/* vi: set sw=4 ts=4: */ /* * A prototype Bourne shell grammar parser. * Intended to follow the original Thompson and Ritchie * "small and simple is beautiful" philosophy, which * incidentally is a good match to today's BusyBox. * * Copyright (C) 2000,2001 Larry Doolittle * Copyright (C) 2008,2009 Denys Vlasenko * * Credits: * The parser routines proper are all original material, first * written Dec 2000 and Jan 2001 by Larry Doolittle. The * execution engine, the builtins, and much of the underlying * support has been adapted from busybox-0.49pre's lash, which is * Copyright (C) 1999-2004 by Erik Andersen * written by Erik Andersen . That, in turn, * is based in part on ladsh.c, by Michael K. Johnson and Erik W. * Troan, which they placed in the public domain. I don't know * how much of the Johnson/Troan code has survived the repeated * rewrites. * * Other credits: * o_addchr derived from similar w_addchar function in glibc-2.2. * parse_redirect, redirect_opt_num, and big chunks of main * and many builtins derived from contributions by Erik Andersen. * Miscellaneous bugfixes from Matt Kraai. * * There are two big (and related) architecture differences between * this parser and the lash parser. One is that this version is * actually designed from the ground up to understand nearly all * of the Bourne grammar. The second, consequential change is that * the parser and input reader have been turned inside out. Now, * the parser is in control, and asks for input as needed. The old * way had the input reader in control, and it asked for parsing to * take place as needed. The new way makes it much easier to properly * handle the recursion implicit in the various substitutions, especially * across continuation lines. * * TODOs: * grep for "TODO" and fix (some of them are easy) * special variables (done: PWD, PPID, RANDOM) * tilde expansion * aliases * follow IFS rules more precisely, including update semantics * builtins mandated by standards we don't support: * [un]alias, command, fc, getopts, newgrp, readonly, times * make complex ${var%...} constructs support optional * make here documents optional * * Bash compat TODO: * redirection of stdout+stderr: &> and >& * brace expansion: one/{two,three,four} * reserved words: function select * advanced test: [[ ]] * process substitution: <(list) and >(list) * =~: regex operator * let EXPR [EXPR...] * Each EXPR is an arithmetic expression (ARITHMETIC EVALUATION) * If the last arg evaluates to 0, let returns 1; 0 otherwise. * NB: let `echo 'a=a + 1'` - error (IOW: multi-word expansion is used) * ((EXPR)) * The EXPR is evaluated according to ARITHMETIC EVALUATION. * This is exactly equivalent to let "EXPR". * $[EXPR]: synonym for $((EXPR)) * export builtin should be special, its arguments are assignments * and therefore expansion of them should be "one-word" expansion: * $ export i=`echo 'a b'` # export has one arg: "i=a b" * compare with: * $ ls i=`echo 'a b'` # ls has two args: "i=a" and "b" * ls: cannot access i=a: No such file or directory * ls: cannot access b: No such file or directory * Note1: same applies to local builtin. * Note2: bash 3.2.33(1) does this only if export word itself * is not quoted: * $ export i=`echo 'aaa bbb'`; echo "$i" * aaa bbb * $ "export" i=`echo 'aaa bbb'`; echo "$i" * aaa * * Licensed under GPLv2 or later, see file LICENSE in this source tree. */ #include "busybox.h" /* for APPLET_IS_NOFORK/NOEXEC */ #include /* for malloc_trim */ #include /* #include */ #if ENABLE_HUSH_CASE # include #endif #include "shell_common.h" #include "math.h" #include "match.h" #if ENABLE_HUSH_RANDOM_SUPPORT # include "random.h" #else # define CLEAR_RANDOM_T(rnd) ((void)0) #endif #ifndef PIPE_BUF # define PIPE_BUF 4096 /* amount of buffering in a pipe */ #endif //applet:IF_HUSH(APPLET(hush, _BB_DIR_BIN, _BB_SUID_DROP)) //applet:IF_MSH(APPLET(msh, _BB_DIR_BIN, _BB_SUID_DROP)) //applet:IF_FEATURE_SH_IS_HUSH(APPLET_ODDNAME(sh, hush, _BB_DIR_BIN, _BB_SUID_DROP, sh)) //applet:IF_FEATURE_BASH_IS_HUSH(APPLET_ODDNAME(bash, hush, _BB_DIR_BIN, _BB_SUID_DROP, bash)) //kbuild:lib-$(CONFIG_HUSH) += hush.o match.o shell_common.o //kbuild:lib-$(CONFIG_HUSH_RANDOM_SUPPORT) += random.o //config:config HUSH //config: bool "hush" //config: default y //config: help //config: hush is a small shell (25k). It handles the normal flow control //config: constructs such as if/then/elif/else/fi, for/in/do/done, while loops, //config: case/esac. Redirections, here documents, $((arithmetic)) //config: and functions are supported. //config: //config: It will compile and work on no-mmu systems. //config: //config: It does not handle select, aliases, brace expansion, //config: tilde expansion, &>file and >&file redirection of stdout+stderr. //config: //config:config HUSH_BASH_COMPAT //config: bool "bash-compatible extensions" //config: default y //config: depends on HUSH //config: help //config: Enable bash-compatible extensions. //config: //config:config HUSH_HELP //config: bool "help builtin" //config: default y //config: depends on HUSH //config: help //config: Enable help builtin in hush. Code size + ~1 kbyte. //config: //config:config HUSH_INTERACTIVE //config: bool "Interactive mode" //config: default y //config: depends on HUSH //config: help //config: Enable interactive mode (prompt and command editing). //config: Without this, hush simply reads and executes commands //config: from stdin just like a shell script from a file. //config: No prompt, no PS1/PS2 magic shell variables. //config: //config:config HUSH_JOB //config: bool "Job control" //config: default y //config: depends on HUSH_INTERACTIVE //config: help //config: Enable job control: Ctrl-Z backgrounds, Ctrl-C interrupts current //config: command (not entire shell), fg/bg builtins work. Without this option, //config: "cmd &" still works by simply spawning a process and immediately //config: prompting for next command (or executing next command in a script), //config: but no separate process group is formed. //config: //config:config HUSH_TICK //config: bool "Process substitution" //config: default y //config: depends on HUSH //config: help //config: Enable process substitution `command` and $(command) in hush. //config: //config:config HUSH_IF //config: bool "Support if/then/elif/else/fi" //config: default y //config: depends on HUSH //config: help //config: Enable if/then/elif/else/fi in hush. //config: //config:config HUSH_LOOPS //config: bool "Support for, while and until loops" //config: default y //config: depends on HUSH //config: help //config: Enable for, while and until loops in hush. //config: //config:config HUSH_CASE //config: bool "Support case ... esac statement" //config: default y //config: depends on HUSH //config: help //config: Enable case ... esac statement in hush. +400 bytes. //config: //config:config HUSH_FUNCTIONS //config: bool "Support funcname() { commands; } syntax" //config: default y //config: depends on HUSH //config: help //config: Enable support for shell functions in hush. +800 bytes. //config: //config:config HUSH_LOCAL //config: bool "Support local builtin" //config: default y //config: depends on HUSH_FUNCTIONS //config: help //config: Enable support for local variables in functions. //config: //config:config HUSH_RANDOM_SUPPORT //config: bool "Pseudorandom generator and $RANDOM variable" //config: default y //config: depends on HUSH //config: help //config: Enable pseudorandom generator and dynamic variable "$RANDOM". //config: Each read of "$RANDOM" will generate a new pseudorandom value. //config: //config:config HUSH_EXPORT_N //config: bool "Support 'export -n' option" //config: default y //config: depends on HUSH //config: help //config: export -n unexports variables. It is a bash extension. //config: //config:config HUSH_MODE_X //config: bool "Support 'hush -x' option and 'set -x' command" //config: default y //config: depends on HUSH //config: help //config: This instructs hush to print commands before execution. //config: Adds ~300 bytes. //config: //config:config MSH //config: bool "msh (deprecated: aliased to hush)" //config: default n //config: select HUSH //config: help //config: msh is deprecated and will be removed, please migrate to hush. //config: //usage:#define hush_trivial_usage NOUSAGE_STR //usage:#define hush_full_usage "" //usage:#define msh_trivial_usage NOUSAGE_STR //usage:#define msh_full_usage "" /* Build knobs */ #define LEAK_HUNTING 0 #define BUILD_AS_NOMMU 0 /* Enable/disable sanity checks. Ok to enable in production, * only adds a bit of bloat. Set to >1 to get non-production level verbosity. * Keeping 1 for now even in released versions. */ #define HUSH_DEBUG 1 /* Slightly bigger (+200 bytes), but faster hush. * So far it only enables a trick with counting SIGCHLDs and forks, * which allows us to do fewer waitpid's. * (we can detect a case where neither forks were done nor SIGCHLDs happened * and therefore waitpid will return the same result as last time) */ #define ENABLE_HUSH_FAST 0 /* TODO: implement simplified code for users which do not need ${var%...} ops * So far ${var%...} ops are always enabled: */ #define ENABLE_HUSH_DOLLAR_OPS 1 #if BUILD_AS_NOMMU # undef BB_MMU # undef USE_FOR_NOMMU # undef USE_FOR_MMU # define BB_MMU 0 # define USE_FOR_NOMMU(...) __VA_ARGS__ # define USE_FOR_MMU(...) #endif #include "NUM_APPLETS.h" #if NUM_APPLETS == 1 /* STANDALONE does not make sense, and won't compile */ # undef CONFIG_FEATURE_SH_STANDALONE # undef ENABLE_FEATURE_SH_STANDALONE # undef IF_FEATURE_SH_STANDALONE # undef IF_NOT_FEATURE_SH_STANDALONE # define ENABLE_FEATURE_SH_STANDALONE 0 # define IF_FEATURE_SH_STANDALONE(...) # define IF_NOT_FEATURE_SH_STANDALONE(...) __VA_ARGS__ #endif #if !ENABLE_HUSH_INTERACTIVE # undef ENABLE_FEATURE_EDITING # define ENABLE_FEATURE_EDITING 0 # undef ENABLE_FEATURE_EDITING_FANCY_PROMPT # define ENABLE_FEATURE_EDITING_FANCY_PROMPT 0 #endif /* Do we support ANY keywords? */ #if ENABLE_HUSH_IF || ENABLE_HUSH_LOOPS || ENABLE_HUSH_CASE # define HAS_KEYWORDS 1 # define IF_HAS_KEYWORDS(...) __VA_ARGS__ # define IF_HAS_NO_KEYWORDS(...) #else # define HAS_KEYWORDS 0 # define IF_HAS_KEYWORDS(...) # define IF_HAS_NO_KEYWORDS(...) __VA_ARGS__ #endif /* If you comment out one of these below, it will be #defined later * to perform debug printfs to stderr: */ #define debug_printf(...) do {} while (0) /* Finer-grained debug switches */ #define debug_printf_parse(...) do {} while (0) #define debug_print_tree(a, b) do {} while (0) #define debug_printf_exec(...) do {} while (0) #define debug_printf_env(...) do {} while (0) #define debug_printf_jobs(...) do {} while (0) #define debug_printf_expand(...) do {} while (0) #define debug_printf_varexp(...) do {} while (0) #define debug_printf_glob(...) do {} while (0) #define debug_printf_list(...) do {} while (0) #define debug_printf_subst(...) do {} while (0) #define debug_printf_clean(...) do {} while (0) #define ERR_PTR ((void*)(long)1) #define JOB_STATUS_FORMAT "[%d] %-22s %.40s\n" #define _SPECIAL_VARS_STR "_*@$!?#" #define SPECIAL_VARS_STR ("_*@$!?#" + 1) #define NUMERIC_SPECVARS_STR ("_*@$!?#" + 3) #if ENABLE_HUSH_BASH_COMPAT /* Support / and // replace ops */ /* Note that // is stored as \ in "encoded" string representation */ # define VAR_ENCODED_SUBST_OPS "\\/%#:-=+?" # define VAR_SUBST_OPS ("\\/%#:-=+?" + 1) # define MINUS_PLUS_EQUAL_QUESTION ("\\/%#:-=+?" + 5) #else # define VAR_ENCODED_SUBST_OPS "%#:-=+?" # define VAR_SUBST_OPS "%#:-=+?" # define MINUS_PLUS_EQUAL_QUESTION ("%#:-=+?" + 3) #endif #define SPECIAL_VAR_SYMBOL 3 struct variable; static const char hush_version_str[] ALIGN1 = "HUSH_VERSION="BB_VER; /* This supports saving pointers malloced in vfork child, * to be freed in the parent. */ #if !BB_MMU typedef struct nommu_save_t { char **new_env; struct variable *old_vars; char **argv; char **argv_from_re_execing; } nommu_save_t; #endif typedef enum reserved_style { RES_NONE = 0, #if ENABLE_HUSH_IF RES_IF , RES_THEN , RES_ELIF , RES_ELSE , RES_FI , #endif #if ENABLE_HUSH_LOOPS RES_FOR , RES_WHILE , RES_UNTIL , RES_DO , RES_DONE , #endif #if ENABLE_HUSH_LOOPS || ENABLE_HUSH_CASE RES_IN , #endif #if ENABLE_HUSH_CASE RES_CASE , /* three pseudo-keywords support contrived "case" syntax: */ RES_CASE_IN, /* "case ... IN", turns into RES_MATCH when IN is observed */ RES_MATCH , /* "word)" */ RES_CASE_BODY, /* "this command is inside CASE" */ RES_ESAC , #endif RES_XXXX , RES_SNTX } reserved_style; typedef struct o_string { char *data; int length; /* position where data is appended */ int maxlen; /* Protect newly added chars against globbing * (by prepending \ to *, ?, [, \) */ smallint o_escape; smallint o_glob; /* At least some part of the string was inside '' or "", * possibly empty one: word"", wo''rd etc. */ smallint has_quoted_part; smallint has_empty_slot; smallint o_assignment; /* 0:maybe, 1:yes, 2:no */ } o_string; enum { MAYBE_ASSIGNMENT = 0, DEFINITELY_ASSIGNMENT = 1, NOT_ASSIGNMENT = 2, WORD_IS_KEYWORD = 3, /* not assigment, but next word may be: "if v=xyz cmd;" */ }; /* Used for initialization: o_string foo = NULL_O_STRING; */ #define NULL_O_STRING { NULL } /* I can almost use ordinary FILE*. Is open_memstream() universally * available? Where is it documented? */ typedef struct in_str { const char *p; /* eof_flag=1: last char in ->p is really an EOF */ char eof_flag; /* meaningless if ->p == NULL */ char peek_buf[2]; #if ENABLE_HUSH_INTERACTIVE smallint promptme; smallint promptmode; /* 0: PS1, 1: PS2 */ #endif FILE *file; int (*get) (struct in_str *) FAST_FUNC; int (*peek) (struct in_str *) FAST_FUNC; } in_str; #define i_getch(input) ((input)->get(input)) #define i_peek(input) ((input)->peek(input)) /* The descrip member of this structure is only used to make * debugging output pretty */ static const struct { int mode; signed char default_fd; char descrip[3]; } redir_table[] = { { O_RDONLY, 0, "<" }, { O_CREAT|O_TRUNC|O_WRONLY, 1, ">" }, { O_CREAT|O_APPEND|O_WRONLY, 1, ">>" }, { O_CREAT|O_RDWR, 1, "<>" }, { O_RDONLY, 0, "<<" }, /* Should not be needed. Bogus default_fd helps in debugging */ /* { O_RDONLY, 77, "<<" }, */ }; struct redir_struct { struct redir_struct *next; char *rd_filename; /* filename */ int rd_fd; /* fd to redirect */ /* fd to redirect to, or -3 if rd_fd is to be closed (n>&-) */ int rd_dup; smallint rd_type; /* (enum redir_type) */ /* note: for heredocs, rd_filename contains heredoc delimiter, * and subsequently heredoc itself; and rd_dup is a bitmask: * bit 0: do we need to trim leading tabs? * bit 1: is heredoc quoted (<<'delim' syntax) ? */ }; typedef enum redir_type { REDIRECT_INPUT = 0, REDIRECT_OVERWRITE = 1, REDIRECT_APPEND = 2, REDIRECT_IO = 3, REDIRECT_HEREDOC = 4, REDIRECT_HEREDOC2 = 5, /* REDIRECT_HEREDOC after heredoc is loaded */ REDIRFD_CLOSE = -3, REDIRFD_SYNTAX_ERR = -2, REDIRFD_TO_FILE = -1, /* otherwise, rd_fd is redirected to rd_dup */ HEREDOC_SKIPTABS = 1, HEREDOC_QUOTED = 2, } redir_type; struct command { pid_t pid; /* 0 if exited */ int assignment_cnt; /* how many argv[i] are assignments? */ smallint is_stopped; /* is the command currently running? */ smallint cmd_type; /* CMD_xxx */ #define CMD_NORMAL 0 #define CMD_SUBSHELL 1 #if ENABLE_HUSH_BASH_COMPAT /* used for "[[ EXPR ]]" */ # define CMD_SINGLEWORD_NOGLOB 2 #endif #if ENABLE_HUSH_FUNCTIONS # define CMD_FUNCDEF 3 #endif /* if non-NULL, this "command" is { list }, ( list ), or a compound statement */ struct pipe *group; #if !BB_MMU char *group_as_string; #endif #if ENABLE_HUSH_FUNCTIONS struct function *child_func; /* This field is used to prevent a bug here: * while...do f1() {a;}; f1; f1() {b;}; f1; done * When we execute "f1() {a;}" cmd, we create new function and clear * cmd->group, cmd->group_as_string, cmd->argv[0]. * When we execute "f1() {b;}", we notice that f1 exists, * and that its "parent cmd" struct is still "alive", * we put those fields back into cmd->xxx * (struct function has ->parent_cmd ptr to facilitate that). * When we loop back, we can execute "f1() {a;}" again and set f1 correctly. * Without this trick, loop would execute a;b;b;b;... * instead of correct sequence a;b;a;b;... * When command is freed, it severs the link * (sets ->child_func->parent_cmd to NULL). */ #endif char **argv; /* command name and arguments */ /* argv vector may contain variable references (^Cvar^C, ^C0^C etc) * and on execution these are substituted with their values. * Substitution can make _several_ words out of one argv[n]! * Example: argv[0]=='.^C*^C.' here: echo .$*. * References of the form ^C`cmd arg^C are `cmd arg` substitutions. */ struct redir_struct *redirects; /* I/O redirections */ }; /* Is there anything in this command at all? */ #define IS_NULL_CMD(cmd) \ (!(cmd)->group && !(cmd)->argv && !(cmd)->redirects) struct pipe { struct pipe *next; int num_cmds; /* total number of commands in pipe */ int alive_cmds; /* number of commands running (not exited) */ int stopped_cmds; /* number of commands alive, but stopped */ #if ENABLE_HUSH_JOB int jobid; /* job number */ pid_t pgrp; /* process group ID for the job */ char *cmdtext; /* name of job */ #endif struct command *cmds; /* array of commands in pipe */ smallint followup; /* PIPE_BG, PIPE_SEQ, PIPE_OR, PIPE_AND */ IF_HAS_KEYWORDS(smallint pi_inverted;) /* "! cmd | cmd" */ IF_HAS_KEYWORDS(smallint res_word;) /* needed for if, for, while, until... */ }; typedef enum pipe_style { PIPE_SEQ = 1, PIPE_AND = 2, PIPE_OR = 3, PIPE_BG = 4, } pipe_style; /* Is there anything in this pipe at all? */ #define IS_NULL_PIPE(pi) \ ((pi)->num_cmds == 0 IF_HAS_KEYWORDS( && (pi)->res_word == RES_NONE)) /* This holds pointers to the various results of parsing */ struct parse_context { /* linked list of pipes */ struct pipe *list_head; /* last pipe (being constructed right now) */ struct pipe *pipe; /* last command in pipe (being constructed right now) */ struct command *command; /* last redirect in command->redirects list */ struct redir_struct *pending_redirect; #if !BB_MMU o_string as_string; #endif #if HAS_KEYWORDS smallint ctx_res_w; smallint ctx_inverted; /* "! cmd | cmd" */ #if ENABLE_HUSH_CASE smallint ctx_dsemicolon; /* ";;" seen */ #endif /* bitmask of FLAG_xxx, for figuring out valid reserved words */ int old_flag; /* group we are enclosed in: * example: "if pipe1; pipe2; then pipe3; fi" * when we see "if" or "then", we malloc and copy current context, * and make ->stack point to it. then we parse pipeN. * when closing "then" / fi" / whatever is found, * we move list_head into ->stack->command->group, * copy ->stack into current context, and delete ->stack. * (parsing of { list } and ( list ) doesn't use this method) */ struct parse_context *stack; #endif }; /* On program start, environ points to initial environment. * putenv adds new pointers into it, unsetenv removes them. * Neither of these (de)allocates the strings. * setenv allocates new strings in malloc space and does putenv, * and thus setenv is unusable (leaky) for shell's purposes */ #define setenv(...) setenv_is_leaky_dont_use() struct variable { struct variable *next; char *varstr; /* points to "name=" portion */ #if ENABLE_HUSH_LOCAL unsigned func_nest_level; #endif int max_len; /* if > 0, name is part of initial env; else name is malloced */ smallint flg_export; /* putenv should be done on this var */ smallint flg_read_only; }; enum { BC_BREAK = 1, BC_CONTINUE = 2, }; #if ENABLE_HUSH_FUNCTIONS struct function { struct function *next; char *name; struct command *parent_cmd; struct pipe *body; # if !BB_MMU char *body_as_string; # endif }; #endif /* "Globals" within this file */ /* Sorted roughly by size (smaller offsets == smaller code) */ struct globals { /* interactive_fd != 0 means we are an interactive shell. * If we are, then saved_tty_pgrp can also be != 0, meaning * that controlling tty is available. With saved_tty_pgrp == 0, * job control still works, but terminal signals * (^C, ^Z, ^Y, ^\) won't work at all, and background * process groups can only be created with "cmd &". * With saved_tty_pgrp != 0, hush will use tcsetpgrp() * to give tty to the foreground process group, * and will take it back when the group is stopped (^Z) * or killed (^C). */ #if ENABLE_HUSH_INTERACTIVE /* 'interactive_fd' is a fd# open to ctty, if we have one * _AND_ if we decided to act interactively */ int interactive_fd; const char *PS1; const char *PS2; # define G_interactive_fd (G.interactive_fd) #else # define G_interactive_fd 0 #endif #if ENABLE_FEATURE_EDITING line_input_t *line_input_state; #endif pid_t root_pid; pid_t root_ppid; pid_t last_bg_pid; #if ENABLE_HUSH_RANDOM_SUPPORT random_t random_gen; #endif #if ENABLE_HUSH_JOB int run_list_level; int last_jobid; pid_t saved_tty_pgrp; struct pipe *job_list; # define G_saved_tty_pgrp (G.saved_tty_pgrp) #else # define G_saved_tty_pgrp 0 #endif smallint flag_SIGINT; #if ENABLE_HUSH_LOOPS smallint flag_break_continue; #endif #if ENABLE_HUSH_FUNCTIONS /* 0: outside of a function (or sourced file) * -1: inside of a function, ok to use return builtin * 1: return is invoked, skip all till end of func */ smallint flag_return_in_progress; #endif smallint n_mode; #if ENABLE_HUSH_MODE_X smallint x_mode; # define G_x_mode (G.x_mode) #else # define G_x_mode 0 #endif smallint exiting; /* used to prevent EXIT trap recursion */ /* These four support $?, $#, and $1 */ smalluint last_exitcode; /* are global_argv and global_argv[1..n] malloced? (note: not [0]) */ smalluint global_args_malloced; smalluint inherited_set_is_saved; /* how many non-NULL argv's we have. NB: $# + 1 */ int global_argc; char **global_argv; #if !BB_MMU char *argv0_for_re_execing; #endif #if ENABLE_HUSH_LOOPS unsigned depth_break_continue; unsigned depth_of_loop; #endif const char *ifs; const char *cwd; struct variable *top_var; /* = &G.shell_ver (set in main()) */ struct variable shell_ver; char **expanded_assignments; #if ENABLE_HUSH_FUNCTIONS struct function *top_func; # if ENABLE_HUSH_LOCAL struct variable **shadowed_vars_pp; unsigned func_nest_level; # endif #endif /* Signal and trap handling */ #if ENABLE_HUSH_FAST unsigned count_SIGCHLD; unsigned handled_SIGCHLD; smallint we_have_children; #endif /* which signals have non-DFL handler (even with no traps set)? */ unsigned non_DFL_mask; char **traps; /* char *traps[NSIG] */ sigset_t blocked_set; sigset_t inherited_set; #if HUSH_DEBUG unsigned long memleak_value; int debug_indent; #endif char user_input_buf[ENABLE_FEATURE_EDITING ? CONFIG_FEATURE_EDITING_MAX_LEN : 2]; }; #define G (*ptr_to_globals) /* Not #defining name to G.name - this quickly gets unwieldy * (too many defines). Also, I actually prefer to see when a variable * is global, thus "G." prefix is a useful hint */ #define INIT_G() do { \ SET_PTR_TO_GLOBALS(xzalloc(sizeof(G))); \ } while (0) /* Function prototypes for builtins */ static int builtin_cd(char **argv) FAST_FUNC; static int builtin_echo(char **argv) FAST_FUNC; static int builtin_eval(char **argv) FAST_FUNC; static int builtin_exec(char **argv) FAST_FUNC; static int builtin_exit(char **argv) FAST_FUNC; static int builtin_export(char **argv) FAST_FUNC; #if ENABLE_HUSH_JOB static int builtin_fg_bg(char **argv) FAST_FUNC; static int builtin_jobs(char **argv) FAST_FUNC; #endif #if ENABLE_HUSH_HELP static int builtin_help(char **argv) FAST_FUNC; #endif #if ENABLE_HUSH_LOCAL static int builtin_local(char **argv) FAST_FUNC; #endif #if HUSH_DEBUG static int builtin_memleak(char **argv) FAST_FUNC; #endif #if ENABLE_PRINTF static int builtin_printf(char **argv) FAST_FUNC; #endif static int builtin_pwd(char **argv) FAST_FUNC; static int builtin_read(char **argv) FAST_FUNC; static int builtin_set(char **argv) FAST_FUNC; static int builtin_shift(char **argv) FAST_FUNC; static int builtin_source(char **argv) FAST_FUNC; static int builtin_test(char **argv) FAST_FUNC; static int builtin_trap(char **argv) FAST_FUNC; static int builtin_type(char **argv) FAST_FUNC; static int builtin_true(char **argv) FAST_FUNC; static int builtin_umask(char **argv) FAST_FUNC; static int builtin_unset(char **argv) FAST_FUNC; static int builtin_wait(char **argv) FAST_FUNC; #if ENABLE_HUSH_LOOPS static int builtin_break(char **argv) FAST_FUNC; static int builtin_continue(char **argv) FAST_FUNC; #endif #if ENABLE_HUSH_FUNCTIONS static int builtin_return(char **argv) FAST_FUNC; #endif /* Table of built-in functions. They can be forked or not, depending on * context: within pipes, they fork. As simple commands, they do not. * When used in non-forking context, they can change global variables * in the parent shell process. If forked, of course they cannot. * For example, 'unset foo | whatever' will parse and run, but foo will * still be set at the end. */ struct built_in_command { const char *b_cmd; int (*b_function)(char **argv) FAST_FUNC; #if ENABLE_HUSH_HELP const char *b_descr; # define BLTIN(cmd, func, help) { cmd, func, help } #else # define BLTIN(cmd, func, help) { cmd, func } #endif }; static const struct built_in_command bltins1[] = { BLTIN("." , builtin_source , "Run commands in a file"), BLTIN(":" , builtin_true , NULL), #if ENABLE_HUSH_JOB BLTIN("bg" , builtin_fg_bg , "Resume a job in the background"), #endif #if ENABLE_HUSH_LOOPS BLTIN("break" , builtin_break , "Exit from a loop"), #endif BLTIN("cd" , builtin_cd , "Change directory"), #if ENABLE_HUSH_LOOPS BLTIN("continue" , builtin_continue, "Start new loop iteration"), #endif BLTIN("eval" , builtin_eval , "Construct and run shell command"), BLTIN("exec" , builtin_exec , "Execute command, don't return to shell"), BLTIN("exit" , builtin_exit , "Exit"), BLTIN("export" , builtin_export , "Set environment variables"), #if ENABLE_HUSH_JOB BLTIN("fg" , builtin_fg_bg , "Bring job into the foreground"), #endif #if ENABLE_HUSH_HELP BLTIN("help" , builtin_help , NULL), #endif #if ENABLE_HUSH_JOB BLTIN("jobs" , builtin_jobs , "List jobs"), #endif #if ENABLE_HUSH_LOCAL BLTIN("local" , builtin_local , "Set local variables"), #endif #if HUSH_DEBUG BLTIN("memleak" , builtin_memleak , NULL), #endif BLTIN("read" , builtin_read , "Input into variable"), #if ENABLE_HUSH_FUNCTIONS BLTIN("return" , builtin_return , "Return from a function"), #endif BLTIN("set" , builtin_set , "Set/unset positional parameters"), BLTIN("shift" , builtin_shift , "Shift positional parameters"), #if ENABLE_HUSH_BASH_COMPAT BLTIN("source" , builtin_source , "Run commands in a file"), #endif BLTIN("trap" , builtin_trap , "Trap signals"), BLTIN("type" , builtin_type , "Show command type"), BLTIN("ulimit" , shell_builtin_ulimit , "Control resource limits"), BLTIN("umask" , builtin_umask , "Set file creation mask"), BLTIN("unset" , builtin_unset , "Unset variables"), BLTIN("wait" , builtin_wait , "Wait for process"), }; /* For now, echo and test are unconditionally enabled. * Maybe make it configurable? */ static const struct built_in_command bltins2[] = { BLTIN("[" , builtin_test , NULL), BLTIN("echo" , builtin_echo , NULL), #if ENABLE_PRINTF BLTIN("printf" , builtin_printf , NULL), #endif BLTIN("pwd" , builtin_pwd , NULL), BLTIN("test" , builtin_test , NULL), }; /* Debug printouts. */ #if HUSH_DEBUG /* prevent disasters with G.debug_indent < 0 */ # define indent() fprintf(stderr, "%*s", (G.debug_indent * 2) & 0xff, "") # define debug_enter() (G.debug_indent++) # define debug_leave() (G.debug_indent--) #else # define indent() ((void)0) # define debug_enter() ((void)0) # define debug_leave() ((void)0) #endif #ifndef debug_printf # define debug_printf(...) (indent(), fprintf(stderr, __VA_ARGS__)) #endif #ifndef debug_printf_parse # define debug_printf_parse(...) (indent(), fprintf(stderr, __VA_ARGS__)) #endif #ifndef debug_printf_exec #define debug_printf_exec(...) (indent(), fprintf(stderr, __VA_ARGS__)) #endif #ifndef debug_printf_env # define debug_printf_env(...) (indent(), fprintf(stderr, __VA_ARGS__)) #endif #ifndef debug_printf_jobs # define debug_printf_jobs(...) (indent(), fprintf(stderr, __VA_ARGS__)) # define DEBUG_JOBS 1 #else # define DEBUG_JOBS 0 #endif #ifndef debug_printf_expand # define debug_printf_expand(...) (indent(), fprintf(stderr, __VA_ARGS__)) # define DEBUG_EXPAND 1 #else # define DEBUG_EXPAND 0 #endif #ifndef debug_printf_varexp # define debug_printf_varexp(...) (indent(), fprintf(stderr, __VA_ARGS__)) #endif #ifndef debug_printf_glob # define debug_printf_glob(...) (indent(), fprintf(stderr, __VA_ARGS__)) # define DEBUG_GLOB 1 #else # define DEBUG_GLOB 0 #endif #ifndef debug_printf_list # define debug_printf_list(...) (indent(), fprintf(stderr, __VA_ARGS__)) #endif #ifndef debug_printf_subst # define debug_printf_subst(...) (indent(), fprintf(stderr, __VA_ARGS__)) #endif #ifndef debug_printf_clean # define debug_printf_clean(...) (indent(), fprintf(stderr, __VA_ARGS__)) # define DEBUG_CLEAN 1 #else # define DEBUG_CLEAN 0 #endif #if DEBUG_EXPAND static void debug_print_strings(const char *prefix, char **vv) { indent(); fprintf(stderr, "%s:\n", prefix); while (*vv) fprintf(stderr, " '%s'\n", *vv++); } #else # define debug_print_strings(prefix, vv) ((void)0) #endif /* Leak hunting. Use hush_leaktool.sh for post-processing. */ #if LEAK_HUNTING static void *xxmalloc(int lineno, size_t size) { void *ptr = xmalloc((size + 0xff) & ~0xff); fdprintf(2, "line %d: malloc %p\n", lineno, ptr); return ptr; } static void *xxrealloc(int lineno, void *ptr, size_t size) { ptr = xrealloc(ptr, (size + 0xff) & ~0xff); fdprintf(2, "line %d: realloc %p\n", lineno, ptr); return ptr; } static char *xxstrdup(int lineno, const char *str) { char *ptr = xstrdup(str); fdprintf(2, "line %d: strdup %p\n", lineno, ptr); return ptr; } static void xxfree(void *ptr) { fdprintf(2, "free %p\n", ptr); free(ptr); } # define xmalloc(s) xxmalloc(__LINE__, s) # define xrealloc(p, s) xxrealloc(__LINE__, p, s) # define xstrdup(s) xxstrdup(__LINE__, s) # define free(p) xxfree(p) #endif /* Syntax and runtime errors. They always abort scripts. * In interactive use they usually discard unparsed and/or unexecuted commands * and return to the prompt. * HUSH_DEBUG >= 2 prints line number in this file where it was detected. */ #if HUSH_DEBUG < 2 # define die_if_script(lineno, ...) die_if_script(__VA_ARGS__) # define syntax_error(lineno, msg) syntax_error(msg) # define syntax_error_at(lineno, msg) syntax_error_at(msg) # define syntax_error_unterm_ch(lineno, ch) syntax_error_unterm_ch(ch) # define syntax_error_unterm_str(lineno, s) syntax_error_unterm_str(s) # define syntax_error_unexpected_ch(lineno, ch) syntax_error_unexpected_ch(ch) #endif static void die_if_script(unsigned lineno, const char *fmt, ...) { va_list p; #if HUSH_DEBUG >= 2 bb_error_msg("hush.c:%u", lineno); #endif va_start(p, fmt); bb_verror_msg(fmt, p, NULL); va_end(p); if (!G_interactive_fd) xfunc_die(); } static void syntax_error(unsigned lineno, const char *msg) { if (msg) die_if_script(lineno, "syntax error: %s", msg); else die_if_script(lineno, "syntax error", NULL); } static void syntax_error_at(unsigned lineno, const char *msg) { die_if_script(lineno, "syntax error at '%s'", msg); } static void syntax_error_unterm_str(unsigned lineno, const char *s) { die_if_script(lineno, "syntax error: unterminated %s", s); } /* It so happens that all such cases are totally fatal * even if shell is interactive: EOF while looking for closing * delimiter. There is nowhere to read stuff from after that, * it's EOF! The only choice is to terminate. */ static void syntax_error_unterm_ch(unsigned lineno, char ch) NORETURN; static void syntax_error_unterm_ch(unsigned lineno, char ch) { char msg[2] = { ch, '\0' }; syntax_error_unterm_str(lineno, msg); xfunc_die(); } static void syntax_error_unexpected_ch(unsigned lineno, int ch) { char msg[2]; msg[0] = ch; msg[1] = '\0'; die_if_script(lineno, "syntax error: unexpected %s", ch == EOF ? "EOF" : msg); } #if HUSH_DEBUG < 2 # undef die_if_script # undef syntax_error # undef syntax_error_at # undef syntax_error_unterm_ch # undef syntax_error_unterm_str # undef syntax_error_unexpected_ch #else # define die_if_script(...) die_if_script(__LINE__, __VA_ARGS__) # define syntax_error(msg) syntax_error(__LINE__, msg) # define syntax_error_at(msg) syntax_error_at(__LINE__, msg) # define syntax_error_unterm_ch(ch) syntax_error_unterm_ch(__LINE__, ch) # define syntax_error_unterm_str(s) syntax_error_unterm_str(__LINE__, s) # define syntax_error_unexpected_ch(ch) syntax_error_unexpected_ch(__LINE__, ch) #endif #if ENABLE_HUSH_INTERACTIVE static void cmdedit_update_prompt(void); #else # define cmdedit_update_prompt() ((void)0) #endif /* Utility functions */ /* Replace each \x with x in place, return ptr past NUL. */ static char *unbackslash(char *src) { char *dst = src = strchrnul(src, '\\'); while (1) { if (*src == '\\') src++; if ((*dst++ = *src++) == '\0') break; } return dst; } static char **add_strings_to_strings(char **strings, char **add, int need_to_dup) { int i; unsigned count1; unsigned count2; char **v; v = strings; count1 = 0; if (v) { while (*v) { count1++; v++; } } count2 = 0; v = add; while (*v) { count2++; v++; } v = xrealloc(strings, (count1 + count2 + 1) * sizeof(char*)); v[count1 + count2] = NULL; i = count2; while (--i >= 0) v[count1 + i] = (need_to_dup ? xstrdup(add[i]) : add[i]); return v; } #if LEAK_HUNTING static char **xx_add_strings_to_strings(int lineno, char **strings, char **add, int need_to_dup) { char **ptr = add_strings_to_strings(strings, add, need_to_dup); fdprintf(2, "line %d: add_strings_to_strings %p\n", lineno, ptr); return ptr; } #define add_strings_to_strings(strings, add, need_to_dup) \ xx_add_strings_to_strings(__LINE__, strings, add, need_to_dup) #endif /* Note: takes ownership of "add" ptr (it is not strdup'ed) */ static char **add_string_to_strings(char **strings, char *add) { char *v[2]; v[0] = add; v[1] = NULL; return add_strings_to_strings(strings, v, /*dup:*/ 0); } #if LEAK_HUNTING static char **xx_add_string_to_strings(int lineno, char **strings, char *add) { char **ptr = add_string_to_strings(strings, add); fdprintf(2, "line %d: add_string_to_strings %p\n", lineno, ptr); return ptr; } #define add_string_to_strings(strings, add) \ xx_add_string_to_strings(__LINE__, strings, add) #endif static void free_strings(char **strings) { char **v; if (!strings) return; v = strings; while (*v) { free(*v); v++; } free(strings); } /* Helpers for setting new $n and restoring them back */ typedef struct save_arg_t { char *sv_argv0; char **sv_g_argv; int sv_g_argc; smallint sv_g_malloced; } save_arg_t; static void save_and_replace_G_args(save_arg_t *sv, char **argv) { int n; sv->sv_argv0 = argv[0]; sv->sv_g_argv = G.global_argv; sv->sv_g_argc = G.global_argc; sv->sv_g_malloced = G.global_args_malloced; argv[0] = G.global_argv[0]; /* retain $0 */ G.global_argv = argv; G.global_args_malloced = 0; n = 1; while (*++argv) n++; G.global_argc = n; } static void restore_G_args(save_arg_t *sv, char **argv) { char **pp; if (G.global_args_malloced) { /* someone ran "set -- arg1 arg2 ...", undo */ pp = G.global_argv; while (*++pp) /* note: does not free $0 */ free(*pp); free(G.global_argv); } argv[0] = sv->sv_argv0; G.global_argv = sv->sv_g_argv; G.global_argc = sv->sv_g_argc; G.global_args_malloced = sv->sv_g_malloced; } /* Basic theory of signal handling in shell * ======================================== * This does not describe what hush does, rather, it is current understanding * what it _should_ do. If it doesn't, it's a bug. * http://www.opengroup.org/onlinepubs/9699919799/utilities/V3_chap02.html#trap * * Signals are handled only after each pipe ("cmd | cmd | cmd" thing) * is finished or backgrounded. It is the same in interactive and * non-interactive shells, and is the same regardless of whether * a user trap handler is installed or a shell special one is in effect. * ^C or ^Z from keyboard seems to execute "at once" because it usually * backgrounds (i.e. stops) or kills all members of currently running * pipe. * * Wait builtin in interruptible by signals for which user trap is set * or by SIGINT in interactive shell. * * Trap handlers will execute even within trap handlers. (right?) * * User trap handlers are forgotten when subshell ("(cmd)") is entered, * except for handlers set to '' (empty string). * * If job control is off, backgrounded commands ("cmd &") * have SIGINT, SIGQUIT set to SIG_IGN. * * Commands which are run in command substitution ("`cmd`") * have SIGTTIN, SIGTTOU, SIGTSTP set to SIG_IGN. * * Ordinary commands have signals set to SIG_IGN/DFL as inherited * by the shell from its parent. * * Signals which differ from SIG_DFL action * (note: child (i.e., [v]forked) shell is not an interactive shell): * * SIGQUIT: ignore * SIGTERM (interactive): ignore * SIGHUP (interactive): * send SIGCONT to stopped jobs, send SIGHUP to all jobs and exit * SIGTTIN, SIGTTOU, SIGTSTP (if job control is on): ignore * Note that ^Z is handled not by trapping SIGTSTP, but by seeing * that all pipe members are stopped. Try this in bash: * while :; do :; done - ^Z does not background it * (while :; do :; done) - ^Z backgrounds it * SIGINT (interactive): wait for last pipe, ignore the rest * of the command line, show prompt. NB: ^C does not send SIGINT * to interactive shell while shell is waiting for a pipe, * since shell is bg'ed (is not in foreground process group). * Example 1: this waits 5 sec, but does not execute ls: * "echo $$; sleep 5; ls -l" + "kill -INT " * Example 2: this does not wait and does not execute ls: * "echo $$; sleep 5 & wait; ls -l" + "kill -INT " * Example 3: this does not wait 5 sec, but executes ls: * "sleep 5; ls -l" + press ^C * * (What happens to signals which are IGN on shell start?) * (What happens with signal mask on shell start?) * * Implementation in hush * ====================== * We use in-kernel pending signal mask to determine which signals were sent. * We block all signals which we don't want to take action immediately, * i.e. we block all signals which need to have special handling as described * above, and all signals which have traps set. * After each pipe execution, we extract any pending signals via sigtimedwait() * and act on them. * * unsigned non_DFL_mask: a mask of such "special" signals * sigset_t blocked_set: current blocked signal set * * "trap - SIGxxx": * clear bit in blocked_set unless it is also in non_DFL_mask * "trap 'cmd' SIGxxx": * set bit in blocked_set (even if 'cmd' is '') * after [v]fork, if we plan to be a shell: * unblock signals with special interactive handling * (child shell is not interactive), * unset all traps except '' (note: regardless of child shell's type - {}, (), etc) * after [v]fork, if we plan to exec: * POSIX says fork clears pending signal mask in child - no need to clear it. * Restore blocked signal set to one inherited by shell just prior to exec. * * Note: as a result, we do not use signal handlers much. The only uses * are to count SIGCHLDs * and to restore tty pgrp on signal-induced exit. * * Note 2 (compat): * Standard says "When a subshell is entered, traps that are not being ignored * are set to the default actions". bash interprets it so that traps which * are set to '' (ignore) are NOT reset to defaults. We do the same. */ enum { SPECIAL_INTERACTIVE_SIGS = 0 | (1 << SIGTERM) | (1 << SIGINT) | (1 << SIGHUP) , SPECIAL_JOB_SIGS = 0 #if ENABLE_HUSH_JOB | (1 << SIGTTIN) | (1 << SIGTTOU) | (1 << SIGTSTP) #endif }; #if ENABLE_HUSH_FAST static void SIGCHLD_handler(int sig UNUSED_PARAM) { G.count_SIGCHLD++; //bb_error_msg("[%d] SIGCHLD_handler: G.count_SIGCHLD:%d G.handled_SIGCHLD:%d", getpid(), G.count_SIGCHLD, G.handled_SIGCHLD); } #endif #if ENABLE_HUSH_JOB /* After [v]fork, in child: do not restore tty pgrp on xfunc death */ # define disable_restore_tty_pgrp_on_exit() (die_sleep = 0) /* After [v]fork, in parent: restore tty pgrp on xfunc death */ # define enable_restore_tty_pgrp_on_exit() (die_sleep = -1) /* Restores tty foreground process group, and exits. * May be called as signal handler for fatal signal * (will resend signal to itself, producing correct exit state) * or called directly with -EXITCODE. * We also call it if xfunc is exiting. */ static void sigexit(int sig) NORETURN; static void sigexit(int sig) { /* Disable all signals: job control, SIGPIPE, etc. */ sigprocmask_allsigs(SIG_BLOCK); /* Careful: we can end up here after [v]fork. Do not restore * tty pgrp then, only top-level shell process does that */ if (G_saved_tty_pgrp && getpid() == G.root_pid) tcsetpgrp(G_interactive_fd, G_saved_tty_pgrp); /* Not a signal, just exit */ if (sig <= 0) _exit(- sig); kill_myself_with_sig(sig); /* does not return */ } #else # define disable_restore_tty_pgrp_on_exit() ((void)0) # define enable_restore_tty_pgrp_on_exit() ((void)0) #endif /* Restores tty foreground process group, and exits. */ static void hush_exit(int exitcode) NORETURN; static void hush_exit(int exitcode) { if (G.exiting <= 0 && G.traps && G.traps[0] && G.traps[0][0]) { /* Prevent recursion: * trap "echo Hi; exit" EXIT; exit */ char *argv[] = { NULL, G.traps[0], NULL }; G.traps[0] = NULL; G.exiting = 1; builtin_eval(argv); free(argv[1]); } #if ENABLE_HUSH_JOB fflush_all(); sigexit(- (exitcode & 0xff)); #else exit(exitcode); #endif } static int check_and_run_traps(int sig) { static const struct timespec zero_timespec; smalluint save_rcode; int last_sig = 0; if (sig) goto jump_in; while (1) { sig = sigtimedwait(&G.blocked_set, NULL, &zero_timespec); if (sig <= 0) break; jump_in: last_sig = sig; if (G.traps && G.traps[sig]) { if (G.traps[sig][0]) { /* We have user-defined handler */ char *argv[] = { NULL, xstrdup(G.traps[sig]), NULL }; save_rcode = G.last_exitcode; builtin_eval(argv); free(argv[1]); G.last_exitcode = save_rcode; } /* else: "" trap, ignoring signal */ continue; } /* not a trap: special action */ switch (sig) { #if ENABLE_HUSH_FAST case SIGCHLD: G.count_SIGCHLD++; //bb_error_msg("[%d] check_and_run_traps: G.count_SIGCHLD:%d G.handled_SIGCHLD:%d", getpid(), G.count_SIGCHLD, G.handled_SIGCHLD); break; #endif case SIGINT: /* Builtin was ^C'ed, make it look prettier: */ bb_putchar('\n'); G.flag_SIGINT = 1; break; #if ENABLE_HUSH_JOB case SIGHUP: { struct pipe *job; /* bash is observed to signal whole process groups, * not individual processes */ for (job = G.job_list; job; job = job->next) { if (job->pgrp <= 0) continue; debug_printf_exec("HUPing pgrp %d\n", job->pgrp); if (kill(- job->pgrp, SIGHUP) == 0) kill(- job->pgrp, SIGCONT); } sigexit(SIGHUP); } #endif default: /* ignored: */ /* SIGTERM, SIGQUIT, SIGTTIN, SIGTTOU, SIGTSTP */ break; } } return last_sig; } static const char *get_cwd(int force) { if (force || G.cwd == NULL) { /* xrealloc_getcwd_or_warn(arg) calls free(arg), * we must not try to free(bb_msg_unknown) */ if (G.cwd == bb_msg_unknown) G.cwd = NULL; G.cwd = xrealloc_getcwd_or_warn((char *)G.cwd); if (!G.cwd) G.cwd = bb_msg_unknown; } return G.cwd; } /* * Shell and environment variable support */ static struct variable **get_ptr_to_local_var(const char *name) { struct variable **pp; struct variable *cur; int len; len = strlen(name); pp = &G.top_var; while ((cur = *pp) != NULL) { if (strncmp(cur->varstr, name, len) == 0 && cur->varstr[len] == '=') return pp; pp = &cur->next; } return NULL; } static struct variable *get_local_var(const char *name) { struct variable **pp = get_ptr_to_local_var(name); if (pp) return *pp; return NULL; } static const char* FAST_FUNC get_local_var_value(const char *name) { struct variable **vpp; if (G.expanded_assignments) { char **cpp = G.expanded_assignments; int len = strlen(name); while (*cpp) { char *cp = *cpp; if (strncmp(cp, name, len) == 0 && cp[len] == '=') return cp + len + 1; cpp++; } } vpp = get_ptr_to_local_var(name); if (vpp) return strchr((*vpp)->varstr, '=') + 1; if (strcmp(name, "PPID") == 0) return utoa(G.root_ppid); // bash compat: UID? EUID? #if ENABLE_HUSH_RANDOM_SUPPORT if (strcmp(name, "RANDOM") == 0) { return utoa(next_random(&G.random_gen)); } #endif return NULL; } /* str holds "NAME=VAL" and is expected to be malloced. * We take ownership of it. * flg_export: * 0: do not change export flag * (if creating new variable, flag will be 0) * 1: set export flag and putenv the variable * -1: clear export flag and unsetenv the variable * flg_read_only is set only when we handle -R var=val */ #if !BB_MMU && ENABLE_HUSH_LOCAL /* all params are used */ #elif BB_MMU && ENABLE_HUSH_LOCAL #define set_local_var(str, flg_export, local_lvl, flg_read_only) \ set_local_var(str, flg_export, local_lvl) #elif BB_MMU && !ENABLE_HUSH_LOCAL #define set_local_var(str, flg_export, local_lvl, flg_read_only) \ set_local_var(str, flg_export) #elif !BB_MMU && !ENABLE_HUSH_LOCAL #define set_local_var(str, flg_export, local_lvl, flg_read_only) \ set_local_var(str, flg_export, flg_read_only) #endif static int set_local_var(char *str, int flg_export, int local_lvl, int flg_read_only) { struct variable **var_pp; struct variable *cur; char *eq_sign; int name_len; eq_sign = strchr(str, '='); if (!eq_sign) { /* not expected to ever happen? */ free(str); return -1; } name_len = eq_sign - str + 1; /* including '=' */ var_pp = &G.top_var; while ((cur = *var_pp) != NULL) { if (strncmp(cur->varstr, str, name_len) != 0) { var_pp = &cur->next; continue; } /* We found an existing var with this name */ if (cur->flg_read_only) { #if !BB_MMU if (!flg_read_only) #endif bb_error_msg("%s: readonly variable", str); free(str); return -1; } if (flg_export == -1) { // "&& cur->flg_export" ? debug_printf_env("%s: unsetenv '%s'\n", __func__, str); *eq_sign = '\0'; unsetenv(str); *eq_sign = '='; } #if ENABLE_HUSH_LOCAL if (cur->func_nest_level < local_lvl) { /* New variable is declared as local, * and existing one is global, or local * from enclosing function. * Remove and save old one: */ *var_pp = cur->next; cur->next = *G.shadowed_vars_pp; *G.shadowed_vars_pp = cur; /* bash 3.2.33(1) and exported vars: * # export z=z * # f() { local z=a; env | grep ^z; } * # f * z=a * # env | grep ^z * z=z */ if (cur->flg_export) flg_export = 1; break; } #endif if (strcmp(cur->varstr + name_len, eq_sign + 1) == 0) { free_and_exp: free(str); goto exp; } if (cur->max_len != 0) { if (cur->max_len >= strlen(str)) { /* This one is from startup env, reuse space */ strcpy(cur->varstr, str); goto free_and_exp; } } else { /* max_len == 0 signifies "malloced" var, which we can * (and has to) free */ free(cur->varstr); } cur->max_len = 0; goto set_str_and_exp; } /* Not found - create new variable struct */ cur = xzalloc(sizeof(*cur)); #if ENABLE_HUSH_LOCAL cur->func_nest_level = local_lvl; #endif cur->next = *var_pp; *var_pp = cur; set_str_and_exp: cur->varstr = str; #if !BB_MMU cur->flg_read_only = flg_read_only; #endif exp: if (flg_export == 1) cur->flg_export = 1; if (name_len == 4 && cur->varstr[0] == 'P' && cur->varstr[1] == 'S') cmdedit_update_prompt(); if (cur->flg_export) { if (flg_export == -1) { cur->flg_export = 0; /* unsetenv was already done */ } else { debug_printf_env("%s: putenv '%s'\n", __func__, cur->varstr); return putenv(cur->varstr); } } return 0; } /* Used at startup and after each cd */ static void set_pwd_var(int exp) { set_local_var(xasprintf("PWD=%s", get_cwd(/*force:*/ 1)), /*exp:*/ exp, /*lvl:*/ 0, /*ro:*/ 0); } static int unset_local_var_len(const char *name, int name_len) { struct variable *cur; struct variable **var_pp; if (!name) return EXIT_SUCCESS; var_pp = &G.top_var; while ((cur = *var_pp) != NULL) { if (strncmp(cur->varstr, name, name_len) == 0 && cur->varstr[name_len] == '=') { if (cur->flg_read_only) { bb_error_msg("%s: readonly variable", name); return EXIT_FAILURE; } *var_pp = cur->next; debug_printf_env("%s: unsetenv '%s'\n", __func__, cur->varstr); bb_unsetenv(cur->varstr); if (name_len == 3 && cur->varstr[0] == 'P' && cur->varstr[1] == 'S') cmdedit_update_prompt(); if (!cur->max_len) free(cur->varstr); free(cur); return EXIT_SUCCESS; } var_pp = &cur->next; } return EXIT_SUCCESS; } static int unset_local_var(const char *name) { return unset_local_var_len(name, strlen(name)); } static void unset_vars(char **strings) { char **v; if (!strings) return; v = strings; while (*v) { const char *eq = strchrnul(*v, '='); unset_local_var_len(*v, (int)(eq - *v)); v++; } free(strings); } #if ENABLE_SH_MATH_SUPPORT # define is_name(c) ((c) == '_' || isalpha((unsigned char)(c))) # define is_in_name(c) ((c) == '_' || isalnum((unsigned char)(c))) static char* FAST_FUNC endofname(const char *name) { char *p; p = (char *) name; if (!is_name(*p)) return p; while (*++p) { if (!is_in_name(*p)) break; } return p; } #endif static void FAST_FUNC set_local_var_from_halves(const char *name, const char *val) { char *var = xasprintf("%s=%s", name, val); set_local_var(var, /*flags:*/ 0, /*lvl:*/ 0, /*ro:*/ 0); } /* * Helpers for "var1=val1 var2=val2 cmd" feature */ static void add_vars(struct variable *var) { struct variable *next; while (var) { next = var->next; var->next = G.top_var; G.top_var = var; if (var->flg_export) { debug_printf_env("%s: restoring exported '%s'\n", __func__, var->varstr); putenv(var->varstr); } else { debug_printf_env("%s: restoring variable '%s'\n", __func__, var->varstr); } var = next; } } static struct variable *set_vars_and_save_old(char **strings) { char **s; struct variable *old = NULL; if (!strings) return old; s = strings; while (*s) { struct variable *var_p; struct variable **var_pp; char *eq; eq = strchr(*s, '='); if (eq) { *eq = '\0'; var_pp = get_ptr_to_local_var(*s); *eq = '='; if (var_pp) { /* Remove variable from global linked list */ var_p = *var_pp; debug_printf_env("%s: removing '%s'\n", __func__, var_p->varstr); *var_pp = var_p->next; /* Add it to returned list */ var_p->next = old; old = var_p; } set_local_var(*s, /*exp:*/ 1, /*lvl:*/ 0, /*ro:*/ 0); } s++; } return old; } /* * in_str support */ static int FAST_FUNC static_get(struct in_str *i) { int ch = *i->p; if (ch != '\0') { i->p++; return ch; } return EOF; } static int FAST_FUNC static_peek(struct in_str *i) { return *i->p; } #if ENABLE_HUSH_INTERACTIVE static void cmdedit_update_prompt(void) { if (ENABLE_FEATURE_EDITING_FANCY_PROMPT) { G.PS1 = get_local_var_value("PS1"); if (G.PS1 == NULL) G.PS1 = "\\w \\$ "; G.PS2 = get_local_var_value("PS2"); } else { G.PS1 = NULL; } if (G.PS2 == NULL) G.PS2 = "> "; } static const char *setup_prompt_string(int promptmode) { const char *prompt_str; debug_printf("setup_prompt_string %d ", promptmode); if (!ENABLE_FEATURE_EDITING_FANCY_PROMPT) { /* Set up the prompt */ if (promptmode == 0) { /* PS1 */ free((char*)G.PS1); /* bash uses $PWD value, even if it is set by user. * It uses current dir only if PWD is unset. * We always use current dir. */ G.PS1 = xasprintf("%s %c ", get_cwd(0), (geteuid() != 0) ? '$' : '#'); prompt_str = G.PS1; } else prompt_str = G.PS2; } else prompt_str = (promptmode == 0) ? G.PS1 : G.PS2; debug_printf("result '%s'\n", prompt_str); return prompt_str; } static void get_user_input(struct in_str *i) { int r; const char *prompt_str; prompt_str = setup_prompt_string(i->promptmode); # if ENABLE_FEATURE_EDITING /* Enable command line editing only while a command line * is actually being read */ do { G.flag_SIGINT = 0; /* buglet: SIGINT will not make new prompt to appear _at once_, * only after . (^C will work) */ r = read_line_input(prompt_str, G.user_input_buf, CONFIG_FEATURE_EDITING_MAX_LEN-1, G.line_input_state); /* catch *SIGINT* etc (^C is handled by read_line_input) */ check_and_run_traps(0); } while (r == 0 || G.flag_SIGINT); /* repeat if ^C or SIGINT */ i->eof_flag = (r < 0); if (i->eof_flag) { /* EOF/error detected */ G.user_input_buf[0] = EOF; /* yes, it will be truncated, it's ok */ G.user_input_buf[1] = '\0'; } # else do { G.flag_SIGINT = 0; fputs(prompt_str, stdout); fflush_all(); G.user_input_buf[0] = r = fgetc(i->file); /*G.user_input_buf[1] = '\0'; - already is and never changed */ //do we need check_and_run_traps(0)? (maybe only if stdin) } while (G.flag_SIGINT); i->eof_flag = (r == EOF); # endif i->p = G.user_input_buf; } #endif /* INTERACTIVE */ /* This is the magic location that prints prompts * and gets data back from the user */ static int FAST_FUNC file_get(struct in_str *i) { int ch; /* If there is data waiting, eat it up */ if (i->p && *i->p) { #if ENABLE_HUSH_INTERACTIVE take_cached: #endif ch = *i->p++; if (i->eof_flag && !*i->p) ch = EOF; /* note: ch is never NUL */ } else { /* need to double check i->file because we might be doing something * more complicated by now, like sourcing or substituting. */ #if ENABLE_HUSH_INTERACTIVE if (G_interactive_fd && i->promptme && i->file == stdin) { do { get_user_input(i); } while (!*i->p); /* need non-empty line */ i->promptmode = 1; /* PS2 */ i->promptme = 0; goto take_cached; } #endif do ch = fgetc(i->file); while (ch == '\0'); } debug_printf("file_get: got '%c' %d\n", ch, ch); #if ENABLE_HUSH_INTERACTIVE if (ch == '\n') i->promptme = 1; #endif return ch; } /* All callers guarantee this routine will never * be used right after a newline, so prompting is not needed. */ static int FAST_FUNC file_peek(struct in_str *i) { int ch; if (i->p && *i->p) { if (i->eof_flag && !i->p[1]) return EOF; return *i->p; /* note: ch is never NUL */ } do ch = fgetc(i->file); while (ch == '\0'); i->eof_flag = (ch == EOF); i->peek_buf[0] = ch; i->peek_buf[1] = '\0'; i->p = i->peek_buf; debug_printf("file_peek: got '%c' %d\n", ch, ch); return ch; } static void setup_file_in_str(struct in_str *i, FILE *f) { i->peek = file_peek; i->get = file_get; #if ENABLE_HUSH_INTERACTIVE i->promptme = 1; i->promptmode = 0; /* PS1 */ #endif i->file = f; i->p = NULL; } static void setup_string_in_str(struct in_str *i, const char *s) { i->peek = static_peek; i->get = static_get; #if ENABLE_HUSH_INTERACTIVE i->promptme = 1; i->promptmode = 0; /* PS1 */ #endif i->p = s; i->eof_flag = 0; } /* * o_string support */ #define B_CHUNK (32 * sizeof(char*)) static void o_reset_to_empty_unquoted(o_string *o) { o->length = 0; o->has_quoted_part = 0; if (o->data) o->data[0] = '\0'; } static void o_free(o_string *o) { free(o->data); memset(o, 0, sizeof(*o)); } static ALWAYS_INLINE void o_free_unsafe(o_string *o) { free(o->data); } static void o_grow_by(o_string *o, int len) { if (o->length + len > o->maxlen) { o->maxlen += (2*len > B_CHUNK ? 2*len : B_CHUNK); o->data = xrealloc(o->data, 1 + o->maxlen); } } static void o_addchr(o_string *o, int ch) { debug_printf("o_addchr: '%c' o->length=%d o=%p\n", ch, o->length, o); o_grow_by(o, 1); o->data[o->length] = ch; o->length++; o->data[o->length] = '\0'; } static void o_addblock(o_string *o, const char *str, int len) { o_grow_by(o, len); memcpy(&o->data[o->length], str, len); o->length += len; o->data[o->length] = '\0'; } static void o_addstr(o_string *o, const char *str) { o_addblock(o, str, strlen(str)); } #if !BB_MMU static void nommu_addchr(o_string *o, int ch) { if (o) o_addchr(o, ch); } #else # define nommu_addchr(o, str) ((void)0) #endif static void o_addstr_with_NUL(o_string *o, const char *str) { o_addblock(o, str, strlen(str) + 1); } static void o_addblock_duplicate_backslash(o_string *o, const char *str, int len) { while (len) { len--; o_addchr(o, *str); if (*str++ == '\\') { /* \z -> \\\z; \ -> \\ */ o_addchr(o, '\\'); if (len) { len--; o_addchr(o, '\\'); o_addchr(o, *str++); } } } } #undef HUSH_BRACE_EXP /* * HUSH_BRACE_EXP code needs corresponding quoting on variable expansion side. * Currently, "v='{q,w}'; echo $v" erroneously expands braces in $v. * Apparently, on unquoted $v bash still does globbing * ("v='*.txt'; echo $v" prints all .txt files), * but NOT brace expansion! Thus, there should be TWO independent * quoting mechanisms on $v expansion side: one protects * $v from brace expansion, and other additionally protects "$v" against globbing. * We have only second one. */ #ifdef HUSH_BRACE_EXP # define MAYBE_BRACES "{}" #else # define MAYBE_BRACES "" #endif /* My analysis of quoting semantics tells me that state information * is associated with a destination, not a source. */ static void o_addqchr(o_string *o, int ch) { int sz = 1; char *found = strchr("*?[\\" MAYBE_BRACES, ch); if (found) sz++; o_grow_by(o, sz); if (found) { o->data[o->length] = '\\'; o->length++; } o->data[o->length] = ch; o->length++; o->data[o->length] = '\0'; } static void o_addQchr(o_string *o, int ch) { int sz = 1; if (o->o_escape && strchr("*?[\\" MAYBE_BRACES, ch)) { sz++; o->data[o->length] = '\\'; o->length++; } o_grow_by(o, sz); o->data[o->length] = ch; o->length++; o->data[o->length] = '\0'; } static void o_addqblock(o_string *o, const char *str, int len) { while (len) { char ch; int sz; int ordinary_cnt = strcspn(str, "*?[\\" MAYBE_BRACES); if (ordinary_cnt > len) /* paranoia */ ordinary_cnt = len; o_addblock(o, str, ordinary_cnt); if (ordinary_cnt == len) return; str += ordinary_cnt; len -= ordinary_cnt + 1; /* we are processing + 1 char below */ ch = *str++; sz = 1; if (ch) { /* it is necessarily one of "*?[\\" MAYBE_BRACES */ sz++; o->data[o->length] = '\\'; o->length++; } o_grow_by(o, sz); o->data[o->length] = ch; o->length++; o->data[o->length] = '\0'; } } static void o_addQblock(o_string *o, const char *str, int len) { if (!o->o_escape) { o_addblock(o, str, len); return; } o_addqblock(o, str, len); } static void o_addQstr(o_string *o, const char *str) { o_addQblock(o, str, strlen(str)); } /* A special kind of o_string for $VAR and `cmd` expansion. * It contains char* list[] at the beginning, which is grown in 16 element * increments. Actual string data starts at the next multiple of 16 * (char*). * list[i] contains an INDEX (int!) into this string data. * It means that if list[] needs to grow, data needs to be moved higher up * but list[i]'s need not be modified. * NB: remembering how many list[i]'s you have there is crucial. * o_finalize_list() operation post-processes this structure - calculates * and stores actual char* ptrs in list[]. Oh, it NULL terminates it as well. */ #if DEBUG_EXPAND || DEBUG_GLOB static void debug_print_list(const char *prefix, o_string *o, int n) { char **list = (char**)o->data; int string_start = ((n + 0xf) & ~0xf) * sizeof(list[0]); int i = 0; indent(); fprintf(stderr, "%s: list:%p n:%d string_start:%d length:%d maxlen:%d glob:%d quoted:%d escape:%d\n", prefix, list, n, string_start, o->length, o->maxlen, o->o_glob, o->has_quoted_part, o->o_escape); while (i < n) { indent(); fprintf(stderr, " list[%d]=%d '%s' %p\n", i, (int)list[i], o->data + (int)list[i] + string_start, o->data + (int)list[i] + string_start); i++; } if (n) { const char *p = o->data + (int)list[n - 1] + string_start; indent(); fprintf(stderr, " total_sz:%ld\n", (long)((p + strlen(p) + 1) - o->data)); } } #else # define debug_print_list(prefix, o, n) ((void)0) #endif /* n = o_save_ptr_helper(str, n) "starts new string" by storing an index value * in list[n] so that it points past last stored byte so far. * It returns n+1. */ static int o_save_ptr_helper(o_string *o, int n) { char **list = (char**)o->data; int string_start; int string_len; if (!o->has_empty_slot) { string_start = ((n + 0xf) & ~0xf) * sizeof(list[0]); string_len = o->length - string_start; if (!(n & 0xf)) { /* 0, 0x10, 0x20...? */ debug_printf_list("list[%d]=%d string_start=%d (growing)\n", n, string_len, string_start); /* list[n] points to string_start, make space for 16 more pointers */ o->maxlen += 0x10 * sizeof(list[0]); o->data = xrealloc(o->data, o->maxlen + 1); list = (char**)o->data; memmove(list + n + 0x10, list + n, string_len); o->length += 0x10 * sizeof(list[0]); } else { debug_printf_list("list[%d]=%d string_start=%d\n", n, string_len, string_start); } } else { /* We have empty slot at list[n], reuse without growth */ string_start = ((n+1 + 0xf) & ~0xf) * sizeof(list[0]); /* NB: n+1! */ string_len = o->length - string_start; debug_printf_list("list[%d]=%d string_start=%d (empty slot)\n", n, string_len, string_start); o->has_empty_slot = 0; } list[n] = (char*)(uintptr_t)string_len; return n + 1; } /* "What was our last o_save_ptr'ed position (byte offset relative o->data)?" */ static int o_get_last_ptr(o_string *o, int n) { char **list = (char**)o->data; int string_start = ((n + 0xf) & ~0xf) * sizeof(list[0]); return ((int)(uintptr_t)list[n-1]) + string_start; } #ifdef HUSH_BRACE_EXP /* There in a GNU extension, GLOB_BRACE, but it is not usable: * first, it processes even {a} (no commas), second, * I didn't manage to make it return strings when they don't match * existing files. Need to re-implement it. */ /* Helper */ static int glob_needed(const char *s) { while (*s) { if (*s == '\\') { if (!s[1]) return 0; s += 2; continue; } if (*s == '*' || *s == '[' || *s == '?' || *s == '{') return 1; s++; } return 0; } /* Return pointer to next closing brace or to comma */ static const char *next_brace_sub(const char *cp) { unsigned depth = 0; cp++; while (*cp != '\0') { if (*cp == '\\') { if (*++cp == '\0') break; cp++; continue; } if ((*cp == '}' && depth-- == 0) || (*cp == ',' && depth == 0)) break; if (*cp++ == '{') depth++; } return *cp != '\0' ? cp : NULL; } /* Recursive brace globber. Note: may garble pattern[]. */ static int glob_brace(char *pattern, o_string *o, int n) { char *new_pattern_buf; const char *begin; const char *next; const char *rest; const char *p; size_t rest_len; debug_printf_glob("glob_brace('%s')\n", pattern); begin = pattern; while (1) { if (*begin == '\0') goto simple_glob; if (*begin == '{') { /* Find the first sub-pattern and at the same time * find the rest after the closing brace */ next = next_brace_sub(begin); if (next == NULL) { /* An illegal expression */ goto simple_glob; } if (*next == '}') { /* "{abc}" with no commas - illegal * brace expr, disregard and skip it */ begin = next + 1; continue; } break; } if (*begin == '\\' && begin[1] != '\0') begin++; begin++; } debug_printf_glob("begin:%s\n", begin); debug_printf_glob("next:%s\n", next); /* Now find the end of the whole brace expression */ rest = next; while (*rest != '}') { rest = next_brace_sub(rest); if (rest == NULL) { /* An illegal expression */ goto simple_glob; } debug_printf_glob("rest:%s\n", rest); } rest_len = strlen(++rest) + 1; /* We are sure the brace expression is well-formed */ /* Allocate working buffer large enough for our work */ new_pattern_buf = xmalloc(strlen(pattern)); /* We have a brace expression. BEGIN points to the opening {, * NEXT points past the terminator of the first element, and REST * points past the final }. We will accumulate result names from * recursive runs for each brace alternative in the buffer using * GLOB_APPEND. */ p = begin + 1; while (1) { /* Construct the new glob expression */ memcpy( mempcpy( mempcpy(new_pattern_buf, /* We know the prefix for all sub-patterns */ pattern, begin - pattern), p, next - p), rest, rest_len); /* Note: glob_brace() may garble new_pattern_buf[]. * That's why we re-copy prefix every time (1st memcpy above). */ n = glob_brace(new_pattern_buf, o, n); if (*next == '}') { /* We saw the last entry */ break; } p = next + 1; next = next_brace_sub(next); } free(new_pattern_buf); return n; simple_glob: { int gr; glob_t globdata; memset(&globdata, 0, sizeof(globdata)); gr = glob(pattern, 0, NULL, &globdata); debug_printf_glob("glob('%s'):%d\n", pattern, gr); if (gr != 0) { if (gr == GLOB_NOMATCH) { globfree(&globdata); /* NB: garbles parameter */ unbackslash(pattern); o_addstr_with_NUL(o, pattern); debug_printf_glob("glob pattern '%s' is literal\n", pattern); return o_save_ptr_helper(o, n); } if (gr == GLOB_NOSPACE) bb_error_msg_and_die(bb_msg_memory_exhausted); /* GLOB_ABORTED? Only happens with GLOB_ERR flag, * but we didn't specify it. Paranoia again. */ bb_error_msg_and_die("glob error %d on '%s'", gr, pattern); } if (globdata.gl_pathv && globdata.gl_pathv[0]) { char **argv = globdata.gl_pathv; while (1) { o_addstr_with_NUL(o, *argv); n = o_save_ptr_helper(o, n); argv++; if (!*argv) break; } } globfree(&globdata); } return n; } /* Performs globbing on last list[], * saving each result as a new list[]. */ static int perform_glob(o_string *o, int n) { char *pattern, *copy; debug_printf_glob("start perform_glob: n:%d o->data:%p\n", n, o->data); if (!o->data) return o_save_ptr_helper(o, n); pattern = o->data + o_get_last_ptr(o, n); debug_printf_glob("glob pattern '%s'\n", pattern); if (!glob_needed(pattern)) { /* unbackslash last string in o in place, fix length */ o->length = unbackslash(pattern) - o->data; debug_printf_glob("glob pattern '%s' is literal\n", pattern); return o_save_ptr_helper(o, n); } copy = xstrdup(pattern); /* "forget" pattern in o */ o->length = pattern - o->data; n = glob_brace(copy, o, n); free(copy); if (DEBUG_GLOB) debug_print_list("perform_glob returning", o, n); return n; } #else /* !HUSH_BRACE_EXP */ /* Helper */ static int glob_needed(const char *s) { while (*s) { if (*s == '\\') { if (!s[1]) return 0; s += 2; continue; } if (*s == '*' || *s == '[' || *s == '?') return 1; s++; } return 0; } /* Performs globbing on last list[], * saving each result as a new list[]. */ static int perform_glob(o_string *o, int n) { glob_t globdata; int gr; char *pattern; debug_printf_glob("start perform_glob: n:%d o->data:%p\n", n, o->data); if (!o->data) return o_save_ptr_helper(o, n); pattern = o->data + o_get_last_ptr(o, n); debug_printf_glob("glob pattern '%s'\n", pattern); if (!glob_needed(pattern)) { literal: /* unbackslash last string in o in place, fix length */ o->length = unbackslash(pattern) - o->data; debug_printf_glob("glob pattern '%s' is literal\n", pattern); return o_save_ptr_helper(o, n); } memset(&globdata, 0, sizeof(globdata)); /* Can't use GLOB_NOCHECK: it does not unescape the string. * If we glob "*.\*" and don't find anything, we need * to fall back to using literal "*.*", but GLOB_NOCHECK * will return "*.\*"! */ gr = glob(pattern, 0, NULL, &globdata); debug_printf_glob("glob('%s'):%d\n", pattern, gr); if (gr != 0) { if (gr == GLOB_NOMATCH) { globfree(&globdata); goto literal; } if (gr == GLOB_NOSPACE) bb_error_msg_and_die(bb_msg_memory_exhausted); /* GLOB_ABORTED? Only happens with GLOB_ERR flag, * but we didn't specify it. Paranoia again. */ bb_error_msg_and_die("glob error %d on '%s'", gr, pattern); } if (globdata.gl_pathv && globdata.gl_pathv[0]) { char **argv = globdata.gl_pathv; /* "forget" pattern in o */ o->length = pattern - o->data; while (1) { o_addstr_with_NUL(o, *argv); n = o_save_ptr_helper(o, n); argv++; if (!*argv) break; } } globfree(&globdata); if (DEBUG_GLOB) debug_print_list("perform_glob returning", o, n); return n; } #endif /* !HUSH_BRACE_EXP */ /* If o->o_glob == 1, glob the string so far remembered. * Otherwise, just finish current list[] and start new */ static int o_save_ptr(o_string *o, int n) { if (o->o_glob) { /* if globbing is requested */ /* If o->has_empty_slot, list[n] was already globbed * (if it was requested back then when it was filled) * so don't do that again! */ if (!o->has_empty_slot) return perform_glob(o, n); /* o_save_ptr_helper is inside */ } return o_save_ptr_helper(o, n); } /* "Please convert list[n] to real char* ptrs, and NULL terminate it." */ static char **o_finalize_list(o_string *o, int n) { char **list; int string_start; n = o_save_ptr(o, n); /* force growth for list[n] if necessary */ if (DEBUG_EXPAND) debug_print_list("finalized", o, n); debug_printf_expand("finalized n:%d\n", n); list = (char**)o->data; string_start = ((n + 0xf) & ~0xf) * sizeof(list[0]); list[--n] = NULL; while (n) { n--; list[n] = o->data + (int)(ptrdiff_t)list[n] + string_start; } return list; } static void free_pipe_list(struct pipe *head); /* Return code is the exit status of the pipe */ static void free_pipe(struct pipe *pi) { char **p; struct command *command; struct redir_struct *r, *rnext; int a, i; if (pi->stopped_cmds > 0) /* why? */ return; debug_printf_clean("run pipe: (pid %d)\n", getpid()); for (i = 0; i < pi->num_cmds; i++) { command = &pi->cmds[i]; debug_printf_clean(" command %d:\n", i); if (command->argv) { for (a = 0, p = command->argv; *p; a++, p++) { debug_printf_clean(" argv[%d] = %s\n", a, *p); } free_strings(command->argv); command->argv = NULL; } /* not "else if": on syntax error, we may have both! */ if (command->group) { debug_printf_clean(" begin group (cmd_type:%d)\n", command->cmd_type); free_pipe_list(command->group); debug_printf_clean(" end group\n"); command->group = NULL; } /* else is crucial here. * If group != NULL, child_func is meaningless */ #if ENABLE_HUSH_FUNCTIONS else if (command->child_func) { debug_printf_exec("cmd %p releases child func at %p\n", command, command->child_func); command->child_func->parent_cmd = NULL; } #endif #if !BB_MMU free(command->group_as_string); command->group_as_string = NULL; #endif for (r = command->redirects; r; r = rnext) { debug_printf_clean(" redirect %d%s", r->rd_fd, redir_table[r->rd_type].descrip); /* guard against the case >$FOO, where foo is unset or blank */ if (r->rd_filename) { debug_printf_clean(" fname:'%s'\n", r->rd_filename); free(r->rd_filename); r->rd_filename = NULL; } debug_printf_clean(" rd_dup:%d\n", r->rd_dup); rnext = r->next; free(r); } command->redirects = NULL; } free(pi->cmds); /* children are an array, they get freed all at once */ pi->cmds = NULL; #if ENABLE_HUSH_JOB free(pi->cmdtext); pi->cmdtext = NULL; #endif } static void free_pipe_list(struct pipe *head) { struct pipe *pi, *next; for (pi = head; pi; pi = next) { #if HAS_KEYWORDS debug_printf_clean(" pipe reserved word %d\n", pi->res_word); #endif free_pipe(pi); debug_printf_clean("pipe followup code %d\n", pi->followup); next = pi->next; /*pi->next = NULL;*/ free(pi); } } /*** Parsing routines ***/ static struct pipe *new_pipe(void) { struct pipe *pi; pi = xzalloc(sizeof(struct pipe)); /*pi->followup = 0; - deliberately invalid value */ /*pi->res_word = RES_NONE; - RES_NONE is 0 anyway */ return pi; } /* Command (member of a pipe) is complete, or we start a new pipe * if ctx->command is NULL. * No errors possible here. */ static int done_command(struct parse_context *ctx) { /* The command is really already in the pipe structure, so * advance the pipe counter and make a new, null command. */ struct pipe *pi = ctx->pipe; struct command *command = ctx->command; if (command) { if (IS_NULL_CMD(command)) { debug_printf_parse("done_command: skipping null cmd, num_cmds=%d\n", pi->num_cmds); goto clear_and_ret; } pi->num_cmds++; debug_printf_parse("done_command: ++num_cmds=%d\n", pi->num_cmds); //debug_print_tree(ctx->list_head, 20); } else { debug_printf_parse("done_command: initializing, num_cmds=%d\n", pi->num_cmds); } /* Only real trickiness here is that the uncommitted * command structure is not counted in pi->num_cmds. */ pi->cmds = xrealloc(pi->cmds, sizeof(*pi->cmds) * (pi->num_cmds+1)); ctx->command = command = &pi->cmds[pi->num_cmds]; clear_and_ret: memset(command, 0, sizeof(*command)); return pi->num_cmds; /* used only for 0/nonzero check */ } static void done_pipe(struct parse_context *ctx, pipe_style type) { int not_null; debug_printf_parse("done_pipe entered, followup %d\n", type); /* Close previous command */ not_null = done_command(ctx); ctx->pipe->followup = type; #if HAS_KEYWORDS ctx->pipe->pi_inverted = ctx->ctx_inverted; ctx->ctx_inverted = 0; ctx->pipe->res_word = ctx->ctx_res_w; #endif /* Without this check, even just on command line generates * tree of three NOPs (!). Which is harmless but annoying. * IOW: it is safe to do it unconditionally. */ if (not_null #if ENABLE_HUSH_IF || ctx->ctx_res_w == RES_FI #endif #if ENABLE_HUSH_LOOPS || ctx->ctx_res_w == RES_DONE || ctx->ctx_res_w == RES_FOR || ctx->ctx_res_w == RES_IN #endif #if ENABLE_HUSH_CASE || ctx->ctx_res_w == RES_ESAC #endif ) { struct pipe *new_p; debug_printf_parse("done_pipe: adding new pipe: " "not_null:%d ctx->ctx_res_w:%d\n", not_null, ctx->ctx_res_w); new_p = new_pipe(); ctx->pipe->next = new_p; ctx->pipe = new_p; /* RES_THEN, RES_DO etc are "sticky" - * they remain set for pipes inside if/while. * This is used to control execution. * RES_FOR and RES_IN are NOT sticky (needed to support * cases where variable or value happens to match a keyword): */ #if ENABLE_HUSH_LOOPS if (ctx->ctx_res_w == RES_FOR || ctx->ctx_res_w == RES_IN) ctx->ctx_res_w = RES_NONE; #endif #if ENABLE_HUSH_CASE if (ctx->ctx_res_w == RES_MATCH) ctx->ctx_res_w = RES_CASE_BODY; if (ctx->ctx_res_w == RES_CASE) ctx->ctx_res_w = RES_CASE_IN; #endif ctx->command = NULL; /* trick done_command below */ /* Create the memory for command, roughly: * ctx->pipe->cmds = new struct command; * ctx->command = &ctx->pipe->cmds[0]; */ done_command(ctx); //debug_print_tree(ctx->list_head, 10); } debug_printf_parse("done_pipe return\n"); } static void initialize_context(struct parse_context *ctx) { memset(ctx, 0, sizeof(*ctx)); ctx->pipe = ctx->list_head = new_pipe(); /* Create the memory for command, roughly: * ctx->pipe->cmds = new struct command; * ctx->command = &ctx->pipe->cmds[0]; */ done_command(ctx); } /* If a reserved word is found and processed, parse context is modified * and 1 is returned. */ #if HAS_KEYWORDS struct reserved_combo { char literal[6]; unsigned char res; unsigned char assignment_flag; int flag; }; enum { FLAG_END = (1 << RES_NONE ), # if ENABLE_HUSH_IF FLAG_IF = (1 << RES_IF ), FLAG_THEN = (1 << RES_THEN ), FLAG_ELIF = (1 << RES_ELIF ), FLAG_ELSE = (1 << RES_ELSE ), FLAG_FI = (1 << RES_FI ), # endif # if ENABLE_HUSH_LOOPS FLAG_FOR = (1 << RES_FOR ), FLAG_WHILE = (1 << RES_WHILE), FLAG_UNTIL = (1 << RES_UNTIL), FLAG_DO = (1 << RES_DO ), FLAG_DONE = (1 << RES_DONE ), FLAG_IN = (1 << RES_IN ), # endif # if ENABLE_HUSH_CASE FLAG_MATCH = (1 << RES_MATCH), FLAG_ESAC = (1 << RES_ESAC ), # endif FLAG_START = (1 << RES_XXXX ), }; static const struct reserved_combo* match_reserved_word(o_string *word) { /* Mostly a list of accepted follow-up reserved words. * FLAG_END means we are done with the sequence, and are ready * to turn the compound list into a command. * FLAG_START means the word must start a new compound list. */ static const struct reserved_combo reserved_list[] = { # if ENABLE_HUSH_IF { "!", RES_NONE, NOT_ASSIGNMENT , 0 }, { "if", RES_IF, WORD_IS_KEYWORD, FLAG_THEN | FLAG_START }, { "then", RES_THEN, WORD_IS_KEYWORD, FLAG_ELIF | FLAG_ELSE | FLAG_FI }, { "elif", RES_ELIF, WORD_IS_KEYWORD, FLAG_THEN }, { "else", RES_ELSE, WORD_IS_KEYWORD, FLAG_FI }, { "fi", RES_FI, NOT_ASSIGNMENT , FLAG_END }, # endif # if ENABLE_HUSH_LOOPS { "for", RES_FOR, NOT_ASSIGNMENT , FLAG_IN | FLAG_DO | FLAG_START }, { "while", RES_WHILE, WORD_IS_KEYWORD, FLAG_DO | FLAG_START }, { "until", RES_UNTIL, WORD_IS_KEYWORD, FLAG_DO | FLAG_START }, { "in", RES_IN, NOT_ASSIGNMENT , FLAG_DO }, { "do", RES_DO, WORD_IS_KEYWORD, FLAG_DONE }, { "done", RES_DONE, NOT_ASSIGNMENT , FLAG_END }, # endif # if ENABLE_HUSH_CASE { "case", RES_CASE, NOT_ASSIGNMENT , FLAG_MATCH | FLAG_START }, { "esac", RES_ESAC, NOT_ASSIGNMENT , FLAG_END }, # endif }; const struct reserved_combo *r; for (r = reserved_list; r < reserved_list + ARRAY_SIZE(reserved_list); r++) { if (strcmp(word->data, r->literal) == 0) return r; } return NULL; } /* Return 0: not a keyword, 1: keyword */ static int reserved_word(o_string *word, struct parse_context *ctx) { # if ENABLE_HUSH_CASE static const struct reserved_combo reserved_match = { "", RES_MATCH, NOT_ASSIGNMENT , FLAG_MATCH | FLAG_ESAC }; # endif const struct reserved_combo *r; if (word->has_quoted_part) return 0; r = match_reserved_word(word); if (!r) return 0; debug_printf("found reserved word %s, res %d\n", r->literal, r->res); # if ENABLE_HUSH_CASE if (r->res == RES_IN && ctx->ctx_res_w == RES_CASE_IN) { /* "case word IN ..." - IN part starts first MATCH part */ r = &reserved_match; } else # endif if (r->flag == 0) { /* '!' */ if (ctx->ctx_inverted) { /* bash doesn't accept '! ! true' */ syntax_error("! ! command"); ctx->ctx_res_w = RES_SNTX; } ctx->ctx_inverted = 1; return 1; } if (r->flag & FLAG_START) { struct parse_context *old; old = xmalloc(sizeof(*old)); debug_printf_parse("push stack %p\n", old); *old = *ctx; /* physical copy */ initialize_context(ctx); ctx->stack = old; } else if (/*ctx->ctx_res_w == RES_NONE ||*/ !(ctx->old_flag & (1 << r->res))) { syntax_error_at(word->data); ctx->ctx_res_w = RES_SNTX; return 1; } else { /* "{...} fi" is ok. "{...} if" is not * Example: * if { echo foo; } then { echo bar; } fi */ if (ctx->command->group) done_pipe(ctx, PIPE_SEQ); } ctx->ctx_res_w = r->res; ctx->old_flag = r->flag; word->o_assignment = r->assignment_flag; if (ctx->old_flag & FLAG_END) { struct parse_context *old; done_pipe(ctx, PIPE_SEQ); debug_printf_parse("pop stack %p\n", ctx->stack); old = ctx->stack; old->command->group = ctx->list_head; old->command->cmd_type = CMD_NORMAL; # if !BB_MMU o_addstr(&old->as_string, ctx->as_string.data); o_free_unsafe(&ctx->as_string); old->command->group_as_string = xstrdup(old->as_string.data); debug_printf_parse("pop, remembering as:'%s'\n", old->command->group_as_string); # endif *ctx = *old; /* physical copy */ free(old); } return 1; } #endif /* HAS_KEYWORDS */ /* Word is complete, look at it and update parsing context. * Normal return is 0. Syntax errors return 1. * Note: on return, word is reset, but not o_free'd! */ static int done_word(o_string *word, struct parse_context *ctx) { struct command *command = ctx->command; debug_printf_parse("done_word entered: '%s' %p\n", word->data, command); if (word->length == 0 && !word->has_quoted_part) { debug_printf_parse("done_word return 0: true null, ignored\n"); return 0; } if (ctx->pending_redirect) { /* We do not glob in e.g. >*.tmp case. bash seems to glob here * only if run as "bash", not "sh" */ /* http://www.opengroup.org/onlinepubs/009695399/utilities/xcu_chap02.html * "2.7 Redirection * ...the word that follows the redirection operator * shall be subjected to tilde expansion, parameter expansion, * command substitution, arithmetic expansion, and quote * removal. Pathname expansion shall not be performed * on the word by a non-interactive shell; an interactive * shell may perform it, but shall do so only when * the expansion would result in one word." */ ctx->pending_redirect->rd_filename = xstrdup(word->data); /* Cater for >\file case: * >\a creates file a; >\\a, >"\a", >"\\a" create file \a * Same with heredocs: * for <<\H delim is H; <<\\H, <<"\H", <<"\\H" - \H */ if (ctx->pending_redirect->rd_type == REDIRECT_HEREDOC) { unbackslash(ctx->pending_redirect->rd_filename); /* Is it <<"HEREDOC"? */ if (word->has_quoted_part) { ctx->pending_redirect->rd_dup |= HEREDOC_QUOTED; } } debug_printf_parse("word stored in rd_filename: '%s'\n", word->data); ctx->pending_redirect = NULL; } else { /* If this word wasn't an assignment, next ones definitely * can't be assignments. Even if they look like ones. */ if (word->o_assignment != DEFINITELY_ASSIGNMENT && word->o_assignment != WORD_IS_KEYWORD ) { word->o_assignment = NOT_ASSIGNMENT; } else { if (word->o_assignment == DEFINITELY_ASSIGNMENT) command->assignment_cnt++; word->o_assignment = MAYBE_ASSIGNMENT; } #if HAS_KEYWORDS # if ENABLE_HUSH_CASE if (ctx->ctx_dsemicolon && strcmp(word->data, "esac") != 0 /* not "... pattern) cmd;; esac" */ ) { /* already done when ctx_dsemicolon was set to 1: */ /* ctx->ctx_res_w = RES_MATCH; */ ctx->ctx_dsemicolon = 0; } else # endif if (!command->argv /* if it's the first word... */ # if ENABLE_HUSH_LOOPS && ctx->ctx_res_w != RES_FOR /* ...not after FOR or IN */ && ctx->ctx_res_w != RES_IN # endif # if ENABLE_HUSH_CASE && ctx->ctx_res_w != RES_CASE # endif ) { debug_printf_parse("checking '%s' for reserved-ness\n", word->data); if (reserved_word(word, ctx)) { o_reset_to_empty_unquoted(word); debug_printf_parse("done_word return %d\n", (ctx->ctx_res_w == RES_SNTX)); return (ctx->ctx_res_w == RES_SNTX); } # if ENABLE_HUSH_BASH_COMPAT if (strcmp(word->data, "[[") == 0) { command->cmd_type = CMD_SINGLEWORD_NOGLOB; } /* fall through */ # endif } #endif if (command->group) { /* "{ echo foo; } echo bar" - bad */ syntax_error_at(word->data); debug_printf_parse("done_word return 1: syntax error, " "groups and arglists don't mix\n"); return 1; } if (word->has_quoted_part /* optimization: and if it's ("" or '') or ($v... or `cmd`...): */ && (word->data[0] == '\0' || word->data[0] == SPECIAL_VAR_SYMBOL) /* (otherwise it's known to be not empty and is already safe) */ ) { /* exclude "$@" - it can expand to no word despite "" */ char *p = word->data; while (p[0] == SPECIAL_VAR_SYMBOL && (p[1] & 0x7f) == '@' && p[2] == SPECIAL_VAR_SYMBOL ) { p += 3; } if (p == word->data || p[0] != '\0') { /* saw no "$@", or not only "$@" but some * real text is there too */ /* insert "empty variable" reference, this makes * e.g. "", $empty"" etc to not disappear */ o_addchr(word, SPECIAL_VAR_SYMBOL); o_addchr(word, SPECIAL_VAR_SYMBOL); } } command->argv = add_string_to_strings(command->argv, xstrdup(word->data)); debug_print_strings("word appended to argv", command->argv); } #if ENABLE_HUSH_LOOPS if (ctx->ctx_res_w == RES_FOR) { if (word->has_quoted_part || !is_well_formed_var_name(command->argv[0], '\0') ) { /* bash says just "not a valid identifier" */ syntax_error("not a valid identifier in for"); return 1; } /* Force FOR to have just one word (variable name) */ /* NB: basically, this makes hush see "for v in ..." * syntax as if it is "for v; in ...". FOR and IN become * two pipe structs in parse tree. */ done_pipe(ctx, PIPE_SEQ); } #endif #if ENABLE_HUSH_CASE /* Force CASE to have just one word */ if (ctx->ctx_res_w == RES_CASE) { done_pipe(ctx, PIPE_SEQ); } #endif o_reset_to_empty_unquoted(word); debug_printf_parse("done_word return 0\n"); return 0; } /* Peek ahead in the input to find out if we have a "&n" construct, * as in "2>&1", that represents duplicating a file descriptor. * Return: * REDIRFD_CLOSE if >&- "close fd" construct is seen, * REDIRFD_SYNTAX_ERR if syntax error, * REDIRFD_TO_FILE if no & was seen, * or the number found. */ #if BB_MMU #define parse_redir_right_fd(as_string, input) \ parse_redir_right_fd(input) #endif static int parse_redir_right_fd(o_string *as_string, struct in_str *input) { int ch, d, ok; ch = i_peek(input); if (ch != '&') return REDIRFD_TO_FILE; ch = i_getch(input); /* get the & */ nommu_addchr(as_string, ch); ch = i_peek(input); if (ch == '-') { ch = i_getch(input); nommu_addchr(as_string, ch); return REDIRFD_CLOSE; } d = 0; ok = 0; while (ch != EOF && isdigit(ch)) { d = d*10 + (ch-'0'); ok = 1; ch = i_getch(input); nommu_addchr(as_string, ch); ch = i_peek(input); } if (ok) return d; //TODO: this is the place to catch ">&file" bashism (redirect both fd 1 and 2) bb_error_msg("ambiguous redirect"); return REDIRFD_SYNTAX_ERR; } /* Return code is 0 normal, 1 if a syntax error is detected */ static int parse_redirect(struct parse_context *ctx, int fd, redir_type style, struct in_str *input) { struct command *command = ctx->command; struct redir_struct *redir; struct redir_struct **redirp; int dup_num; dup_num = REDIRFD_TO_FILE; if (style != REDIRECT_HEREDOC) { /* Check for a '>&1' type redirect */ dup_num = parse_redir_right_fd(&ctx->as_string, input); if (dup_num == REDIRFD_SYNTAX_ERR) return 1; } else { int ch = i_peek(input); dup_num = (ch == '-'); /* HEREDOC_SKIPTABS bit is 1 */ if (dup_num) { /* <<-... */ ch = i_getch(input); nommu_addchr(&ctx->as_string, ch); ch = i_peek(input); } } if (style == REDIRECT_OVERWRITE && dup_num == REDIRFD_TO_FILE) { int ch = i_peek(input); if (ch == '|') { /* >|FILE redirect ("clobbering" >). * Since we do not support "set -o noclobber" yet, * >| and > are the same for now. Just eat |. */ ch = i_getch(input); nommu_addchr(&ctx->as_string, ch); } } /* Create a new redir_struct and append it to the linked list */ redirp = &command->redirects; while ((redir = *redirp) != NULL) { redirp = &(redir->next); } *redirp = redir = xzalloc(sizeof(*redir)); /* redir->next = NULL; */ /* redir->rd_filename = NULL; */ redir->rd_type = style; redir->rd_fd = (fd == -1) ? redir_table[style].default_fd : fd; debug_printf_parse("redirect type %d %s\n", redir->rd_fd, redir_table[style].descrip); redir->rd_dup = dup_num; if (style != REDIRECT_HEREDOC && dup_num != REDIRFD_TO_FILE) { /* Erik had a check here that the file descriptor in question * is legit; I postpone that to "run time" * A "-" representation of "close me" shows up as a -3 here */ debug_printf_parse("duplicating redirect '%d>&%d'\n", redir->rd_fd, redir->rd_dup); } else { /* Set ctx->pending_redirect, so we know what to do at the * end of the next parsed word. */ ctx->pending_redirect = redir; } return 0; } /* If a redirect is immediately preceded by a number, that number is * supposed to tell which file descriptor to redirect. This routine * looks for such preceding numbers. In an ideal world this routine * needs to handle all the following classes of redirects... * echo 2>foo # redirects fd 2 to file "foo", nothing passed to echo * echo 49>foo # redirects fd 49 to file "foo", nothing passed to echo * echo -2>foo # redirects fd 1 to file "foo", "-2" passed to echo * echo 49x>foo # redirects fd 1 to file "foo", "49x" passed to echo * * http://www.opengroup.org/onlinepubs/009695399/utilities/xcu_chap02.html * "2.7 Redirection * ... If n is quoted, the number shall not be recognized as part of * the redirection expression. For example: * echo \2>a * writes the character 2 into file a" * We are getting it right by setting ->has_quoted_part on any \ * * A -1 return means no valid number was found, * the caller should use the appropriate default for this redirection. */ static int redirect_opt_num(o_string *o) { int num; if (o->data == NULL) return -1; num = bb_strtou(o->data, NULL, 10); if (errno || num < 0) return -1; o_reset_to_empty_unquoted(o); return num; } #if BB_MMU #define fetch_till_str(as_string, input, word, skip_tabs) \ fetch_till_str(input, word, skip_tabs) #endif static char *fetch_till_str(o_string *as_string, struct in_str *input, const char *word, int heredoc_flags) { o_string heredoc = NULL_O_STRING; int past_EOL = 0; int prev = 0; /* not \ */ int ch; goto jump_in; while (1) { ch = i_getch(input); nommu_addchr(as_string, ch); if (ch == '\n' /* TODO: or EOF? (heredoc delimiter may end with , not only ) */ && ((heredoc_flags & HEREDOC_QUOTED) || prev != '\\') ) { if (strcmp(heredoc.data + past_EOL, word) == 0) { heredoc.data[past_EOL] = '\0'; debug_printf_parse("parsed heredoc '%s'\n", heredoc.data); return heredoc.data; } do { o_addchr(&heredoc, '\n'); prev = 0; /* not \ */ past_EOL = heredoc.length; jump_in: do { ch = i_getch(input); nommu_addchr(as_string, ch); } while ((heredoc_flags & HEREDOC_SKIPTABS) && ch == '\t'); } while (ch == '\n'); } if (ch == EOF) { o_free_unsafe(&heredoc); return NULL; } o_addchr(&heredoc, ch); if (prev == '\\' && ch == '\\') /* Correctly handle foo\\ (not a line cont.) */ prev = 0; /* not \ */ else prev = ch; nommu_addchr(as_string, ch); } } /* Look at entire parse tree for not-yet-loaded REDIRECT_HEREDOCs * and load them all. There should be exactly heredoc_cnt of them. */ static int fetch_heredocs(int heredoc_cnt, struct parse_context *ctx, struct in_str *input) { struct pipe *pi = ctx->list_head; while (pi && heredoc_cnt) { int i; struct command *cmd = pi->cmds; debug_printf_parse("fetch_heredocs: num_cmds:%d cmd argv0:'%s'\n", pi->num_cmds, cmd->argv ? cmd->argv[0] : "NONE"); for (i = 0; i < pi->num_cmds; i++) { struct redir_struct *redir = cmd->redirects; debug_printf_parse("fetch_heredocs: %d cmd argv0:'%s'\n", i, cmd->argv ? cmd->argv[0] : "NONE"); while (redir) { if (redir->rd_type == REDIRECT_HEREDOC) { char *p; redir->rd_type = REDIRECT_HEREDOC2; /* redir->rd_dup is (ab)used to indicate <<- */ p = fetch_till_str(&ctx->as_string, input, redir->rd_filename, redir->rd_dup); if (!p) { syntax_error("unexpected EOF in here document"); return 1; } free(redir->rd_filename); redir->rd_filename = p; heredoc_cnt--; } redir = redir->next; } cmd++; } pi = pi->next; } #if 0 /* Should be 0. If it isn't, it's a parse error */ if (heredoc_cnt) bb_error_msg_and_die("heredoc BUG 2"); #endif return 0; } static int run_list(struct pipe *pi); #if BB_MMU #define parse_stream(pstring, input, end_trigger) \ parse_stream(input, end_trigger) #endif static struct pipe *parse_stream(char **pstring, struct in_str *input, int end_trigger); #if !ENABLE_HUSH_FUNCTIONS #define parse_group(dest, ctx, input, ch) \ parse_group(ctx, input, ch) #endif static int parse_group(o_string *dest, struct parse_context *ctx, struct in_str *input, int ch) { /* dest contains characters seen prior to ( or {. * Typically it's empty, but for function defs, * it contains function name (without '()'). */ struct pipe *pipe_list; int endch; struct command *command = ctx->command; debug_printf_parse("parse_group entered\n"); #if ENABLE_HUSH_FUNCTIONS if (ch == '(' && !dest->has_quoted_part) { if (dest->length) if (done_word(dest, ctx)) return 1; if (!command->argv) goto skip; /* (... */ if (command->argv[1]) { /* word word ... (... */ syntax_error_unexpected_ch('('); return 1; } /* it is "word(..." or "word (..." */ do ch = i_getch(input); while (ch == ' ' || ch == '\t'); if (ch != ')') { syntax_error_unexpected_ch(ch); return 1; } nommu_addchr(&ctx->as_string, ch); do ch = i_getch(input); while (ch == ' ' || ch == '\t' || ch == '\n'); if (ch != '{') { syntax_error_unexpected_ch(ch); return 1; } nommu_addchr(&ctx->as_string, ch); command->cmd_type = CMD_FUNCDEF; goto skip; } #endif #if 0 /* Prevented by caller */ if (command->argv /* word [word]{... */ || dest->length /* word{... */ || dest->has_quoted_part /* ""{... */ ) { syntax_error(NULL); debug_printf_parse("parse_group return 1: " "syntax error, groups and arglists don't mix\n"); return 1; } #endif #if ENABLE_HUSH_FUNCTIONS skip: #endif endch = '}'; if (ch == '(') { endch = ')'; command->cmd_type = CMD_SUBSHELL; } else { /* bash does not allow "{echo...", requires whitespace */ ch = i_getch(input); if (ch != ' ' && ch != '\t' && ch != '\n') { syntax_error_unexpected_ch(ch); return 1; } nommu_addchr(&ctx->as_string, ch); } { #if BB_MMU # define as_string NULL #else char *as_string = NULL; #endif pipe_list = parse_stream(&as_string, input, endch); #if !BB_MMU if (as_string) o_addstr(&ctx->as_string, as_string); #endif /* empty ()/{} or parse error? */ if (!pipe_list || pipe_list == ERR_PTR) { /* parse_stream already emitted error msg */ if (!BB_MMU) free(as_string); debug_printf_parse("parse_group return 1: " "parse_stream returned %p\n", pipe_list); return 1; } command->group = pipe_list; #if !BB_MMU as_string[strlen(as_string) - 1] = '\0'; /* plink ')' or '}' */ command->group_as_string = as_string; debug_printf_parse("end of group, remembering as:'%s'\n", command->group_as_string); #endif #undef as_string } debug_printf_parse("parse_group return 0\n"); return 0; /* command remains "open", available for possible redirects */ } #if ENABLE_HUSH_TICK || ENABLE_SH_MATH_SUPPORT || ENABLE_HUSH_DOLLAR_OPS /* Subroutines for copying $(...) and `...` things */ static void add_till_backquote(o_string *dest, struct in_str *input); /* '...' */ static void add_till_single_quote(o_string *dest, struct in_str *input) { while (1) { int ch = i_getch(input); if (ch == EOF) { syntax_error_unterm_ch('\''); /*xfunc_die(); - redundant */ } if (ch == '\'') return; o_addchr(dest, ch); } } /* "...\"...`..`...." - do we need to handle "...$(..)..." too? */ static void add_till_double_quote(o_string *dest, struct in_str *input) { while (1) { int ch = i_getch(input); if (ch == EOF) { syntax_error_unterm_ch('"'); /*xfunc_die(); - redundant */ } if (ch == '"') return; if (ch == '\\') { /* \x. Copy both chars. */ o_addchr(dest, ch); ch = i_getch(input); } o_addchr(dest, ch); if (ch == '`') { add_till_backquote(dest, input); o_addchr(dest, ch); continue; } //if (ch == '$') ... } } /* Process `cmd` - copy contents until "`" is seen. Complicated by * \` quoting. * "Within the backquoted style of command substitution, backslash * shall retain its literal meaning, except when followed by: '$', '`', or '\'. * The search for the matching backquote shall be satisfied by the first * backquote found without a preceding backslash; during this search, * if a non-escaped backquote is encountered within a shell comment, * a here-document, an embedded command substitution of the $(command) * form, or a quoted string, undefined results occur. A single-quoted * or double-quoted string that begins, but does not end, within the * "`...`" sequence produces undefined results." * Example Output * echo `echo '\'TEST\`echo ZZ\`BEST` \TESTZZBEST */ static void add_till_backquote(o_string *dest, struct in_str *input) { while (1) { int ch = i_getch(input); if (ch == EOF) { syntax_error_unterm_ch('`'); /*xfunc_die(); - redundant */ } if (ch == '`') return; if (ch == '\\') { /* \x. Copy both chars unless it is \` */ int ch2 = i_getch(input); if (ch2 == EOF) { syntax_error_unterm_ch('`'); /*xfunc_die(); - redundant */ } if (ch2 != '`' && ch2 != '$' && ch2 != '\\') o_addchr(dest, ch); ch = ch2; } o_addchr(dest, ch); } } /* Process $(cmd) - copy contents until ")" is seen. Complicated by * quoting and nested ()s. * "With the $(command) style of command substitution, all characters * following the open parenthesis to the matching closing parenthesis * constitute the command. Any valid shell script can be used for command, * except a script consisting solely of redirections which produces * unspecified results." * Example Output * echo $(echo '(TEST)' BEST) (TEST) BEST * echo $(echo 'TEST)' BEST) TEST) BEST * echo $(echo \(\(TEST\) BEST) ((TEST) BEST * * Also adapted to eat ${var%...} and $((...)) constructs, since ... part * can contain arbitrary constructs, just like $(cmd). * In bash compat mode, it needs to also be able to stop on ':' or '/' * for ${var:N[:M]} and ${var/P[/R]} parsing. */ #define DOUBLE_CLOSE_CHAR_FLAG 0x80 static int add_till_closing_bracket(o_string *dest, struct in_str *input, unsigned end_ch) { int ch; char dbl = end_ch & DOUBLE_CLOSE_CHAR_FLAG; # if ENABLE_HUSH_BASH_COMPAT char end_char2 = end_ch >> 8; # endif end_ch &= (DOUBLE_CLOSE_CHAR_FLAG - 1); while (1) { ch = i_getch(input); if (ch == EOF) { syntax_error_unterm_ch(end_ch); /*xfunc_die(); - redundant */ } if (ch == end_ch IF_HUSH_BASH_COMPAT( || ch == end_char2)) { if (!dbl) break; /* we look for closing )) of $((EXPR)) */ if (i_peek(input) == end_ch) { i_getch(input); /* eat second ')' */ break; } } o_addchr(dest, ch); if (ch == '(' || ch == '{') { ch = (ch == '(' ? ')' : '}'); add_till_closing_bracket(dest, input, ch); o_addchr(dest, ch); continue; } if (ch == '\'') { add_till_single_quote(dest, input); o_addchr(dest, ch); continue; } if (ch == '"') { add_till_double_quote(dest, input); o_addchr(dest, ch); continue; } if (ch == '`') { add_till_backquote(dest, input); o_addchr(dest, ch); continue; } if (ch == '\\') { /* \x. Copy verbatim. Important for \(, \) */ ch = i_getch(input); if (ch == EOF) { syntax_error_unterm_ch(')'); /*xfunc_die(); - redundant */ } o_addchr(dest, ch); continue; } } return ch; } #endif /* ENABLE_HUSH_TICK || ENABLE_SH_MATH_SUPPORT || ENABLE_HUSH_DOLLAR_OPS */ /* Return code: 0 for OK, 1 for syntax error */ #if BB_MMU #define parse_dollar(as_string, dest, input) \ parse_dollar(dest, input) #define as_string NULL #endif static int parse_dollar(o_string *as_string, o_string *dest, struct in_str *input) { int ch = i_peek(input); /* first character after the $ */ unsigned char quote_mask = dest->o_escape ? 0x80 : 0; debug_printf_parse("parse_dollar entered: ch='%c'\n", ch); if (isalpha(ch)) { ch = i_getch(input); nommu_addchr(as_string, ch); make_var: o_addchr(dest, SPECIAL_VAR_SYMBOL); while (1) { debug_printf_parse(": '%c'\n", ch); o_addchr(dest, ch | quote_mask); quote_mask = 0; ch = i_peek(input); if (!isalnum(ch) && ch != '_') break; ch = i_getch(input); nommu_addchr(as_string, ch); } o_addchr(dest, SPECIAL_VAR_SYMBOL); } else if (isdigit(ch)) { make_one_char_var: ch = i_getch(input); nommu_addchr(as_string, ch); o_addchr(dest, SPECIAL_VAR_SYMBOL); debug_printf_parse(": '%c'\n", ch); o_addchr(dest, ch | quote_mask); o_addchr(dest, SPECIAL_VAR_SYMBOL); } else switch (ch) { case '$': /* pid */ case '!': /* last bg pid */ case '?': /* last exit code */ case '#': /* number of args */ case '*': /* args */ case '@': /* args */ goto make_one_char_var; case '{': { o_addchr(dest, SPECIAL_VAR_SYMBOL); ch = i_getch(input); /* eat '{' */ nommu_addchr(as_string, ch); ch = i_getch(input); /* first char after '{' */ nommu_addchr(as_string, ch); /* It should be ${?}, or ${#var}, * or even ${?+subst} - operator acting on a special variable, * or the beginning of variable name. */ if (!strchr(_SPECIAL_VARS_STR, ch) && !isalnum(ch)) { /* not one of those */ bad_dollar_syntax: syntax_error_unterm_str("${name}"); debug_printf_parse("parse_dollar return 1: unterminated ${name}\n"); return 1; } ch |= quote_mask; /* It's possible to just call add_till_closing_bracket() at this point. * However, this regresses some of our testsuite cases * which check invalid constructs like ${%}. * Oh well... let's check that the var name part is fine... */ while (1) { unsigned pos; o_addchr(dest, ch); debug_printf_parse(": '%c'\n", ch); ch = i_getch(input); nommu_addchr(as_string, ch); if (ch == '}') break; if (!isalnum(ch) && ch != '_') { unsigned end_ch; unsigned char last_ch; /* handle parameter expansions * http://www.opengroup.org/onlinepubs/009695399/utilities/xcu_chap02.html#tag_02_06_02 */ if (!strchr(VAR_SUBST_OPS, ch)) /* ${var... */ goto bad_dollar_syntax; /* Eat everything until closing '}' (or ':') */ end_ch = '}'; if (ENABLE_HUSH_BASH_COMPAT && ch == ':' && !strchr(MINUS_PLUS_EQUAL_QUESTION, i_peek(input)) ) { /* It's ${var:N[:M]} thing */ end_ch = '}' * 0x100 + ':'; } if (ENABLE_HUSH_BASH_COMPAT && ch == '/' ) { /* It's ${var/[/]pattern[/repl]} thing */ if (i_peek(input) == '/') { /* ${var//pattern[/repl]}? */ i_getch(input); nommu_addchr(as_string, '/'); ch = '\\'; } end_ch = '}' * 0x100 + '/'; } o_addchr(dest, ch); again: if (!BB_MMU) pos = dest->length; #if ENABLE_HUSH_DOLLAR_OPS last_ch = add_till_closing_bracket(dest, input, end_ch); #else #error Simple code to only allow ${var} is not implemented #endif if (as_string) { o_addstr(as_string, dest->data + pos); o_addchr(as_string, last_ch); } if (ENABLE_HUSH_BASH_COMPAT && (end_ch & 0xff00)) { /* close the first block: */ o_addchr(dest, SPECIAL_VAR_SYMBOL); /* while parsing N from ${var:N[:M]} * or pattern from ${var/[/]pattern[/repl]} */ if ((end_ch & 0xff) == last_ch) { /* got ':' or '/'- parse the rest */ end_ch = '}'; goto again; } /* got '}' */ if (end_ch == '}' * 0x100 + ':') { /* it's ${var:N} - emulate :999999999 */ o_addstr(dest, "999999999"); } /* else: it's ${var/[/]pattern} */ } break; } } o_addchr(dest, SPECIAL_VAR_SYMBOL); break; } #if ENABLE_SH_MATH_SUPPORT || ENABLE_HUSH_TICK case '(': { unsigned pos; ch = i_getch(input); nommu_addchr(as_string, ch); # if ENABLE_SH_MATH_SUPPORT if (i_peek(input) == '(') { ch = i_getch(input); nommu_addchr(as_string, ch); o_addchr(dest, SPECIAL_VAR_SYMBOL); o_addchr(dest, /*quote_mask |*/ '+'); if (!BB_MMU) pos = dest->length; add_till_closing_bracket(dest, input, ')' | DOUBLE_CLOSE_CHAR_FLAG); if (as_string) { o_addstr(as_string, dest->data + pos); o_addchr(as_string, ')'); o_addchr(as_string, ')'); } o_addchr(dest, SPECIAL_VAR_SYMBOL); break; } # endif # if ENABLE_HUSH_TICK o_addchr(dest, SPECIAL_VAR_SYMBOL); o_addchr(dest, quote_mask | '`'); if (!BB_MMU) pos = dest->length; add_till_closing_bracket(dest, input, ')'); if (as_string) { o_addstr(as_string, dest->data + pos); o_addchr(as_string, ')'); } o_addchr(dest, SPECIAL_VAR_SYMBOL); # endif break; } #endif case '_': ch = i_getch(input); nommu_addchr(as_string, ch); ch = i_peek(input); if (isalnum(ch)) { /* it's $_name or $_123 */ ch = '_'; goto make_var; } /* else: it's $_ */ /* TODO: $_ and $-: */ /* $_ Shell or shell script name; or last argument of last command * (if last command wasn't a pipe; if it was, bash sets $_ to ""); * but in command's env, set to full pathname used to invoke it */ /* $- Option flags set by set builtin or shell options (-i etc) */ default: o_addQchr(dest, '$'); } debug_printf_parse("parse_dollar return 0\n"); return 0; #undef as_string } #if BB_MMU #define parse_stream_dquoted(as_string, dest, input, dquote_end) \ parse_stream_dquoted(dest, input, dquote_end) #define as_string NULL #endif static int parse_stream_dquoted(o_string *as_string, o_string *dest, struct in_str *input, int dquote_end) { int ch; int next; again: ch = i_getch(input); if (ch != EOF) nommu_addchr(as_string, ch); if (ch == dquote_end) { /* may be only '"' or EOF */ if (dest->o_assignment == NOT_ASSIGNMENT) dest->o_escape ^= 1; debug_printf_parse("parse_stream_dquoted return 0\n"); return 0; } /* note: can't move it above ch == dquote_end check! */ if (ch == EOF) { syntax_error_unterm_ch('"'); /*xfunc_die(); - redundant */ } next = '\0'; if (ch != '\n') { next = i_peek(input); } debug_printf_parse("\" ch=%c (%d) escape=%d\n", ch, ch, dest->o_escape); if (ch == '\\') { if (next == EOF) { syntax_error("\\"); xfunc_die(); } /* bash: * "The backslash retains its special meaning [in "..."] * only when followed by one of the following characters: * $, `, ", \, or . A double quote may be quoted * within double quotes by preceding it with a backslash." * NB: in (unquoted) heredoc, above does not apply to ". */ if (next == dquote_end || strchr("$`\\\n", next) != NULL) { ch = i_getch(input); if (ch != '\n') { o_addqchr(dest, ch); nommu_addchr(as_string, ch); } } else { o_addqchr(dest, '\\'); nommu_addchr(as_string, '\\'); } goto again; } if (ch == '$') { if (parse_dollar(as_string, dest, input) != 0) { debug_printf_parse("parse_stream_dquoted return 1: " "parse_dollar returned non-0\n"); return 1; } goto again; } #if ENABLE_HUSH_TICK if (ch == '`') { //unsigned pos = dest->length; o_addchr(dest, SPECIAL_VAR_SYMBOL); o_addchr(dest, 0x80 | '`'); add_till_backquote(dest, input); o_addchr(dest, SPECIAL_VAR_SYMBOL); //debug_printf_subst("SUBST RES3 '%s'\n", dest->data + pos); goto again; } #endif o_addQchr(dest, ch); if (ch == '=' && (dest->o_assignment == MAYBE_ASSIGNMENT || dest->o_assignment == WORD_IS_KEYWORD) && is_well_formed_var_name(dest->data, '=') ) { dest->o_assignment = DEFINITELY_ASSIGNMENT; } goto again; #undef as_string } /* * Scan input until EOF or end_trigger char. * Return a list of pipes to execute, or NULL on EOF * or if end_trigger character is met. * On syntax error, exit is shell is not interactive, * reset parsing machinery and start parsing anew, * or return ERR_PTR. */ static struct pipe *parse_stream(char **pstring, struct in_str *input, int end_trigger) { struct parse_context ctx; o_string dest = NULL_O_STRING; int is_in_dquote; int heredoc_cnt; /* Double-quote state is handled in the state variable is_in_dquote. * A single-quote triggers a bypass of the main loop until its mate is * found. When recursing, quote state is passed in via dest->o_escape. */ debug_printf_parse("parse_stream entered, end_trigger='%c'\n", end_trigger ? end_trigger : 'X'); debug_enter(); /* If very first arg is "" or '', dest.data may end up NULL. * Preventing this: */ o_addchr(&dest, '\0'); dest.length = 0; G.ifs = get_local_var_value("IFS"); if (G.ifs == NULL) G.ifs = defifs; reset: #if ENABLE_HUSH_INTERACTIVE input->promptmode = 0; /* PS1 */ #endif /* dest.o_assignment = MAYBE_ASSIGNMENT; - already is */ initialize_context(&ctx); is_in_dquote = 0; heredoc_cnt = 0; while (1) { const char *is_ifs; const char *is_special; int ch; int next; int redir_fd; redir_type redir_style; if (is_in_dquote) { /* dest.has_quoted_part = 1; - already is (see below) */ if (parse_stream_dquoted(&ctx.as_string, &dest, input, '"')) { goto parse_error; } /* We reached closing '"' */ is_in_dquote = 0; } ch = i_getch(input); debug_printf_parse(": ch=%c (%d) escape=%d\n", ch, ch, dest.o_escape); if (ch == EOF) { struct pipe *pi; if (heredoc_cnt) { syntax_error_unterm_str("here document"); goto parse_error; } /* end_trigger == '}' case errors out earlier, * checking only ')' */ if (end_trigger == ')') { syntax_error_unterm_ch('('); /* exits */ /* goto parse_error; */ } if (done_word(&dest, &ctx)) { goto parse_error; } o_free(&dest); done_pipe(&ctx, PIPE_SEQ); pi = ctx.list_head; /* If we got nothing... */ /* (this makes bare "&" cmd a no-op. * bash says: "syntax error near unexpected token '&'") */ if (pi->num_cmds == 0 IF_HAS_KEYWORDS( && pi->res_word == RES_NONE) ) { free_pipe_list(pi); pi = NULL; } #if !BB_MMU debug_printf_parse("as_string '%s'\n", ctx.as_string.data); if (pstring) *pstring = ctx.as_string.data; else o_free_unsafe(&ctx.as_string); #endif debug_leave(); debug_printf_parse("parse_stream return %p\n", pi); return pi; } nommu_addchr(&ctx.as_string, ch); next = '\0'; if (ch != '\n') next = i_peek(input); is_special = "{}<>;&|()#'" /* special outside of "str" */ "\\$\"" IF_HUSH_TICK("`"); /* always special */ /* Are { and } special here? */ if (ctx.command->argv /* word [word]{... - non-special */ || dest.length /* word{... - non-special */ || dest.has_quoted_part /* ""{... - non-special */ || (next != ';' /* }; - special */ && next != ')' /* }) - special */ && next != '&' /* }& and }&& ... - special */ && next != '|' /* }|| ... - special */ && !strchr(G.ifs, next) /* {word - non-special */ ) ) { /* They are not special, skip "{}" */ is_special += 2; } is_special = strchr(is_special, ch); is_ifs = strchr(G.ifs, ch); if (!is_special && !is_ifs) { /* ordinary char */ ordinary_char: o_addQchr(&dest, ch); if ((dest.o_assignment == MAYBE_ASSIGNMENT || dest.o_assignment == WORD_IS_KEYWORD) && ch == '=' && is_well_formed_var_name(dest.data, '=') ) { dest.o_assignment = DEFINITELY_ASSIGNMENT; } continue; } if (is_ifs) { if (done_word(&dest, &ctx)) { goto parse_error; } if (ch == '\n') { #if ENABLE_HUSH_CASE /* "case ... in word) ..." - * newlines are ignored (but ';' wouldn't be) */ if (ctx.command->argv == NULL && ctx.ctx_res_w == RES_MATCH ) { continue; } #endif /* Treat newline as a command separator. */ done_pipe(&ctx, PIPE_SEQ); debug_printf_parse("heredoc_cnt:%d\n", heredoc_cnt); if (heredoc_cnt) { if (fetch_heredocs(heredoc_cnt, &ctx, input)) { goto parse_error; } heredoc_cnt = 0; } dest.o_assignment = MAYBE_ASSIGNMENT; ch = ';'; /* note: if (is_ifs) continue; * will still trigger for us */ } } /* "cmd}" or "cmd }..." without semicolon or &: * } is an ordinary char in this case, even inside { cmd; } * Pathological example: { ""}; } should exec "}" cmd */ if (ch == '}') { if (!IS_NULL_CMD(ctx.command) /* cmd } */ || dest.length != 0 /* word} */ || dest.has_quoted_part /* ""} */ ) { goto ordinary_char; } if (!IS_NULL_PIPE(ctx.pipe)) /* cmd | } */ goto skip_end_trigger; /* else: } does terminate a group */ } if (end_trigger && end_trigger == ch && (ch != ';' || heredoc_cnt == 0) #if ENABLE_HUSH_CASE && (ch != ')' || ctx.ctx_res_w != RES_MATCH || (!dest.has_quoted_part && strcmp(dest.data, "esac") == 0) ) #endif ) { if (heredoc_cnt) { /* This is technically valid: * { cat < before deciding whether this word is * an assignment. a=1 2>z b=2: b=2 is still assignment */ switch (ch) { case '>': redir_fd = redirect_opt_num(&dest); if (done_word(&dest, &ctx)) { goto parse_error; } redir_style = REDIRECT_OVERWRITE; if (next == '>') { redir_style = REDIRECT_APPEND; ch = i_getch(input); nommu_addchr(&ctx.as_string, ch); } #if 0 else if (next == '(') { syntax_error(">(process) not supported"); goto parse_error; } #endif if (parse_redirect(&ctx, redir_fd, redir_style, input)) goto parse_error; continue; /* back to top of while (1) */ case '<': redir_fd = redirect_opt_num(&dest); if (done_word(&dest, &ctx)) { goto parse_error; } redir_style = REDIRECT_INPUT; if (next == '<') { redir_style = REDIRECT_HEREDOC; heredoc_cnt++; debug_printf_parse("++heredoc_cnt=%d\n", heredoc_cnt); ch = i_getch(input); nommu_addchr(&ctx.as_string, ch); } else if (next == '>') { redir_style = REDIRECT_IO; ch = i_getch(input); nommu_addchr(&ctx.as_string, ch); } #if 0 else if (next == '(') { syntax_error("<(process) not supported"); goto parse_error; } #endif if (parse_redirect(&ctx, redir_fd, redir_style, input)) goto parse_error; continue; /* back to top of while (1) */ } if (dest.o_assignment == MAYBE_ASSIGNMENT /* check that we are not in word in "a=1 2>word b=1": */ && !ctx.pending_redirect ) { /* ch is a special char and thus this word * cannot be an assignment */ dest.o_assignment = NOT_ASSIGNMENT; } /* Note: nommu_addchr(&ctx.as_string, ch) is already done */ switch (ch) { case '#': if (dest.length == 0) { while (1) { ch = i_peek(input); if (ch == EOF || ch == '\n') break; i_getch(input); /* note: we do not add it to &ctx.as_string */ } nommu_addchr(&ctx.as_string, '\n'); } else { o_addQchr(&dest, ch); } break; case '\\': if (next == EOF) { syntax_error("\\"); xfunc_die(); } ch = i_getch(input); if (ch != '\n') { o_addchr(&dest, '\\'); /*nommu_addchr(&ctx.as_string, '\\'); - already done */ o_addchr(&dest, ch); nommu_addchr(&ctx.as_string, ch); /* Example: echo Hello \2>file * we need to know that word 2 is quoted */ dest.has_quoted_part = 1; } #if !BB_MMU else { /* It's "\". Remove trailing '\' from ctx.as_string */ ctx.as_string.data[--ctx.as_string.length] = '\0'; } #endif break; case '$': if (parse_dollar(&ctx.as_string, &dest, input) != 0) { debug_printf_parse("parse_stream parse error: " "parse_dollar returned non-0\n"); goto parse_error; } break; case '\'': dest.has_quoted_part = 1; while (1) { ch = i_getch(input); if (ch == EOF) { syntax_error_unterm_ch('\''); /*xfunc_die(); - redundant */ } nommu_addchr(&ctx.as_string, ch); if (ch == '\'') break; o_addqchr(&dest, ch); } break; case '"': dest.has_quoted_part = 1; is_in_dquote ^= 1; /* invert */ if (dest.o_assignment == NOT_ASSIGNMENT) dest.o_escape ^= 1; break; #if ENABLE_HUSH_TICK case '`': { unsigned pos; o_addchr(&dest, SPECIAL_VAR_SYMBOL); o_addchr(&dest, '`'); pos = dest.length; add_till_backquote(&dest, input); # if !BB_MMU o_addstr(&ctx.as_string, dest.data + pos); o_addchr(&ctx.as_string, '`'); # endif o_addchr(&dest, SPECIAL_VAR_SYMBOL); //debug_printf_subst("SUBST RES3 '%s'\n", dest.data + pos); break; } #endif case ';': #if ENABLE_HUSH_CASE case_semi: #endif if (done_word(&dest, &ctx)) { goto parse_error; } done_pipe(&ctx, PIPE_SEQ); #if ENABLE_HUSH_CASE /* Eat multiple semicolons, detect * whether it means something special */ while (1) { ch = i_peek(input); if (ch != ';') break; ch = i_getch(input); nommu_addchr(&ctx.as_string, ch); if (ctx.ctx_res_w == RES_CASE_BODY) { ctx.ctx_dsemicolon = 1; ctx.ctx_res_w = RES_MATCH; break; } } #endif new_cmd: /* We just finished a cmd. New one may start * with an assignment */ dest.o_assignment = MAYBE_ASSIGNMENT; break; case '&': if (done_word(&dest, &ctx)) { goto parse_error; } if (next == '&') { ch = i_getch(input); nommu_addchr(&ctx.as_string, ch); done_pipe(&ctx, PIPE_AND); } else { done_pipe(&ctx, PIPE_BG); } goto new_cmd; case '|': if (done_word(&dest, &ctx)) { goto parse_error; } #if ENABLE_HUSH_CASE if (ctx.ctx_res_w == RES_MATCH) break; /* we are in case's "word | word)" */ #endif if (next == '|') { /* || */ ch = i_getch(input); nommu_addchr(&ctx.as_string, ch); done_pipe(&ctx, PIPE_OR); } else { /* we could pick up a file descriptor choice here * with redirect_opt_num(), but bash doesn't do it. * "echo foo 2| cat" yields "foo 2". */ done_command(&ctx); #if !BB_MMU o_reset_to_empty_unquoted(&ctx.as_string); #endif } goto new_cmd; case '(': #if ENABLE_HUSH_CASE /* "case... in [(]word)..." - skip '(' */ if (ctx.ctx_res_w == RES_MATCH && ctx.command->argv == NULL /* not (word|(... */ && dest.length == 0 /* not word(... */ && dest.has_quoted_part == 0 /* not ""(... */ ) { continue; } #endif case '{': if (parse_group(&dest, &ctx, input, ch) != 0) { goto parse_error; } goto new_cmd; case ')': #if ENABLE_HUSH_CASE if (ctx.ctx_res_w == RES_MATCH) goto case_semi; #endif case '}': /* proper use of this character is caught by end_trigger: * if we see {, we call parse_group(..., end_trigger='}') * and it will match } earlier (not here). */ syntax_error_unexpected_ch(ch); goto parse_error; default: if (HUSH_DEBUG) bb_error_msg_and_die("BUG: unexpected %c\n", ch); } } /* while (1) */ parse_error: { struct parse_context *pctx; IF_HAS_KEYWORDS(struct parse_context *p2;) /* Clean up allocated tree. * Sample for finding leaks on syntax error recovery path. * Run it from interactive shell, watch pmap `pidof hush`. * while if false; then false; fi; do break; fi * Samples to catch leaks at execution: * while if (true | {true;}); then echo ok; fi; do break; done * while if (true | {true;}); then echo ok; fi; do (if echo ok; break; then :; fi) | cat; break; done */ pctx = &ctx; do { /* Update pipe/command counts, * otherwise freeing may miss some */ done_pipe(pctx, PIPE_SEQ); debug_printf_clean("freeing list %p from ctx %p\n", pctx->list_head, pctx); debug_print_tree(pctx->list_head, 0); free_pipe_list(pctx->list_head); debug_printf_clean("freed list %p\n", pctx->list_head); #if !BB_MMU o_free_unsafe(&pctx->as_string); #endif IF_HAS_KEYWORDS(p2 = pctx->stack;) if (pctx != &ctx) { free(pctx); } IF_HAS_KEYWORDS(pctx = p2;) } while (HAS_KEYWORDS && pctx); /* Free text, clear all dest fields */ o_free(&dest); /* If we are not in top-level parse, we return, * our caller will propagate error. */ if (end_trigger != ';') { #if !BB_MMU if (pstring) *pstring = NULL; #endif debug_leave(); return ERR_PTR; } /* Discard cached input, force prompt */ input->p = NULL; IF_HUSH_INTERACTIVE(input->promptme = 1;) goto reset; } } /*** Execution routines ***/ /* Expansion can recurse, need forward decls: */ static char *expand_string_to_string(const char *str); static int process_command_subs(o_string *dest, const char *s); /* expand_strvec_to_strvec() takes a list of strings, expands * all variable references within and returns a pointer to * a list of expanded strings, possibly with larger number * of strings. (Think VAR="a b"; echo $VAR). * This new list is allocated as a single malloc block. * NULL-terminated list of char* pointers is at the beginning of it, * followed by strings themself. * Caller can deallocate entire list by single free(list). */ /* Store given string, finalizing the word and starting new one whenever * we encounter IFS char(s). This is used for expanding variable values. * End-of-string does NOT finalize word: think about 'echo -$VAR-' */ static int expand_on_ifs(o_string *output, int n, const char *str) { while (1) { int word_len = strcspn(str, G.ifs); if (word_len) { if (output->o_escape) o_addqblock(output, str, word_len); else if (!output->o_glob) o_addblock(output, str, word_len); else /* if (!escape && glob) */ { /* Protect backslashes against globbing up :) * Example: "v='\*'; echo b$v" */ o_addblock_duplicate_backslash(output, str, word_len); /*/ Why can't we do it easier? */ /*o_addblock(output, str, word_len); - WRONG: "v='\*'; echo Z$v" prints "Z*" instead of "Z\*" */ /*o_addqblock(output, str, word_len); - WRONG: "v='*'; echo Z$v" prints "Z*" instead of Z* files */ } str += word_len; } if (!*str) /* EOL - do not finalize word */ break; o_addchr(output, '\0'); debug_print_list("expand_on_ifs", output, n); n = o_save_ptr(output, n); str += strspn(str, G.ifs); /* skip ifs chars */ } debug_print_list("expand_on_ifs[1]", output, n); return n; } /* Helper to expand $((...)) and heredoc body. These act as if * they are in double quotes, with the exception that they are not :). * Just the rules are similar: "expand only $var and `cmd`" * * Returns malloced string. * As an optimization, we return NULL if expansion is not needed. */ static char *expand_pseudo_dquoted(const char *str) { char *exp_str; struct in_str input; o_string dest = NULL_O_STRING; if (!strchr(str, '$') && !strchr(str, '\\') #if ENABLE_HUSH_TICK && !strchr(str, '`') #endif ) { return NULL; } /* We need to expand. Example: * echo $(($a + `echo 1`)) $((1 + $((2)) )) */ setup_string_in_str(&input, str); parse_stream_dquoted(NULL, &dest, &input, EOF); //bb_error_msg("'%s' -> '%s'", str, dest.data); exp_str = expand_string_to_string(dest.data); //bb_error_msg("'%s' -> '%s'", dest.data, exp_str); o_free_unsafe(&dest); return exp_str; } #if ENABLE_SH_MATH_SUPPORT static arith_t expand_and_evaluate_arith(const char *arg, int *errcode_p) { arith_eval_hooks_t hooks; arith_t res; char *exp_str; hooks.lookupvar = get_local_var_value; hooks.setvar = set_local_var_from_halves; hooks.endofname = endofname; exp_str = expand_pseudo_dquoted(arg); res = arith(exp_str ? exp_str : arg, errcode_p, &hooks); free(exp_str); return res; } #endif #if ENABLE_HUSH_BASH_COMPAT /* ${var/[/]pattern[/repl]} helpers */ static char *strstr_pattern(char *val, const char *pattern, int *size) { while (1) { char *end = scan_and_match(val, pattern, SCAN_MOVE_FROM_RIGHT + SCAN_MATCH_LEFT_HALF); debug_printf_varexp("val:'%s' pattern:'%s' end:'%s'\n", val, pattern, end); if (end) { *size = end - val; return val; } if (*val == '\0') return NULL; /* Optimization: if "*pat" did not match the start of "string", * we know that "tring", "ring" etc will not match too: */ if (pattern[0] == '*') return NULL; val++; } } static char *replace_pattern(char *val, const char *pattern, const char *repl, char exp_op) { char *result = NULL; unsigned res_len = 0; unsigned repl_len = strlen(repl); while (1) { int size; char *s = strstr_pattern(val, pattern, &size); if (!s) break; result = xrealloc(result, res_len + (s - val) + repl_len + 1); memcpy(result + res_len, val, s - val); res_len += s - val; strcpy(result + res_len, repl); res_len += repl_len; debug_printf_varexp("val:'%s' s:'%s' result:'%s'\n", val, s, result); val = s + size; if (exp_op == '/') break; } if (val[0] && result) { result = xrealloc(result, res_len + strlen(val) + 1); strcpy(result + res_len, val); debug_printf_varexp("val:'%s' result:'%s'\n", val, result); } debug_printf_varexp("result:'%s'\n", result); return result; } #endif /* Helper: * Handles varname... construct. */ static NOINLINE const char *expand_one_var(char **to_be_freed_pp, char *arg, char **pp, char first_ch) { const char *val = NULL; char *to_be_freed = NULL; char *p = *pp; char *var; char first_char; char exp_op; char exp_save = exp_save; /* for compiler */ char *exp_saveptr; /* points to expansion operator */ char *exp_word = exp_word; /* for compiler */ var = arg; *p = '\0'; exp_saveptr = arg[1] ? strchr(VAR_ENCODED_SUBST_OPS, arg[1]) : NULL; first_char = arg[0] = first_ch & 0x7f; exp_op = 0; if (first_char == '#' && arg[1] && !exp_saveptr) { /* handle length expansion ${#var} */ var++; exp_op = 'L'; } else { /* maybe handle parameter expansion */ if (exp_saveptr /* if 2nd char is one of expansion operators */ && strchr(NUMERIC_SPECVARS_STR, first_char) /* 1st char is special variable */ ) { /* ${?:0}, ${#[:]%0} etc */ exp_saveptr = var + 1; } else { /* ${?}, ${var}, ${var:0}, ${var[:]%0} etc */ exp_saveptr = var+1 + strcspn(var+1, VAR_ENCODED_SUBST_OPS); } exp_op = exp_save = *exp_saveptr; if (exp_op) { exp_word = exp_saveptr + 1; if (exp_op == ':') { exp_op = *exp_word++; if (ENABLE_HUSH_BASH_COMPAT && (exp_op == '\0' || !strchr(MINUS_PLUS_EQUAL_QUESTION, exp_op)) ) { /* oops... it's ${var:N[:M]}, not ${var:?xxx} or some such */ exp_op = ':'; exp_word--; } } *exp_saveptr = '\0'; } /* else: it's not an expansion op, but bare ${var} */ } /* lookup the variable in question */ if (isdigit(var[0])) { /* parse_dollar() should have vetted var for us */ int n = xatoi_positive(var); if (n < G.global_argc) val = G.global_argv[n]; /* else val remains NULL: $N with too big N */ } else { switch (var[0]) { case '$': /* pid */ val = utoa(G.root_pid); break; case '!': /* bg pid */ val = G.last_bg_pid ? utoa(G.last_bg_pid) : ""; break; case '?': /* exitcode */ val = utoa(G.last_exitcode); break; case '#': /* argc */ val = utoa(G.global_argc ? G.global_argc-1 : 0); break; default: val = get_local_var_value(var); } } /* Handle any expansions */ if (exp_op == 'L') { debug_printf_expand("expand: length(%s)=", val); val = utoa(val ? strlen(val) : 0); debug_printf_expand("%s\n", val); } else if (exp_op) { if (exp_op == '%' || exp_op == '#') { /* Standard-mandated substring removal ops: * ${parameter%word} - remove smallest suffix pattern * ${parameter%%word} - remove largest suffix pattern * ${parameter#word} - remove smallest prefix pattern * ${parameter##word} - remove largest prefix pattern * * Word is expanded to produce a glob pattern. * Then var's value is matched to it and matching part removed. */ if (val && val[0]) { char *exp_exp_word; char *loc; unsigned scan_flags = pick_scan(exp_op, *exp_word); if (exp_op == *exp_word) /* ## or %% */ exp_word++; //TODO: avoid xstrdup unless needed // (see HACK ALERT below for an example) val = to_be_freed = xstrdup(val); //TODO: fix expansion rules: exp_exp_word = expand_pseudo_dquoted(exp_word); if (exp_exp_word) exp_word = exp_exp_word; loc = scan_and_match(to_be_freed, exp_word, scan_flags); //bb_error_msg("op:%c str:'%s' pat:'%s' res:'%s'", // exp_op, to_be_freed, exp_word, loc); free(exp_exp_word); if (loc) { /* match was found */ if (scan_flags & SCAN_MATCH_LEFT_HALF) /* #[#] */ val = loc; else /* %[%] */ *loc = '\0'; } } } #if ENABLE_HUSH_BASH_COMPAT else if (exp_op == '/' || exp_op == '\\') { /* It's ${var/[/]pattern[/repl]} thing. * Note that in encoded form it has TWO parts: * var/patternrepl */ /* Empty variable always gives nothing: */ // "v=''; echo ${v/*/w}" prints "", not "w" if (val && val[0]) { /* It's ${var/[/]pattern[/repl]} thing */ /* * Pattern is taken literally, while * repl should be de-backslased and globbed * by the usual expansion rules: * >az; >bz; * v='a bz'; echo "${v/a*z/a*z}" prints "a*z" * v='a bz'; echo "${v/a*z/\z}" prints "\z" * v='a bz'; echo ${v/a*z/a*z} prints "az" * v='a bz'; echo ${v/a*z/\z} prints "z" * (note that a*z _pattern_ is never globbed!) */ //TODO: fix expansion rules: char *pattern, *repl, *t; pattern = expand_pseudo_dquoted(exp_word); if (!pattern) pattern = xstrdup(exp_word); debug_printf_varexp("pattern:'%s'->'%s'\n", exp_word, pattern); *p++ = SPECIAL_VAR_SYMBOL; exp_word = p; p = strchr(p, SPECIAL_VAR_SYMBOL); *p = '\0'; repl = expand_pseudo_dquoted(exp_word); debug_printf_varexp("repl:'%s'->'%s'\n", exp_word, repl); /* HACK ALERT. We depend here on the fact that * G.global_argv and results of utoa and get_local_var_value * are actually in writable memory: * replace_pattern momentarily stores NULs there. */ t = (char*)val; to_be_freed = replace_pattern(t, pattern, (repl ? repl : exp_word), exp_op); if (to_be_freed) /* at least one replace happened */ val = to_be_freed; free(pattern); free(repl); } } #endif else if (exp_op == ':') { #if ENABLE_HUSH_BASH_COMPAT && ENABLE_SH_MATH_SUPPORT /* It's ${var:N[:M]} bashism. * Note that in encoded form it has TWO parts: * var:NM */ arith_t beg, len; int errcode = 0; beg = expand_and_evaluate_arith(exp_word, &errcode); debug_printf_varexp("beg:'%s'=%lld\n", exp_word, (long long)beg); *p++ = SPECIAL_VAR_SYMBOL; exp_word = p; p = strchr(p, SPECIAL_VAR_SYMBOL); *p = '\0'; len = expand_and_evaluate_arith(exp_word, &errcode); debug_printf_varexp("len:'%s'=%lld\n", exp_word, (long long)len); if (errcode >= 0 && len >= 0) { /* bash compat: len < 0 is illegal */ if (beg < 0) /* bash compat */ beg = 0; debug_printf_varexp("from val:'%s'\n", val); if (len == 0 || !val || beg >= strlen(val)) val = ""; else { /* Paranoia. What if user entered 9999999999999 * which fits in arith_t but not int? */ if (len >= INT_MAX) len = INT_MAX; val = to_be_freed = xstrndup(val + beg, len); } debug_printf_varexp("val:'%s'\n", val); } else #endif { die_if_script("malformed ${%s:...}", var); val = ""; } } else { /* one of "-=+?" */ /* Standard-mandated substitution ops: * ${var?word} - indicate error if unset * If var is unset, word (or a message indicating it is unset * if word is null) is written to standard error * and the shell exits with a non-zero exit status. * Otherwise, the value of var is substituted. * ${var-word} - use default value * If var is unset, word is substituted. * ${var=word} - assign and use default value * If var is unset, word is assigned to var. * In all cases, final value of var is substituted. * ${var+word} - use alternative value * If var is unset, null is substituted. * Otherwise, word is substituted. * * Word is subjected to tilde expansion, parameter expansion, * command substitution, and arithmetic expansion. * If word is not needed, it is not expanded. * * Colon forms (${var:-word}, ${var:=word} etc) do the same, * but also treat null var as if it is unset. */ int use_word = (!val || ((exp_save == ':') && !val[0])); if (exp_op == '+') use_word = !use_word; debug_printf_expand("expand: op:%c (null:%s) test:%i\n", exp_op, (exp_save == ':') ? "true" : "false", use_word); if (use_word) { to_be_freed = expand_pseudo_dquoted(exp_word); if (to_be_freed) exp_word = to_be_freed; if (exp_op == '?') { /* mimic bash message */ die_if_script("%s: %s", var, exp_word[0] ? exp_word : "parameter null or not set" ); //TODO: how interactive bash aborts expansion mid-command? } else { val = exp_word; } if (exp_op == '=') { /* ${var=[word]} or ${var:=[word]} */ if (isdigit(var[0]) || var[0] == '#') { /* mimic bash message */ die_if_script("$%s: cannot assign in this way", var); val = NULL; } else { char *new_var = xasprintf("%s=%s", var, val); set_local_var(new_var, /*exp:*/ 0, /*lvl:*/ 0, /*ro:*/ 0); } } } } /* one of "-=+?" */ *exp_saveptr = exp_save; } /* if (exp_op) */ arg[0] = first_ch; *pp = p; *to_be_freed_pp = to_be_freed; return val; } /* Expand all variable references in given string, adding words to list[] * at n, n+1,... positions. Return updated n (so that list[n] is next one * to be filled). This routine is extremely tricky: has to deal with * variables/parameters with whitespace, $* and $@, and constructs like * 'echo -$*-'. If you play here, you must run testsuite afterwards! */ static NOINLINE int expand_vars_to_list(o_string *output, int n, char *arg, char or_mask) { /* or_mask is either 0 (normal case) or 0x80 - * expansion of right-hand side of assignment == 1-element expand. * It will also do no globbing, and thus we must not backslash-quote! */ char ored_ch; char *p; ored_ch = 0; debug_printf_expand("expand_vars_to_list: arg:'%s' or_mask:%x\n", arg, or_mask); debug_print_list("expand_vars_to_list", output, n); n = o_save_ptr(output, n); debug_print_list("expand_vars_to_list[0]", output, n); while ((p = strchr(arg, SPECIAL_VAR_SYMBOL)) != NULL) { char first_ch; char *to_be_freed = NULL; const char *val = NULL; #if ENABLE_HUSH_TICK o_string subst_result = NULL_O_STRING; #endif #if ENABLE_SH_MATH_SUPPORT char arith_buf[sizeof(arith_t)*3 + 2]; #endif o_addblock(output, arg, p - arg); debug_print_list("expand_vars_to_list[1]", output, n); arg = ++p; p = strchr(p, SPECIAL_VAR_SYMBOL); first_ch = arg[0] | or_mask; /* forced to "quoted" if or_mask = 0x80 */ /* "$@" is special. Even if quoted, it can still * expand to nothing (not even an empty string) */ if ((first_ch & 0x7f) != '@') ored_ch |= first_ch; switch (first_ch & 0x7f) { /* Highest bit in first_ch indicates that var is double-quoted */ case '*': case '@': { int i; if (!G.global_argv[1]) break; i = 1; ored_ch |= first_ch; /* do it for "$@" _now_, when we know it's not empty */ if (!(first_ch & 0x80)) { /* unquoted $* or $@ */ smallint sv = output->o_escape; /* unquoted var's contents should be globbed, so don't escape */ output->o_escape = 0; while (G.global_argv[i]) { n = expand_on_ifs(output, n, G.global_argv[i]); debug_printf_expand("expand_vars_to_list: argv %d (last %d)\n", i, G.global_argc - 1); if (G.global_argv[i++][0] && G.global_argv[i]) { /* this argv[] is not empty and not last: * put terminating NUL, start new word */ o_addchr(output, '\0'); debug_print_list("expand_vars_to_list[2]", output, n); n = o_save_ptr(output, n); debug_print_list("expand_vars_to_list[3]", output, n); } } output->o_escape = sv; } else /* If or_mask is nonzero, we handle assignment 'a=....$@.....' * and in this case should treat it like '$*' - see 'else...' below */ if (first_ch == ('@'|0x80) && !or_mask) { /* quoted $@ */ while (1) { o_addQstr(output, G.global_argv[i]); if (++i >= G.global_argc) break; o_addchr(output, '\0'); debug_print_list("expand_vars_to_list[4]", output, n); n = o_save_ptr(output, n); } } else { /* quoted $*: add as one word */ while (1) { o_addQstr(output, G.global_argv[i]); if (!G.global_argv[++i]) break; if (G.ifs[0]) o_addchr(output, G.ifs[0]); } } break; } case SPECIAL_VAR_SYMBOL: /* */ /* "Empty variable", used to make "" etc to not disappear */ arg++; ored_ch = 0x80; break; #if ENABLE_HUSH_TICK case '`': /* `cmd */ *p = '\0'; arg++; /* Can't just stuff it into output o_string, * expanded result may need to be globbed * and $IFS-splitted */ debug_printf_subst("SUBST '%s' first_ch %x\n", arg, first_ch); G.last_exitcode = process_command_subs(&subst_result, arg); debug_printf_subst("SUBST RES:%d '%s'\n", G.last_exitcode, subst_result.data); val = subst_result.data; goto store_val; #endif #if ENABLE_SH_MATH_SUPPORT case '+': { /* +cmd */ arith_t res; int errcode; arg++; /* skip '+' */ *p = '\0'; /* replace trailing */ debug_printf_subst("ARITH '%s' first_ch %x\n", arg, first_ch); res = expand_and_evaluate_arith(arg, &errcode); if (errcode < 0) { const char *msg = "error in arithmetic"; switch (errcode) { case -3: msg = "exponent less than 0"; break; case -2: msg = "divide by 0"; break; case -5: msg = "expression recursion loop detected"; break; } die_if_script(msg); } debug_printf_subst("ARITH RES '"arith_t_fmt"'\n", res); sprintf(arith_buf, arith_t_fmt, res); val = arith_buf; break; } #endif default: val = expand_one_var(&to_be_freed, arg, &p, first_ch); IF_HUSH_TICK(store_val:) if (!(first_ch & 0x80)) { /* unquoted $VAR */ debug_printf_expand("unquoted '%s', output->o_escape:%d\n", val, output->o_escape); if (val && val[0]) { /* unquoted var's contents should be globbed, so don't escape */ smallint sv = output->o_escape; output->o_escape = 0; n = expand_on_ifs(output, n, val); val = NULL; output->o_escape = sv; } } else { /* quoted $VAR, val will be appended below */ debug_printf_expand("quoted '%s', output->o_escape:%d\n", val, output->o_escape); } break; } /* switch (char after ) */ if (val && val[0]) { o_addQstr(output, val); } free(to_be_freed); /* Do the check to avoid writing to a const string */ if (*p != SPECIAL_VAR_SYMBOL) *p = SPECIAL_VAR_SYMBOL; #if ENABLE_HUSH_TICK o_free(&subst_result); #endif arg = ++p; } /* end of "while (SPECIAL_VAR_SYMBOL is found) ..." */ if (arg[0]) { debug_print_list("expand_vars_to_list[a]", output, n); /* this part is literal, and it was already pre-quoted * if needed (much earlier), do not use o_addQstr here! */ o_addstr_with_NUL(output, arg); debug_print_list("expand_vars_to_list[b]", output, n); } else if (output->length == o_get_last_ptr(output, n) /* expansion is empty */ && !(ored_ch & 0x80) /* and all vars were not quoted. */ ) { n--; /* allow to reuse list[n] later without re-growth */ output->has_empty_slot = 1; } else { o_addchr(output, '\0'); } return n; } enum { EXPVAR_FLAG_GLOB = 0x200, EXPVAR_FLAG_ESCAPE_VARS = 0x100, EXPVAR_FLAG_SINGLEWORD = 0x80, /* must be 0x80 */ }; static char **expand_variables(char **argv, unsigned or_mask) { int n; char **list; o_string output = NULL_O_STRING; /* protect against globbing for "$var"? */ /* (unquoted $var will temporarily switch it off) */ output.o_escape = 1 & (or_mask / EXPVAR_FLAG_ESCAPE_VARS); output.o_glob = 1 & (or_mask / EXPVAR_FLAG_GLOB); n = 0; while (*argv) { n = expand_vars_to_list(&output, n, *argv, (unsigned char)or_mask); argv++; } debug_print_list("expand_variables", &output, n); /* output.data (malloced in one block) gets returned in "list" */ list = o_finalize_list(&output, n); debug_print_strings("expand_variables[1]", list); return list; } static char **expand_strvec_to_strvec(char **argv) { return expand_variables(argv, EXPVAR_FLAG_GLOB | EXPVAR_FLAG_ESCAPE_VARS); } #if ENABLE_HUSH_BASH_COMPAT static char **expand_strvec_to_strvec_singleword_noglob(char **argv) { return expand_variables(argv, EXPVAR_FLAG_SINGLEWORD); } #endif /* Used for expansion of right hand of assignments */ /* NB: should NOT do globbing! * "export v=/bin/c*; env | grep ^v=" outputs "v=/bin/c*" */ static char *expand_string_to_string(const char *str) { char *argv[2], **list; /* This is generally an optimization, but it also * handles "", which otherwise trips over !list[0] check below. * (is this ever happens that we actually get str="" here?) */ if (!strchr(str, SPECIAL_VAR_SYMBOL) && !strchr(str, '\\')) { //TODO: Can use on strings with \ too, just unbackslash() them? debug_printf_expand("string_to_string(fast)='%s'\n", str); return xstrdup(str); } argv[0] = (char*)str; argv[1] = NULL; list = expand_variables(argv, EXPVAR_FLAG_ESCAPE_VARS | EXPVAR_FLAG_SINGLEWORD); if (HUSH_DEBUG) if (!list[0] || list[1]) bb_error_msg_and_die("BUG in varexp2"); /* actually, just move string 2*sizeof(char*) bytes back */ overlapping_strcpy((char*)list, list[0]); unbackslash((char*)list); debug_printf_expand("string_to_string='%s'\n", (char*)list); return (char*)list; } /* Used for "eval" builtin */ static char* expand_strvec_to_string(char **argv) { char **list; list = expand_variables(argv, EXPVAR_FLAG_SINGLEWORD); /* Convert all NULs to spaces */ if (list[0]) { int n = 1; while (list[n]) { if (HUSH_DEBUG) if (list[n-1] + strlen(list[n-1]) + 1 != list[n]) bb_error_msg_and_die("BUG in varexp3"); /* bash uses ' ' regardless of $IFS contents */ list[n][-1] = ' '; n++; } } overlapping_strcpy((char*)list, list[0]); debug_printf_expand("strvec_to_string='%s'\n", (char*)list); return (char*)list; } static char **expand_assignments(char **argv, int count) { int i; char **p; G.expanded_assignments = p = NULL; /* Expand assignments into one string each */ for (i = 0; i < count; i++) { G.expanded_assignments = p = add_string_to_strings(p, expand_string_to_string(argv[i])); } G.expanded_assignments = NULL; return p; } #if BB_MMU /* never called */ void re_execute_shell(char ***to_free, const char *s, char *g_argv0, char **g_argv, char **builtin_argv) NORETURN; static void reset_traps_to_defaults(void) { /* This function is always called in a child shell * after fork (not vfork, NOMMU doesn't use this function). */ unsigned sig; unsigned mask; /* Child shells are not interactive. * SIGTTIN/SIGTTOU/SIGTSTP should not have special handling. * Testcase: (while :; do :; done) + ^Z should background. * Same goes for SIGTERM, SIGHUP, SIGINT. */ if (!G.traps && !(G.non_DFL_mask & SPECIAL_INTERACTIVE_SIGS)) return; /* already no traps and no SPECIAL_INTERACTIVE_SIGS */ /* Switching off SPECIAL_INTERACTIVE_SIGS. * Stupid. It can be done with *single* &= op, but we can't use * the fact that G.blocked_set is implemented as a bitmask * in libc... */ mask = (SPECIAL_INTERACTIVE_SIGS >> 1); sig = 1; while (1) { if (mask & 1) { /* Careful. Only if no trap or trap is not "" */ if (!G.traps || !G.traps[sig] || G.traps[sig][0]) sigdelset(&G.blocked_set, sig); } mask >>= 1; if (!mask) break; sig++; } /* Our homegrown sig mask is saner to work with :) */ G.non_DFL_mask &= ~SPECIAL_INTERACTIVE_SIGS; /* Resetting all traps to default except empty ones */ mask = G.non_DFL_mask; if (G.traps) for (sig = 0; sig < NSIG; sig++, mask >>= 1) { if (!G.traps[sig] || !G.traps[sig][0]) continue; free(G.traps[sig]); G.traps[sig] = NULL; /* There is no signal for 0 (EXIT) */ if (sig == 0) continue; /* There was a trap handler, we just removed it. * But if sig still has non-DFL handling, * we should not unblock the sig. */ if (mask & 1) continue; sigdelset(&G.blocked_set, sig); } sigprocmask(SIG_SETMASK, &G.blocked_set, NULL); } #else /* !BB_MMU */ static void re_execute_shell(char ***to_free, const char *s, char *g_argv0, char **g_argv, char **builtin_argv) NORETURN; static void re_execute_shell(char ***to_free, const char *s, char *g_argv0, char **g_argv, char **builtin_argv) { # define NOMMU_HACK_FMT ("-$%x:%x:%x:%x:%x:%llx" IF_HUSH_LOOPS(":%x")) /* delims + 2 * (number of bytes in printed hex numbers) */ char param_buf[sizeof(NOMMU_HACK_FMT) + 2 * (sizeof(int)*6 + sizeof(long long)*1)]; char *heredoc_argv[4]; struct variable *cur; # if ENABLE_HUSH_FUNCTIONS struct function *funcp; # endif char **argv, **pp; unsigned cnt; unsigned long long empty_trap_mask; if (!g_argv0) { /* heredoc */ argv = heredoc_argv; argv[0] = (char *) G.argv0_for_re_execing; argv[1] = (char *) "-<"; argv[2] = (char *) s; argv[3] = NULL; pp = &argv[3]; /* used as pointer to empty environment */ goto do_exec; } cnt = 0; pp = builtin_argv; if (pp) while (*pp++) cnt++; empty_trap_mask = 0; if (G.traps) { int sig; for (sig = 1; sig < NSIG; sig++) { if (G.traps[sig] && !G.traps[sig][0]) empty_trap_mask |= 1LL << sig; } } sprintf(param_buf, NOMMU_HACK_FMT , (unsigned) G.root_pid , (unsigned) G.root_ppid , (unsigned) G.last_bg_pid , (unsigned) G.last_exitcode , cnt , empty_trap_mask IF_HUSH_LOOPS(, G.depth_of_loop) ); # undef NOMMU_HACK_FMT /* 1:hush 2:-$::: * 3:-c 4: 5: 6:NULL */ cnt += 6; for (cur = G.top_var; cur; cur = cur->next) { if (!cur->flg_export || cur->flg_read_only) cnt += 2; } # if ENABLE_HUSH_FUNCTIONS for (funcp = G.top_func; funcp; funcp = funcp->next) cnt += 3; # endif pp = g_argv; while (*pp++) cnt++; *to_free = argv = pp = xzalloc(sizeof(argv[0]) * cnt); *pp++ = (char *) G.argv0_for_re_execing; *pp++ = param_buf; for (cur = G.top_var; cur; cur = cur->next) { if (strcmp(cur->varstr, hush_version_str) == 0) continue; if (cur->flg_read_only) { *pp++ = (char *) "-R"; *pp++ = cur->varstr; } else if (!cur->flg_export) { *pp++ = (char *) "-V"; *pp++ = cur->varstr; } } # if ENABLE_HUSH_FUNCTIONS for (funcp = G.top_func; funcp; funcp = funcp->next) { *pp++ = (char *) "-F"; *pp++ = funcp->name; *pp++ = funcp->body_as_string; } # endif /* We can pass activated traps here. Say, -Tnn:trap_string * * However, POSIX says that subshells reset signals with traps * to SIG_DFL. * I tested bash-3.2 and it not only does that with true subshells * of the form ( list ), but with any forked children shells. * I set trap "echo W" WINCH; and then tried: * * { echo 1; sleep 20; echo 2; } & * while true; do echo 1; sleep 20; echo 2; break; done & * true | { echo 1; sleep 20; echo 2; } | cat * * In all these cases sending SIGWINCH to the child shell * did not run the trap. If I add trap "echo V" WINCH; * _inside_ group (just before echo 1), it works. * * I conclude it means we don't need to pass active traps here. * Even if we would use signal handlers instead of signal masking * in order to implement trap handling, * exec syscall below resets signals to SIG_DFL for us. */ *pp++ = (char *) "-c"; *pp++ = (char *) s; if (builtin_argv) { while (*++builtin_argv) *pp++ = *builtin_argv; *pp++ = (char *) ""; } *pp++ = g_argv0; while (*g_argv) *pp++ = *g_argv++; /* *pp = NULL; - is already there */ pp = environ; do_exec: debug_printf_exec("re_execute_shell pid:%d cmd:'%s'\n", getpid(), s); sigprocmask(SIG_SETMASK, &G.inherited_set, NULL); execve(bb_busybox_exec_path, argv, pp); /* Fallback. Useful for init=/bin/hush usage etc */ if (argv[0][0] == '/') execve(argv[0], argv, pp); xfunc_error_retval = 127; bb_error_msg_and_die("can't re-execute the shell"); } #endif /* !BB_MMU */ static int run_and_free_list(struct pipe *pi); /* Executing from string: eval, sh -c '...' * or from file: /etc/profile, . file, sh