1829,6 +1839,21 @@ int parse_arguments(int *argc_p, const char ***argv_p) } } + if (old_style_args < 0) { + if (!am_server && protect_args <= 0 && (arg = getenv("RSYNC_OLD_ARGS")) != NULL && *arg) { + protect_args = 0; + old_style_args = atoi(arg); + } else + old_style_args = 0; + } else if (old_style_args) { + if (protect_args > 0) { + snprintf(err_buf, sizeof err_buf, + "--protect-args conflicts with --old-args.\n"); + return 0; + } + protect_args = 0; + } + if (protect_args < 0) { if (am_server) protect_args = 0; @@ -2381,6 +2406,71 @@ int parse_arguments(int *argc_p, const char ***argv_p) } +static char SPLIT_ARG_WHEN_OLD[1]; + +/** + * Do backslash quoting of any weird chars in "arg", append the resulting + * string to the end of the "opt" (which gets a "=" appended if it is not + * an empty or NULL string), and return the (perhaps malloced) result. + * If opt is NULL, arg is considered a filename arg that allows wildcards. + * If it is "" or any other value, it is considered an option. + **/ +char *safe_arg(const char *opt, const char *arg) +{ +#define SHELL_CHARS "!#$&;|<>(){}\"' \t\\" +#define WILD_CHARS "*?[]" /* We don't allow remote brace expansion */ + BOOL is_filename_arg = !opt; + char *escapes = is_filename_arg ? SHELL_CHARS : WILD_CHARS SHELL_CHARS; + BOOL escape_leading_dash = is_filename_arg && *arg == '-'; + BOOL escape_leading_tilde = 0; + int len1 = opt && *opt ? strlen(opt) + 1 : 0; + int len2 = strlen(arg); + int extras = escape_leading_dash ? 2 : 0; + char *ret; + if (!protect_args && old_style_args < 2 && (!old_style_args || (!is_filename_arg && opt != SPLIT_ARG_WHEN_OLD))) { + const char *f; + if (!old_style_args && *arg == '~' && (relative_paths || !strchr(arg, '/'))) { + extras++; + escape_leading_tilde = 1; + } + for (f = arg; *f; f++) { + if (strchr(escapes, *f)) + extras++; + } + } + if (!len1 && !extras) + return (char*)arg; + ret = new_array(char, len1 + len2 + extras + 1); + if (len1) { + memcpy(ret, opt, len1-1); + ret[len1-1] = '='; + } + if (escape_leading_dash) { + ret[len1++] = '.'; + ret[len1++] = '/'; + extras -= 2; + } + if (!extras) + memcpy(ret + len1, arg, len2); + else { + const char *f = arg; + char *t = ret + len1; + if (escape_leading_tilde) + *t++ = '\\'; + while (*f) { + if (*f == '\\') { + if (!is_filename_arg || !strchr(WILD_CHARS, f[1])) + *t++ = '\\'; + } else if (strchr(escapes, *f)) + *t++ = '\\'; + *t++ = *f++; + } + } + ret[len1+len2+extras] = '\0'; + return ret; +} + + /** * Construct a filtered list of options to pass through from the * client to the server. @@ -2556,9 +2646,7 @@ void server_options(char **args, int *argc_p) set++; else set = iconv_opt; - if (asprintf(&arg, "--iconv=%s", set) < 0) - goto oom; - args[ac++] = arg; + args[ac++] = safe_arg("--iconv", set); } #endif @@ -2625,23 +2713,17 @@ void server_options(char **args, int *argc_p) } if (backup_dir) { + /* This split idiom allows for ~/path expansion via the shell. */ args[ac++] = "--backup-dir"; - args[ac++] = backup_dir; + args[ac++] = safe_arg("", backup_dir); } /* Only send --suffix if it specifies a non-default value. */ - if (strcmp(backup_suffix, backup_dir ? "" : BACKUP_SUFFIX) != 0) { - /* We use the following syntax to avoid weirdness with '~'. */ - if (asprintf(&arg, "--suffix=%s", backup_suffix) < 0) - goto oom; - args[ac++] = arg; - } + if (strcmp(backup_suffix, backup_dir ? "" : BACKUP_SUFFIX) != 0) + args[ac++] = safe_arg("--suffix", backup_suffix); - if (checksum_choice) { - if (asprintf(&arg, "--checksum-choice=%s", checksum_choice) < 0) - goto oom; - args[ac++] = arg; - } + if (checksum_choice) + args[ac++] = safe_arg("--checksum-choice", checksum_choice); if (am_sender) { if (max_delete > 0) { @@ -2650,14 +2732,10 @@ void server_options(char **args, int *argc_p) args[ac++] = arg; } else if (max_delete == 0) args[ac++] = "--max-delete=-1"; - if (min_size >= 0) { - args[ac++] = "--min-size"; - args[ac++] = min_size_arg; - } - if (max_size >= 0) { - args[ac++] = "--max-size"; - args[ac++] = max_size_arg; - } + if (min_size >= 0) + args[ac++] = safe_arg("--min-size", min_size_arg); + if (max_size >= 0) + args[ac++] = safe_arg("--max-size", max_size_arg); if (delete_before) args[ac++] = "--delete-before"; else if (delete_during == 2) @@ -2681,11 +2759,8 @@ void server_options(char **args, int *argc_p) if (do_stats) args[ac++] = "--stats"; } else { - if (skip_compress) { - if (asprintf(&arg, "--skip-compress=%s", skip_compress) < 0) - goto oom; - args[ac++] = arg; - } + if (skip_compress) + args[ac++] = safe_arg("--skip-compress", skip_compress); } /* --delete-missing-args needs the cooperation of both sides, but @@ -2711,7 +2786,7 @@ void server_options(char **args, int *argc_p) if (partial_dir && am_sender) { if (partial_dir != tmp_partialdir) { args[ac++] = "--partial-dir"; - args[ac++] = partial_dir; + args[ac++] = safe_arg("", partial_dir); } if (delay_updates) args[ac++] = "--delay-updates"; @@ -2734,17 +2809,11 @@ void server_options(char **args, int *argc_p) args[ac++] = "--use-qsort"; if (am_sender) { - if (usermap) { - if (asprintf(&arg, "--usermap=%s", usermap) < 0) - goto oom; - args[ac++] = arg; - } + if (usermap) + args[ac++] = safe_arg("--usermap", usermap); - if (groupmap) { - if (asprintf(&arg, "--groupmap=%s", groupmap) < 0) - goto oom; - args[ac++] = arg; - } + if (groupmap) + args[ac++] = safe_arg("--groupmap", groupmap); if (ignore_existing) args[ac++] = "--ignore-existing"; @@ -2755,7 +2824,7 @@ void server_options(char **args, int *argc_p) if (tmpdir) { args[ac++] = "--temp-dir"; - args[ac++] = tmpdir; + args[ac++] = safe_arg("", tmpdir); } if (basis_dir[0]) { @@ -2765,7 +2834,7 @@ void server_options(char **args, int *argc_p) */ for (i = 0; i < basis_dir_cnt; i++) { args[ac++] = dest_option; - args[ac++] = basis_dir[i]; + args[ac++] = safe_arg("", basis_dir[i]); } } } @@ -2790,7 +2859,7 @@ void server_options(char **args, int *argc_p) if (files_from && (!am_sender || filesfrom_host)) { if (filesfrom_host) { args[ac++] = "--files-from"; - args[ac++] = files_from; + args[ac++] = safe_arg("", files_from); if (eol_nulls) args[ac++] = "--from0"; } else { @@ -2830,7 +2899,7 @@ void server_options(char **args, int *argc_p) exit_cleanup(RERR_SYNTAX); } for (j = 1; j <= remote_option_cnt; j++) - args[ac++] = (char*)remote_options[j]; + args[ac++] = safe_arg(SPLIT_ARG_WHEN_OLD, remote_options[j]); } *argc_p = ac; diff --git a/clientserver.c b/clientserver.c index e2e2dc0..c18c024 100644 --- a/clientserver.c +++ b/clientserver.c @@ -45,6 +45,7 @@ extern int protocol_version; extern int io_timeout; extern int no_detach; extern int write_batch; +extern int old_style_args; extern int default_af_hint; extern int logfile_format_has_i; extern int logfile_format_has_o_or_i; @@ -255,20 +256,45 @@ int start_inband_exchange(int f_in, int f_out, const char *user, int argc, char sargs[sargc++] = "."; + if (!old_style_args) + snprintf(line, sizeof line, " %.*s/", modlen, modname); + while (argc > 0) { if (sargc >= MAX_ARGS - 1) { arg_overflow: rprintf(FERROR, "internal: args[] overflowed in do_cmd()\n"); exit_cleanup(RERR_SYNTAX); } - if (strncmp(*argv, modname, modlen) == 0 - && argv[0][modlen] == '\0') + if (strncmp(*argv, modname, modlen) == 0 && argv[0][modlen] == '\0') sargs[sargc++] = modname; /* we send "modname/" */ - else if (**argv == '-') { - if (asprintf(sargs + sargc++, "./%s", *argv) < 0) - out_of_memory("start_inband_exchange"); - } else - sargs[sargc++] = *argv; + else { + char *arg = *argv; + int extra_chars = *arg == '-' ? 2 : 0; /* a leading dash needs a "./" prefix. */ + /* If --old-args was not specified, make sure that the arg won't split at a mod name! */ + if (!old_style_args && (p = strstr(arg, line)) != NULL) { + do { + extra_chars += 2; + } while ((p = strstr(p+1, line)) != NULL); + } + if (extra_chars) { + char *f = arg; + char *t = arg = new_array(char, strlen(arg) + extra_chars + 1); + if (*f == '-') { + *t++ = '.'; + *t++ = '/'; + } + while (*f) { + if (*f == ' ' && strncmp(f, line, modlen+2) == 0) { + *t++ = '['; + *t++ = *f++; + *t++ = ']'; + } else + *t++ = *f++; + } + *t = '\0'; + } + sargs[sargc++] = arg; + } argv++; argc--; } diff --git a/rsync.1 b/rsync.1 index cf2f573..839f5ad 100644 --- a/rsync.1 +++ b/rsync.1 @@ -197,7 +197,7 @@ or with the hostname omitted. For instance, all these work: .br \f(CWrsync \-av host::modname/file{1,2} host::modname/file3 /dest/\fP .br -\f(CWrsync \-av host::modname/file1 ::modname/file{3,4}\fP +\f(CWrsync \-av host::modname/file1 ::modname/file{3,4} /dest/\fP .RE .PP @@ -211,18 +211,23 @@ examples: .RE .PP -This word\-splitting still works (by default) in the latest rsync, but is -not as easy to use as the first method. -.PP -If you need to transfer a filename that contains whitespace, you can either -specify the \fB\-\-protect\-args\fP (\fB\-s\fP) option, or you\(cq\&ll need to escape -the whitespace in a way that the remote shell will understand. For -instance: -.PP -.RS -\f(CWrsync \-av host:'\&file\e name\e with\e spaces'\& /dest\fP +Starting this version of rsync, filenames are passed to a remote shell +in such a way as to preserve the characters you give it. +Thus, if you ask for a file with spaces in the name, that's what the +remote rsync looks for: +.PP +.RS +\f(CWrsync \-aiv host:'\&a simple file.pdf'\& /dest/\fP .RE +.PP +If you use scripts that have been written to manually apply extra quoting to +the remote rsync args (or to require remote arg splitting), you can ask rsync +to let your script handle the extra escaping. This is done by either adding +the \fB\-\-old\-args\fP option to the rsync runs in the script (which requires +a new rsync) or exporting \fBRSYNC_OLD_ARGS\fP=1 and \fBRSYNC_PROTECT_ARGS\fP=0 +(which works with old or new rsync versions). + .PP .SH "CONNECTING TO AN RSYNC DAEMON" @@ -429,6 +434,7 @@ to the detailed description below for a complete description. \-\-append append data onto shorter files \-\-append\-verify \-\-append w/old data in file checksum \-d, \-\-dirs transfer directories without recursing + \-\-old\-dirs, \-\-old\-d works like --dirs when talking to old rsync \-l, \-\-links copy symlinks as symlinks \-L, \-\-copy\-links transform symlink into referent file/dir \-\-copy\-unsafe\-links only \(dq\&unsafe\(dq\& symlinks are transformed @@ -511,6 +517,7 @@ to the detailed description below for a complete description. \-\-include\-from=FILE read include patterns from FILE \-\-files\-from=FILE read list of source\-file names from FILE \-0, \-\-from0 all *from/filter files are delimited by 0s + \-\-old\-args disable the modern arg-protection idiom \-s, \-\-protect\-args no space\-splitting; wildcard chars only \-\-address=ADDRESS bind address for outgoing socket to daemon \-\-port=PORT specify double\-colon alternate port number @@ -1857,10 +1864,10 @@ Be cautious using this, as it is possible to toggle an option that will cause rsync to have a different idea about what data to expect next over the socket, and that will make it fail in a cryptic fashion. .IP -Note that it is best to use a separate \fB\-\-remote\-option\fP for each option you -want to pass. This makes your useage compatible with the \fB\-\-protect\-args\fP -option. If that option is off, any spaces in your remote options will be split -by the remote shell unless you take steps to protect them. +Note that you should use a separate \fB\-M\fP for each remote option you +want to pass. On older rsync versions, the presence of any spaces in the +remote-option arg could cause it to be split into separate remote args, but +this requires the use of \fB\-\-old\-args\fP in this version of rsync. .IP When performing a local transfer, the \(dq\&local\(dq\& side is the sender and the \(dq\&remote\(dq\& side is the receiver. @@ -2054,32 +2061,64 @@ merged files specified in a \fB\-\-filter\fP rule. It does not affect \fB\-\-cvs\-exclude\fP (since all names read from a .cvsignore file are split on whitespace). .IP +.IP "\fB\-\-old\-args\fP" +This option tells rsync to stop trying to protect the arg values from +unintended word-splitting or other misinterpretation by using its new +backslash-escape idiom. The newest default is for remote filenames to only +allow wildcards characters to be interpretated by the shell while +protecting other shell-interpreted characters (and the args of options get +even wildcards escaped). The only active wildcard characters on the remote +side are: `*`, `?`, `[`, & `]`. +.IP +If you have a script that wants to use old-style arg splitting in the +filenames, specify this option once. If the remote shell has a problem +with any backslash escapes, specify the option twice. +.IP +You may also control this setting via the RSYNC_OLD_ARGS environment +variable. If it has the value "1", rsync will default to a single-option +setting. If it has the value "2" (or more), rsync will default to a +repeated-option setting. If it is "0", you'll get the default escaping +behavior. The environment is always overridden by manually specified +positive or negative options (the negative is \fB\-\-no\-old\-args\fP). +.IP +Note that this option also disables the extra safety check added in this +version of rsync, +that ensures that a remote sender isn't including extra top-level items in +the file-list that you didn't request. This side-effect is necessary +because we can't know for sure what names to expect when the remote shell +is interpreting the args. +.IP +This option conflicts with the \fB\-\-protect\-args\fP option. +.IP .IP "\fB\-s, \-\-protect\-args\fP" -This option sends all filenames and most options to -the remote rsync without allowing the remote shell to interpret them. This -means that spaces are not split in names, and any non\-wildcard special -characters are not translated (such as ~, $, ;, &, etc.). Wildcards are -expanded on the remote host by rsync (instead of the shell doing it). +This option sends all filenames and most options to the remote rsync +without allowing the remote shell to interpret them. Wildcards are +expanded on the remote host by rsync instead of the shell doing it. +.IP +This is similar to the new-style backslash-escaping of args that was added +in this version of rsync, but supports some extra features and doesn't +rely on backslash escaping in the remote shell. .IP If you use this option with \fB\-\-iconv\fP, the args related to the remote side will also be translated from the local to the remote character\-set. The translation happens before wild\-cards are expanded. See also the \fB\-\-files\-from\fP option. .IP -You may also control this option via the RSYNC_PROTECT_ARGS environment -variable. If this variable has a non\-zero value, this option will be enabled +You may also control this setting via the RSYNC_PROTECT_ARGS environment +variable. If it has a non-zero value, this setting will be enabled by default, otherwise it will be disabled by default. Either state is overridden by a manually specified positive or negative version of this option (note that \fB\-\-no\-s\fP and \fB\-\-no\-protect\-args\fP are the negative versions). -Since this option was first introduced in 3.0.0, you\(cq\&ll need to make sure it\(cq\&s -disabled if you ever need to interact with a remote rsync that is older than -that. -.IP -Rsync can also be configured (at build time) to have this option enabled by -default (with is overridden by both the environment and the command\-line). -This option will eventually become a new default setting at some -as\-yet\-undetermined point in the future. +This environment variable is also superseded by a non-zero \fBRSYNC_OLD_ARGS\fP export. .IP +You may need to disable this option when interacting with an older rsync +(one prior to 3.0.0). +.IP +This option conflicts with the \fB\-\-old\-args\fP option. +.IP +Note that this option is incompatible with the use of the restricted rsync +script (`rrsync`) since it hides options from the script's inspection. +.IP .IP "\fB\-T, \-\-temp\-dir=DIR\fP" This option instructs rsync to use DIR as a scratch directory when creating temporary copies of the files transferred @@ -2371,7 +2410,11 @@ as a super\-user (see also the \fB\-\-fake\-super\fP option). For the \fB\-\-gr option to have any effect, the \fB\-g\fP (\fB\-\-groups\fP) option must be used (or implied), and the receiver will need to have permissions to set that group. -.IP +.IP +An older rsync client may need to use \fB\-\-protect\-args\fP (\fB\-s\fP) +to avoid a complaint about wildcard characters, but a modern rsync handles +this automatically. +.IP .IP "\fB\-\-chown=USER:GROUP\fP" This option forces all files to be owned by USER with group GROUP. This is a simpler interface than using \fB\-\-usermap\fP and @@ -2382,6 +2425,10 @@ be omitted, but if USER is empty, a leading colon must be supplied. .IP If you specify \(dq\&\-\-chown=foo:bar, this is exactly the same as specifying \(dq\&\-\-usermap=*:foo \-\-groupmap=*:bar\(dq\&, only easier. +.IP +An older rsync client may need to use \fB\-\-protect\-args\fP (\fB\-s\fP) to avoid a +complaint about wildcard characters, but a modern rsync handles this +automatically. .IP .IP "\fB\-\-timeout=TIMEOUT\fP" This option allows you to set a maximum I/O @@ -3983,10 +4030,24 @@ more details. .IP "\fBRSYNC_ICONV\fP" Specify a default \fB\-\-iconv\fP setting using this environment variable. (First supported in 3.0.0.) +.IP "\fBRSYNC_OLD_ARGS\fP" +Specify a "1" if you want the \fB\-\-old\-args\fP option to be enabled by default, +a "2" (or more) if you want it to be enabled in the option-repeated state, +or a "0" to make sure that it is disabled by default. When this environment +variable is set to a non-zero value, it supersedes the \fBRSYNC_PROTECT_ARGS\fP +variable. +.IP +This variable is ignored if \fB\-\-old\-args\fP, \fB\-\-no\-old\-args\fP, or +\fB\-\-protect\-args\fP is specified on the command line. .IP "\fBRSYNC_PROTECT_ARGS\fP" Specify a non\-zero numeric value if you want the \fB\-\-protect\-args\fP option to be enabled by default, or a zero value to make sure that it is disabled by default. (First supported in 3.1.0.) +.IP +This variable is ignored if \fB\-\-protect\-args\fP, \fB\-\-no\-protect\-args\fP, +or \fB\-\-old\-args\fP is specified on the command line. +.IP +This variable is ignored if \fBRSYNC_OLD_ARGS\fP is set to a non-zero value. .IP "\fBRSYNC_RSH\fP" The RSYNC_RSH environment variable allows you to override the default shell used as the transport for rsync. Command line