glib/gshell.c

   1 /* gshell.c - Shell-related utilities
   2  *
   3  *  Copyright 2000 Red Hat, Inc.
   4  *  g_execvpe implementation based on GNU libc execvp:
   5  *   Copyright 1991, 92, 95, 96, 97, 98, 99 Free Software Foundation, Inc.
   6  *
   7  * GLib is free software; you can redistribute it and/or
   8  * modify it under the terms of the GNU Lesser General Public License as
   9  * published by the Free Software Foundation; either version 2 of the
  10  * License, or (at your option) any later version.
  11  *
  12  * GLib is distributed in the hope that it will be useful,
  13  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  14  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  15  * Lesser General Public License for more details.
  16  *
  17  * You should have received a copy of the GNU Lesser General Public
  18  * License along with GLib; see the file COPYING.LIB.  If not, write
  19  * to the Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  20  * Boston, MA 02111-1307, USA.
  21  */
  22
  23 #include "config.h"
  24
  25 #include <string.h>
  26
  27 #include "gshell.h"
  28
  29 #include "gslist.h"
  30 #include "gstrfuncs.h"
  31 #include "gstring.h"
  32 #include "gtestutils.h"
  33 #include "glibintl.h"
  34 #include "gthread.h"
  35
  36 /**
  37  * SECTION:shell
  38  * @title: Shell-related Utilities
  39  * @short_description: shell-like commandline handling
  40  **/
  41
  42 /**
  43  * G_SHELL_ERROR:
  44  *
  45  * Error domain for shell functions. Errors in this domain will be from
  46  * the #GShellError enumeration. See #GError for information on error
  47  * domains.
  48  **/
  49
  50 /**
  51  * GShellError:
  52  * @G_SHELL_ERROR_BAD_QUOTING: Mismatched or otherwise mangled quoting.
  53  * @G_SHELL_ERROR_EMPTY_STRING: String to be parsed was empty.
  54  * @G_SHELL_ERROR_FAILED: Some other error.
  55  *
  56  * Error codes returned by shell functions.
  57  **/
  58 G_DEFINE_QUARK (g-shell-error-quark, g_shell_error)
  59
  60 /* Single quotes preserve the literal string exactly. escape
  61  * sequences are not allowed; not even \' - if you want a '
  62  * in the quoted text, you have to do something like 'foo'\''bar'
  63  *
  64  * Double quotes allow $ ` " \ and newline to be escaped with backslash.
  65  * Otherwise double quotes preserve things literally.
  66  */
  67
  68 static gboolean
  69 unquote_string_inplace (gchar* str, gchar** end, GError** err)
  70 {
  71   gchar* dest;
  72   gchar* s;
  73   gchar quote_char;
  74
  75   g_return_val_if_fail(end != NULL, FALSE);
  76   g_return_val_if_fail(err == NULL || *err == NULL, FALSE);
  77   g_return_val_if_fail(str != NULL, FALSE);
  78
  79   dest = s = str;
  80
  81   quote_char = *s;
  82
  83   if (!(*s == '"' || *s == '\''))
  84     {
  85       g_set_error_literal (err,
  86                            G_SHELL_ERROR,
  87                            G_SHELL_ERROR_BAD_QUOTING,
  88                            _("Quoted text doesn't begin with a quotation mark"));
  89       *end = str;
  90       return FALSE;
  91     }
  92
  93   /* Skip the initial quote mark */
  94   ++s;
  95
  96   if (quote_char == '"')
  97     {
  98       while (*s)
  99         {
 100           g_assert(s > dest); /* loop invariant */
 101
 102           switch (*s)
 103             {
 104             case '"':
 105               /* End of the string, return now */
 106               *dest = '\0';
 107               ++s;
 108               *end = s;
 109               return TRUE;
 110               break;
 111
 112             case '\\':
 113               /* Possible escaped quote or \ */
 114               ++s;
 115               switch (*s)
 116                 {
 117                 case '"':
 118                 case '\\':
 119                 case '`':
 120                 case '$':
 121                 case '\n':
 122                   *dest = *s;
 123                   ++s;
 124                   ++dest;
 125                   break;
 126
 127                 default:
 128                   /* not an escaped char */
 129                   *dest = '\\';
 130                   ++dest;
 131                   /* ++s already done. */
 132                   break;
 133                 }
 134               break;
 135
 136             default:
 137               *dest = *s;
 138               ++dest;
 139               ++s;
 140               break;
 141             }
 142
 143           g_assert(s > dest); /* loop invariant */
 144         }
 145     }
 146   else
 147     {
 148       while (*s)
 149         {
 150           g_assert(s > dest); /* loop invariant */
 151
 152           if (*s == '\'')
 153             {
 154               /* End of the string, return now */
 155               *dest = '\0';
 156               ++s;
 157               *end = s;
 158               return TRUE;
 159             }
 160           else
 161             {
 162               *dest = *s;
 163               ++dest;
 164               ++s;
 165             }
 166
 167           g_assert(s > dest); /* loop invariant */
 168         }
 169     }
 170
 171   /* If we reach here this means the close quote was never encountered */
 172
 173   *dest = '\0';
 174
 175   g_set_error_literal (err,
 176                        G_SHELL_ERROR,
 177                        G_SHELL_ERROR_BAD_QUOTING,
 178                        _("Unmatched quotation mark in command line or other shell-quoted text"));
 179   *end = s;
 180   return FALSE;
 181 }
 182
 183 /**
 184  * g_shell_quote:
 185  * @unquoted_string: a literal string
 186  *
 187  * Quotes a string so that the shell (/bin/sh) will interpret the
 188  * quoted string to mean @unquoted_string. If you pass a filename to
 189  * the shell, for example, you should first quote it with this
 190  * function.  The return value must be freed with g_free(). The
 191  * quoting style used is undefined (single or double quotes may be
 192  * used).
 193  *
 194  * Return value: quoted string
 195  **/
 196 gchar*
 197 g_shell_quote (const gchar *unquoted_string)
 198 {
 199   /* We always use single quotes, because the algorithm is cheesier.
 200    * We could use double if we felt like it, that might be more
 201    * human-readable.
 202    */
 203
 204   const gchar *p;
 205   GString *dest;
 206
 207   g_return_val_if_fail (unquoted_string != NULL, NULL);
 208
 209   dest = g_string_new ("'");
 210
 211   p = unquoted_string;
 212
 213   /* could speed this up a lot by appending chunks of text at a
 214    * time.
 215    */
 216   while (*p)
 217     {
 218       /* Replace literal ' with a close ', a \', and a open ' */
 219       if (*p == '\'')
 220         g_string_append (dest, "'\\''");
 221       else
 222         g_string_append_c (dest, *p);
 223
 224       ++p;
 225     }
 226
 227   /* close the quote */
 228   g_string_append_c (dest, '\'');
 229
 230   return g_string_free (dest, FALSE);
 231 }
 232
 233 /**
 234  * g_shell_unquote:
 235  * @quoted_string: shell-quoted string
 236  * @error: error return location or NULL
 237  *
 238  * Unquotes a string as the shell (/bin/sh) would. Only handles
 239  * quotes; if a string contains file globs, arithmetic operators,
 240  * variables, backticks, redirections, or other special-to-the-shell
 241  * features, the result will be different from the result a real shell
 242  * would produce (the variables, backticks, etc. will be passed
 243  * through literally instead of being expanded). This function is
 244  * guaranteed to succeed if applied to the result of
 245  * g_shell_quote(). If it fails, it returns %NULL and sets the
 246  * error. The @quoted_string need not actually contain quoted or
 247  * escaped text; g_shell_unquote() simply goes through the string and
 248  * unquotes/unescapes anything that the shell would. Both single and
 249  * double quotes are handled, as are escapes including escaped
 250  * newlines. The return value must be freed with g_free(). Possible
 251  * errors are in the #G_SHELL_ERROR domain.
 252  *
 253  * Shell quoting rules are a bit strange. Single quotes preserve the
 254  * literal string exactly. escape sequences are not allowed; not even
 255  * \' - if you want a ' in the quoted text, you have to do something
 256  * like 'foo'\''bar'.  Double quotes allow $, `, ", \, and newline to
 257  * be escaped with backslash. Otherwise double quotes preserve things
 258  * literally.
 259  *
 260  * Return value: an unquoted string
 261  **/
 262 gchar*
 263 g_shell_unquote (const gchar *quoted_string,
 264                  GError     **error)
 265 {
 266   gchar *unquoted;
 267   gchar *end;
 268   gchar *start;
 269   GString *retval;
 270
 271   g_return_val_if_fail (quoted_string != NULL, NULL);
 272
 273   unquoted = g_strdup (quoted_string);
 274
 275   start = unquoted;
 276   end = unquoted;
 277   retval = g_string_new (NULL);
 278
 279   /* The loop allows cases such as
 280    * "foo"blah blah'bar'woo foo"baz"la la la\'\''foo'
 281    */
 282   while (*start)
 283     {
 284       /* Append all non-quoted chars, honoring backslash escape
 285        */
 286
 287       while (*start && !(*start == '"' || *start == '\''))
 288         {
 289           if (*start == '\\')
 290             {
 291               /* all characters can get escaped by backslash,
 292                * except newline, which is removed if it follows
 293                * a backslash outside of quotes
 294                */
 295
 296               ++start;
 297               if (*start)
 298                 {
 299                   if (*start != '\n')
 300                     g_string_append_c (retval, *start);
 301                   ++start;
 302                 }
 303             }
 304           else
 305             {
 306               g_string_append_c (retval, *start);
 307               ++start;
 308             }
 309         }
 310
 311       if (*start)
 312         {
 313           if (!unquote_string_inplace (start, &end, error))
 314             {
 315               goto error;
 316             }
 317           else
 318             {
 319               g_string_append (retval, start);
 320               start = end;
 321             }
 322         }
 323     }
 324
 325   g_free (unquoted);
 326   return g_string_free (retval, FALSE);
 327
 328  error:
 329   g_assert (error == NULL || *error != NULL);
 330
 331   g_free (unquoted);
 332   g_string_free (retval, TRUE);
 333   return NULL;
 334 }
 335
 336 /* g_parse_argv() does a semi-arbitrary weird subset of the way
 337  * the shell parses a command line. We don't do variable expansion,
 338  * don't understand that operators are tokens, don't do tilde expansion,
 339  * don't do command substitution, no arithmetic expansion, IFS gets ignored,
 340  * don't do filename globs, don't remove redirection stuff, etc.
 341  *
 342  * READ THE UNIX98 SPEC on "Shell Command Language" before changing
 343  * the behavior of this code.
 344  *
 345  * Steps to parsing the argv string:
 346  *
 347  *  - tokenize the string (but since we ignore operators,
 348  *    our tokenization may diverge from what the shell would do)
 349  *    note that tokenization ignores the internals of a quoted
 350  *    word and it always splits on spaces, not on IFS even
 351  *    if we used IFS. We also ignore "end of input indicator"
 352  *    (I guess this is control-D?)
 353  *
 354  *    Tokenization steps, from UNIX98 with operator stuff removed,
 355  *    are:
 356  *
 357  *    1) "If the current character is backslash, single-quote or
 358  *        double-quote (\, ' or ") and it is not quoted, it will affect
 359  *        quoting for subsequent characters up to the end of the quoted
 360  *        text. The rules for quoting are as described in Quoting
 361  *        . During token recognition no substitutions will be actually
 362  *        performed, and the result token will contain exactly the
 363  *        characters that appear in the input (except for newline
 364  *        character joining), unmodified, including any embedded or
 365  *        enclosing quotes or substitution operators, between the quote
 366  *        mark and the end of the quoted text. The token will not be
 367  *        delimited by the end of the quoted field."
 368  *
 369  *    2) "If the current character is an unquoted newline character,
 370  *        the current token will be delimited."
 371  *
 372  *    3) "If the current character is an unquoted blank character, any
 373  *        token containing the previous character is delimited and the
 374  *        current character will be discarded."
 375  *
 376  *    4) "If the previous character was part of a word, the current
 377  *        character will be appended to that word."
 378  *
 379  *    5) "If the current character is a "#", it and all subsequent
 380  *        characters up to, but excluding, the next newline character
 381  *        will be discarded as a comment. The newline character that
 382  *        ends the line is not considered part of the comment. The
 383  *        "#" starts a comment only when it is at the beginning of a
 384  *        token. Since the search for the end-of-comment does not
 385  *        consider an escaped newline character specially, a comment
 386  *        cannot be continued to the next line."
 387  *
 388  *    6) "The current character will be used as the start of a new word."
 389  *
 390  *
 391  *  - for each token (word), perform portions of word expansion, namely
 392  *    field splitting (using default whitespace IFS) and quote
 393  *    removal.  Field splitting may increase the number of words.
 394  *    Quote removal does not increase the number of words.
 395  *
 396  *   "If the complete expansion appropriate for a word results in an
 397  *   empty field, that empty field will be deleted from the list of
 398  *   fields that form the completely expanded command, unless the
 399  *   original word contained single-quote or double-quote characters."
 400  *    - UNIX98 spec
 401  *
 402  *
 403  */
 404
 405 static inline void
 406 ensure_token (GString **token)
 407 {
 408   if (*token == NULL)
 409     *token = g_string_new (NULL);
 410 }
 411
 412 static void
 413 delimit_token (GString **token,
 414                GSList **retval)
 415 {
 416   if (*token == NULL)
 417     return;
 418
 419   *retval = g_slist_prepend (*retval, g_string_free (*token, FALSE));
 420
 421   *token = NULL;
 422 }
 423
 424 static GSList*
 425 tokenize_command_line (const gchar *command_line,
 426                        GError **error)
 427 {
 428   gchar current_quote;
 429   const gchar *p;
 430   GString *current_token = NULL;
 431   GSList *retval = NULL;
 432   gboolean quoted;
 433
 434   current_quote = '\0';
 435   quoted = FALSE;
 436   p = command_line;
 437
 438   while (*p)
 439     {
 440       if (current_quote == '\\')
 441         {
 442           if (*p == '\n')
 443             {
 444               /* we append nothing; backslash-newline become nothing */
 445             }
 446           else
 447             {
 448               /* we append the backslash and the current char,
 449                * to be interpreted later after tokenization
 450                */
 451               ensure_token (&current_token);
 452               g_string_append_c (current_token, '\\');
 453               g_string_append_c (current_token, *p);
 454             }
 455
 456           current_quote = '\0';
 457         }
 458       else if (current_quote == '#')
 459         {
 460           /* Discard up to and including next newline */
 461           while (*p && *p != '\n')
 462             ++p;
 463
 464           current_quote = '\0';
 465
 466           if (*p == '\0')
 467             break;
 468         }
 469       else if (current_quote)
 470         {
 471           if (*p == current_quote &&
 472               /* check that it isn't an escaped double quote */
 473               !(current_quote == '"' && quoted))
 474             {
 475               /* close the quote */
 476               current_quote = '\0';
 477             }
 478
 479           /* Everything inside quotes, and the close quote,
 480            * gets appended literally.
 481            */
 482
 483           ensure_token (&current_token);
 484           g_string_append_c (current_token, *p);
 485         }
 486       else
 487         {
 488           switch (*p)
 489             {
 490             case '\n':
 491               delimit_token (&current_token, &retval);
 492               break;
 493
 494             case ' ':
 495             case '\t':
 496               /* If the current token contains the previous char, delimit
 497                * the current token. A nonzero length
 498                * token should always contain the previous char.
 499                */
 500               if (current_token &&
 501                   current_token->len > 0)
 502                 {
 503                   delimit_token (&current_token, &retval);
 504                 }
 505
 506               /* discard all unquoted blanks (don't add them to a token) */
 507               break;
 508
 509
 510               /* single/double quotes are appended to the token,
 511                * escapes are maybe appended next time through the loop,
 512                * comment chars are never appended.
 513                */
 514
 515             case '\'':
 516             case '"':
 517               ensure_token (&current_token);
 518               g_string_append_c (current_token, *p);
 519
 520               /* FALL THRU */
 521             case '\\':
 522               current_quote = *p;
 523               break;
 524
 525             case '#':
 526               if (p == command_line)
 527                 { /* '#' was the first char */
 528                   current_quote = *p;
 529                   break;
 530                 }
 531               switch(*(p-1))
 532                 {
 533                   case ' ':
 534                   case '\n':
 535                   case '\0':
 536                     current_quote = *p;
 537                     break;
 538                   default:
 539                     ensure_token (&current_token);
 540                     g_string_append_c (current_token, *p);
 541                     break;
 542                 }
 543               break;
 544
 545             default:
 546               /* Combines rules 4) and 6) - if we have a token, append to it,
 547                * otherwise create a new token.
 548                */
 549               ensure_token (&current_token);
 550               g_string_append_c (current_token, *p);
 551               break;
 552             }
 553         }
 554
 555       /* We need to count consecutive backslashes mod 2,
 556        * to detect escaped doublequotes.
 557        */
 558       if (*p != '\\')
 559         quoted = FALSE;
 560       else
 561         quoted = !quoted;
 562
 563       ++p;
 564     }
 565
 566   delimit_token (&current_token, &retval);
 567
 568   if (current_quote)
 569     {
 570       if (current_quote == '\\')
 571         g_set_error (error,
 572                      G_SHELL_ERROR,
 573                      G_SHELL_ERROR_BAD_QUOTING,
 574                      _("Text ended just after a '\\' character."
 575                        " (The text was '%s')"),
 576                      command_line);
 577       else
 578         g_set_error (error,
 579                      G_SHELL_ERROR,
 580                      G_SHELL_ERROR_BAD_QUOTING,
 581                      _("Text ended before matching quote was found for %c."
 582                        " (The text was '%s')"),
 583                      current_quote, command_line);
 584
 585       goto error;
 586     }
 587
 588   if (retval == NULL)
 589     {
 590       g_set_error_literal (error,
 591                            G_SHELL_ERROR,
 592                            G_SHELL_ERROR_EMPTY_STRING,
 593                            _("Text was empty (or contained only whitespace)"));
 594
 595       goto error;
 596     }
 597
 598   /* we appended backward */
 599   retval = g_slist_reverse (retval);
 600
 601   return retval;
 602
 603  error:
 604   g_assert (error == NULL || *error != NULL);
 605
 606   g_slist_free_full (retval, g_free);
 607
 608   return NULL;
 609 }
 610
 611 /**
 612  * g_shell_parse_argv:
 613  * @command_line: command line to parse
 614  * @argcp: (out): return location for number of args
 615  * @argvp: (out) (array length=argcp zero-terminated=1): return location for array of args
 616  * @error: return location for error
 617  *
 618  * Parses a command line into an argument vector, in much the same way
 619  * the shell would, but without many of the expansions the shell would
 620  * perform (variable expansion, globs, operators, filename expansion,
 621  * etc. are not supported). The results are defined to be the same as
 622  * those you would get from a UNIX98 /bin/sh, as long as the input
 623  * contains none of the unsupported shell expansions. If the input
 624  * does contain such expansions, they are passed through
 625  * literally. Possible errors are those from the #G_SHELL_ERROR
 626  * domain. Free the returned vector with g_strfreev().
 627  *
 628  * Return value: %TRUE on success, %FALSE if error set
 629  **/
 630 gboolean
 631 g_shell_parse_argv (const gchar *command_line,
 632                     gint        *argcp,
 633                     gchar     ***argvp,
 634                     GError     **error)
 635 {
 636   /* Code based on poptParseArgvString() from libpopt */
 637   gint argc = 0;
 638   gchar **argv = NULL;
 639   GSList *tokens = NULL;
 640   gint i;
 641   GSList *tmp_list;
 642
 643   g_return_val_if_fail (command_line != NULL, FALSE);
 644
 645   tokens = tokenize_command_line (command_line, error);
 646   if (tokens == NULL)
 647     return FALSE;
 648
 649   /* Because we can't have introduced any new blank space into the
 650    * tokens (we didn't do any new expansions), we don't need to
 651    * perform field splitting. If we were going to honor IFS or do any
 652    * expansions, we would have to do field splitting on each word
 653    * here. Also, if we were going to do any expansion we would need to
 654    * remove any zero-length words that didn't contain quotes
 655    * originally; but since there's no expansion we know all words have
 656    * nonzero length, unless they contain quotes.
 657    *
 658    * So, we simply remove quotes, and don't do any field splitting or
 659    * empty word removal, since we know there was no way to introduce
 660    * such things.
 661    */
 662
 663   argc = g_slist_length (tokens);
 664   argv = g_new0 (gchar*, argc + 1);
 665   i = 0;
 666   tmp_list = tokens;
 667   while (tmp_list)
 668     {
 669       argv[i] = g_shell_unquote (tmp_list->data, error);
 670
 671       /* Since we already checked that quotes matched up in the
 672        * tokenizer, this shouldn't be possible to reach I guess.
 673        */
 674       if (argv[i] == NULL)
 675         goto failed;
 676
 677       tmp_list = g_slist_next (tmp_list);
 678       ++i;
 679     }
 680
 681   g_slist_free_full (tokens, g_free);
 682
 683   if (argcp)
 684     *argcp = argc;
 685
 686   if (argvp)
 687     *argvp = argv;
 688   else
 689     g_strfreev (argv);
 690
 691   return TRUE;
 692
 693  failed:
 694
 695   g_assert (error == NULL || *error != NULL);
 696   g_strfreev (argv);
 697   g_slist_free_full (tokens, g_free);
 698
 699   return FALSE;
 700 }