/* 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 * * Licensed under GPLv2 or later, see file LICENSE in this source tree. * * 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) * make complex ${var%...} constructs support optional * make here documents optional * special variables (done: PWD, PPID, RANDOM) * follow IFS rules more precisely, including update semantics * tilde expansion * aliases * "command" missing features: * command -p CMD: run CMD using default $PATH * (can use this to override standalone shell as well?) * command BLTIN: disables special-ness (e.g. errors do not abort) * command -V CMD1 CMD2 CMD3 (multiple args) (not in standard) * builtins mandated by standards we don't support: * [un]alias, fc: * fc -l[nr] [BEG] [END]: list range of commands in history * fc [-e EDITOR] [BEG] [END]: edit/rerun range of commands * fc -s [PAT=REP] [CMD]: rerun CMD, replacing PAT with REP * * Bash compat TODO: * redirection of stdout+stderr: &> and >& * reserved words: function select * advanced test: [[ ]] * process substitution: <(list) and >(list) * 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)) * indirect expansion: ${!VAR} * substring op on @: ${@:n:m} * * Won't do: * Some builtins mandated by standards: * newgrp [GRP]: not a builtin in bash but a suid binary * which spawns a new shell with new group ID * * Status of [[ support: * [[ args ]] are CMD_SINGLEWORD_NOGLOB: * v='a b'; [[ $v = 'a b' ]]; echo 0:$? * [[ /bin/n* ]]; echo 0:$? * = is glob match operator, not equality operator: STR = GLOB * == same as = * =~ is regex match operator: STR =~ REGEX * TODO: * quoting needs to be considered (-f is an operator, "-f" and ""-f are not; etc) * in word = GLOB, quoting should be significant on char-by-char basis: a*cd"*" */ //config:config HUSH //config: bool "hush (70 kb)" //config: default y //config: select SHELL_HUSH //config: help //config: hush is a small shell. 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, tilde expansion, //config: &>file and >&file redirection of stdout+stderr. //config: // This option is visible (has a description) to make it possible to select // a "scripted" applet (such as NOLOGIN) but avoid selecting any shells: //config:config SHELL_HUSH //config: bool "Internal shell for embedded script support" //config: default n //config: //config:# hush options //config:# It's only needed to get "nice" menuconfig indenting. //config:if SHELL_HUSH || HUSH || SH_IS_HUSH || BASH_IS_HUSH //config: //config:config HUSH_BASH_COMPAT //config: bool "bash-compatible extensions" //config: default y //config: depends on SHELL_HUSH //config: //config:config HUSH_BRACE_EXPANSION //config: bool "Brace expansion" //config: default y //config: depends on HUSH_BASH_COMPAT //config: help //config: Enable {abc,def} extension. //config: //config:config HUSH_BASH_SOURCE_CURDIR //config: bool "'source' and '.' builtins search current directory after $PATH" //config: default n # do not encourage non-standard behavior //config: depends on HUSH_BASH_COMPAT //config: help //config: This is not compliant with standards. Avoid if possible. //config: //config:config HUSH_LINENO_VAR //config: bool "$LINENO variable (bashism)" //config: default y //config: depends on SHELL_HUSH //config: //config:config HUSH_INTERACTIVE //config: bool "Interactive mode" //config: default y //config: depends on SHELL_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_SAVEHISTORY //config: bool "Save command history to .hush_history" //config: default y //config: depends on HUSH_INTERACTIVE && FEATURE_EDITING_SAVEHISTORY //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 "Support command substitution" //config: default y //config: depends on SHELL_HUSH //config: help //config: Enable `command` and $(command). //config: //config:config HUSH_IF //config: bool "Support if/then/elif/else/fi" //config: default y //config: depends on SHELL_HUSH //config: //config:config HUSH_LOOPS //config: bool "Support for, while and until loops" //config: default y //config: depends on SHELL_HUSH //config: //config:config HUSH_CASE //config: bool "Support case ... esac statement" //config: default y //config: depends on SHELL_HUSH //config: help //config: Enable case ... esac statement. +400 bytes. //config: //config:config HUSH_FUNCTIONS //config: bool "Support funcname() { commands; } syntax" //config: default y //config: depends on SHELL_HUSH //config: help //config: Enable support for shell functions. +800 bytes. //config: //config:config HUSH_LOCAL //config: bool "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 $RANDOM variable" //config: default y //config: depends on SHELL_HUSH //config: help //config: Enable pseudorandom generator and dynamic variable "$RANDOM". //config: Each read of "$RANDOM" will generate a new pseudorandom value. //config: //config:config HUSH_MODE_X //config: bool "Support 'hush -x' option and 'set -x' command" //config: default y //config: depends on SHELL_HUSH //config: help //config: This instructs hush to print commands before execution. //config: Adds ~300 bytes. //config: //config:config HUSH_ECHO //config: bool "echo builtin" //config: default y //config: depends on SHELL_HUSH //config: //config:config HUSH_PRINTF //config: bool "printf builtin" //config: default y //config: depends on SHELL_HUSH //config: //config:config HUSH_TEST //config: bool "test builtin" //config: default y //config: depends on SHELL_HUSH //config: //config:config HUSH_HELP //config: bool "help builtin" //config: default y //config: depends on SHELL_HUSH //config: //config:config HUSH_EXPORT //config: bool "export builtin" //config: default y //config: depends on SHELL_HUSH //config: //config:config HUSH_EXPORT_N //config: bool "Support 'export -n' option" //config: default y //config: depends on HUSH_EXPORT //config: help //config: export -n unexports variables. It is a bash extension. //config: //config:config HUSH_READONLY //config: bool "readonly builtin" //config: default y //config: depends on SHELL_HUSH //config: help //config: Enable support for read-only variables. //config: //config:config HUSH_KILL //config: bool "kill builtin (supports kill %jobspec)" //config: default y //config: depends on SHELL_HUSH //config: //config:config HUSH_WAIT //config: bool "wait builtin" //config: default y //config: depends on SHELL_HUSH //config: //config:config HUSH_COMMAND //config: bool "command builtin" //config: default y //config: depends on SHELL_HUSH //config: //config:config HUSH_TRAP //config: bool "trap builtin" //config: default y //config: depends on SHELL_HUSH //config: //config:config HUSH_TYPE //config: bool "type builtin" //config: default y //config: depends on SHELL_HUSH //config: //config:config HUSH_TIMES //config: bool "times builtin" //config: default y //config: depends on SHELL_HUSH //config: //config:config HUSH_READ //config: bool "read builtin" //config: default y //config: depends on SHELL_HUSH //config: //config:config HUSH_SET //config: bool "set builtin" //config: default y //config: depends on SHELL_HUSH //config: //config:config HUSH_UNSET //config: bool "unset builtin" //config: default y //config: depends on SHELL_HUSH //config: //config:config HUSH_ULIMIT //config: bool "ulimit builtin" //config: default y //config: depends on SHELL_HUSH //config: //config:config HUSH_UMASK //config: bool "umask builtin" //config: default y //config: depends on SHELL_HUSH //config: //config:config HUSH_GETOPTS //config: bool "getopts builtin" //config: default y //config: depends on SHELL_HUSH //config: //config:config HUSH_MEMLEAK //config: bool "memleak builtin (debugging)" //config: default n //config: depends on SHELL_HUSH //config: //config:endif # hush options //applet:IF_HUSH(APPLET(hush, BB_DIR_BIN, BB_SUID_DROP)) // APPLET_ODDNAME:name main location suid_type help //applet:IF_SH_IS_HUSH( APPLET_ODDNAME(sh, hush, BB_DIR_BIN, BB_SUID_DROP, hush)) //applet:IF_BASH_IS_HUSH(APPLET_ODDNAME(bash, hush, BB_DIR_BIN, BB_SUID_DROP, hush)) //kbuild:lib-$(CONFIG_SHELL_HUSH) += hush.o match.o shell_common.o //kbuild:lib-$(CONFIG_HUSH_RANDOM_SUPPORT) += random.o /* -i (interactive) is also accepted, * but does nothing, therefore not shown in help. * NOMMU-specific options are not meant to be used by users, * therefore we don't show them either. */ //usage:#define hush_trivial_usage //usage: "[-enxl] [-c 'SCRIPT' [ARG0 ARGS] | FILE ARGS | -s ARGS]" //usage:#define hush_full_usage "\n\n" //usage: "Unix shell interpreter" #if !(defined(__FreeBSD__) || defined(__OpenBSD__) || defined(__NetBSD__) \ || defined(__APPLE__) \ ) # include /* for malloc_trim */ #endif #include /* #include */ #if ENABLE_HUSH_CASE # include #endif #include #include /* for setting $HOSTNAME */ #include "busybox.h" /* for APPLET_IS_NOFORK/NOEXEC */ #include "unicode.h" #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 O_CLOEXEC # define O_CLOEXEC 0 #endif #ifndef F_DUPFD_CLOEXEC # define F_DUPFD_CLOEXEC F_DUPFD #endif #if ENABLE_FEATURE_SH_EMBEDDED_SCRIPTS && !ENABLE_SHELL_ASH # include "embedded_scripts.h" #else # define NUM_SCRIPTS 0 #endif /* So far, all bash compat is controlled by one config option */ /* Separate defines document which part of code implements what */ #define BASH_PATTERN_SUBST ENABLE_HUSH_BASH_COMPAT #define BASH_SUBSTR ENABLE_HUSH_BASH_COMPAT #define BASH_SOURCE ENABLE_HUSH_BASH_COMPAT #define BASH_DOLLAR_SQUOTE ENABLE_HUSH_BASH_COMPAT #define BASH_HOSTNAME_VAR ENABLE_HUSH_BASH_COMPAT #define BASH_EPOCH_VARS ENABLE_HUSH_BASH_COMPAT #define BASH_TEST2 (ENABLE_HUSH_BASH_COMPAT && ENABLE_HUSH_TEST) #define BASH_READ_D ENABLE_HUSH_BASH_COMPAT /* 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 # undef ENABLE_FEATURE_EDITING_SAVE_ON_EXIT # define ENABLE_FEATURE_EDITING_SAVE_ON_EXIT 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_printf_heredoc(...) 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_redir(...) do {} while (0) #define debug_printf_list(...) do {} while (0) #define debug_printf_subst(...) do {} while (0) #define debug_printf_prompt(...) do {} while (0) #define debug_printf_clean(...) do {} while (0) #define ERR_PTR ((void*)(long)1) #define JOB_STATUS_FORMAT "[%u] %-22s %.40s\n" #define _SPECIAL_VARS_STR "_*@$!?#-" #define SPECIAL_VARS_STR ("_*@$!?#-" + 1) #define NUMERIC_SPECVARS_STR ("_*@$!?#-" + 3) #if BASH_PATTERN_SUBST /* 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_STR "\3" #define SPECIAL_VAR_SYMBOL 3 /* The "variable" with name "\1" emits string "\3". Testcase: "echo ^C" */ #define SPECIAL_VAR_QUOTED_SVS 1 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 { struct variable *old_vars; char **argv; char **argv_from_re_execing; } nommu_save_t; #endif enum { 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 }; typedef struct o_string { char *data; int length; /* position where data is appended */ int maxlen; int o_expflags; /* 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 ended_in_ifs; } o_string; enum { EXP_FLAG_SINGLEWORD = 0x80, /* must be 0x80 */ EXP_FLAG_GLOB = 0x2, /* Protect newly added chars against globbing * by prepending \ to *, ?, [, \ */ EXP_FLAG_ESC_GLOB_CHARS = 0x1, }; /* Used for initialization: o_string foo = NULL_O_STRING; */ #define NULL_O_STRING { NULL } #ifndef debug_printf_parse static const char *const assignment_flag[] ALIGN_PTR = { "MAYBE_ASSIGNMENT", "DEFINITELY_ASSIGNMENT", "NOT_ASSIGNMENT", "WORD_IS_KEYWORD", }; #endif /* We almost can use standard FILE api, but we need an ability to move * its fd when redirects coincide with it. No api exists for that * (RFE for it at https://sourceware.org/bugzilla/show_bug.cgi?id=21902). * HFILE is our internal alternative. Only supports reading. * Since we now can, we incorporate linked list of all opened HFILEs * into the struct (used to be a separate mini-list). */ typedef struct HFILE { char *cur; char *end; struct HFILE *next_hfile; int fd; char buf[1024]; } HFILE; typedef struct in_str { const char *p; int peek_buf[2]; int last_char; HFILE *file; } in_str; /* The descrip member of this structure is only used to make * debugging output pretty */ static const struct { int32_t mode; signed char default_fd; char descrip[3]; } redir_table[] ALIGN4 = { { 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 */ unsigned assignment_cnt; /* how many argv[i] are assignments? */ #if ENABLE_HUSH_LINENO_VAR unsigned lineno; #endif smallint cmd_type; /* CMD_xxx */ #define CMD_NORMAL 0 #define CMD_SUBSHELL 1 #if BASH_TEST2 /* used for "[[ EXPR ]]" */ # define CMD_TEST2_SINGLEWORD_NOGLOB 2 #endif #if BASH_TEST2 || ENABLE_HUSH_LOCAL || ENABLE_HUSH_EXPORT || ENABLE_HUSH_READONLY /* used to prevent word splitting and globbing in "export v=t*" */ # define CMD_SINGLEWORD_NOGLOB 3 #endif #if ENABLE_HUSH_FUNCTIONS # define CMD_FUNCDEF 4 #endif smalluint cmd_exitcode; /* 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 unsigned 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 = 0, PIPE_AND = 1, PIPE_OR = 2, PIPE_BG = 3, } 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; o_string word; #if !BB_MMU o_string as_string; #endif smallint is_assignment; /* 0:maybe, 1:yes, 2:no, 3:keyword */ #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 }; enum { MAYBE_ASSIGNMENT = 0, DEFINITELY_ASSIGNMENT = 1, NOT_ASSIGNMENT = 2, /* Not an assignment, but next word may be: "if v=xyz cmd;" */ WORD_IS_KEYWORD = 3, }; /* 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 */ int max_len; /* if > 0, name is part of initial env; else name is malloced */ uint16_t var_nest_level; 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 /* set -/+o OPT support. (TODO: make it optional) * bash supports the following opts: * allexport off * braceexpand on * emacs on * errexit off * errtrace off * functrace off * hashall on * histexpand off * history on * ignoreeof off * interactive-comments on * keyword off * monitor on * noclobber off * noexec off * noglob off * nolog off * notify off * nounset off * onecmd off * physical off * pipefail off * posix off * privileged off * verbose off * vi off * xtrace off */ static const char o_opt_strings[] ALIGN1 = "pipefail\0" "noexec\0" "errexit\0" #if ENABLE_HUSH_MODE_X "xtrace\0" #endif ; enum { OPT_O_PIPEFAIL, OPT_O_NOEXEC, OPT_O_ERREXIT, #if ENABLE_HUSH_MODE_X OPT_O_XTRACE, #endif NUM_OPT_O }; /* "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; IF_NOT_FEATURE_EDITING_FANCY_PROMPT(char *PS1;) # 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; unsigned 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 /* How deeply are we in context where "set -e" is ignored */ int errexit_depth; /* "set -e" rules (do we follow them correctly?): * Exit if pipe, list, or compound command exits with a non-zero status. * Shell does not exit if failed command is part of condition in * if/while, part of && or || list except the last command, any command * in a pipe but the last, or if the command's return value is being * inverted with !. If a compound command other than a subshell returns a * non-zero status because a command failed while -e was being ignored, the * shell does not exit. A trap on ERR, if set, is executed before the shell * exits [ERR is a bashism]. * * If a compound command or function executes in a context where -e is * ignored, none of the commands executed within are affected by the -e * setting. If a compound command or function sets -e while executing in a * context where -e is ignored, that setting does not have any effect until * the compound command or the command containing the function call completes. */ char o_opt[NUM_OPT_O]; #if ENABLE_HUSH_MODE_X # define G_x_mode (G.o_opt[OPT_O_XTRACE]) #else # define G_x_mode 0 #endif char opt_s; char opt_c; #if ENABLE_HUSH_INTERACTIVE smallint promptmode; /* 0: PS1, 1: PS2 */ #endif /* set by signal handler if SIGINT is received _and_ its trap is not set */ 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; # define G_flag_return_in_progress (G.flag_return_in_progress) #else # define G_flag_return_in_progress 0 #endif smallint exiting; /* used to prevent EXIT trap recursion */ /* These support $? */ smalluint last_exitcode; smalluint expand_exitcode; smalluint last_bg_pid_exitcode; #if ENABLE_HUSH_SET /* are global_argv and global_argv[1..n] malloced? (note: not [0]) */ smalluint global_args_malloced; # define G_global_args_malloced (G.global_args_malloced) #else # define G_global_args_malloced 0 #endif #if ENABLE_HUSH_BASH_COMPAT int dead_job_exitcode; /* for "wait -n" */ #endif /* 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 #if ENABLE_HUSH_GETOPTS unsigned getopt_count; #endif const char *ifs; char *ifs_whitespace; /* = G.ifs or malloced */ const char *cwd; struct variable *top_var; char **expanded_assignments; struct variable **shadowed_vars_pp; unsigned var_nest_level; #if ENABLE_HUSH_FUNCTIONS # if ENABLE_HUSH_LOCAL unsigned func_nest_level; /* solely to prevent "local v" in non-functions */ # endif struct function *top_func; #endif /* Signal and trap handling */ #if ENABLE_HUSH_FAST unsigned count_SIGCHLD; unsigned handled_SIGCHLD; smallint we_have_children; #endif #if ENABLE_HUSH_LINENO_VAR unsigned parse_lineno; unsigned execute_lineno; #endif HFILE *HFILE_list; HFILE *HFILE_stdin; /* Which signals have non-DFL handler (even with no traps set)? * Set at the start to: * (SIGQUIT + maybe SPECIAL_INTERACTIVE_SIGS + maybe SPECIAL_JOBSTOP_SIGS) * SPECIAL_INTERACTIVE_SIGS are cleared after fork. * The rest is cleared right before execv syscalls. * Other than these two times, never modified. */ unsigned special_sig_mask; #if ENABLE_HUSH_JOB unsigned fatal_sig_mask; # define G_fatal_sig_mask (G.fatal_sig_mask) #else # define G_fatal_sig_mask 0 #endif #if ENABLE_HUSH_TRAP int pre_trap_exitcode; # if ENABLE_HUSH_FUNCTIONS int return_exitcode; # endif char **traps; /* char *traps[NSIG] */ # define G_traps G.traps #else # define G_traps ((char**)NULL) #endif sigset_t pending_set; #if ENABLE_HUSH_MEMLEAK unsigned long memleak_value; #endif #if ENABLE_HUSH_MODE_X unsigned x_mode_depth; /* "set -x" output should not be redirectable with subsequent 2>FILE. * We dup fd#2 to x_mode_fd when "set -x" is executed, and use it * for all subsequent output. */ int x_mode_fd; o_string x_mode_buf; #endif #if HUSH_DEBUG >= 2 int debug_indent; #endif #if ENABLE_HUSH_TEST || BASH_TEST2 /* Cached supplementary group array (for testing executable'ity of files) */ struct cached_groupinfo groupinfo; # define GROUPINFO_INIT { G.groupinfo.euid = -1; G.groupinfo.egid = -1; } #else # define GROUPINFO_INIT /* nothing */ #endif struct sigaction sa; char optstring_buf[sizeof("eixcs")]; #if BASH_EPOCH_VARS char epoch_buf[sizeof("%llu.nnnnnn") + sizeof(long long)*3]; #endif #if ENABLE_FEATURE_EDITING char user_input_buf[CONFIG_FEATURE_EDITING_MAX_LEN]; #endif }; #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))); \ /* memset(&G.sa, 0, sizeof(G.sa)); */ \ sigfillset(&G.sa.sa_mask); \ G.sa.sa_flags = SA_RESTART; \ GROUPINFO_INIT; \ } while (0) /* Function prototypes for builtins */ static int builtin_cd(char **argv) FAST_FUNC; #if ENABLE_HUSH_ECHO static int builtin_echo(char **argv) FAST_FUNC; #endif static int builtin_eval(char **argv) FAST_FUNC; static int builtin_exec(char **argv) FAST_FUNC; static int builtin_exit(char **argv) FAST_FUNC; #if ENABLE_HUSH_EXPORT static int builtin_export(char **argv) FAST_FUNC; #endif #if ENABLE_HUSH_READONLY static int builtin_readonly(char **argv) FAST_FUNC; #endif static int builtin_false(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_GETOPTS static int builtin_getopts(char **argv) FAST_FUNC; #endif #if ENABLE_HUSH_HELP static int builtin_help(char **argv) FAST_FUNC; #endif #if MAX_HISTORY && ENABLE_FEATURE_EDITING static int builtin_history(char **argv) FAST_FUNC; #endif #if ENABLE_HUSH_LOCAL static int builtin_local(char **argv) FAST_FUNC; #endif #if ENABLE_HUSH_MEMLEAK static int builtin_memleak(char **argv) FAST_FUNC; #endif #if ENABLE_HUSH_PRINTF static int builtin_printf(char **argv) FAST_FUNC; #endif static int builtin_pwd(char **argv) FAST_FUNC; #if ENABLE_HUSH_READ static int builtin_read(char **argv) FAST_FUNC; #endif #if ENABLE_HUSH_SET static int builtin_set(char **argv) FAST_FUNC; #endif static int builtin_shift(char **argv) FAST_FUNC; static int builtin_source(char **argv) FAST_FUNC; #if ENABLE_HUSH_TEST || BASH_TEST2 static int builtin_test(char **argv) FAST_FUNC; #endif #if ENABLE_HUSH_TRAP static int builtin_trap(char **argv) FAST_FUNC; #endif #if ENABLE_HUSH_TYPE static int builtin_type(char **argv) FAST_FUNC; #endif #if ENABLE_HUSH_TIMES static int builtin_times(char **argv) FAST_FUNC; #endif static int builtin_true(char **argv) FAST_FUNC; #if ENABLE_HUSH_UMASK static int builtin_umask(char **argv) FAST_FUNC; #endif #if ENABLE_HUSH_UNSET static int builtin_unset(char **argv) FAST_FUNC; #endif #if ENABLE_HUSH_KILL static int builtin_kill(char **argv) FAST_FUNC; #endif #if ENABLE_HUSH_WAIT static int builtin_wait(char **argv) FAST_FUNC; #endif #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[] ALIGN_PTR = { BLTIN("." , builtin_source , "Run commands in file"), BLTIN(":" , builtin_true , NULL), #if ENABLE_HUSH_JOB BLTIN("bg" , builtin_fg_bg , "Resume job in background"), #endif #if ENABLE_HUSH_LOOPS BLTIN("break" , builtin_break , "Exit 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 , NULL), #if ENABLE_HUSH_EXPORT BLTIN("export" , builtin_export , "Set environment variables"), #endif BLTIN("false" , builtin_false , NULL), #if ENABLE_HUSH_JOB BLTIN("fg" , builtin_fg_bg , "Bring job to foreground"), #endif #if ENABLE_HUSH_GETOPTS BLTIN("getopts" , builtin_getopts , NULL), #endif #if ENABLE_HUSH_HELP BLTIN("help" , builtin_help , NULL), #endif #if MAX_HISTORY && ENABLE_FEATURE_EDITING BLTIN("history" , builtin_history , "Show history"), #endif #if ENABLE_HUSH_JOB BLTIN("jobs" , builtin_jobs , "List jobs"), #endif #if ENABLE_HUSH_KILL BLTIN("kill" , builtin_kill , "Send signals to processes"), #endif #if ENABLE_HUSH_LOCAL BLTIN("local" , builtin_local , "Set local variables"), #endif #if ENABLE_HUSH_MEMLEAK BLTIN("memleak" , builtin_memleak , NULL), #endif #if ENABLE_HUSH_READ BLTIN("read" , builtin_read , "Input into variable"), #endif #if ENABLE_HUSH_READONLY BLTIN("readonly" , builtin_readonly, "Make variables read-only"), #endif #if ENABLE_HUSH_FUNCTIONS BLTIN("return" , builtin_return , "Return from function"), #endif #if ENABLE_HUSH_SET BLTIN("set" , builtin_set , "Set positional parameters"), #endif BLTIN("shift" , builtin_shift , "Shift positional parameters"), #if BASH_SOURCE BLTIN("source" , builtin_source , NULL), #endif #if ENABLE_HUSH_TIMES BLTIN("times" , builtin_times , NULL), #endif #if ENABLE_HUSH_TRAP BLTIN("trap" , builtin_trap , "Trap signals"), #endif BLTIN("true" , builtin_true , NULL), #if ENABLE_HUSH_TYPE BLTIN("type" , builtin_type , "Show command type"), #endif #if ENABLE_HUSH_ULIMIT BLTIN("ulimit" , shell_builtin_ulimit, "Control resource limits"), #endif #if ENABLE_HUSH_UMASK BLTIN("umask" , builtin_umask , "Set file creation mask"), #endif #if ENABLE_HUSH_UNSET BLTIN("unset" , builtin_unset , "Unset variables"), #endif #if ENABLE_HUSH_WAIT BLTIN("wait" , builtin_wait , "Wait for process to finish"), #endif }; /* These builtins won't be used if we are on NOMMU and need to re-exec * (it's cheaper to run an external program in this case): */ static const struct built_in_command bltins2[] ALIGN_PTR = { #if ENABLE_HUSH_TEST BLTIN("[" , builtin_test , NULL), #endif #if BASH_TEST2 BLTIN("[[" , builtin_test , NULL), #endif #if ENABLE_HUSH_ECHO BLTIN("echo" , builtin_echo , NULL), #endif #if ENABLE_HUSH_PRINTF BLTIN("printf" , builtin_printf , NULL), #endif BLTIN("pwd" , builtin_pwd , NULL), #if ENABLE_HUSH_TEST BLTIN("test" , builtin_test , NULL), #endif }; /* * Debug printouts. */ #if HUSH_DEBUG >= 2 /* prevent disasters with G.debug_indent < 0 */ # define indent() fdprintf(2, "%*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(), fdprintf(2, __VA_ARGS__)) #endif #ifndef debug_printf_parse # define debug_printf_parse(...) (indent(), fdprintf(2, __VA_ARGS__)) #endif #ifndef debug_printf_heredoc # define debug_printf_heredoc(...) (indent(), fdprintf(2, __VA_ARGS__)) #endif #ifndef debug_printf_exec #define debug_printf_exec(...) (indent(), fdprintf(2, __VA_ARGS__)) #endif #ifndef debug_printf_env # define debug_printf_env(...) (indent(), fdprintf(2, __VA_ARGS__)) #endif #ifndef debug_printf_jobs # define debug_printf_jobs(...) (indent(), fdprintf(2, __VA_ARGS__)) # define DEBUG_JOBS 1 #else # define DEBUG_JOBS 0 #endif #ifndef debug_printf_expand # define debug_printf_expand(...) (indent(), fdprintf(2, __VA_ARGS__)) # define DEBUG_EXPAND 1 #else # define DEBUG_EXPAND 0 #endif #ifndef debug_printf_varexp # define debug_printf_varexp(...) (indent(), fdprintf(2, __VA_ARGS__)) #endif #ifndef debug_printf_glob # define debug_printf_glob(...) (indent(), fdprintf(2, __VA_ARGS__)) # define DEBUG_GLOB 1 #else # define DEBUG_GLOB 0 #endif #ifndef debug_printf_redir # define debug_printf_redir(...) (indent(), fdprintf(2, __VA_ARGS__)) #endif #ifndef debug_printf_list # define debug_printf_list(...) (indent(), fdprintf(2, __VA_ARGS__)) #endif #ifndef debug_printf_subst # define debug_printf_subst(...) (indent(), fdprintf(2, __VA_ARGS__)) #endif #ifndef debug_printf_prompt # define debug_printf_prompt(...) (indent(), fdprintf(2, __VA_ARGS__)) #endif #ifndef debug_printf_clean # define debug_printf_clean(...) (indent(), fdprintf(2, __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(); fdprintf(2, "%s:\n", prefix); while (*vv) fdprintf(2, " '%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 msg_and_die_if_script(lineno, ...) msg_and_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(void) { if (!G_interactive_fd) { if (G.last_exitcode) /* sometines it's 2, not 1 (bash compat) */ xfunc_error_retval = G.last_exitcode; xfunc_die(); } } static void msg_and_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); die_if_script(); } static void syntax_error(unsigned lineno UNUSED_PARAM, const char *msg) { if (msg) bb_error_msg("syntax error: %s", msg); else bb_simple_error_msg("syntax error"); die_if_script(); } static void syntax_error_at(unsigned lineno UNUSED_PARAM, const char *msg) { bb_error_msg("syntax error at '%s'", msg); die_if_script(); } static void syntax_error_unterm_str(unsigned lineno UNUSED_PARAM, const char *s) { bb_error_msg("syntax error: unterminated %s", s); //? source4.tests fails: in bash, echo ${^} in script does not terminate the script // (but bash --posix, or if bash is run as "sh", does terminate in script, so maybe uncomment this?) // die_if_script(); } static void syntax_error_unterm_ch(unsigned lineno, char ch) { char msg[2] = { ch, '\0' }; syntax_error_unterm_str(lineno, msg); } static void syntax_error_unexpected_ch(unsigned lineno UNUSED_PARAM, int ch) { char msg[2]; msg[0] = ch; msg[1] = '\0'; #if HUSH_DEBUG >= 2 bb_error_msg("hush.c:%u", lineno); #endif bb_error_msg("syntax error: unexpected %s", ch == EOF ? "EOF" : msg); die_if_script(); } #if HUSH_DEBUG < 2 # undef msg_and_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 msg_and_die_if_script(...) msg_and_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 /* * 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 (*src != '\0') { /* \x -> x */ *dst++ = *src++; continue; } /* else: "\". Do not delete this backslash. * Testcase: eval 'echo ok\' */ *dst++ = '\\'; /* fallthrough */ } 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); } static int dup_CLOEXEC(int fd, int avoid_fd) { int newfd; repeat: newfd = fcntl(fd, F_DUPFD_CLOEXEC, avoid_fd + 1); if (newfd >= 0) { if (F_DUPFD_CLOEXEC == F_DUPFD) /* if old libc (w/o F_DUPFD_CLOEXEC) */ close_on_exec_on(newfd); } else { /* newfd < 0 */ if (errno == EBUSY) goto repeat; if (errno == EINTR) goto repeat; if (errno != EBADF) { /* "echo >&9999" gets EINVAL trying to save fd 1 to above 9999. * We could try saving it _below_ 9999 instead (how?), but * this probably means that dup2(9999,1) to effectuate >&9999 * would also not work: fd 9999 can't exist. * (This differs from "echo >&99" where saving works, but * subsequent dup2(99,1) fails if fd 99 is not open). */ bb_perror_msg("fcntl(%d,F_DUPFD,%d)", fd, avoid_fd + 1); } } return newfd; } static int xdup_CLOEXEC_and_close(int fd, int avoid_fd) { int newfd; repeat: newfd = fcntl(fd, F_DUPFD_CLOEXEC, avoid_fd + 1); if (newfd < 0) { if (errno == EBUSY) goto repeat; if (errno == EINTR) goto repeat; /* fd was not open? */ if (errno == EBADF) return fd; xfunc_die(); } if (F_DUPFD_CLOEXEC == F_DUPFD) /* if old libc (w/o F_DUPFD_CLOEXEC) */ close_on_exec_on(newfd); close(fd); return newfd; } /* * Manipulating HFILEs */ static HFILE *hfopen(const char *name) { HFILE *fp; int fd; fd = STDIN_FILENO; if (name) { fd = open(name, O_RDONLY | O_CLOEXEC); if (fd < 0) return NULL; if (O_CLOEXEC == 0) /* ancient libc */ close_on_exec_on(fd); } fp = xmalloc(sizeof(*fp)); if (name == NULL) G.HFILE_stdin = fp; fp->fd = fd; fp->cur = fp->end = fp->buf; fp->next_hfile = G.HFILE_list; G.HFILE_list = fp; return fp; } static void hfclose(HFILE *fp) { HFILE **pp = &G.HFILE_list; while (*pp) { HFILE *cur = *pp; if (cur == fp) { *pp = cur->next_hfile; break; } pp = &cur->next_hfile; } if (fp->fd >= 0) close(fp->fd); free(fp); } static int refill_HFILE_and_getc(HFILE *fp) { int n; if (fp->fd < 0) { /* Already saw EOF */ return EOF; } #if ENABLE_HUSH_INTERACTIVE && !ENABLE_FEATURE_EDITING /* If user presses ^C, read() restarts after SIGINT (we use SA_RESTART). * IOW: ^C will not immediately stop line input. * But poll() is different: it does NOT restart after signals. */ if (fp == G.HFILE_stdin) { struct pollfd pfd[1]; pfd[0].fd = fp->fd; pfd[0].events = POLLIN; n = poll(pfd, 1, -1); if (n < 0 /*&& errno == EINTR - assumed true */ && sigismember(&G.pending_set, SIGINT) ) { return '\0'; } } #else /* if FEATURE_EDITING=y, we do not use this routine for interactive input */ #endif /* Try to buffer more input */ n = safe_read(fp->fd, fp->buf, sizeof(fp->buf)); if (n < 0) { bb_simple_perror_msg("read error"); n = 0; } fp->cur = fp->buf; fp->end = fp->buf + n; if (n == 0) { /* EOF/error */ close(fp->fd); fp->fd = -1; return EOF; } return (unsigned char)(*fp->cur++); } /* Inlined for common case of non-empty buffer. */ static ALWAYS_INLINE int hfgetc(HFILE *fp) { if (fp->cur < fp->end) return (unsigned char)(*fp->cur++); /* Buffer empty */ return refill_HFILE_and_getc(fp); } static int move_HFILEs_on_redirect(int fd, int avoid_fd) { HFILE *fl = G.HFILE_list; while (fl) { if (fd == fl->fd) { /* We use it only on script files, they are all CLOEXEC */ fl->fd = xdup_CLOEXEC_and_close(fd, avoid_fd); debug_printf_redir("redirect_fd %d: matches a script fd, moving it to %d\n", fd, fl->fd); return 1; /* "found and moved" */ } fl = fl->next_hfile; } #if ENABLE_HUSH_MODE_X if (G.x_mode_fd > 0 && fd == G.x_mode_fd) { G.x_mode_fd = xdup_CLOEXEC_and_close(fd, avoid_fd); return 1; /* "found and moved" */ } #endif return 0; /* "not in the list" */ } #if ENABLE_FEATURE_SH_STANDALONE && BB_MMU static void close_all_HFILE_list(void) { HFILE *fl = G.HFILE_list; while (fl) { /* hfclose would also free HFILE object. * It is disastrous if we share memory with a vforked parent. * I'm not sure we never come here after vfork. * Therefore just close fd, nothing more. * * ">" instead of ">=": we don't close fd#0, * interactive shell uses hfopen(NULL) as stdin input * which has fl->fd == 0, but fd#0 gets redirected in pipes. * If we'd close it here, then e.g. interactive "set | sort" * with NOFORKed sort, would have sort's input fd closed. */ if (fl->fd > 0) /*hfclose(fl); - unsafe */ close(fl->fd); fl = fl->next_hfile; } } #endif static int fd_in_HFILEs(int fd) { HFILE *fl = G.HFILE_list; while (fl) { if (fl->fd == fd) return 1; fl = fl->next_hfile; } return 0; } /* * 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; IF_HUSH_SET(smallint sv_g_malloced;) } save_arg_t; static void save_and_replace_G_args(save_arg_t *sv, char **argv) { sv->sv_argv0 = argv[0]; sv->sv_g_argv = G.global_argv; sv->sv_g_argc = G.global_argc; IF_HUSH_SET(sv->sv_g_malloced = G.global_args_malloced;) argv[0] = G.global_argv[0]; /* retain $0 */ G.global_argv = argv; IF_HUSH_SET(G.global_args_malloced = 0;) G.global_argc = 1 + string_array_len(argv + 1); } static void restore_G_args(save_arg_t *sv, char **argv) { #if ENABLE_HUSH_SET if (G.global_args_malloced) { /* someone ran "set -- arg1 arg2 ...", undo */ char **pp = G.global_argv; while (*++pp) /* note: does not free $0 */ free(*pp); free(G.global_argv); } #endif argv[0] = sv->sv_argv0; G.global_argv = sv->sv_g_argv; G.global_argc = sv->sv_g_argc; IF_HUSH_SET(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 is 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. * Kernel would do this for us ("orphaned process group" handling * according to POSIX) if we are a session leader and thus our death * frees the controlling tty, but to be bash-compatible, we also do it * for every interactive shell's death by SIGHUP. * (Also, we need to restore tty pgrp, otherwise e.g. Midnight Commander * backgrounds when hush started from it gets killed by SIGHUP). * 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 * Example 4: this does not wait and does not execute ls: * "sleep 5 & wait; ls -l" + press ^C * * (What happens to signals which are IGN on shell start?) * (What happens with signal mask on shell start?) * * Old implementation * ================== * 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 special_sig_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 special_sig_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. * * Problem: the above approach makes it unwieldy to catch signals while * we are in read builtin, or while we read commands from stdin: * masked signals are not visible! * * New implementation * ================== * We record each signal we are interested in by installing signal handler * for them - a bit like emulating kernel pending signal mask in userspace. * We are interested in: signals which need to have special handling * as described above, and all signals which have traps set. * Signals are recorded in pending_set. * After each pipe execution, we extract any pending signals * and act on them. * * unsigned special_sig_mask: a mask of shell-special signals. * unsigned fatal_sig_mask: a mask of signals on which we restore tty pgrp. * char *traps[sig] if trap for sig is set (even if it's ''). * sigset_t pending_set: set of sigs we received. * * "trap - SIGxxx": * if sig is in special_sig_mask, set handler back to: * record_pending_signo, or to IGN if it's a tty stop signal * if sig is in fatal_sig_mask, set handler back to sigexit. * else: set handler back to SIG_DFL * "trap 'cmd' SIGxxx": * set handler to record_pending_signo. * "trap '' SIGxxx": * set handler to SIG_IGN. * after [v]fork, if we plan to be a shell: * set signals with special interactive handling to SIG_DFL * (because 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. * * To make wait builtin interruptible, we handle SIGCHLD as special signal, * otherwise (if we leave it SIG_DFL) sigsuspend in wait builtin will not wake up on it. * * Note (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_JOBSTOP_SIGS = 0 #if ENABLE_HUSH_JOB | (1 << SIGTTIN) | (1 << SIGTTOU) | (1 << SIGTSTP) #endif , }; static void record_pending_signo(int sig) { sigaddset(&G.pending_set, sig); #if ENABLE_FEATURE_EDITING if (sig != SIGCHLD || (G_traps && G_traps[SIGCHLD] && G_traps[SIGCHLD][0]) /* ^^^ if SIGCHLD, interrupt line reading only if it has a trap */ ) { bb_got_signal = sig; /* for read_line_input: "we got a signal" */ } #endif #if ENABLE_HUSH_FAST if (sig == SIGCHLD) { 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 } static sighandler_t install_sighandler(int sig, sighandler_t handler) { struct sigaction old_sa; /* We could use signal() to install handlers... almost: * except that we need to mask ALL signals while handlers run. * I saw signal nesting in strace, race window isn't small. * SA_RESTART is also needed, but in Linux, signal() * sets SA_RESTART too. */ /* memset(&G.sa, 0, sizeof(G.sa)); - already done */ /* sigfillset(&G.sa.sa_mask); - already done */ /* G.sa.sa_flags = SA_RESTART; - already done */ G.sa.sa_handler = handler; sigaction(sig, &G.sa, &old_sa); return old_sa.sa_handler; } static void hush_exit(int exitcode) NORETURN; static void restore_ttypgrp_and__exit(void) NORETURN; static void restore_ttypgrp_and__exit(void) { /* xfunc has failed! die die die */ /* no EXIT traps, this is an escape hatch! */ G.exiting = 1; hush_exit(xfunc_error_retval); } #if ENABLE_HUSH_JOB /* Needed only on some libc: * It was observed that on exit(), fgetc'ed buffered data * gets "unwound" via lseek(fd, -NUM, SEEK_CUR). * With the net effect that even after fork(), not vfork(), * exit() in NOEXECed applet in "sh SCRIPT": * noexec_applet_here * echo END_OF_SCRIPT * lseeks fd in input FILE object from EOF to "e" in "echo END_OF_SCRIPT". * This makes "echo END_OF_SCRIPT" executed twice. * Similar problems can be seen with msg_and_die_if_script() -> xfunc_die() * and in `cmd` handling. * If set as die_func(), this makes xfunc_die() exit via _exit(), not exit(): */ static void fflush_and__exit(void) NORETURN; static void fflush_and__exit(void) { fflush_all(); _exit(xfunc_error_retval); } /* After [v]fork, in child: do not restore tty pgrp on xfunc death */ # define disable_restore_tty_pgrp_on_exit() (die_func = fflush_and__exit) /* After [v]fork, in parent: restore tty pgrp on xfunc death */ # define enable_restore_tty_pgrp_on_exit() (die_func = restore_ttypgrp_and__exit) /* 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) { /* 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) { /* Disable all signals: job control, SIGPIPE, etc. * Mostly paranoid measure, to prevent infinite SIGTTOU. */ sigprocmask_allsigs(SIG_BLOCK); 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 static sighandler_t pick_sighandler(unsigned sig) { sighandler_t handler = SIG_DFL; if (sig < sizeof(unsigned)*8) { unsigned sigmask = (1 << sig); #if ENABLE_HUSH_JOB /* is sig fatal? */ if (G_fatal_sig_mask & sigmask) handler = sigexit; else #endif /* sig has special handling? */ if (G.special_sig_mask & sigmask) { handler = record_pending_signo; /* TTIN/TTOU/TSTP can't be set to record_pending_signo * in order to ignore them: they will be raised * in an endless loop when we try to do some * terminal ioctls! We do have to _ignore_ these. */ if (SPECIAL_JOBSTOP_SIGS & sigmask) handler = SIG_IGN; } } return handler; } /* Restores tty foreground process group, and exits. */ static void hush_exit(int exitcode) { #if ENABLE_FEATURE_EDITING_SAVE_ON_EXIT save_history(G.line_input_state); /* may be NULL */ #endif fflush_all(); if (G.exiting <= 0 && G_traps && G_traps[0] && G_traps[0][0]) { char *argv[3]; /* argv[0] is unused */ argv[1] = xstrdup(G_traps[0]); /* copy, since EXIT trap handler may modify G_traps[0] */ argv[2] = NULL; G.exiting = 1; /* prevent EXIT trap recursion */ /* Note: G_traps[0] is not cleared! * "trap" will still show it, if executed * in the handler */ builtin_eval(argv); } #if ENABLE_FEATURE_CLEAN_UP { struct variable *cur_var; if (G.cwd != bb_msg_unknown) free((char*)G.cwd); cur_var = G.top_var; while (cur_var) { struct variable *tmp = cur_var; if (!cur_var->max_len) free(cur_var->varstr); cur_var = cur_var->next; free(tmp); } } #endif fflush_all(); #if ENABLE_HUSH_JOB sigexit(- (exitcode & 0xff)); #else _exit(exitcode); #endif } //TODO: return a mask of ALL handled sigs? static int check_and_run_traps(void) { int last_sig = 0; while (1) { int sig; if (sigisemptyset(&G.pending_set)) break; sig = 0; do { sig++; if (sigismember(&G.pending_set, sig)) { sigdelset(&G.pending_set, sig); goto got_sig; } } while (sig < NSIG); break; got_sig: #if ENABLE_HUSH_TRAP if (G_traps && G_traps[sig]) { debug_printf_exec("%s: sig:%d handler:'%s'\n", __func__, sig, G.traps[sig]); if (G_traps[sig][0]) { /* We have user-defined handler */ smalluint save_rcode; int save_pre; char *argv[3]; /* argv[0] is unused */ argv[1] = xstrdup(G_traps[sig]); /* why strdup? trap can modify itself: trap 'trap "echo oops" INT' INT */ argv[2] = NULL; save_pre = G.pre_trap_exitcode; G.pre_trap_exitcode = save_rcode = G.last_exitcode; builtin_eval(argv); free(argv[1]); G.pre_trap_exitcode = save_pre; G.last_exitcode = save_rcode; # if ENABLE_HUSH_FUNCTIONS if (G.return_exitcode >= 0) { debug_printf_exec("trap exitcode:%d\n", G.return_exitcode); G.last_exitcode = G.return_exitcode; } # endif last_sig = sig; } /* else: "" trap, ignoring signal */ continue; } #endif /* not a trap: special action */ switch (sig) { case SIGINT: debug_printf_exec("%s: sig:%d default SIGINT handler\n", __func__, sig); G.flag_SIGINT = 1; last_sig = sig; break; #if ENABLE_HUSH_JOB case SIGHUP: { /* if (G_interactive_fd) - no need to check, the handler * is only installed if we *are* interactive */ { /* bash compat: "Before exiting, an interactive * shell resends the SIGHUP to all jobs, running * or stopped. Stopped jobs are sent SIGCONT * to ensure that they receive the SIGHUP." */ struct pipe *job; debug_printf_exec("%s: sig:%d default SIGHUP handler\n", __func__, sig); /* 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); } } /* this restores tty pgrp, then kills us with SIGHUP */ sigexit(SIGHUP); } #endif #if ENABLE_HUSH_FAST case SIGCHLD: debug_printf_exec("%s: sig:%d default SIGCHLD handler\n", __func__, sig); 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); /* Note: * We don't do 'last_sig = sig' here -> NOT returning this sig. * This simplifies wait builtin a bit. */ break; #endif default: /* ignored: */ debug_printf_exec("%s: sig:%d default handling is to ignore\n", __func__, sig); /* SIGTERM, SIGQUIT, SIGTTIN, SIGTTOU, SIGTSTP */ /* Note: * We don't do 'last_sig = sig' here -> NOT returning this sig. * Example: wait is not interrupted by TERM * in interactive shell, because TERM is ignored. */ break; } } return last_sig; } /* * Shell and environment variable support */ 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; } static struct variable **get_ptr_to_local_var(const char *name) { struct variable **pp; struct variable *cur; pp = &G.top_var; while ((cur = *pp) != NULL) { if (varcmp(cur->varstr, name) == 0) return pp; pp = &cur->next; } 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; while (*cpp) { char *cp = *cpp; if (varcmp(cp, name) == 0) return strchr(cp, '=') + 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 #if ENABLE_HUSH_LINENO_VAR if (strcmp(name, "LINENO") == 0) return utoa(G.execute_lineno); #endif #if BASH_EPOCH_VARS { const char *fmt = NULL; if (strcmp(name, "EPOCHSECONDS") == 0) fmt = "%llu"; else if (strcmp(name, "EPOCHREALTIME") == 0) fmt = "%llu.%06u"; if (fmt) { struct timeval tv; xgettimeofday(&tv); sprintf(G.epoch_buf, fmt, (unsigned long long)tv.tv_sec, (unsigned)tv.tv_usec); return G.epoch_buf; } } #endif return NULL; } #if ENABLE_HUSH_GETOPTS static void handle_changed_special_names(const char *name) { if (varcmp(name, "OPTIND") == 0) { G.getopt_count = 0; return; } } #else /* Do not even bother evaluating arguments */ # define handle_changed_special_names(...) ((void)0) #endif /* str holds "NAME=VAL" and is expected to be malloced. * We take ownership of it. */ #define SETFLAG_EXPORT (1 << 0) #define SETFLAG_UNEXPORT (1 << 1) #define SETFLAG_MAKE_RO (1 << 2) #define SETFLAG_VARLVL_SHIFT 3 static int set_local_var(char *str, unsigned flags) { struct variable **cur_pp; struct variable *cur; char *free_me = NULL; char *eq_sign; int name_len; int retval; unsigned local_lvl = (flags >> SETFLAG_VARLVL_SHIFT); eq_sign = strchr(str, '='); if (HUSH_DEBUG && !eq_sign) bb_simple_error_msg_and_die("BUG in setvar"); name_len = eq_sign - str + 1; /* including '=' */ cur_pp = &G.top_var; while ((cur = *cur_pp) != NULL) { if (strncmp(cur->varstr, str, name_len) != 0) { cur_pp = &cur->next; continue; } /* We found an existing var with this name */ if (cur->flg_read_only) { bb_error_msg("%s: readonly variable", str); free(str); //NOTE: in bash, assignment in "export READONLY_VAR=Z" fails, and sets $?=1, //but export per se succeeds (does put the var in env). We don't mimic that. return -1; } if (flags & SETFLAG_UNEXPORT) { // && cur->flg_export ? debug_printf_env("%s: unsetenv '%s'\n", __func__, str); *eq_sign = '\0'; unsetenv(str); *eq_sign = '='; } if (cur->var_nest_level < local_lvl) { /* 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) flags |= SETFLAG_EXPORT; /* New variable is local ("local VAR=VAL" or * "VAR=VAL cmd") * and existing one is global, or local * on a lower level that new one. * Remove it from global variable list: */ *cur_pp = cur->next; if (G.shadowed_vars_pp) { /* Save in "shadowed" list */ debug_printf_env("shadowing %s'%s'/%u by '%s'/%u\n", cur->flg_export ? "exported " : "", cur->varstr, cur->var_nest_level, str, local_lvl ); cur->next = *G.shadowed_vars_pp; *G.shadowed_vars_pp = cur; } else { /* Came from pseudo_exec_argv(), no need to save: delete it */ debug_printf_env("shadow-deleting %s'%s'/%u by '%s'/%u\n", cur->flg_export ? "exported " : "", cur->varstr, cur->var_nest_level, str, local_lvl ); if (cur->max_len == 0) /* allocated "VAR=VAL"? */ free_me = cur->varstr; /* then free it later */ free(cur); } break; } if (strcmp(cur->varstr + name_len, eq_sign + 1) == 0) { debug_printf_env("assignement '%s' does not change anything\n", str); free_and_exp: free(str); goto exp; } /* Replace the value in the found "struct variable" */ if (cur->max_len != 0) { if (cur->max_len >= strnlen(str, cur->max_len + 1)) { /* This one is from startup env, reuse space */ debug_printf_env("reusing startup env for '%s'\n", str); strcpy(cur->varstr, str); goto free_and_exp; } /* Can't reuse */ cur->max_len = 0; goto set_str_and_exp; } /* max_len == 0 signifies "malloced" var, which we can * (and have to) free. But we can't free(cur->varstr) here: * if cur->flg_export is 1, it is in the environment. * We should either unsetenv+free, or wait until putenv, * then putenv(new)+free(old). */ free_me = cur->varstr; goto set_str_and_exp; } /* Not found or shadowed - create new variable struct */ debug_printf_env("%s: alloc new var '%s'/%u\n", __func__, str, local_lvl); cur = xzalloc(sizeof(*cur)); cur->var_nest_level = local_lvl; cur->next = *cur_pp; *cur_pp = cur; set_str_and_exp: cur->varstr = str; exp: #if !BB_MMU || ENABLE_HUSH_READONLY if (flags & SETFLAG_MAKE_RO) { cur->flg_read_only = 1; } #endif if (flags & SETFLAG_EXPORT) cur->flg_export = 1; retval = 0; if (cur->flg_export) { if (flags & SETFLAG_UNEXPORT) { cur->flg_export = 0; /* unsetenv was already done */ } else { debug_printf_env("%s: putenv '%s'/%u\n", __func__, cur->varstr, cur->var_nest_level); retval = putenv(cur->varstr); /* fall through to "free(free_me)" - * only now we can free old exported malloced string */ } } free(free_me); handle_changed_special_names(cur->varstr); return retval; } static int set_local_var0(char *str) { return set_local_var(str, 0); } static void FAST_FUNC set_local_var_from_halves(const char *name, const char *val) { char *var = xasprintf("%s=%s", name, val); set_local_var0(var); } /* Used at startup and after each cd */ static void set_pwd_var(unsigned flag) { set_local_var(xasprintf("PWD=%s", get_cwd(/*force:*/ 1)), flag); } #if ENABLE_HUSH_UNSET || ENABLE_HUSH_GETOPTS static int unset_local_var(const char *name) { struct variable *cur; struct variable **cur_pp; cur_pp = &G.top_var; while ((cur = *cur_pp) != NULL) { if (varcmp(cur->varstr, name) == 0) { if (cur->flg_read_only) { bb_error_msg("%s: readonly variable", name); return EXIT_FAILURE; } *cur_pp = cur->next; debug_printf_env("%s: unsetenv '%s'\n", __func__, cur->varstr); bb_unsetenv(cur->varstr); if (!cur->max_len) free(cur->varstr); free(cur); break; } cur_pp = &cur->next; } /* Handle "unset LINENO" et al even if did not find the variable to unset */ handle_changed_special_names(name); return EXIT_SUCCESS; } #endif /* * 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'/%u\n", __func__, var->varstr, var->var_nest_level); putenv(var->varstr); } else { debug_printf_env("%s: restoring variable '%s'/%u\n", __func__, var->varstr, var->var_nest_level); } var = next; } } /* We put strings[i] into variable table and possibly putenv them. * If variable is read only, we can free the strings[i] * which attempts to overwrite it. * The strings[] vector itself is freed. */ static void set_vars_and_save_old(char **strings) { char **s; if (!strings) return; s = strings; while (*s) { struct variable *var_p; struct variable **var_pp; char *eq; eq = strchr(*s, '='); if (HUSH_DEBUG && !eq) bb_simple_error_msg_and_die("BUG in varexp4"); var_pp = get_ptr_to_local_var(*s); if (var_pp) { var_p = *var_pp; if (var_p->flg_read_only) { char **p; bb_error_msg("%s: readonly variable", *s); /* * "VAR=V BLTIN" unsets VARs after BLTIN completes. * If VAR is readonly, leaving it in the list * after asssignment error (msg above) * causes doubled error message later, on unset. */ debug_printf_env("removing/freeing '%s' element\n", *s); free(*s); p = s; do { *p = p[1]; p++; } while (*p); goto next; } /* below, set_local_var() with nest level will * "shadow" (remove) this variable from * global linked list. */ } debug_printf_env("%s: env override '%s'/%u\n", __func__, *s, G.var_nest_level); set_local_var(*s, (G.var_nest_level << SETFLAG_VARLVL_SHIFT) | SETFLAG_EXPORT); s++; next: ; } free(strings); } /* * Unicode helper */ static void reinit_unicode_for_hush(void) { /* Unicode support should be activated even if LANG is set * _during_ shell execution, not only if it was set when * shell was started. Therefore, re-check LANG every time: */ if (ENABLE_FEATURE_CHECK_UNICODE_IN_ENV || ENABLE_UNICODE_USING_LOCALE ) { const char *s = get_local_var_value("LC_ALL"); if (!s) s = get_local_var_value("LC_CTYPE"); if (!s) s = get_local_var_value("LANG"); reinit_unicode(s); } } /* * in_str support (strings, and "strings" read from files). */ #if ENABLE_HUSH_INTERACTIVE /* To test correct lineedit/interactive behavior, type from command line: * echo $P\ * \ * AT\ * H\ * \ * It exercises a lot of corner cases. */ static const char *setup_prompt_string(void) { const char *prompt_str; debug_printf_prompt("%s promptmode:%d\n", __func__, G.promptmode); # if ENABLE_FEATURE_EDITING_FANCY_PROMPT prompt_str = get_local_var_value(G.promptmode == 0 ? "PS1" : "PS2"); if (!prompt_str) prompt_str = ""; # else prompt_str = "> "; /* if PS2, else... */ if (G.promptmode == 0) { /* PS1 */ /* No fancy prompts supported, (re)generate "CURDIR $ " by hand */ free(G.PS1); /* bash uses $PWD value, even if it is set by user. * It uses current dir only if PWD is unset. * We always use current dir. */ prompt_str = G.PS1 = xasprintf("%s %c ", get_cwd(0), (geteuid() != 0) ? '$' : '#'); } # endif debug_printf("prompt_str '%s'\n", prompt_str); return prompt_str; } static int get_user_input(struct in_str *i) { # if ENABLE_FEATURE_EDITING /* In EDITING case, this function reads next input line, * saves it in i->p, then returns 1st char of it. */ int r; const char *prompt_str; prompt_str = setup_prompt_string(); for (;;) { reinit_unicode_for_hush(); G.flag_SIGINT = 0; bb_got_signal = 0; if (!sigisemptyset(&G.pending_set)) { /* Whoops, already got a signal, do not call read_line_input */ bb_got_signal = r = -1; } else { /* For shell, LI_INTERRUPTIBLE is set: * read_line_input will abort on either * getting EINTR in poll() and bb_got_signal became != 0, * or if it sees bb_got_signal != 0 * (IOW: if signal arrives before poll() is reached). * Interactive testcases: * (while kill -INT $$; do sleep 1; done) & * #^^^ prints ^C, prints prompt, repeats * trap 'echo I' int; (while kill -INT $$; do sleep 1; done) & * #^^^ prints ^C, prints "I", prints prompt, repeats * trap 'echo T' term; (while kill $$; do sleep 1; done) & * #^^^ prints "T", prints prompt, repeats * #(bash 5.0.17 exits after first "T", looks like a bug) */ r = read_line_input(G.line_input_state, prompt_str, G.user_input_buf, CONFIG_FEATURE_EDITING_MAX_LEN-1 ); /* read_line_input intercepts ^C, "convert" it to SIGINT */ if (r == 0) raise(SIGINT); } /* bash prints ^C (before running a trap, if any) * both on keyboard ^C and on real SIGINT (non-kbd generated). */ if (sigismember(&G.pending_set, SIGINT)) { write(STDOUT_FILENO, "^C\n", 3); G.last_exitcode = 128 | SIGINT; } check_and_run_traps(); if (r == 0) /* keyboard ^C? */ continue; /* go back, read another input line */ if (r > 0) /* normal input? (no ^C, no ^D, no signals) */ break; if (!bb_got_signal) { /* r < 0: ^D/EOF/error detected (but not signal) */ /* ^D on interactive input goes to next line before exiting: */ write(STDOUT_FILENO, "\n", 1); i->p = NULL; i->peek_buf[0] = r = EOF; return r; } /* it was a signal: go back, read another input line */ } i->p = G.user_input_buf; return (unsigned char)*i->p++; # else /* In !EDITING case, this function gets called for every char. * Buffering happens deeper in the call chain, in hfgetc(i->file). */ int r; for (;;) { G.flag_SIGINT = 0; if (i->last_char == '\0' || i->last_char == '\n') { const char *prompt_str = setup_prompt_string(); /* Why check_and_run_traps here? Try this interactively: * $ trap 'echo INT' INT; (sleep 2; kill -INT $$) & * $ <[enter], repeatedly...> * Without check_and_run_traps, handler never runs. */ check_and_run_traps(); fputs_stdout(prompt_str); fflush_all(); } r = hfgetc(i->file); /* In !ENABLE_FEATURE_EDITING we don't use read_line_input, * no ^C masking happens during fgetc, no special code for ^C: * it generates SIGINT as usual. */ check_and_run_traps(); if (r != '\0' && !G.flag_SIGINT) break; if (G.flag_SIGINT) { /* ^C or SIGINT: repeat */ /* bash prints ^C even on real SIGINT (non-kbd generated) */ /* kernel prints "^C" itself, just print newline: */ write(STDOUT_FILENO, "\n", 1); G.last_exitcode = 128 | SIGINT; } } return r; # endif } /* This is the magic location that prints prompts * and gets data back from the user */ static int fgetc_interactive(struct in_str *i) { int ch; /* If it's interactive stdin, get new line. */ if (G_interactive_fd && i->file == G.HFILE_stdin) { /* Returns first char (or EOF), the rest is in i->p[] */ ch = get_user_input(i); G.promptmode = 1; /* PS2 */ debug_printf_prompt("%s promptmode=%d\n", __func__, G.promptmode); } else { /* Not stdin: script file, sourced file, etc */ do ch = hfgetc(i->file); while (ch == '\0'); } return ch; } #else /* !INTERACTIVE */ static ALWAYS_INLINE int fgetc_interactive(struct in_str *i) { int ch; do ch = hfgetc(i->file); while (ch == '\0'); return ch; } #endif /* !INTERACTIVE */ static int i_getch(struct in_str *i) { int ch; if (!i->file) { /* string-based in_str */ ch = (unsigned char)*i->p; if (ch != '\0') { i->p++; i->last_char = ch; #if ENABLE_HUSH_LINENO_VAR if (ch == '\n') { G.parse_lineno++; debug_printf_parse("G.parse_lineno++ = %u\n", G.parse_lineno); } #endif return ch; } return EOF; } /* FILE-based in_str */ #if ENABLE_FEATURE_EDITING /* This can be stdin, check line editing char[] buffer */ if (i->p && *i->p != '\0') { ch = (unsigned char)*i->p++; goto out; } #endif /* peek_buf[] is an int array, not char. Can contain EOF. */ ch = i->peek_buf[0]; if (ch != 0) { int ch2 = i->peek_buf[1]; i->peek_buf[0] = ch2; if (ch2 == 0) /* very likely, avoid redundant write */ goto out; i->peek_buf[1] = 0; goto out; } ch = fgetc_interactive(i); out: debug_printf("file_get: got '%c' %d\n", ch, ch); i->last_char = ch; #if ENABLE_HUSH_LINENO_VAR if (ch == '\n') { G.parse_lineno++; debug_printf_parse("G.parse_lineno++ = %u\n", G.parse_lineno); } #endif return ch; } static int i_peek(struct in_str *i) { int ch; if (!i->file) { /* string-based in_str */ /* Doesn't report EOF on NUL. None of the callers care. */ return (unsigned char)*i->p; } /* FILE-based in_str */ #if ENABLE_FEATURE_EDITING && ENABLE_HUSH_INTERACTIVE /* This can be stdin, check line editing char[] buffer */ if (i->p && *i->p != '\0') return (unsigned char)*i->p; #endif /* peek_buf[] is an int array, not char. Can contain EOF. */ ch = i->peek_buf[0]; if (ch != 0) return ch; /* Need to get a new char */ ch = fgetc_interactive(i); debug_printf("file_peek: got '%c' %d\n", ch, ch); /* Save it by either rolling back line editing buffer, or in i->peek_buf[0] */ #if ENABLE_FEATURE_EDITING && ENABLE_HUSH_INTERACTIVE if (i->p) { i->p -= 1; return ch; } #endif i->peek_buf[0] = ch; /*i->peek_buf[1] = 0; - already is */ return ch; } /* Only ever called if i_peek() was called, and did not return EOF. * IOW: we know the previous peek saw an ordinary char, not EOF, not NUL, * not end-of-line. Therefore we never need to read a new editing line here. */ static int i_peek2(struct in_str *i) { int ch; /* There are two cases when i->p[] buffer exists. * (1) it's a string in_str. * (2) It's a file, and we have a saved line editing buffer. * In both cases, we know that i->p[0] exists and not NUL, and * the peek2 result is in i->p[1]. */ if (i->p) return (unsigned char)i->p[1]; /* Now we know it is a file-based in_str. */ /* peek_buf[] is an int array, not char. Can contain EOF. */ /* Is there 2nd char? */ ch = i->peek_buf[1]; if (ch == 0) { /* We did not read it yet, get it now */ do ch = hfgetc(i->file); while (ch == '\0'); i->peek_buf[1] = ch; } debug_printf("file_peek2: got '%c' %d\n", ch, ch); return ch; } static int i_getch_and_eat_bkslash_nl(struct in_str *input) { for (;;) { int ch, ch2; ch = i_getch(input); if (ch != '\\') return ch; ch2 = i_peek(input); if (ch2 != '\n') return ch; /* backslash+newline, skip it */ i_getch(input); } } /* Note: this function _eats_ \ pairs, safe to use plain * i_getch() after it instead of i_getch_and_eat_bkslash_nl(). */ static int i_peek_and_eat_bkslash_nl(struct in_str *input) { for (;;) { int ch, ch2; ch = i_peek(input); if (ch != '\\') return ch; ch2 = i_peek2(input); if (ch2 != '\n') return ch; /* backslash+newline, skip it */ i_getch(input); i_getch(input); } } static void setup_file_in_str(struct in_str *i, HFILE *fp) { memset(i, 0, sizeof(*i)); i->file = fp; /* i->p = NULL; */ } static void setup_string_in_str(struct in_str *i, const char *s) { memset(i, 0, sizeof(*i)); /*i->file = NULL */; i->p = s; } /* * 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_and_set_NULL(o_string *o) { free(o->data); memset(o, 0, sizeof(*o)); } static ALWAYS_INLINE void o_free(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-1); 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); if (o->length < o->maxlen) { /* likely. avoid o_grow_by() call */ add: o->data[o->length] = ch; o->length++; o->data[o->length] = '\0'; return; } o_grow_by(o, 1); goto add; } #if 0 /* Valid only if we know o_string is not empty */ static void o_delchr(o_string *o) { o->length--; o->data[o->length] = '\0'; } #endif static void o_addblock(o_string *o, const char *str, int len) { o_grow_by(o, len); ((char*)mempcpy(&o->data[o->length], str, len))[0] = '\0'; o->length += len; } static void o_addstr(o_string *o, const char *str) { o_addblock(o, str, strlen(str)); } static void o_addstr_with_NUL(o_string *o, const char *str) { o_addblock(o, str, strlen(str) + 1); } #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 #if ENABLE_HUSH_MODE_X static void x_mode_addchr(int ch) { o_addchr(&G.x_mode_buf, ch); } static void x_mode_addstr(const char *str) { o_addstr(&G.x_mode_buf, str); } static void x_mode_addblock(const char *str, int len) { o_addblock(&G.x_mode_buf, str, len); } static void x_mode_prefix(void) { int n = G.x_mode_depth; do x_mode_addchr('+'); while (--n >= 0); } static void x_mode_flush(void) { int len = G.x_mode_buf.length; if (len <= 0) return; if (G.x_mode_fd > 0) { G.x_mode_buf.data[len] = '\n'; full_write(G.x_mode_fd, G.x_mode_buf.data, len + 1); } G.x_mode_buf.length = 0; } #endif /* * HUSH_BRACE_EXPANSION 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. */ #if ENABLE_HUSH_BRACE_EXPANSION # 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; /* '-' is included because of this case: * >filename0 >filename1 >filename9; v='-'; echo filename[0"$v"9] */ 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_expflags & EXP_FLAG_ESC_GLOB_CHARS) && 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; /* NUL is already added by o_addblock */ 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_expflags & EXP_FLAG_ESC_GLOB_CHARS)) { 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(); fdprintf(2, "%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_expflags & EXP_FLAG_GLOB), o->has_quoted_part, !!(o->o_expflags & EXP_FLAG_ESC_GLOB_CHARS)); while (i < n) { indent(); fdprintf(2, " list[%d]=%d '%s' %p\n", i, (int)(uintptr_t)list[i], o->data + (int)(uintptr_t)list[i] + string_start, o->data + (int)(uintptr_t)list[i] + string_start); i++; } if (n) { const char *p = o->data + (int)(uintptr_t)list[n - 1] + string_start; indent(); fdprintf(2, " 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); /* * expand_on_ifs() has a "previous argv[] ends in IFS?" * check. (grep for -prev-ifs-check-). * Ensure that argv[-1][last] is not garbage * but zero bytes, to save index check there. */ list[n + 0x10 - 1] = 0; 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; } o->has_quoted_part = 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; } /* * Globbing routines. * * Most words in commands need to be globbed, even ones which are * (single or double) quoted. This stems from the possiblity of * constructs like "abc"* and 'abc'* - these should be globbed. * Having a different code path for fully-quoted strings ("abc", * 'abc') would only help performance-wise, but we still need * code for partially-quoted strings. * * Unfortunately, if we want to match bash and ash behavior in all cases, * the logic can't be "shell-syntax argument is first transformed * to a string, then globbed, and if globbing does not match anything, * it is used verbatim". Here are two examples where it fails: * * echo 'b\*'? * * The globbing can't be avoided (because of '?' at the end). * The glob pattern is: b\\\*? - IOW, both \ and * are literals * and are glob-escaped. If this does not match, bash/ash print b\*? * - IOW: they "unbackslash" the glob pattern. * Now, look at this: * * v='\\\*'; echo b$v? * * The glob pattern is the same here: b\\\*? - the unquoted $v expansion * should be used as glob pattern with no changes. However, if glob * does not match, bash/ash print b\\\*? - NOT THE SAME as first example! * * ash implements this by having an encoded representation of the word * to glob, which IS NOT THE SAME as the glob pattern - it has more data. * Glob pattern is derived from it. If glob fails, the decision what result * should be is made using that encoded representation. Not glob pattern. */ #if ENABLE_HUSH_BRACE_EXPANSION /* 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_die_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_EXPANSION */ /* 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_die_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_EXPANSION */ /* If o->o_expflags & EXP_FLAG_GLOB, 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_expflags & EXP_FLAG_GLOB) { /* 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; 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)(uintptr_t)list[n] + string_start; } return list; } static void free_pipe_list(struct pipe *pi); /* Returns pi->next - next pipe in the list */ static struct pipe *free_pipe(struct pipe *pi) { struct pipe *next; int i; debug_printf_clean("free_pipe (pid %d)\n", getpid()); for (i = 0; i < pi->num_cmds; i++) { struct command *command; struct redir_struct *r, *rnext; command = &pi->cmds[i]; debug_printf_clean(" command %d:\n", i); if (command->argv) { if (DEBUG_CLEAN) { int a; char **p; 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 next = pi->next; free(pi); return next; } static void free_pipe_list(struct pipe *pi) { while (pi) { #if HAS_KEYWORDS debug_printf_clean("pipe reserved word %d\n", pi->res_word); #endif debug_printf_clean("pipe followup code %d\n", pi->followup); pi = free_pipe(pi); } } /* * Parsing routines */ #ifndef debug_print_tree static void debug_print_tree(struct pipe *pi, int lvl) { static const char *const PIPE[] ALIGN_PTR = { [PIPE_SEQ] = "SEQ", [PIPE_AND] = "AND", [PIPE_OR ] = "OR" , [PIPE_BG ] = "BG" , }; static const char *RES[] = { [RES_NONE ] = "NONE" , # if ENABLE_HUSH_IF [RES_IF ] = "IF" , [RES_THEN ] = "THEN" , [RES_ELIF ] = "ELIF" , [RES_ELSE ] = "ELSE" , [RES_FI ] = "FI" , # endif # if ENABLE_HUSH_LOOPS [RES_FOR ] = "FOR" , [RES_WHILE] = "WHILE", [RES_UNTIL] = "UNTIL", [RES_DO ] = "DO" , [RES_DONE ] = "DONE" , # endif # if ENABLE_HUSH_LOOPS || ENABLE_HUSH_CASE [RES_IN ] = "IN" , # endif # if ENABLE_HUSH_CASE [RES_CASE ] = "CASE" , [RES_CASE_IN ] = "CASE_IN" , [RES_MATCH] = "MATCH", [RES_CASE_BODY] = "CASE_BODY", [RES_ESAC ] = "ESAC" , # endif [RES_XXXX ] = "XXXX" , [RES_SNTX ] = "SNTX" , }; static const char *const CMDTYPE[] ALIGN_PTR = { "{}", "()", "[noglob]", # if ENABLE_HUSH_FUNCTIONS "func()", # endif }; int pin, prn; pin = 0; while (pi) { fdprintf(2, "%*spipe %d #cmds:%d %sres_word=%s followup=%d %s\n", lvl*2, "", pin, pi->num_cmds, (IF_HAS_KEYWORDS(pi->pi_inverted ? "! " :) ""), RES[pi->res_word], pi->followup, PIPE[pi->followup] ); prn = 0; while (prn < pi->num_cmds) { struct command *command = &pi->cmds[prn]; char **argv = command->argv; fdprintf(2, "%*s cmd %d assignment_cnt:%d", lvl*2, "", prn, command->assignment_cnt); # if ENABLE_HUSH_LINENO_VAR fdprintf(2, " LINENO:%u", command->lineno); # endif if (command->group) { fdprintf(2, " group %s: (argv=%p)%s%s\n", CMDTYPE[command->cmd_type], argv # if !BB_MMU , " group_as_string:", command->group_as_string # else , "", "" # endif ); debug_print_tree(command->group, lvl+1); prn++; continue; } if (argv) while (*argv) { fdprintf(2, " '%s'", *argv); argv++; } if (command->redirects) fdprintf(2, " {redir}"); fdprintf(2, "\n"); prn++; } pi = pi->next; pin++; } } #endif /* debug_print_tree */ static struct pipe *new_pipe(void) { struct pipe *pi; pi = xzalloc(sizeof(struct pipe)); /*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 0 /* Instead we emit error message at run time */ if (ctx->pending_redirect) { /* For example, "cmd >" (no filename to redirect to) */ syntax_error("invalid redirect"); ctx->pending_redirect = NULL; } #endif 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)); #if ENABLE_HUSH_LINENO_VAR command->lineno = G.parse_lineno; debug_printf_parse("command->lineno = G.parse_lineno (%u)\n", G.parse_lineno); #endif 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); #if HAS_KEYWORDS ctx->pipe->pi_inverted = ctx->ctx_inverted; ctx->ctx_inverted = 0; ctx->pipe->res_word = ctx->ctx_res_w; #endif if (type == PIPE_BG && ctx->list_head != ctx->pipe) { /* Necessary since && and || have precedence over &: * "cmd1 && cmd2 &" must spawn both cmds, not only cmd2, * in a backgrounded subshell. */ struct pipe *pi; struct command *command; /* Is this actually this construct, all pipes end with && or ||? */ pi = ctx->list_head; while (pi != ctx->pipe) { if (pi->followup != PIPE_AND && pi->followup != PIPE_OR) goto no_conv; pi = pi->next; } debug_printf_parse("BG with more than one pipe, converting to { p1 &&...pN; } &\n"); pi->followup = PIPE_SEQ; /* close pN _not_ with "&"! */ pi = xzalloc(sizeof(*pi)); pi->followup = PIPE_BG; pi->num_cmds = 1; pi->cmds = xzalloc(sizeof(pi->cmds[0])); command = &pi->cmds[0]; if (CMD_NORMAL != 0) /* "if xzalloc didn't do that already" */ command->cmd_type = CMD_NORMAL; command->group = ctx->list_head; #if !BB_MMU command->group_as_string = xstrndup( ctx->as_string.data, ctx->as_string.length - 1 /* do not copy last char, "&" */ ); #endif /* Replace all pipes in ctx with one newly created */ ctx->list_head = ctx->pipe = pi; /* for cases like "cmd && &", do not be tricked by last command * being null - the entire {...} & is NOT null! */ not_null = 1; } else { no_conv: ctx->pipe->followup = type; } /* 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)); if (MAYBE_ASSIGNMENT != 0) ctx->is_assignment = MAYBE_ASSIGNMENT; 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; uint32_t 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[] ALIGN4 = { # if ENABLE_HUSH_IF { "!", RES_NONE, NOT_ASSIGNMENT , 0 }, { "if", RES_IF, MAYBE_ASSIGNMENT, FLAG_THEN | FLAG_START }, { "then", RES_THEN, MAYBE_ASSIGNMENT, FLAG_ELIF | FLAG_ELSE | FLAG_FI }, { "elif", RES_ELIF, MAYBE_ASSIGNMENT, FLAG_THEN }, { "else", RES_ELSE, MAYBE_ASSIGNMENT, 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, MAYBE_ASSIGNMENT, FLAG_DO | FLAG_START }, { "until", RES_UNTIL, MAYBE_ASSIGNMENT, FLAG_DO | FLAG_START }, { "in", RES_IN, NOT_ASSIGNMENT , FLAG_DO }, { "do", RES_DO, MAYBE_ASSIGNMENT, 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 NULL: not a keyword, else: keyword */ static const struct reserved_combo* reserved_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 (ctx->word.has_quoted_part) return 0; r = match_reserved_word(&ctx->word); if (!r) return r; /* NULL */ 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 r; } if (r->flag & FLAG_START) { struct parse_context *old; old = xmemdup(ctx, sizeof(*ctx)); debug_printf_parse("push stack %p\n", old); initialize_context(ctx); ctx->stack = old; } else if (/*ctx->ctx_res_w == RES_NONE ||*/ !(ctx->old_flag & (1 << r->res))) { syntax_error_at(ctx->word.data); ctx->ctx_res_w = RES_SNTX; return r; } 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; ctx->is_assignment = r->assignment_flag; debug_printf_parse("ctx->is_assignment='%s'\n", assignment_flag[ctx->is_assignment]); 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 /* At this point, the compound command's string is in * ctx->as_string... except for the leading keyword! * Consider this example: "echo a | if true; then echo a; fi" * ctx->as_string will contain "true; then echo a; fi", * with "if " remaining in old->as_string! */ { char *str; int len = old->as_string.length; /* Concatenate halves */ o_addstr(&old->as_string, ctx->as_string.data); o_free(&ctx->as_string); /* Find where leading keyword starts in first half */ str = old->as_string.data + len; if (str > old->as_string.data) str--; /* skip whitespace after keyword */ while (str > old->as_string.data && isalpha(str[-1])) str--; /* Ugh, we're done with this horrid hack */ old->command->group_as_string = xstrdup(str); debug_printf_parse("pop, remembering as:'%s'\n", old->command->group_as_string); } # endif *ctx = *old; /* physical copy */ free(old); } return r; } #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(struct parse_context *ctx) { struct command *command = ctx->command; debug_printf_parse("done_word entered: '%s' %p\n", ctx->word.data, command); if (ctx->word.length == 0 && !ctx->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://pubs.opengroup.org/onlinepubs/9699919799/utilities/V3_chap02.html * "2.7 Redirection * If the redirection operator is "<<" or "<<-", the word * that follows the redirection operator shall be * subjected to quote removal; it is unspecified whether * any of the other expansions occur. For the other * redirection operators, 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." */ //bash does not do parameter/command substitution or arithmetic expansion //for _heredoc_ redirection word: these constructs look for exact eof marker // as written: // <pending_redirect->rd_filename = xstrdup(ctx->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 (ctx->word.has_quoted_part) { ctx->pending_redirect->rd_dup |= HEREDOC_QUOTED; } } debug_printf_parse("word stored in rd_filename: '%s'\n", ctx->word.data); ctx->pending_redirect = NULL; } else { #if HAS_KEYWORDS # if ENABLE_HUSH_CASE if (ctx->ctx_dsemicolon && strcmp(ctx->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 defined(CMD_TEST2_SINGLEWORD_NOGLOB) if (command->cmd_type == CMD_TEST2_SINGLEWORD_NOGLOB && strcmp(ctx->word.data, "]]") == 0 ) { /* allow "[[ ]] >file" etc */ command->cmd_type = CMD_SINGLEWORD_NOGLOB; } 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 ) { const struct reserved_combo *reserved; reserved = reserved_word(ctx); debug_printf_parse("checking for reserved-ness: %d\n", !!reserved); if (reserved) { # if ENABLE_HUSH_LINENO_VAR /* Case: * "while ...; do * cmd ..." * If we don't close the pipe _now_, immediately after "do", lineno logic * sees "cmd" as starting at "do" - i.e., at the previous line. */ if (0 IF_HUSH_IF(|| reserved->res == RES_THEN) IF_HUSH_IF(|| reserved->res == RES_ELIF) IF_HUSH_IF(|| reserved->res == RES_ELSE) IF_HUSH_LOOPS(|| reserved->res == RES_DO) ) { done_pipe(ctx, PIPE_SEQ); } # endif o_reset_to_empty_unquoted(&ctx->word); debug_printf_parse("done_word return %d\n", (ctx->ctx_res_w == RES_SNTX)); return (ctx->ctx_res_w == RES_SNTX); } # if defined(CMD_TEST2_SINGLEWORD_NOGLOB) if (strcmp(ctx->word.data, "[[") == 0) { command->cmd_type = CMD_TEST2_SINGLEWORD_NOGLOB; } else # endif # if defined(CMD_SINGLEWORD_NOGLOB) if (0 /* In bash, local/export/readonly are special, args * are assignments and therefore expansion of them * should be "one-word" expansion: * $ export i=`echo 'a b'` # one arg: "i=a b" * compare with: * $ ls i=`echo 'a b'` # 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 * Note: 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 */ IF_HUSH_LOCAL( || strcmp(ctx->word.data, "local") == 0) IF_HUSH_EXPORT( || strcmp(ctx->word.data, "export") == 0) IF_HUSH_READONLY(|| strcmp(ctx->word.data, "readonly") == 0) ) { command->cmd_type = CMD_SINGLEWORD_NOGLOB; } # else { /* empty block to pair "if ... else" */ } # endif } #endif /* HAS_KEYWORDS */ if (command->group) { /* "{ echo foo; } echo bar" - bad */ syntax_error_at(ctx->word.data); debug_printf_parse("done_word return 1: syntax error, " "groups and arglists don't mix\n"); return 1; } /* If this word wasn't an assignment, next ones definitely * can't be assignments. Even if they look like ones. */ if (ctx->is_assignment != DEFINITELY_ASSIGNMENT && ctx->is_assignment != WORD_IS_KEYWORD ) { ctx->is_assignment = NOT_ASSIGNMENT; } else { if (ctx->is_assignment == DEFINITELY_ASSIGNMENT) { command->assignment_cnt++; debug_printf_parse("++assignment_cnt=%d\n", command->assignment_cnt); } debug_printf_parse("ctx->is_assignment was:'%s'\n", assignment_flag[ctx->is_assignment]); ctx->is_assignment = MAYBE_ASSIGNMENT; } debug_printf_parse("ctx->is_assignment='%s'\n", assignment_flag[ctx->is_assignment]); command->argv = add_string_to_strings(command->argv, xstrdup(ctx->word.data)); debug_print_strings("word appended to argv", command->argv); } #if ENABLE_HUSH_LOOPS if (ctx->ctx_res_w == RES_FOR) { if (ctx->word.has_quoted_part || endofname(command->argv[0])[0] != '\0' ) { /* bash says just "not a valid identifier" */ syntax_error("bad for loop variable"); 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(&ctx->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_simple_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_and_eat_bkslash_nl(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_and_eat_bkslash_nl(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 { #if 0 /* Instead we emit error message at run time */ if (ctx->pending_redirect) { /* For example, "cmd > 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; unsigned past_EOL; int prev = 0; /* not \ */ int ch; /* Starting with "" is necessary for this case: * cat < (not a line cont.) */ prev = 0; /* not '\' */ else prev = ch; } } /* Look at entire parse tree for not-yet-loaded REDIRECT_HEREDOCs * and load them all. There should be exactly heredoc_cnt of them. */ #if BB_MMU #define fetch_heredocs(as_string, pi, heredoc_cnt, input) \ fetch_heredocs(pi, heredoc_cnt, input) #endif static int fetch_heredocs(o_string *as_string, struct pipe *pi, int heredoc_cnt, struct in_str *input) { while (pi && heredoc_cnt) { int i; struct command *cmd = pi->cmds; debug_printf_heredoc("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_heredoc("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(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; } if (cmd->group) { //bb_error_msg("%s:%u heredoc_cnt:%d", __func__, __LINE__, heredoc_cnt); heredoc_cnt = fetch_heredocs(as_string, cmd->group, heredoc_cnt, input); //bb_error_msg("%s:%u heredoc_cnt:%d", __func__, __LINE__, heredoc_cnt); if (heredoc_cnt < 0) return heredoc_cnt; /* error */ } cmd++; } pi = pi->next; } return heredoc_cnt; } static int run_list(struct pipe *pi); #if BB_MMU #define parse_stream(pstring, heredoc_cnt_ptr, input, end_trigger) \ parse_stream(heredoc_cnt_ptr, input, end_trigger) #endif static struct pipe *parse_stream(char **pstring, int *heredoc_cnt_ptr, struct in_str *input, int end_trigger); /* Returns number of heredocs not yet consumed, * or -1 on error. */ static int parse_group(struct parse_context *ctx, struct in_str *input, int ch) { /* ctx->word contains characters seen prior to ( or {. * Typically it's empty, but for function defs, * it contains function name (without '()'). */ #if BB_MMU # define as_string NULL #else char *as_string = NULL; #endif struct pipe *pipe_list; int heredoc_cnt = 0; int endch; struct command *command = ctx->command; debug_printf_parse("parse_group entered\n"); #if ENABLE_HUSH_FUNCTIONS if (ch == '(' && !ctx->word.has_quoted_part) { if (ctx->word.length) if (done_word(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 != '{' && ch != '(') { syntax_error_unexpected_ch(ch); return -1; } //bash allows functions named "123", "..", "return"! // if (endofname(command->argv[0])[0] != '\0') { // syntax_error("bad function name"); // 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]{... */ || ctx->word.length /* word{... */ || ctx->word.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_HUSH_FUNCTIONS(skip:) endch = '}'; if (ch == '(') { endch = ')'; IF_HUSH_FUNCTIONS(if (command->cmd_type != CMD_FUNCDEF)) command->cmd_type = CMD_SUBSHELL; } else { /* bash does not allow "{echo...", requires whitespace */ ch = i_peek(input); if (ch != ' ' && ch != '\t' && ch != '\n' && ch != '(' /* but "{(..." is allowed (without whitespace) */ ) { syntax_error_unexpected_ch(ch); return -1; } if (ch != '(') { ch = i_getch(input); nommu_addchr(&ctx->as_string, ch); } } debug_printf_heredoc("calling parse_stream, heredoc_cnt:%d\n", heredoc_cnt); pipe_list = parse_stream(&as_string, &heredoc_cnt, input, endch); debug_printf_heredoc("parse_stream returned: heredoc_cnt:%d\n", heredoc_cnt); #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; } #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 #if ENABLE_HUSH_FUNCTIONS /* Convert "f() (cmds)" to "f() {(cmds)}" */ if (command->cmd_type == CMD_FUNCDEF && endch == ')') { struct command *cmd2; cmd2 = xzalloc(sizeof(*cmd2)); cmd2->cmd_type = CMD_SUBSHELL; cmd2->group = pipe_list; # if !BB_MMU //UNTESTED! cmd2->group_as_string = command->group_as_string; command->group_as_string = xasprintf("(%s)", command->group_as_string); # endif pipe_list = new_pipe(); pipe_list->cmds = cmd2; pipe_list->num_cmds = 1; } #endif command->group = pipe_list; debug_printf_parse("parse_group return %d\n", heredoc_cnt); return heredoc_cnt; /* command remains "open", available for possible redirects */ #undef as_string } #if ENABLE_HUSH_TICK || ENABLE_FEATURE_SH_MATH || ENABLE_HUSH_DOLLAR_OPS /* Subroutines for copying $(...) and `...` things */ /* '...' */ static int add_till_single_quote(o_string *dest, struct in_str *input) { while (1) { int ch = i_getch(input); if (ch == EOF) { syntax_error_unterm_ch('\''); return 0; } if (ch == '\'') return 1; o_addchr(dest, ch); } } static int add_till_single_quote_dquoted(o_string *dest, struct in_str *input) { while (1) { int ch = i_getch(input); if (ch == EOF) { syntax_error_unterm_ch('\''); return 0; } if (ch == '\'') return 1; o_addqchr(dest, ch); } } /* "...\"...`..`...." - do we need to handle "...$(..)..." too? */ static int add_till_backquote(o_string *dest, struct in_str *input, int in_dquote); static int 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('"'); return 0; } if (ch == '"') return 1; if (ch == '\\') { /* \x. Copy both chars. */ o_addchr(dest, ch); ch = i_getch(input); } o_addchr(dest, ch); if (ch == '`') { if (!add_till_backquote(dest, input, /*in_dquote:*/ 1)) return 0; 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 $(command) * form, or a quoted string, undefined results occur. A single-quoted * or double-quoted string that begins, but does not end, within the * "`...`" sequence produces undefined results." * Example Output * echo `echo '\'TEST\`echo ZZ\`BEST` \TESTZZBEST */ static int add_till_backquote(o_string *dest, struct in_str *input, int in_dquote) { while (1) { int ch = i_getch(input); if (ch == '`') return 1; if (ch == '\\') { /* \x. Copy both unless it is \`, \$, \\ and maybe \" */ ch = i_getch(input); if (ch != '`' && ch != '$' && ch != '\\' && (!in_dquote || ch != '"') ) { o_addchr(dest, '\\'); } } if (ch == EOF) { syntax_error_unterm_ch('`'); return 0; } 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 BASH_SUBSTR || BASH_PATTERN_SUBST char end_char2 = end_ch >> 8; # endif end_ch &= (DOUBLE_CLOSE_CHAR_FLAG - 1); # if ENABLE_HUSH_INTERACTIVE G.promptmode = 1; /* PS2 */ # endif debug_printf_prompt("%s promptmode=%d\n", __func__, G.promptmode); while (1) { ch = i_getch(input); if (ch == EOF) { syntax_error_unterm_ch(end_ch); return 0; } if (ch == end_ch # if BASH_SUBSTR || BASH_PATTERN_SUBST || ch == end_char2 # endif ) { if (!dbl) break; /* we look for closing )) of $((EXPR)) */ if (i_peek_and_eat_bkslash_nl(input) == end_ch) { i_getch(input); /* eat second ')' */ break; } } o_addchr(dest, ch); //bb_error_msg("%s:o_addchr('%c')", __func__, ch); if (ch == '(' || ch == '{') { ch = (ch == '(' ? ')' : '}'); if (!add_till_closing_bracket(dest, input, ch)) return 0; o_addchr(dest, ch); continue; } if (ch == '\'') { if (!add_till_single_quote(dest, input)) return 0; o_addchr(dest, ch); continue; } if (ch == '"') { if (!add_till_double_quote(dest, input)) return 0; o_addchr(dest, ch); continue; } if (ch == '`') { if (!add_till_backquote(dest, input, /*in_dquote:*/ 0)) return 0; o_addchr(dest, ch); continue; } if (ch == '\\') { /* \x. Copy verbatim. Important for \(, \) */ ch = i_getch(input); if (ch == EOF) { syntax_error_unterm_ch(end_ch); return 0; } # if 0 if (ch == '\n') { /* "backslash+newline", ignore both */ o_delchr(dest); /* undo insertion of '\' */ continue; } # endif o_addchr(dest, ch); //bb_error_msg("%s:o_addchr('%c') after '\\'", __func__, ch); continue; } } debug_printf_parse("%s return '%s' ch:'%c'\n", __func__, dest->data, ch); return ch; } #endif /* ENABLE_HUSH_TICK || ENABLE_FEATURE_SH_MATH || ENABLE_HUSH_DOLLAR_OPS */ #if BASH_DOLLAR_SQUOTE /* Return code: 1 for "found and parsed", 0 for "seen something else" */ # if BB_MMU #define parse_dollar_squote(as_string, dest, input) \ parse_dollar_squote(dest, input) #define as_string NULL # endif static int parse_dollar_squote(o_string *as_string, o_string *dest, struct in_str *input) { int start; int ch = i_peek_and_eat_bkslash_nl(input); /* first character after the $ */ debug_printf_parse("parse_dollar_squote entered: ch='%c'\n", ch); if (ch != '\'') return 0; dest->has_quoted_part = 1; start = dest->length; ch = i_getch(input); /* eat ' */ nommu_addchr(as_string, ch); while (1) { ch = i_getch(input); nommu_addchr(as_string, ch); if (ch == EOF) { syntax_error_unterm_ch('\''); return 0; } if (ch == '\'') break; if (ch == SPECIAL_VAR_SYMBOL) { /* Convert raw ^C to corresponding special variable reference */ o_addchr(dest, SPECIAL_VAR_SYMBOL); o_addchr(dest, SPECIAL_VAR_QUOTED_SVS); /* will addchr() another SPECIAL_VAR_SYMBOL (see after the if() block) */ } else if (ch == '\\') { static const char C_escapes[] ALIGN1 = "nrbtfav""x\\01234567"; ch = i_getch(input); nommu_addchr(as_string, ch); if (strchr(C_escapes, ch)) { char buf[4]; char *p = buf; int cnt = 2; buf[0] = ch; if ((unsigned char)(ch - '0') <= 7) { /* \ooo */ do { ch = i_peek(input); if ((unsigned char)(ch - '0') > 7) break; *++p = ch = i_getch(input); nommu_addchr(as_string, ch); } while (--cnt != 0); } else if (ch == 'x') { /* \xHH */ do { ch = i_peek(input); if (!isxdigit(ch)) break; *++p = ch = i_getch(input); nommu_addchr(as_string, ch); } while (--cnt != 0); if (cnt == 2) { /* \x but next char is "bad" */ ch = 'x'; goto unrecognized; } } /* else simple seq like \\ or \t */ *++p = '\0'; p = buf; ch = bb_process_escape_sequence((void*)&p); //bb_error_msg("buf:'%s' ch:%x", buf, ch); if (ch == '\0') continue; /* bash compat: $'...\0...' emits nothing */ } else { /* unrecognized "\z": encode both chars unless ' or " */ if (ch != '\'' && ch != '"') { unrecognized: o_addqchr(dest, '\\'); } } } /* if (\...) */ o_addqchr(dest, ch); } if (dest->length == start) { /* $'', $'\0', $'\000\x00' and the like */ o_addchr(dest, SPECIAL_VAR_SYMBOL); o_addchr(dest, SPECIAL_VAR_SYMBOL); } return 1; # undef as_string } #else # define parse_dollar_squote(as_string, dest, input) 0 #endif /* BASH_DOLLAR_SQUOTE */ /* Return code: 0 for OK, 1 for syntax error */ #if BB_MMU #define parse_dollar(as_string, dest, input, quote_mask) \ parse_dollar(dest, input, quote_mask) #define as_string NULL #endif static int parse_dollar(o_string *as_string, o_string *dest, struct in_str *input, unsigned char quote_mask) { int ch = i_peek_and_eat_bkslash_nl(input); /* first character after the $ */ debug_printf_parse("parse_dollar entered: ch='%c' quote_mask:0x%x\n", ch, quote_mask); if (isalpha(ch)) { make_var: ch = i_getch(input); nommu_addchr(as_string, ch); /*make_var1:*/ 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_and_eat_bkslash_nl(input); if (!isalnum(ch) && ch != '_') { /* End of variable name reached */ 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 */ case '-': /* $- option flags set by set builtin or shell options (-i etc) */ goto make_one_char_var; case '{': { char len_single_ch; o_addchr(dest, SPECIAL_VAR_SYMBOL); ch = i_getch(input); /* eat '{' */ nommu_addchr(as_string, ch); ch = i_getch_and_eat_bkslash_nl(input); /* first char after '{' */ /* It should be ${?}, or ${#var}, * or even ${?+subst} - operator acting on a special variable, * or the beginning of variable name. */ if (ch == EOF || (!strchr(_SPECIAL_VARS_STR, ch) && !isalnum(ch)) /* not one of those */ ) { bad_dollar_syntax: syntax_error_unterm_str("${name}"); debug_printf_parse("parse_dollar return 0: unterminated ${name}\n"); return 0; } nommu_addchr(as_string, ch); len_single_ch = ch; 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... */ if (isdigit(len_single_ch) || (len_single_ch == '#' && isdigit(i_peek_and_eat_bkslash_nl(input))) ) { /* Execution engine uses plain xatoi_positive() * to interpret ${NNN} and {#NNN}, * check syntax here in the parser. * (bash does not support expressions in ${#NN}, * e.g. ${#$var} and {#1:+WORD} are not supported). */ unsigned cnt = 9; /* max 9 digits for ${NN} and 8 for {#NN} */ while (1) { o_addchr(dest, ch); debug_printf_parse(": '%c'\n", ch); ch = i_getch_and_eat_bkslash_nl(input); nommu_addchr(as_string, ch); if (ch == '}') break; if (--cnt == 0) goto bad_dollar_syntax; if (len_single_ch != '#' && strchr(VAR_SUBST_OPS, ch)) /* ${NN...} is valid */ goto eat_until_closing; if (!isdigit(ch)) goto bad_dollar_syntax; } } else 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... */ if (len_single_ch != '#' /*|| !strchr(SPECIAL_VARS_STR, ch) - disallow errors like ${#+} ? */ || i_peek(input) != '}' ) { goto bad_dollar_syntax; } /* else: it's "length of C" ${#C} op, * where C is a single char * special var name, e.g. ${#!}. */ } eat_until_closing: /* Eat everything until closing '}' (or ':') */ end_ch = '}'; if (BASH_SUBSTR && ch == ':' && !strchr(MINUS_PLUS_EQUAL_QUESTION, i_peek(input)) ) { /* It's ${var:N[:M]} thing */ end_ch = '}' * 0x100 + ':'; } if (BASH_PATTERN_SUBST && ch == '/' ) { /* It's ${var/[/]pattern[/repl]} thing */ if (i_peek(input) == '/') { /* ${var//pattern[/repl]}? */ i_getch(input); nommu_addchr(as_string, '/'); ch = '\\'; } end_ch = '}' * 0x100 + '/'; } o_addchr(dest, ch); /* The pattern can't be empty. * IOW: if the first char after "${v//" is a slash, * it does not terminate the pattern - it's the first char of the pattern: * v=/dev/ram; echo ${v////-} prints -dev-ram (pattern is "/") * v=/dev/ram; echo ${v///r/-} prints /dev-am (pattern is "/r") */ if (i_peek(input) == '/') { o_addchr(dest, i_getch(input)); } again: if (!BB_MMU) pos = dest->length; #if ENABLE_HUSH_DOLLAR_OPS last_ch = add_till_closing_bracket(dest, input, end_ch); if (last_ch == 0) /* error? */ return 0; #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 ((BASH_SUBSTR || BASH_PATTERN_SUBST) && (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 (BASH_SUBSTR && end_ch == '}' * 0x100 + ':') { /* it's ${var:N} - emulate :999999999 */ o_addstr(dest, "999999999"); } /* else: it's ${var/[/]pattern} */ } break; } len_single_ch = 0; /* it can't be ${#C} op */ } o_addchr(dest, SPECIAL_VAR_SYMBOL); break; } #if ENABLE_FEATURE_SH_MATH || ENABLE_HUSH_TICK case '(': { unsigned pos; ch = i_getch(input); nommu_addchr(as_string, ch); # if ENABLE_FEATURE_SH_MATH if (i_peek_and_eat_bkslash_nl(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; if (!add_till_closing_bracket(dest, input, ')' | DOUBLE_CLOSE_CHAR_FLAG)) return 0; /* error */ 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; if (!add_till_closing_bracket(dest, input, ')')) return 0; /* error */ if (as_string) { o_addstr(as_string, dest->data + pos); o_addchr(as_string, ')'); } o_addchr(dest, SPECIAL_VAR_SYMBOL); # endif break; } #endif case '_': goto make_var; #if 0 /* TODO: $_: */ /* $_ 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 */ ch = i_getch(input); nommu_addchr(as_string, ch); ch = i_peek_and_eat_bkslash_nl(input); if (isalnum(ch)) { /* it's $_name or $_123 */ ch = '_'; goto make_var1; } /* else: it's $_ */ #endif default: o_addQchr(dest, '$'); } debug_printf_parse("parse_dollar return 1 (ok)\n"); return 1; #undef as_string } #if BB_MMU #define encode_string(as_string, dest, input, dquote_end) \ encode_string(dest, input, dquote_end) #define as_string NULL #endif static int encode_string(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 */ debug_printf_parse("encode_string return 1 (ok)\n"); return 1; } /* note: can't move it above ch == dquote_end check! */ if (ch == EOF) { syntax_error_unterm_ch('"'); return 0; /* error */ } next = '\0'; if (ch != '\n') { next = i_peek(input); } debug_printf_parse("\" ch=%c (%d) escape=%d\n", ch, ch, !!(dest->o_expflags & EXP_FLAG_ESC_GLOB_CHARS)); if (ch == '\\') { if (next == EOF) { /* Testcase: in interactive shell a file with * echo "unterminated string\ * is sourced. */ syntax_error_unterm_ch('"'); return 0; /* error */ } /* bash: * "The backslash retains its special meaning [in "..."] * only when followed by one of the following characters: * $, `, ", \, or . A double quote may be quoted * within double quotes by preceding it with a backslash." * NB: in (unquoted) heredoc, above does not apply to ", * therefore we check for it by "next == dquote_end" cond. */ if (next == dquote_end || strchr("$`\\\n", next)) { ch = i_getch(input); /* eat next */ if (ch == '\n') goto again; /* skip \ */ } /* else: ch remains == '\\', and we double it below: */ o_addqchr(dest, ch); /* \c if c is a glob char, else just c */ nommu_addchr(as_string, ch); goto again; } if (ch == '$') { //if (parse_dollar_squote(as_string, dest, input)) // goto again; if (!parse_dollar(as_string, dest, input, /*quote_mask:*/ 0x80)) { debug_printf_parse("encode_string return 0: " "parse_dollar returned 0 (error)\n"); return 0; } goto again; } #if ENABLE_HUSH_TICK if (ch == '`') { //unsigned pos = dest->length; o_addchr(dest, SPECIAL_VAR_SYMBOL); o_addchr(dest, 0x80 | '`'); if (!add_till_backquote(dest, input, /*in_dquote:*/ dquote_end == '"')) return 0; /* error */ 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 == SPECIAL_VAR_SYMBOL) { /* Convert "^C" to corresponding special variable reference */ o_addchr(dest, SPECIAL_VAR_QUOTED_SVS); o_addchr(dest, SPECIAL_VAR_SYMBOL); } 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 if shell is not interactive, * reset parsing machinery and start parsing anew, * or return ERR_PTR. */ static struct pipe *parse_stream(char **pstring, int *heredoc_cnt_ptr, struct in_str *input, int end_trigger) { struct parse_context ctx; int heredoc_cnt; /* Single-quote triggers a bypass of the main loop until its mate is * found. When recursing, quote state is passed in via ctx.word.o_expflags. */ debug_printf_parse("parse_stream entered, end_trigger='%c'\n", end_trigger ? end_trigger : 'X'); debug_enter(); initialize_context(&ctx); /* If very first arg is "" or '', ctx.word.data may end up NULL. * Preventing this: */ ctx.word.data = xzalloc(1); /* start as "", not as NULL */ /* We used to separate words on $IFS here. This was wrong. * $IFS is used only for word splitting when $var is expanded, * here we should use blank chars as separators, not $IFS */ heredoc_cnt = 0; while (1) { const char *is_blank; const char *is_special; int ch; int next; int redir_fd; redir_type redir_style; ch = i_getch(input); debug_printf_parse(": ch=%c (%d) escape=%d\n", ch, ch, !!(ctx.word.o_expflags & EXP_FLAG_ESC_GLOB_CHARS)); if (ch == EOF) { struct pipe *pi; if (heredoc_cnt) { syntax_error_unterm_str("here document"); goto parse_error_exitcode1; } if (end_trigger == ')') { syntax_error_unterm_ch('('); goto parse_error_exitcode1; } if (end_trigger == '}') { syntax_error_unterm_ch('{'); goto parse_error_exitcode1; } if (done_word(&ctx)) { goto parse_error_exitcode1; } o_free_and_set_NULL(&ctx.word); done_pipe(&ctx, PIPE_SEQ); /* Do we sit inside of any if's, loops or case's? */ if (HAS_KEYWORDS IF_HAS_KEYWORDS(&& (ctx.ctx_res_w != RES_NONE || ctx.old_flag != 0)) ) { syntax_error_unterm_str("compound statement"); goto parse_error_exitcode1; } 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_string1 '%s'\n", ctx.as_string.data); if (pstring) *pstring = ctx.as_string.data; else o_free(&ctx.as_string); #endif // heredoc_cnt must be 0 here anyway //if (heredoc_cnt_ptr) // *heredoc_cnt_ptr = heredoc_cnt; debug_leave(); debug_printf_heredoc("parse_stream return heredoc_cnt:%d\n", heredoc_cnt); debug_printf_parse("parse_stream return %p: EOF\n", pi); return pi; } /* Handle "'" and "\" first, as they won't play nice with * i_peek_and_eat_bkslash_nl() anyway: * echo z\\ * and * echo '\ * ' * would break. */ if (ch == '\\') { ch = i_getch(input); if (ch == '\n') continue; /* drop \, get next char */ nommu_addchr(&ctx.as_string, '\\'); if (ch == SPECIAL_VAR_SYMBOL) { nommu_addchr(&ctx.as_string, ch); /* Convert \^C to corresponding special variable reference */ goto case_SPECIAL_VAR_SYMBOL; } o_addchr(&ctx.word, '\\'); if (ch == EOF) { /* Testcase: eval 'echo Ok\' */ /* bash-4.3.43 was removing backslash, * but 4.4.19 retains it, most other shells too */ continue; /* get next char */ } /* Example: echo Hello \2>file * we need to know that word 2 is quoted */ ctx.word.has_quoted_part = 1; nommu_addchr(&ctx.as_string, ch); o_addchr(&ctx.word, ch); continue; /* get next char */ } nommu_addchr(&ctx.as_string, ch); if (ch == '\'') { ctx.word.has_quoted_part = 1; next = i_getch(input); if (next == '\'' && !ctx.pending_redirect) goto insert_empty_quoted_str_marker; ch = next; while (1) { if (ch == EOF) { syntax_error_unterm_ch('\''); goto parse_error_exitcode1; } nommu_addchr(&ctx.as_string, ch); if (ch == '\'') break; if (ch == SPECIAL_VAR_SYMBOL) { /* Convert raw ^C to corresponding special variable reference */ o_addchr(&ctx.word, SPECIAL_VAR_SYMBOL); o_addchr(&ctx.word, SPECIAL_VAR_QUOTED_SVS); } o_addqchr(&ctx.word, ch); ch = i_getch(input); } continue; /* get next char */ } next = '\0'; if (ch != '\n') next = i_peek_and_eat_bkslash_nl(input); is_special = "{}<>&|();#" /* special outside of "str" */ "$\"" IF_HUSH_TICK("`") /* always special */ SPECIAL_VAR_SYMBOL_STR; #if defined(CMD_TEST2_SINGLEWORD_NOGLOB) if (ctx.command->cmd_type == CMD_TEST2_SINGLEWORD_NOGLOB) { /* In [[ ]], {}<>&|() are not special */ is_special += 8; } else #endif /* Are { and } special here? */ if (ctx.command->argv /* word [word]{... - non-special */ || ctx.word.length /* word{... - non-special */ || ctx.word.has_quoted_part /* ""{... - non-special */ || (next != ';' /* }; - special */ && next != ')' /* }) - special */ && next != '(' /* {( - special */ && next != '&' /* }& and }&& ... - special */ && next != '|' /* }|| ... - special */ && !strchr(defifs, next) /* {word - non-special */ ) ) { /* They are not special, skip "{}" */ is_special += 2; } is_special = strchr(is_special, ch); is_blank = strchr(defifs, ch); if (!is_special && !is_blank) { /* ordinary char */ ordinary_char: o_addQchr(&ctx.word, ch); if ((ctx.is_assignment == MAYBE_ASSIGNMENT || ctx.is_assignment == WORD_IS_KEYWORD) && ch == '=' && endofname(ctx.word.data)[0] == '=' ) { ctx.is_assignment = DEFINITELY_ASSIGNMENT; debug_printf_parse("ctx.is_assignment='%s'\n", assignment_flag[ctx.is_assignment]); } continue; } if (is_blank) { #if ENABLE_HUSH_LINENO_VAR /* Case: * "while ...; do * cmd ..." * would think that "cmd" starts in - * i.e., at the previous line. * We need to skip all whitespace before newlines. */ while (ch != '\n') { next = i_peek(input); if (next != ' ' && next != '\t' && next != '\n') break; /* next char is not ws */ ch = i_getch(input); } /* ch == last eaten whitespace char */ #endif if (done_word(&ctx)) { goto parse_error_exitcode1; } if (ch == '\n') { /* Is this a case when newline is simply ignored? * Some examples: * "cmd | cmd ..." * "case ... in word) ..." */ if (IS_NULL_CMD(ctx.command) && ctx.word.length == 0 && !ctx.word.has_quoted_part && heredoc_cnt == 0 ) { /* This newline can be ignored. But... * Without check #1, interactive shell * ignores even bare , * and shows the continuation prompt: * ps1_prompt$ * ps2> _ <=== wrong, should be ps1 * Without check #2, "cmd & " * is similarly mistreated. * (BTW, this makes "cmd & cmd" * and "cmd && cmd" non-orthogonal. * Really, ask yourself, why * "cmd && " doesn't start * cmd but waits for more input? * The only reason is that it might be * a "cmd1 && cmd2 &" construct, * cmd1 may need to run in BG). */ struct pipe *pi = ctx.list_head; if (pi->num_cmds != 0 /* check #1 */ && pi->followup != PIPE_BG /* check #2 */ ) { continue; } } /* Treat newline as a command separator. */ done_pipe(&ctx, PIPE_SEQ); debug_printf_heredoc("heredoc_cnt:%d\n", heredoc_cnt); if (heredoc_cnt) { heredoc_cnt = fetch_heredocs(&ctx.as_string, ctx.list_head, heredoc_cnt, input); if (heredoc_cnt != 0) goto parse_error_exitcode1; } ctx.is_assignment = MAYBE_ASSIGNMENT; debug_printf_parse("ctx.is_assignment='%s'\n", assignment_flag[ctx.is_assignment]); ch = ';'; /* note: if (is_blank) 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 (ctx.word.length != 0 /* word} */ || ctx.word.has_quoted_part /* ""} */ ) { goto ordinary_char; } if (!IS_NULL_CMD(ctx.command)) { /* cmd } */ /* Generally, there should be semicolon: "cmd; }" * However, bash allows to omit it if "cmd" is * a group. Examples: * { { echo 1; } } * {(echo 1)} * { echo 0 >&2 | { echo 1; } } * { while false; do :; done } * { case a in b) ;; esac } */ if (ctx.command->group) goto term_group; goto ordinary_char; } if (!IS_NULL_PIPE(ctx.pipe)) /* cmd | } */ /* Can't be an end of {cmd}, skip the check */ goto skip_end_trigger; /* else: } does terminate a group */ } term_group: if (end_trigger && end_trigger == ch && (ch != ';' || heredoc_cnt == 0) #if ENABLE_HUSH_CASE && (ch != ')' || ctx.ctx_res_w != RES_MATCH || (!ctx.word.has_quoted_part && strcmp(ctx.word.data, "esac") == 0) ) #endif ) { if (done_word(&ctx)) { goto parse_error_exitcode1; } done_pipe(&ctx, PIPE_SEQ); ctx.is_assignment = MAYBE_ASSIGNMENT; debug_printf_parse("ctx.is_assignment='%s'\n", assignment_flag[ctx.is_assignment]); /* Do we sit outside of any if's, loops or case's? */ if (!HAS_KEYWORDS IF_HAS_KEYWORDS(|| (ctx.ctx_res_w == RES_NONE && ctx.old_flag == 0)) ) { o_free_and_set_NULL(&ctx.word); #if !BB_MMU debug_printf_parse("as_string2 '%s'\n", ctx.as_string.data); if (pstring) *pstring = ctx.as_string.data; else o_free(&ctx.as_string); #endif if (ch != ';' && IS_NULL_PIPE(ctx.list_head)) { /* Example: bare "{ }", "()" */ G.last_exitcode = 2; /* bash compat */ syntax_error_unexpected_ch(ch); goto parse_error; } if (heredoc_cnt_ptr) *heredoc_cnt_ptr = heredoc_cnt; debug_printf_heredoc("parse_stream return heredoc_cnt:%d\n", heredoc_cnt); debug_printf_parse("parse_stream return %p: " "end_trigger char found\n", ctx.list_head); debug_leave(); return ctx.list_head; } } if (is_blank) continue; /* Catch <, > 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(&ctx.word); if (done_word(&ctx)) { goto parse_error_exitcode1; } 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_exitcode1; } #endif if (parse_redirect(&ctx, redir_fd, redir_style, input)) goto parse_error_exitcode1; continue; /* get next char */ case '<': redir_fd = redirect_opt_num(&ctx.word); if (done_word(&ctx)) { goto parse_error_exitcode1; } redir_style = REDIRECT_INPUT; if (next == '<') { redir_style = REDIRECT_HEREDOC; heredoc_cnt++; debug_printf_heredoc("++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_exitcode1; } #endif if (parse_redirect(&ctx, redir_fd, redir_style, input)) goto parse_error_exitcode1; continue; /* get next char */ case '#': if (ctx.word.length == 0 && !ctx.word.has_quoted_part) { /* skip "#comment" */ /* note: we do not add it to &ctx.as_string */ /* TODO: in bash: * comment inside $() goes to the next \n, even inside quoted string (!): * cmd "$(cmd2 #comment)" - syntax error * cmd "`cmd2 #comment`" - ok * We accept both (comment ends where command subst ends, in both cases). */ while (1) { ch = i_peek(input); if (ch == '\n') { nommu_addchr(&ctx.as_string, '\n'); break; } ch = i_getch(input); if (ch == EOF) break; } continue; /* get next char */ } break; } skip_end_trigger: if (ctx.is_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 */ ctx.is_assignment = NOT_ASSIGNMENT; debug_printf_parse("ctx.is_assignment='%s'\n", assignment_flag[ctx.is_assignment]); } /* Note: nommu_addchr(&ctx.as_string, ch) is already done */ switch (ch) { case_SPECIAL_VAR_SYMBOL: case SPECIAL_VAR_SYMBOL: /* Convert raw ^C to corresponding special variable reference */ o_addchr(&ctx.word, SPECIAL_VAR_SYMBOL); o_addchr(&ctx.word, SPECIAL_VAR_QUOTED_SVS); /* fall through */ case '#': /* non-comment #: "echo a#b" etc */ o_addchr(&ctx.word, ch); continue; /* get next char */ case '$': if (parse_dollar_squote(&ctx.as_string, &ctx.word, input)) continue; /* get next char */ if (!parse_dollar(&ctx.as_string, &ctx.word, input, /*quote_mask:*/ 0)) { debug_printf_parse("parse_stream parse error: " "parse_dollar returned 0 (error)\n"); goto parse_error_exitcode1; } continue; /* get next char */ case '"': ctx.word.has_quoted_part = 1; if (next == '"' && !ctx.pending_redirect) { i_getch(input); /* eat second " */ insert_empty_quoted_str_marker: nommu_addchr(&ctx.as_string, next); o_addchr(&ctx.word, SPECIAL_VAR_SYMBOL); o_addchr(&ctx.word, SPECIAL_VAR_SYMBOL); continue; /* get next char */ } if (ctx.is_assignment == NOT_ASSIGNMENT) ctx.word.o_expflags |= EXP_FLAG_ESC_GLOB_CHARS; if (!encode_string(&ctx.as_string, &ctx.word, input, '"')) goto parse_error_exitcode1; ctx.word.o_expflags &= ~EXP_FLAG_ESC_GLOB_CHARS; continue; /* get next char */ #if ENABLE_HUSH_TICK case '`': { USE_FOR_NOMMU(unsigned pos;) o_addchr(&ctx.word, SPECIAL_VAR_SYMBOL); o_addchr(&ctx.word, '`'); USE_FOR_NOMMU(pos = ctx.word.length;) if (!add_till_backquote(&ctx.word, input, /*in_dquote:*/ 0)) goto parse_error_exitcode1; # if !BB_MMU o_addstr(&ctx.as_string, ctx.word.data + pos); o_addchr(&ctx.as_string, '`'); # endif o_addchr(&ctx.word, SPECIAL_VAR_SYMBOL); //debug_printf_subst("SUBST RES3 '%s'\n", ctx.word.data + pos); continue; /* get next char */ } #endif case ';': #if ENABLE_HUSH_CASE case_semi: #endif if (done_word(&ctx)) { goto parse_error_exitcode1; } done_pipe(&ctx, PIPE_SEQ); #if ENABLE_HUSH_CASE /* Eat multiple semicolons, detect * whether it means something special */ while (1) { ch = i_peek_and_eat_bkslash_nl(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 */ ctx.is_assignment = MAYBE_ASSIGNMENT; debug_printf_parse("ctx.is_assignment='%s'\n", assignment_flag[ctx.is_assignment]); continue; /* get next char */ case '&': if (done_word(&ctx)) { goto parse_error_exitcode1; } 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(&ctx)) { goto parse_error_exitcode1; } #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); } 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|(... */ && ctx.word.length == 0 /* not word(... */ && ctx.word.has_quoted_part == 0 /* not ""(... */ ) { continue; /* get next char */ } #endif /* fall through */ case '{': { int n = parse_group(&ctx, input, ch); if (n < 0) { goto parse_error_exitcode1; } debug_printf_heredoc("parse_group done, needs heredocs:%d\n", n); heredoc_cnt += n; 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). */ G.last_exitcode = 2; syntax_error_unexpected_ch(ch); goto parse_error; default: if (HUSH_DEBUG) bb_error_msg_and_die("BUG: unexpected %c", ch); } } /* while (1) */ parse_error_exitcode1: G.last_exitcode = 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(&pctx->as_string); #endif IF_HAS_KEYWORDS(p2 = pctx->stack;) if (pctx != &ctx) { free(pctx); } IF_HAS_KEYWORDS(pctx = p2;) } while (HAS_KEYWORDS && pctx); o_free(&ctx.word); #if !BB_MMU if (pstring) *pstring = NULL; #endif debug_leave(); return ERR_PTR; } } /* * Execution routines */ /* Expansion can recurse, need forward decls: */ #if !BASH_PATTERN_SUBST && !ENABLE_HUSH_CASE #define expand_string_to_string(str, EXP_flags, do_unbackslash) \ expand_string_to_string(str) #endif static char *expand_string_to_string(const char *str, int EXP_flags, int do_unbackslash); #if ENABLE_HUSH_TICK static int process_command_subs(o_string *dest, const char *s); #endif static int expand_vars_to_list(o_string *output, int n, char *arg); /* 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 $VAR). * This new list is allocated as a single malloc block. * NULL-terminated list of char* pointers is at the beginning of it, * followed by strings themselves. * Caller can deallocate entire list by single free(list). */ /* A horde of its helpers come first: */ static void o_addblock_duplicate_backslash(o_string *o, const char *str, int len) { while (--len >= 0) { char c = *str++; #if ENABLE_HUSH_BRACE_EXPANSION if (c == '{' || c == '}') { /* { -> \{, } -> \} */ o_addchr(o, '\\'); /* And now we want to add { or } and continue: * o_addchr(o, c); * continue; * luckily, just falling through achieves this. */ } #endif o_addchr(o, c); if (c == '\\') { /* \z -> \\\z; \ -> \\ */ o_addchr(o, '\\'); if (len) { len--; o_addchr(o, '\\'); o_addchr(o, *str++); } } } } /* Store given string, finalizing the word and starting new one whenever * we encounter IFS char(s). This is used for expanding variable values. * End-of-string does NOT finalize word: think about 'echo -$VAR-'. * Return in output->ended_in_ifs: * 1 - ended with IFS char, else 0 (this includes case of empty str). */ static int expand_on_ifs(o_string *output, int n, const char *str) { int last_is_ifs = 0; while (1) { int word_len; if (!*str) /* EOL - do not finalize word */ break; word_len = strcspn(str, G.ifs); if (word_len) { /* We have WORD_LEN leading non-IFS chars */ if (!(output->o_expflags & EXP_FLAG_GLOB)) { o_addblock(output, str, word_len); } else { /* Protect backslashes against globbing up :) * Example: "v='\*'; echo b$v" prints "b\*" * (and does not try to glob on "*") */ o_addblock_duplicate_backslash(output, str, word_len); /*/ Why can't we do it easier? */ /*o_addblock(output, str, word_len); - WRONG: "v='\*'; echo Z$v" prints "Z*" instead of "Z\*" */ /*o_addqblock(output, str, word_len); - WRONG: "v='*'; echo Z$v" prints "Z*" instead of Z* files */ } last_is_ifs = 0; str += word_len; if (!*str) /* EOL - do not finalize word */ break; } /* We know str here points to at least one IFS char */ last_is_ifs = 1; str += strspn(str, G.ifs_whitespace); /* skip IFS whitespace chars */ if (!*str) /* EOL - do not finalize word */ break; if (G.ifs_whitespace != G.ifs /* usually false ($IFS is usually all whitespace), */ && strchr(G.ifs, *str) /* the second check would fail */ ) { /* This is a non-whitespace $IFS char */ /* Skip it and IFS whitespace chars, start new word */ str++; str += strspn(str, G.ifs_whitespace); goto new_word; } /* Start new word... but not always! */ /* Case "v=' a'; echo ''$v": we do need to finalize empty word: */ if (output->has_quoted_part /* * Case "v=' a'; echo $v": * here nothing precedes the space in $v expansion, * therefore we should not finish the word * (IOW: if there *is* word to finalize, only then do it): * It's okay if this accesses the byte before first argv[]: * past call to o_save_ptr() cleared it to zero byte * (grep for -prev-ifs-check-). */ || output->data[output->length - 1] ) { new_word: o_addchr(output, '\0'); debug_print_list("expand_on_ifs", output, n); n = o_save_ptr(output, n); } } output->ended_in_ifs = last_is_ifs; 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 $var and `cmd`" * * Returns malloced string. * As an optimization, we return NULL if expansion is not needed. */ static char *encode_then_expand_string(const char *str) { char *exp_str; struct in_str input; o_string dest = NULL_O_STRING; const char *cp; cp = str; for (;;) { if (!*cp) return NULL; /* string has no special chars */ if (*cp == '$') break; if (*cp == '\\') break; #if ENABLE_HUSH_TICK if (*cp == '`') break; #endif cp++; } /* We need to expand. Example: * echo $(($a + `echo 1`)) $((1 + $((2)) )) */ setup_string_in_str(&input, str); encode_string(NULL, &dest, &input, EOF); //TODO: error check (encode_string returns 0 on error)? //bb_error_msg("'%s' -> '%s'", str, dest.data); exp_str = expand_string_to_string(dest.data, EXP_FLAG_ESC_GLOB_CHARS, /*unbackslash:*/ 1 ); //bb_error_msg("'%s' -> '%s'", dest.data, exp_str); o_free(&dest); return exp_str; } static const char *first_special_char_in_vararg(const char *cp) { for (;;) { if (!*cp) return NULL; /* string has no special chars */ if (*cp == '$') return cp; if (*cp == '\\') return cp; if (*cp == '\'') return cp; if (*cp == '"') return cp; #if ENABLE_HUSH_TICK if (*cp == '`') return cp; #endif /* dquoted "${x:+ARG}" should not glob, therefore * '*' et al require some non-literal processing: */ if (*cp == '*') return cp; if (*cp == '?') return cp; if (*cp == '[') return cp; cp++; } } /* Expanding ARG in ${var#ARG}, ${var%ARG}, or ${var/ARG/ARG}. * These can contain single- and double-quoted strings, * and treated as if the ARG string is initially unquoted. IOW: * ${var#ARG} and "${var#ARG}" treat ARG the same (ARG can even be * a dquoted string: "${var#"zz"}"), the difference only comes later * (word splitting and globbing of the ${var...} result). */ #if !BASH_PATTERN_SUBST #define encode_then_expand_vararg(str, handle_squotes, do_unbackslash) \ encode_then_expand_vararg(str, handle_squotes) #endif static char *encode_then_expand_vararg(const char *str, int handle_squotes, int do_unbackslash) { #if !BASH_PATTERN_SUBST && ENABLE_HUSH_CASE const int do_unbackslash = 0; #endif char *exp_str; struct in_str input; o_string dest = NULL_O_STRING; if (!first_special_char_in_vararg(str)) { /* string has no special chars */ return NULL; } setup_string_in_str(&input, str); dest.data = xzalloc(1); /* start as "", not as NULL */ exp_str = NULL; for (;;) { int ch; ch = i_getch(&input); debug_printf_parse("%s: ch=%c (%d) escape=%d\n", __func__, ch, ch, !!dest.o_expflags); if (!dest.o_expflags) { if (ch == EOF) break; if (handle_squotes && ch == '\'') { if (!add_till_single_quote_dquoted(&dest, &input)) goto ret; /* error */ continue; } } if (ch == EOF) { syntax_error_unterm_ch('"'); goto ret; /* error */ } if (ch == '"') { dest.o_expflags ^= EXP_FLAG_ESC_GLOB_CHARS; continue; } if (ch == '\\') { ch = i_getch(&input); if (ch == EOF) { //example? error message? syntax_error_unterm_ch('"'); debug_printf_parse("%s: error: \\\n", __func__); goto ret; } o_addqchr(&dest, ch); continue; } if (ch == '$') { if (parse_dollar_squote(NULL, &dest, &input)) continue; if (!parse_dollar(NULL, &dest, &input, /*quote_mask:*/ 0x80)) { debug_printf_parse("%s: error: parse_dollar returned 0 (error)\n", __func__); goto ret; } continue; } #if ENABLE_HUSH_TICK if (ch == '`') { //unsigned pos = dest->length; o_addchr(&dest, SPECIAL_VAR_SYMBOL); o_addchr(&dest, 0x80 | '`'); if (!add_till_backquote(&dest, &input, /*in_dquote:*/ dest.o_expflags /* nonzero if EXP_FLAG_ESC_GLOB_CHARS set */ ) ) { goto ret; /* error */ } o_addchr(&dest, SPECIAL_VAR_SYMBOL); //debug_printf_subst("SUBST RES3 '%s'\n", dest->data + pos); continue; } #endif o_addQchr(&dest, ch); } /* for (;;) */ debug_printf_parse("encode: '%s' -> '%s'\n", str, dest.data); exp_str = expand_string_to_string(dest.data, do_unbackslash ? EXP_FLAG_ESC_GLOB_CHARS : 0, do_unbackslash ); ret: debug_printf_parse("expand: '%s' -> '%s'\n", dest.data, exp_str); o_free(&dest); return exp_str; } /* Expanding ARG in ${var+ARG}, ${var-ARG} */ static NOINLINE int encode_then_append_var_plusminus(o_string *output, int n, char *str, int dquoted) { struct in_str input; o_string dest = NULL_O_STRING; if (!first_special_char_in_vararg(str) && '\0' == str[strcspn(str, G.ifs)] ) { /* string has no special chars * && string has no $IFS chars */ if (dquoted) { /* Prints 1 (quoted expansion is a "" word, not nothing): * set -- "${notexist-}"; echo $# */ output->has_quoted_part = 1; } return expand_vars_to_list(output, n, str); } setup_string_in_str(&input, str); for (;;) { int ch; ch = i_getch(&input); debug_printf_parse("%s: ch=%c (%d) escape=%x\n", __func__, ch, ch, dest.o_expflags); if (!dest.o_expflags) { if (ch == EOF) break; if (!dquoted && !(output->o_expflags & EXP_FLAG_SINGLEWORD) && strchr(G.ifs, ch)) { /* PREFIX${x:d${e}f ...} and we met space: expand "d${e}f" and start new word. * do not assume we are at the start of the word (PREFIX above). */ if (dest.data) { n = expand_vars_to_list(output, n, dest.data); o_free_and_set_NULL(&dest); o_addchr(output, '\0'); n = o_save_ptr(output, n); /* create next word */ } else if (output->length != o_get_last_ptr(output, n) || output->has_quoted_part ) { /* For these cases: * f() { for i; do echo "|$i|"; done; }; x=x * f a${x:+ }b # 1st condition * |a| * |b| * f ""${x:+ }b # 2nd condition * || * |b| */ o_addchr(output, '\0'); n = o_save_ptr(output, n); /* create next word */ } continue; } if (!dquoted && ch == '\'') { if (!add_till_single_quote_dquoted(&dest, &input)) goto ret; /* error */ o_addchr(&dest, SPECIAL_VAR_SYMBOL); o_addchr(&dest, SPECIAL_VAR_SYMBOL); continue; } } if (ch == EOF) { syntax_error_unterm_ch('"'); goto ret; /* error */ } if (ch == '"') { dest.o_expflags ^= EXP_FLAG_ESC_GLOB_CHARS; if (dest.o_expflags) { o_addchr(&dest, SPECIAL_VAR_SYMBOL); o_addchr(&dest, SPECIAL_VAR_SYMBOL); } continue; } if (ch == '\\') { ch = i_getch(&input); if (ch == EOF) { //example? error message? syntax_error_unterm_ch('"'); debug_printf_parse("%s: error: \\\n", __func__); goto ret; } o_addqchr(&dest, ch); continue; } if (ch == '$') { if (!parse_dollar(NULL, &dest, &input, /*quote_mask:*/ (dest.o_expflags || dquoted) ? 0x80 : 0)) { debug_printf_parse("%s: error: parse_dollar returned 0 (error)\n", __func__); goto ret; } continue; } #if ENABLE_HUSH_TICK if (ch == '`') { //unsigned pos = dest->length; o_addchr(&dest, SPECIAL_VAR_SYMBOL); o_addchr(&dest, (dest.o_expflags || dquoted) ? 0x80 | '`' : '`'); if (!add_till_backquote(&dest, &input, /*in_dquote:*/ dest.o_expflags /* nonzero if EXP_FLAG_ESC_GLOB_CHARS set */ ) ) { goto ret; /* error */ } o_addchr(&dest, SPECIAL_VAR_SYMBOL); //debug_printf_subst("SUBST RES3 '%s'\n", dest->data + pos); continue; } #endif if (dquoted) { /* Always glob-protect if in dquotes: * x=x; echo "${x:+/bin/c*}" - prints: /bin/c* * x=x; echo "${x:+"/bin/c*"}" - prints: /bin/c* */ o_addqchr(&dest, ch); } else { /* Glob-protect only if char is quoted: * x=x; echo ${x:+/bin/c*} - prints many filenames * x=x; echo ${x:+"/bin/c*"} - prints: /bin/c* */ o_addQchr(&dest, ch); } } /* for (;;) */ if (dest.data) { n = expand_vars_to_list(output, n, dest.data); } ret: o_free(&dest); return n; } #if ENABLE_FEATURE_SH_MATH static arith_t expand_and_evaluate_arith(const char *arg, const char **errmsg_p) { arith_state_t math_state; arith_t res; char *exp_str; math_state.lookupvar = get_local_var_value; math_state.setvar = set_local_var_from_halves; //math_state.endofname = endofname; exp_str = encode_then_expand_string(arg); res = arith(&math_state, exp_str ? exp_str : arg); free(exp_str); if (errmsg_p) *errmsg_p = math_state.errmsg; if (math_state.errmsg) msg_and_die_if_script(math_state.errmsg); return res; } #endif #if BASH_PATTERN_SUBST /* ${var/[/]pattern[/repl]} helpers */ static char *strstr_pattern(char *val, const char *pattern, int *size) { int first_escaped = (pattern[0] == '\\' && pattern[1]); /* "first_escaped" trick allows to treat e.g. "\*no_glob_chars" * as literal too (as it is semi-common, and easy to accomodate * by just using str + 1). */ int sz = strcspn(pattern + first_escaped * 2, "*?[\\"); if ((pattern + first_escaped * 2)[sz] == '\0') { /* Optimization for trivial patterns. * Testcase for very slow replace (performs about 22k replaces): * x=:::::::::::::::::::::: * x=$x$x;x=$x$x;x=$x$x;x=$x$x;x=$x$x;x=$x$x;x=$x$x;x=$x$x;x=$x$x;x=$x$x;echo ${#x} * echo "${x//:/|}" */ *size = sz + first_escaped; return strstr(val, pattern + first_escaped); } 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); /* Null pattern never matches, including if "var" is empty */ if (!pattern[0]) return result; /* NULL, no replaces happened */ while (1) { int size; char *s = strstr_pattern(val, pattern, &size); if (!s) break; result = xrealloc(result, res_len + (s - val) + repl_len + 1); strcpy(mempcpy(result + res_len, val, s - val), repl); res_len += (s - val) + repl_len; debug_printf_varexp("val:'%s' s:'%s' result:'%s'\n", val, s, result); val = s + size; if (exp_op == '/') break; } if (*val && 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 /* BASH_PATTERN_SUBST */ static int append_str_maybe_ifs_split(o_string *output, int n, int first_ch, const char *val) { if (!(first_ch & 0x80)) { /* unquoted $VAR */ debug_printf_expand("unquoted '%s', output->o_escape:%d\n", val, !!(output->o_expflags & EXP_FLAG_ESC_GLOB_CHARS)); if (val && val[0]) n = expand_on_ifs(output, n, val); } else { /* quoted "$VAR" */ output->has_quoted_part = 1; debug_printf_expand("quoted '%s', output->o_escape:%d\n", val, !!(output->o_expflags & EXP_FLAG_ESC_GLOB_CHARS)); if (val && val[0]) o_addQstr(output, val); } return n; } /* Handle varname... construct. */ static NOINLINE int expand_one_var(o_string *output, int n, int first_ch, char *arg, char **pp) { const char *val; char *to_be_freed; char *p; char *var; char exp_op; char exp_save = exp_save; /* for compiler */ char *exp_saveptr; /* points to expansion operator */ char *exp_word = exp_word; /* for compiler */ char arg0; val = NULL; to_be_freed = NULL; p = *pp; *p = '\0'; /* replace trailing SPECIAL_VAR_SYMBOL */ var = arg; exp_saveptr = arg[1] ? strchr(VAR_ENCODED_SUBST_OPS, arg[1]) : NULL; arg0 = arg[0]; arg[0] = (arg0 & 0x7f); exp_op = 0; if (arg[0] == '#' && arg[1] /* ${#...} but not ${#} */ && (!exp_saveptr /* and ( not(${#...}) */ || (arg[2] == '\0' && strchr(SPECIAL_VARS_STR, arg[1])) /* or ${#C} "len of $C" ) */ ) /* NB: skipping ^^^specvar check mishandles ${#::2} */ ) { /* It must be length operator: ${#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, arg[0]) /* 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++; //TODO: try ${var:} and ${var:bogus} in non-bash config if (BASH_SUBSTR && (!exp_op || !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} */ } /* Look up the variable in question */ if (isdigit(var[0])) { /* parse_dollar should have vetted var for us */ int nn = xatoi_positive(var); if (nn < G.global_argc) val = G.global_argv[nn]; /* 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; case '-': { /* active options */ /* Check set_mode() to see what option chars we support */ char *cp; val = cp = G.optstring_buf; if (G.o_opt[OPT_O_ERREXIT]) *cp++ = 'e'; if (G_interactive_fd) *cp++ = 'i'; if (G_x_mode) *cp++ = 'x'; /* If G.o_opt[OPT_O_NOEXEC] is true, * commands read but are not executed, * so $- can not execute too, 'n' is never seen in $-. */ if (G.opt_c) *cp++ = 'c'; if (G.opt_s) *cp++ = 's'; *cp = '\0'; break; } default: val = get_local_var_value(var); } } /* Handle any expansions */ if (exp_op == 'L') { reinit_unicode_for_hush(); debug_printf_expand("expand: length(%s)=", val); val = utoa(val ? unicode_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. */ /* bash compat: if x is "" and no shrinking of it is possible, * inner ${...} is not evaluated. Example: * unset b; : ${a%${b=B}}; echo $b * assignment b=B only happens if $a is not "". */ if (val && val[0]) { char *t; char *exp_exp_word; char *loc; unsigned scan_flags = pick_scan(exp_op, *exp_word); if (exp_op == *exp_word) /* ## or %% */ exp_word++; debug_printf_expand("expand: exp_word:'%s'\n", exp_word); exp_exp_word = encode_then_expand_vararg(exp_word, /*handle_squotes:*/ 1, /*unbackslash:*/ 0); if (exp_exp_word) exp_word = exp_exp_word; debug_printf_expand("expand: exp_word:'%s'\n", exp_word); /* * 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: * scan_and_match momentarily stores NULs there. */ t = (char*)val; loc = scan_and_match(t, exp_word, scan_flags); debug_printf_expand("op:%c str:'%s' pat:'%s' res:'%s'\n", exp_op, t, exp_word, loc); free(exp_exp_word); if (loc) { /* match was found */ if (scan_flags & SCAN_MATCH_LEFT_HALF) /* #[#] */ val = loc; /* take right part */ else /* %[%] */ val = to_be_freed = xstrndup(val, loc - val); /* left */ } } } #if BASH_PATTERN_SUBST else if (exp_op == '/' || exp_op == '\\') { /* It's ${var/[/]pattern[/repl]} thing. * Note that in encoded form it has TWO parts: * var/patternrepl * and if // is used, it is encoded as \: * var\patternrepl */ /* bash compat: if var is "", both pattern and repl * are still evaluated, if it is unset, then not: * unset b; a=; : ${a/z/${b=3}}; echo $b # b=3 * unset b; unset a; : ${a/z/${b=3}}; echo $b # b not set */ if (val /*&& val[0]*/) { /* pattern uses non-standard expansion. * repl should be unbackslashed and globbed * by the usual expansion rules: * >az >bz * v='a bz'; echo "${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!) */ char *pattern, *repl, *t; pattern = encode_then_expand_vararg(exp_word, /*handle_squotes:*/ 1, /*unbackslash:*/ 0); 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 = encode_then_expand_vararg(exp_word, /*handle_squotes:*/ 1, /*unbackslash:*/ 1); 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); } else { /* Unset variable always gives nothing */ // a=; echo ${a/*/w} # "w" // unset a; echo ${a/*/w} # "" /* Just skip "replace" part */ *p++ = SPECIAL_VAR_SYMBOL; p = strchr(p, SPECIAL_VAR_SYMBOL); *p = '\0'; } } #endif /* BASH_PATTERN_SUBST */ else if (exp_op == ':') { #if BASH_SUBSTR && ENABLE_FEATURE_SH_MATH /* It's ${var:N[:M]} bashism. * Note that in encoded form it has TWO parts: * var:NM */ arith_t beg, len; unsigned vallen; const char *errmsg; beg = expand_and_evaluate_arith(exp_word, &errmsg); if (errmsg) goto empty_result; 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'; vallen = val ? strlen(val) : 0; if (beg < 0) { /* negative beg counts from the end */ beg = (arith_t)vallen + beg; } /* If expansion will be empty, do not even evaluate len */ if (!val || beg < 0 || beg > vallen) { /* Why > vallen, not >=? bash: * unset b; a=ab; : ${a:2:${b=3}}; echo $b # "", b=3 (!!!) * unset b; a=a; : ${a:2:${b=3}}; echo $b # "", b not set */ goto empty_result; } len = expand_and_evaluate_arith(exp_word, &errmsg); if (errmsg) goto empty_result; debug_printf_varexp("len:'%s'=%lld\n", exp_word, (long long)len); debug_printf_varexp("from val:'%s'\n", val); if (len < 0) { /* in bash, len=-n means strlen()-n */ len = (arith_t)vallen - beg + len; if (len < 0) /* bash compat */ msg_and_die_if_script("%s: substring expression < 0", var); } if (len <= 0 || !val /*|| beg >= vallen*/) { empty_result: val = NULL; } 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 /* not (HUSH_SUBSTR_EXPANSION && FEATURE_SH_MATH) */ msg_and_die_if_script("malformed ${%s:...}", var); val = NULL; #endif } 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. * * Word-splitting and single quote behavior: * * $ f() { for i; do echo "|$i|"; done; } * * $ x=; f ${x:?'x y' z}; echo $? * bash: x: x y z # neither f nor "echo $?" executes * (if interactive, bash does not exit, but merely aborts to prompt. $? is set to 1) * $ x=; f "${x:?'x y' z}" * bash: x: x y z # dash prints: dash: x: 'x y' z * * $ x=; f ${x:='x y' z} * |x| * |y| * |z| * $ x=; f "${x:='x y' z}" * |'x y' z| * * $ x=x; f ${x:+'x y' z} * |x y| * |z| * $ x=x; f "${x:+'x y' z}" * |'x y' z| * * $ x=; f ${x:-'x y' z} * |x y| * |z| * $ x=; f "${x:-'x y' z}" * |'x y' z| */ 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) { if (exp_op == '+' || exp_op == '-') { /* ${var+word} - use alternative value */ /* ${var-word} - use default value */ n = encode_then_append_var_plusminus(output, n, exp_word, /*dquoted:*/ (arg0 & 0x80) ); val = NULL; } else { /* ${var?word} - indicate error if unset */ /* ${var=word} - assign and use default value */ to_be_freed = encode_then_expand_vararg(exp_word, /*handle_squotes:*/ !(arg0 & 0x80), /*unbackslash:*/ 0 ); if (to_be_freed) exp_word = to_be_freed; if (exp_op == '?') { /* mimic bash message */ msg_and_die_if_script("%s: %s", var, exp_word[0] ? exp_word : "parameter null or not set" /* ash has more specific messages, a-la: */ /*: (exp_save == ':' ? "parameter null or not set" : "parameter not set")*/ ); //TODO: how interactive bash aborts expansion mid-command? //It aborts the entire line, returns to prompt: // $ f() { for i; do echo "|$i|"; done; }; x=; f "${x:?'x y' z}"; echo YO // bash: x: x y z // $ // ("echo YO" is not executed, neither the f function call) } else { val = exp_word; } if (exp_op == '=') { /* ${var=[word]} or ${var:=[word]} */ if (isdigit(var[0]) || var[0] == '#') { /* mimic bash message */ msg_and_die_if_script("$%s: cannot assign in this way", var); val = NULL; } else { char *new_var = xasprintf("%s=%s", var, val); set_local_var0(new_var); } } } } } /* one of "-=+?" */ *exp_saveptr = exp_save; } /* if (exp_op) */ arg[0] = arg0; *pp = p; n = append_str_maybe_ifs_split(output, n, first_ch, val); free(to_be_freed); return n; } /* 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) { /* output->o_expflags & EXP_FLAG_SINGLEWORD (0x80) if we are in * expansion of right-hand side of assignment == 1-element expand. */ char cant_be_null = 0; /* only bit 0x80 matters */ char *p; debug_printf_expand("expand_vars_to_list: arg:'%s' singleword:%x\n", arg, !!(output->o_expflags & EXP_FLAG_SINGLEWORD)); debug_print_list("expand_vars_to_list[0]", output, n); while ((p = strchr(arg, SPECIAL_VAR_SYMBOL)) != NULL) { char first_ch; #if ENABLE_FEATURE_SH_MATH char arith_buf[sizeof(arith_t)*3 + 2]; #endif if (output->ended_in_ifs) { o_addchr(output, '\0'); n = o_save_ptr(output, n); output->ended_in_ifs = 0; } o_addblock(output, arg, p - arg); debug_print_list("expand_vars_to_list[1]", output, n); arg = ++p; p = strchr(p, SPECIAL_VAR_SYMBOL); /* Fetch special var name (if it is indeed one of them) * and quote bit, force the bit on if singleword expansion - * important for not getting v=$@ expand to many words. */ first_ch = arg[0] | (output->o_expflags & EXP_FLAG_SINGLEWORD); /* Is this variable quoted and thus expansion can't be null? * "$@" is special. Even if quoted, it can still * expand to nothing (not even an empty string), * thus it is excluded. */ if ((first_ch & 0x7f) != '@') cant_be_null |= 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; cant_be_null |= first_ch; /* do it for "$@" _now_, when we know it's not empty */ if (!(first_ch & 0x80)) { /* unquoted $* or $@ */ while (G.global_argv[i]) { n = expand_on_ifs(output, n, G.global_argv[i]); debug_printf_expand("expand_vars_to_list: argv %d (last %d)\n", i, G.global_argc - 1); if (G.global_argv[i++][0] && G.global_argv[i]) { /* this argv[] is not empty and not last: * put terminating NUL, start new word */ o_addchr(output, '\0'); debug_print_list("expand_vars_to_list[2]", output, n); n = o_save_ptr(output, n); debug_print_list("expand_vars_to_list[3]", output, n); } } } else /* If EXP_FLAG_SINGLEWORD, we handle assignment 'a=....$@.....' * and in this case should treat it like '$*' - see 'else...' below */ if (first_ch == (char)('@'|0x80) /* quoted $@ */ && !(output->o_expflags & EXP_FLAG_SINGLEWORD) /* not v="$@" case */ ) { 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 $* (or v="$@" case): 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]); } output->has_quoted_part = 1; } break; } case SPECIAL_VAR_SYMBOL: { /* */ /* "Empty variable", used to make "" etc to not disappear */ output->has_quoted_part = 1; cant_be_null = 0x80; arg++; break; } case SPECIAL_VAR_QUOTED_SVS: /* */ /* "^C variable", represents literal ^C char (possible in scripts) */ o_addchr(output, SPECIAL_VAR_SYMBOL); arg++; break; #if ENABLE_HUSH_TICK case '`': { /* `cmd */ o_string subst_result = NULL_O_STRING; *p = '\0'; /* replace trailing */ arg++; /* Can't just stuff it into output o_string, * expanded result may need to be globbed * and $IFS-split */ debug_printf_subst("SUBST '%s' first_ch %x\n", arg, first_ch); G.last_exitcode = process_command_subs(&subst_result, arg); G.expand_exitcode = G.last_exitcode; debug_printf_subst("SUBST RES:%d '%s'\n", G.last_exitcode, subst_result.data); n = append_str_maybe_ifs_split(output, n, first_ch, subst_result.data); o_free(&subst_result); break; } #endif #if ENABLE_FEATURE_SH_MATH case '+': { /* +arith */ arith_t res; arg++; /* skip '+' */ *p = '\0'; /* replace trailing */ debug_printf_subst("ARITH '%s' first_ch %x\n", arg, first_ch); res = expand_and_evaluate_arith(arg, NULL); debug_printf_subst("ARITH RES '"ARITH_FMT"'\n", res); sprintf(arith_buf, ARITH_FMT, res); if (res < 0 && first_ch == (char)('+'|0x80) /* && (output->o_expflags & EXP_FLAG_ESC_GLOB_CHARS) */ ) { /* Quoted negative ariths, like filename[0"$((-9))"], * should not be interpreted as glob ranges. * Convert leading '-' to '\-': */ o_grow_by(output, 1); output->data[output->length++] = '\\'; } o_addstr(output, arith_buf); break; } #endif default: /* varname[ops] */ n = expand_one_var(output, n, first_ch, arg, &p); break; } /* switch (char after ) */ /* Restore NULL'ed SPECIAL_VAR_SYMBOL. * Do the check to avoid writing to a const string. */ if (*p != SPECIAL_VAR_SYMBOL) *p = SPECIAL_VAR_SYMBOL; arg = ++p; } /* end of "while (SPECIAL_VAR_SYMBOL is found) ..." */ if (*arg) { /* handle trailing string */ if (output->ended_in_ifs) { o_addchr(output, '\0'); n = o_save_ptr(output, n); } 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(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 */ && !(cant_be_null & 0x80) /* and all vars were not quoted */ && !output->has_quoted_part ) { n--; /* allow to reuse list[n] later without re-growth */ output->has_empty_slot = 1; } return n; } static char **expand_variables(char **argv, unsigned expflags) { int n; char **list; o_string output = NULL_O_STRING; output.o_expflags = expflags; n = 0; for (;;) { /* go to next list[n] */ output.ended_in_ifs = 0; n = o_save_ptr(&output, n); if (!*argv) break; /* expand argv[i] */ n = expand_vars_to_list(&output, n, *argv++); /* if (!output->has_empty_slot) -- need this?? */ o_addchr(&output, '\0'); } 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, EXP_FLAG_GLOB | EXP_FLAG_ESC_GLOB_CHARS); } #if defined(CMD_SINGLEWORD_NOGLOB) || defined(CMD_TEST2_SINGLEWORD_NOGLOB) static char **expand_strvec_to_strvec_singleword_noglob(char **argv) { return expand_variables(argv, EXP_FLAG_SINGLEWORD); } #endif /* Used for expansion of right hand of assignments, * $((...)), heredocs, variable expansion parts. * * 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, int EXP_flags, int do_unbackslash) { #if !BASH_PATTERN_SUBST && !ENABLE_HUSH_CASE const int do_unbackslash = 1; const int EXP_flags = EXP_FLAG_ESC_GLOB_CHARS; #endif char *argv[2], **list; debug_printf_expand("string_to_string<='%s'\n", str); /* 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, EXP_flags | EXP_FLAG_SINGLEWORD); if (!list[0]) { /* Example where it happens: * x=; echo ${x:-"$@"} */ ((char*)list)[0] = '\0'; } else { if (HUSH_DEBUG) if (list[1]) bb_simple_error_msg_and_die("BUG in varexp2"); /* actually, just move string 2*sizeof(char*) bytes back */ overlapping_strcpy((char*)list, list[0]); if (do_unbackslash) unbackslash((char*)list); } debug_printf_expand("string_to_string=>'%s'\n", (char*)list); return (char*)list; } #if 0 static char* expand_strvec_to_string(char **argv) { char **list; list = expand_variables(argv, EXP_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] ? list[0] : ""); debug_printf_expand("strvec_to_string='%s'\n", (char*)list); return (char*)list; } #endif 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++) { p = add_string_to_strings(p, expand_string_to_string(argv[i], EXP_FLAG_ESC_GLOB_CHARS, /*unbackslash:*/ 1 ) ); G.expanded_assignments = p; } G.expanded_assignments = NULL; return p; } static void switch_off_special_sigs(unsigned mask) { unsigned sig = 0; while ((mask >>= 1) != 0) { sig++; if (!(mask & 1)) continue; #if ENABLE_HUSH_TRAP if (G_traps) { if (G_traps[sig] && !G_traps[sig][0]) /* trap is '', has to remain SIG_IGN */ continue; free(G_traps[sig]); G_traps[sig] = NULL; } #endif /* We are here only if no trap or trap was not '' */ install_sighandler(sig, SIG_DFL); } } #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). */ IF_HUSH_TRAP(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. */ mask = (G.special_sig_mask & SPECIAL_INTERACTIVE_SIGS) | G_fatal_sig_mask; if (!G_traps && !mask) return; /* already no traps and no special sigs */ /* Switch off special sigs */ switch_off_special_sigs(mask); # if ENABLE_HUSH_JOB G_fatal_sig_mask = 0; # endif G.special_sig_mask &= ~SPECIAL_INTERACTIVE_SIGS; /* SIGQUIT,SIGCHLD and maybe SPECIAL_JOBSTOP_SIGS * remain set in G.special_sig_mask */ # if ENABLE_HUSH_TRAP if (!G_traps) return; /* Reset all sigs to default except ones with empty traps */ for (sig = 0; sig < NSIG; sig++) { if (!G_traps[sig]) continue; /* no trap: nothing to do */ if (!G_traps[sig][0]) continue; /* empty trap: has to remain SIG_IGN */ /* sig has non-empty trap, reset it: */ free(G_traps[sig]); G_traps[sig] = NULL; /* There is no signal for trap 0 (EXIT) */ if (sig == 0) continue; install_sighandler(sig, pick_sighandler(sig)); } # endif } #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_LINENO_VAR cnt += 2; # endif # 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_LINENO_VAR *pp++ = (char *) "-L"; *pp++ = utoa(G.execute_lineno); # endif # 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. */ *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); /* Don't propagate SIG_IGN to the child */ if (SPECIAL_JOBSTOP_SIGS != 0) switch_off_special_sigs(G.special_sig_mask & SPECIAL_JOBSTOP_SIGS); 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_simple_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