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