Index: env.c =================================================================== RCS file: /home/ncvs/src/usr.bin/env/env.c,v retrieving revision 1.11 diff -u -r1.11 env.c --- env.c 4 Sep 2002 23:28:59 -0000 1.11 +++ env.c 15 Jun 2005 02:25:00 -0000 @@ -46,6 +46,9 @@ #include __FBSDID("$FreeBSD: src/usr.bin/env/env.c,v 1.11 2002/09/04 23:28:59 dwmalone Exp $"); +#include +#include +#include #include #include #include @@ -53,31 +56,84 @@ #include #include -extern char **environ; +extern char **environ; -static void usage(void); +/* + * isspace() takes a parameter of 'int', but expect values in the range + * of unsigned char. Define a wrapper which takes a value of type 'char', + * whether signed or unsigned, and ensure it ends up in the right range. + */ +#define isalnumch(Anychar) isalnum((u_char)(Anychar)) +#define isalphach(Anychar) isalpha((u_char)(Anychar)) +#define isspacech(Anychar) isspace((u_char)(Anychar)) + +static int verbose; + +static void expand_vars(char **thisarg_p, char **dest_p, const char + **src_p); +static int is_there(char *candidate); +static void search_paths(char *path, char **argv); +static void split_spaces(const char *str, int *origind, int *origc, + char ***origv); +static void usage(void); int main(int argc, char **argv) { - char **ep, *p; + char **ep, *p, **parg, *altpath; char *cleanenv[1]; - int ch; + int ch, want_clear; - while ((ch = getopt(argc, argv, "-i")) != -1) + altpath = NULL; + want_clear = 0; + while ((ch = getopt(argc, argv, "-iP:S:v")) != -1) switch(ch) { case '-': case 'i': - environ = cleanenv; - cleanenv[0] = NULL; + want_clear = 1; + break; + case 'P': + altpath = strdup(optarg); + break; + case 'S': + /* + * The -S option, for "split string on spaces, with + * support for some simple substitutions"... + */ + split_spaces(optarg, &optind, &argc, &argv); + break; + case 'v': + verbose++; + if (verbose > 1) + fprintf(stderr, "#env verbosity now at %d\n", + verbose); break; case '?': default: usage(); } - for (argv += optind; *argv && (p = strchr(*argv, '=')); ++argv) + if (want_clear) { + environ = cleanenv; + cleanenv[0] = NULL; + if (verbose) + fprintf(stderr, "#env clearing environ\n"); + } + for (argv += optind; *argv && (p = strchr(*argv, '=')); ++argv) { + if (verbose) + fprintf(stderr, "#env setenv:\t%s\n", *argv); (void)setenv(*argv, ++p, 1); + } if (*argv) { + if (altpath) + search_paths(altpath, argv); + if (verbose) { + fprintf(stderr, "#env executing:\t%s\n", *argv); + for (parg = argv, argc = 0; *parg; parg++, argc++) + fprintf(stderr, "#env arg[%d]=\t'%s'\n", + argc, *parg); + if (verbose > 1) + sleep(1); + } execvp(*argv, argv); err(errno == ENOENT ? 127 : 126, "%s", *argv); } @@ -86,10 +142,374 @@ exit(0); } +/* + * Routine to determine if a given fully-qualified filename is executable. + * This routine is copied almost verbatim from usr.bin/which/which.c. + */ +static int +is_there(char *candidate) +{ + struct stat fin; + + /* XXX work around access(2) false positives for superuser */ + if (access(candidate, X_OK) == 0 && + stat(candidate, &fin) == 0 && + S_ISREG(fin.st_mode) && + (getuid() != 0 || + (fin.st_mode & (S_IXUSR | S_IXGRP | S_IXOTH)) != 0)) { + if (verbose > 1) + fprintf(stderr, "#env matched:\t'%s'\n", candidate); + return (1); + } + return (0); +} + +/** + * Routine to search through an alternate path-list, looking for a given + * filename to execute. If the file is found, replace the original + * unqualified name with a fully-qualified path. This allows `env' to + * execute programs from a specific strict list of possible paths, without + * changing the value of PATH seen by the program which will be executed. + * E.G.: + * #!/usr/bin/env -S-P/usr/local/bin:/usr/bin perl + * will execute /usr/local/bin/perl or /usr/bin/perl (whichever is found + * first), no matter what the current value of PATH is, and without + * changing the value of PATH that the script will see when it runs. + * + * This is similar to the print_matches() routine in usr.bin/which/which.c. + */ +static void +search_paths(char *path, char **argv) +{ + char candidate[PATH_MAX]; + const char *d; + char *filename, *fqname; + + /* If the file has a `/' in it, then no search is done */ + filename = *argv; + if (strchr(filename, '/') != NULL) + return; + + if (verbose > 1) { + fprintf(stderr, "#env Searching:\t'%s'\n", path); + fprintf(stderr, "#env for file:\t'%s'\n", filename); + } + + fqname = NULL; + while ((d = strsep(&path, ":")) != NULL) { + if (*d == '\0') + d = "."; + if (snprintf(candidate, sizeof(candidate), "%s/%s", d, + filename) >= (int)sizeof(candidate)) + continue; + if (is_there(candidate)) { + fqname = candidate; + break; + } + } + + if (fqname == NULL) { + errno = ENOENT; + err(127, "%s", filename); + } + *argv = strdup(candidate); +} + +/** + * Routine to split a string into multiple parameters, while recognizing a + * few special characters. It recognizes both single and double-quoted + * strings. This processing is designed entirely for the benefit of the + * parsing of "#!"-lines (aka "shebang" lines == the first line of an + * executable script). Different operating systems parse that line in very + * different ways, and this split-on-spaces processing is meant to provide + * ways to specify arbitrary arguments on that line, no matter how the OS + * parses it. + * + * Within a single-quoted string, the two characters "\'" are treated as + * a literal "'" character to add to the string, and "\\" are treated as + * a literal "\" character to add. Other than that, all characters are + * copied until the processing gets to a terminating "'". + * + * Within a double-quoted string, many more "\"-style escape sequences + * are recognized, mostly copied from what is recognized in the `printf' + * command. Some OS's will not allow a literal blank character to be + * included in the one argument that they recognize on a shebang-line, + * so a few additional escape-sequences are defined to provide ways to + * specify blanks. + * + * Within a double-quoted string "\_" is turned into a literal blank. + * (Inside of a single-quoted string, the two characters are just copied) + * Outside of a quoted string, "\_" is treated as both a blank, and the + * end of the current argument. So with a shelbang-line of: + * #!/usr/bin/env -SA=avalue\_perl + * the -S value would be broken up into arguments "A=avalue" and "perl". + */ +static void +split_spaces(const char *str, int *origind, int *origc, char ***origv) +{ + const char *bq_src, *src; + char *dest, **newargv, *newstr, **nextarg, **oldarg; + int addcount, bq_destlen, copychar, found_sep, in_arg, in_dq, in_sq; + + /* + * Ignore leading space on the string, and then malloc enough room + * to build a copy of it. The copy might end up shorter than the + * original, due to quoted strings and '\'-processing. + */ + while (isspacech(*str)) + str++; + if (*str == '\0') + return; + newstr = malloc(strlen(str) + 1); + + /* + * Allocate plenty of space for the new array of arg-pointers, + * and start that array off with the first element of the old + * array. + */ + newargv = malloc((*origc + (strlen(str) / 2) + 2) * sizeof(char *)); + nextarg = newargv; + *nextarg++ = **origv; + + /* Come up with the new args by splitting up the given string. */ + addcount = 0; + bq_destlen = in_arg = in_dq = in_sq = 0; + bq_src = NULL; + for (src = str, dest = newstr; *src != '\0'; src++) { + copychar = found_sep = 0; + switch (*src) { + case '"': + if (in_sq) + copychar = *src; + else if (in_dq) + in_dq = 0; + else { + in_dq = 1; + bq_destlen = dest - *(nextarg - 1); + bq_src = src; + } + break; + case '$': + if (in_sq) + copychar = *src; + else { + expand_vars((nextarg - 1), &dest, &src); + } + break; + case '\'': + if (in_dq) + copychar = *src; + else if (in_sq) + in_sq = 0; + else { + in_sq = 1; + bq_destlen = dest - *(nextarg - 1); + bq_src = src; + } + break; + case '\\': + if (in_sq) { + /* + * Inside single-quoted strings, only the + * "\'" and "\\" are recognized as special + * strings. + */ + copychar = *(src + 1); + if (copychar == '\'' || copychar == '\\') + src++; + else + copychar = *src; + break; + } + src++; + switch (*src) { + case '"': + case '#': + case '$': + case '\'': + case '\\': + copychar = *src; + break; + case '_': + /* + * Alternate way to get a blank, which allows + * that blank be used to separate arguments + * when it is not inside a quoted string. + */ + if (in_dq) + copychar = ' '; + else { + found_sep = 1; + src++; + } + break; + case 'c': + /* + * Ignore remaining characters in the -S string. + * This would not make sense if found in the + * middle of a quoted string. + */ + if (in_dq) + errx(1, "Sequence '\\%c' is not allowed" + " in quoted strings", *src); + goto str_done; + case 'f': + copychar = '\f'; + break; + case 'n': + copychar = '\n'; + break; + case 'r': + copychar = '\r'; + break; + case 't': + copychar = '\t'; + break; + case 'v': + copychar = '\v'; + break; + default: + if (isspacech(*src)) + copychar = *src; + else + errx(1, "Invalid sequence '\\%c' in -S", + *src); + } + break; + default: + if ((in_dq || in_sq) && in_arg) + copychar = *src; + else if (in_arg && isspacech(*src)) + found_sep = 1; + else { + /* + * If the first character of a new argument + * is `#', then ignore the remaining chars. + */ + if (!in_arg && *src == '#') + goto str_done; + copychar = *src; + } + } + if (copychar) { + if (!in_arg) { + /* This is the first byte of a new argument */ + *nextarg++ = dest; + addcount++; + in_arg = 1; + } + *dest++ = (char)copychar; + } else if (found_sep) { + *dest++ = '\0'; + while (isspacech(*src)) + src++; + --src; + in_arg = 0; + } + } +str_done: + *dest = '\0'; + *nextarg = NULL; + if (in_dq || in_sq) { + errx(1, "No terminating quote for string: %.*s%s", + bq_destlen, *(nextarg - 1), bq_src); + } + if (verbose > 1) { + fprintf(stderr, "#env split -S:\t'%s'\n", str); + oldarg = newargv + 1; + fprintf(stderr, "#env into:\t'%s'\n", *oldarg); + for (oldarg++; *oldarg; oldarg++) + fprintf(stderr, "#env &\t'%s'\n", *oldarg); + } + + /* Copy the unprocessed arg-pointers from the original array */ + for (oldarg = *origv + *origind; *oldarg; oldarg++) + *nextarg++ = *oldarg; + *nextarg = NULL; + + /* Update optind/argc/argv in the calling routine */ + *origind = 1; + *origc += addcount; + *origv = newargv; +} + +/** + * Routine to split expand any environment variables referenced in the string + * that -S is processing. For now it only supports the form ${VARNAME}. It + * explicitly does not support $VARNAME, and obviously can not handle special + * shell-variables such as $?, $*, $1, etc. It is called with *src_p pointing + * at the initial '$', and if successful it will update *src_p, *dest_p, and + * possibly *thisarg_p in the calling routine. + */ +static void +expand_vars(char **thisarg_p, char **dest_p, const char **src_p) +{ + const char *vbegin, *vend, *vvalue; + char *edest, *newstr, *vname; + int bad_reference; + size_t namelen, newlen; + + bad_reference = 1; + vbegin = vend = (*src_p) + 1; + if (*vbegin++ == '{') + if (*vbegin == '_' || isalphach(*vbegin)) { + vend = vbegin + 1; + while (*vend == '_' || isalnumch(*vend)) + vend++; + if (*vend == '}') + bad_reference = 0; + } + if (bad_reference) + errx(1, "Only ${VARNAME} expansion is supported, error at: %s", + *src_p); + + /* + * We now know we have a valid environment variable name, so update + * the caller's source-pointer to the last character in that reference, + * and then pick up the matching value. If the variable is not found, + * or if it has a null value, then our work here is done. + */ + *src_p = vend; + namelen = vend - vbegin + 1; + vname = malloc(namelen); + strlcpy(vname, vbegin, namelen); + vvalue = getenv(vname); + if (vvalue == NULL || *vvalue == '\0') + return; + + /* + * There is some value to copy to the destination. If the value is + * shorter than the ${VARNAME} reference that it replaces, then we + * can just copy the value to the existing destination. + */ + edest = *dest_p; + if (strlen(vname) + 3 >= strlen(vvalue)) { + while (*vvalue != '\0') + *edest++ = *vvalue++; + *dest_p = edest; + return; + } + + /* + * The value is longer than the string it replaces, which means the + * present destination area is too small to hold it. Create a new + * destination area, copy the present 'thisarg' value and the value + * of the referenced-variable to it, and then update the caller's + * 'thisarg' and 'dest' variables to match. + */ + *edest = '\0'; /* Provide terminator for 'thisarg' */ + newlen = strlen(*thisarg_p) + strlen(vvalue) + strlen(*src_p) + 1; + newstr = malloc(newlen); + strcpy(newstr, *thisarg_p); + strcat(newstr, vvalue); + *thisarg_p = newstr; + *dest_p = strchr(newstr, '\0'); +} + static void usage(void) { (void)fprintf(stderr, - "usage: env [-i] [name=value ...] [utility [argument ...]]\n"); + "usage: env [-iv] [-P utilpath] [-S string] [name=value ...] [utility [argument ...]]\n"); exit(1); }