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