/* find -- search for files in a directory hierarchy Copyright (C) 1990, 1991, 1992, 1993, 1994, 2000, 2003, 2004, 2005, 2007, 2008, 2009, 2010, 2011 Free Software Foundation, Inc. This program is free software: you can redistribute it and/or modify it under the terms of the GNU General Public License as published by the Free Software Foundation, either version 3 of the License, or (at your option) any later version. This program is distributed in the hope that it will be useful, but WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details. You should have received a copy of the GNU General Public License along with this program. If not, see . */ /* GNU find was written by Eric Decker , with enhancements by David MacKenzie , Jay Plett , and Tim Wood . The idea for -print0 and xargs -0 came from Dan Bernstein . Improvements have been made by James Youngman . */ /* config.h must be included first. */ #include /* system headers. */ #include #include #include #include #include /* gnulib headers. */ #include "canonicalize.h" #include "closein.h" #include "dirent-safer.h" #include "dirname.h" #include "error.h" #include "fcntl--.h" #include "gettext.h" #include "human.h" #include "progname.h" #include "save-cwd.h" #include "xalloc.h" #include "xgetcwd.h" /* find headers. */ #include "buildcmd.h" #include "defs.h" #include "fdleak.h" #undef STAT_MOUNTPOINTS #ifdef CLOSEDIR_VOID /* Fake a return value. */ # define CLOSEDIR(d) (closedir (d), 0) #else # define CLOSEDIR(d) closedir (d) #endif enum { NOT_AN_INODE_NUMBER = 0 }; #ifdef D_INO_IN_DIRENT # define D_INO(dp) (dp)->d_ino #else /* Some systems don't have inodes, so fake them to avoid lots of ifdefs. */ # define D_INO(dp) NOT_AN_INODE_NUMBER #endif #if ENABLE_NLS # include # define _(Text) gettext (Text) #else # define _(Text) Text #define textdomain(Domain) #define bindtextdomain(Package, Directory) #define ngettext(singular,plural,n) ((1==n) ? singular : plural) #endif #ifdef gettext_noop # define N_(String) gettext_noop (String) #else /* See locate.c for explanation as to why not use (String) */ # define N_(String) String #endif #ifdef STAT_MOUNTPOINTS static void init_mounted_dev_list (int mandatory); #endif static void process_top_path (char *pathname, mode_t mode, ino_t inum); static int process_path (char *pathname, char *name, bool leaf, char *parent, mode_t type, ino_t inum); static void process_dir (char *pathname, char *name, int pathlen, const struct stat *statp, char *parent); /* A file descriptor open to the initial working directory. Doing it this way allows us to work when the i.w.d. has unreadable parents. */ extern int starting_desc; /* The stat buffer of the initial working directory. */ static struct stat starting_stat_buf; enum ChdirSymlinkHandling { SymlinkHandleDefault, /* Normally the right choice */ SymlinkFollowOk /* see comment in process_top_path() */ }; enum TraversalDirection { TraversingUp, TraversingDown }; enum WdSanityCheckFatality { FATAL_IF_SANITY_CHECK_FAILS, RETRY_IF_SANITY_CHECK_FAILS, NON_FATAL_IF_SANITY_CHECK_FAILS }; #if defined HAVE_STRUCT_DIRENT_D_TYPE /* Convert the value of struct dirent.d_type into a value for * struct stat.st_mode (at least the file type bits), or zero * if the type is DT_UNKNOWN or is a value we don't know about. */ static mode_t type_to_mode (unsigned type) { switch (type) { #ifdef DT_FIFO case DT_FIFO: return S_IFIFO; #endif #ifdef DT_CHR case DT_CHR: return S_IFCHR; #endif #ifdef DT_DIR case DT_DIR: return S_IFDIR; #endif #ifdef DT_BLK case DT_BLK: return S_IFBLK; #endif #ifdef DT_REG case DT_REG: return S_IFREG; #endif #ifdef DT_LNK case DT_LNK: return S_IFLNK; #endif #ifdef DT_SOCK case DT_SOCK: return S_IFSOCK; #endif default: return 0; /* Unknown. */ } } #endif int get_current_dirfd (void) { return AT_FDCWD; } /* CAUTION: this is the entry point for the oldfind executable, which is not the binary that * will actually get installed. See ftsfind.c. */ int main (int argc, char **argv) { int i; int end_of_leading_options = 0; /* First arg after any -H/-L etc. */ struct predicate *eval_tree; if (argv[0]) set_program_name (argv[0]); else set_program_name ("find"); state.exit_status = 0; if (fd_leak_check_is_enabled ()) { remember_non_cloexec_fds (); } record_initial_cwd (); state.already_issued_stat_error_msg = false; state.shared_files = sharefile_init ("w"); if (NULL == state.shared_files) { error (EXIT_FAILURE, errno, _("Failed to initialize shared-file hash table")); } /* Set the option defaults before we do the locale * initialisation as check_nofollow () needs to be executed in the * POSIX locale. */ set_option_defaults (&options); #ifdef HAVE_SETLOCALE setlocale (LC_ALL, ""); #endif bindtextdomain (PACKAGE, LOCALEDIR); textdomain (PACKAGE); if (atexit (close_stdin)) { error (EXIT_FAILURE, errno, _("The atexit library function failed")); } /* Check for -P, -H or -L options. */ end_of_leading_options = process_leading_options (argc, argv); if (options.debug_options & DebugStat) options.xstat = debug_stat; #ifdef DEBUG fprintf (stderr, "cur_day_start = %s", ctime (&options.cur_day_start)); #endif /* DEBUG */ /* state.cwd_dir_fd has to be initialized before we call build_expression_tree () * because command-line parsing may lead us to stat some files. */ state.cwd_dir_fd = AT_FDCWD; /* We are now processing the part of the "find" command line * after the -H/-L options (if any). */ eval_tree = build_expression_tree (argc, argv, end_of_leading_options); /* safely_chdir () needs to check that it has ended up in the right place. * To avoid bailing out when something gets automounted, it checks if * the target directory appears to have had a directory mounted on it as * we chdir ()ed. The problem with this is that in order to notice that * a file system was mounted, we would need to lstat () all the mount points. * That strategy loses if our machine is a client of a dead NFS server. * * Hence if safely_chdir () and wd_sanity_check () can manage without needing * to know the mounted device list, we do that. */ if (!options.open_nofollow_available) { #ifdef STAT_MOUNTPOINTS init_mounted_dev_list (0); #endif } set_stat_placeholders (&starting_stat_buf); if ((*options.xstat) (".", &starting_stat_buf) != 0) error (EXIT_FAILURE, errno, _("cannot stat current directory")); /* If no paths are given, default to ".". */ for (i = end_of_leading_options; i < argc && !looks_like_expression (argv[i], true); i++) { process_top_path (argv[i], 0, starting_stat_buf.st_ino); } /* If there were no path arguments, default to ".". */ if (i == end_of_leading_options) { /* * We use a temporary variable here because some actions modify * the path temporarily. Hence if we use a string constant, * we get a coredump. The best example of this is if we say * "find -printf %H" (note, not "find . -printf %H"). */ char defaultpath[2] = "."; process_top_path (defaultpath, 0, starting_stat_buf.st_ino); } /* If "-exec ... {} +" has been used, there may be some * partially-full command lines which have been built, * but which are not yet complete. Execute those now. */ show_success_rates (eval_tree); cleanup (); return state.exit_status; } bool is_fts_enabled (int *ftsoptions) { /* this version of find (i.e. this main ()) does not use fts. */ *ftsoptions = 0; return false; } static char * specific_dirname (const char *dir) { char dirbuf[1024]; if (0 == strcmp (".", dir)) { /* OK, what's '.'? */ if (NULL != getcwd (dirbuf, sizeof (dirbuf))) { return strdup (dirbuf); } else { return strdup (dir); } } else { char *result = canonicalize_filename_mode (dir, CAN_EXISTING); if (NULL == result) return strdup (dir); else return result; } } /* Return non-zero if FS is the name of a file system that is likely to * be automounted */ static int fs_likely_to_be_automounted (const char *fs) { return ( (0==strcmp (fs, "nfs")) || (0==strcmp (fs, "autofs")) || (0==strcmp (fs, "subfs"))); } #ifdef STAT_MOUNTPOINTS static dev_t *mounted_devices = NULL; static size_t num_mounted_devices = 0u; static void init_mounted_dev_list (int mandatory) { assert (NULL == mounted_devices); assert (0 == num_mounted_devices); mounted_devices = get_mounted_devices (&num_mounted_devices); if (mandatory && (NULL == mounted_devices)) { error (EXIT_FAILURE, 0, _("Cannot read list of mounted devices.")); } } static void refresh_mounted_dev_list (void) { if (mounted_devices) { free (mounted_devices); mounted_devices = 0; } num_mounted_devices = 0u; init_mounted_dev_list (1); } /* Search for device DEV in the array LIST, which is of size N. */ static int dev_present (dev_t dev, const dev_t *list, size_t n) { if (list) { while (n-- > 0u) { if ( (*list++) == dev ) return 1; } } return 0; } enum MountPointStateChange { MountPointRecentlyMounted, MountPointRecentlyUnmounted, MountPointStateUnchanged }; static enum MountPointStateChange get_mount_state (dev_t newdev) { int new_is_present, new_was_present; new_was_present = dev_present (newdev, mounted_devices, num_mounted_devices); refresh_mounted_dev_list (); new_is_present = dev_present (newdev, mounted_devices, num_mounted_devices); if (new_was_present == new_is_present) return MountPointStateUnchanged; else if (new_is_present) return MountPointRecentlyMounted; else return MountPointRecentlyUnmounted; } /* We stat()ed a directory, chdir()ed into it (we know this * since direction is TraversingDown), stat()ed it again, * and noticed that the device numbers are different. Check * if the file system was recently mounted. * * If it was, it looks like chdir()ing into the directory * caused a file system to be mounted. Maybe automount is * running. Anyway, that's probably OK - but it happens * only when we are moving downward. * * We also allow for the possibility that a similar thing * has happened with the unmounting of a file system. This * is much rarer, as it relies on an automounter timeout * occurring at exactly the wrong moment. */ static enum WdSanityCheckFatality dirchange_is_fatal (const char *specific_what, enum WdSanityCheckFatality isfatal, int silent, struct stat *newinfo) { enum MountPointStateChange transition = get_mount_state (newinfo->st_dev); switch (transition) { case MountPointRecentlyUnmounted: isfatal = NON_FATAL_IF_SANITY_CHECK_FAILS; if (!silent) { error (0, 0, _("WARNING: file system %s has recently been unmounted."), safely_quote_err_filename (0, specific_what)); } break; case MountPointRecentlyMounted: isfatal = NON_FATAL_IF_SANITY_CHECK_FAILS; if (!silent) { error (0, 0, _("WARNING: file system %s has recently been mounted."), safely_quote_err_filename (0, specific_what)); } break; case MountPointStateUnchanged: /* leave isfatal as it is */ break; } return isfatal; } #endif /* Examine the results of the stat() of a directory from before we * entered or left it, with the results of stat()ing it afterward. If * these are different, the file system tree has been modified while we * were traversing it. That might be an attempt to use a race * condition to persuade find to do something it didn't intend * (e.g. an attempt by an ordinary user to exploit the fact that root * sometimes runs find on the whole file system). However, this can * also happen if automount is running (certainly on Solaris). With * automount, moving into a directory can cause a file system to be * mounted there. * * To cope sensibly with this, we will raise an error if we see the * device number change unless we are chdir()ing into a subdirectory, * and the directory we moved into has been mounted or unmounted "recently". * Here "recently" means since we started "find" or we last re-read * the /etc/mnttab file. * * If the device number does not change but the inode does, that is a * problem. * * If the device number and inode are both the same, we are happy. * * If a file system is (un)mounted as we chdir() into the directory, that * may mean that we're now examining a section of the file system that might * have been excluded from consideration (via -prune or -quit for example). * Hence we print a warning message to indicate that the output of find * might be inconsistent due to the change in the file system. */ static bool wd_sanity_check (const char *thing_to_stat, const char *progname, const char *what, dev_t old_dev, ino_t old_ino, struct stat *newinfo, int parent, int line_no, enum TraversalDirection direction, enum WdSanityCheckFatality isfatal, bool *changed) /* output parameter */ { const char *fstype; char *specific_what = NULL; int silent = 0; const char *current_dir = "."; *changed = false; set_stat_placeholders (newinfo); if ((*options.xstat) (current_dir, newinfo) != 0) fatal_target_file_error (errno, thing_to_stat); if (old_dev != newinfo->st_dev) { *changed = true; specific_what = specific_dirname (what); fstype = filesystem_type (newinfo, current_dir); silent = fs_likely_to_be_automounted (fstype); /* This condition is rare, so once we are here it is * reasonable to perform an expensive computation to * determine if we should continue or fail. */ if (TraversingDown == direction) { #ifdef STAT_MOUNTPOINTS isfatal = dirchange_is_fatal (specific_what,isfatal,silent,newinfo); #else (void) silent; isfatal = RETRY_IF_SANITY_CHECK_FAILS; #endif } switch (isfatal) { case FATAL_IF_SANITY_CHECK_FAILS: { fstype = filesystem_type (newinfo, current_dir); error (EXIT_FAILURE, 0, _("%s%s changed during execution of %s (old device number %ld, new device number %ld, file system type is %s) [ref %ld]"), safely_quote_err_filename (0, specific_what), parent ? "/.." : "", safely_quote_err_filename (1, progname), (long) old_dev, (long) newinfo->st_dev, fstype, (long)line_no); /*NOTREACHED*/ return false; } case NON_FATAL_IF_SANITY_CHECK_FAILS: { /* Since the device has changed under us, the inode number * will almost certainly also be different. However, we have * already decided that this is not a problem. Hence we return * without checking the inode number. */ free (specific_what); return true; } case RETRY_IF_SANITY_CHECK_FAILS: return false; } } /* Device number was the same, check if the inode has changed. */ if (old_ino != newinfo->st_ino) { *changed = true; specific_what = specific_dirname (what); fstype = filesystem_type (newinfo, current_dir); error ((isfatal == FATAL_IF_SANITY_CHECK_FAILS) ? 1 : 0, 0, /* no relevant errno value */ _("%s%s changed during execution of %s " "(old inode number %" PRIuMAX ", new inode number %" PRIuMAX ", file system type is %s) [ref %ld]"), safely_quote_err_filename (0, specific_what), parent ? "/.." : "", safely_quote_err_filename (1, progname), (uintmax_t) old_ino, (uintmax_t) newinfo->st_ino, fstype, (long)line_no); free (specific_what); return false; } return true; } enum SafeChdirStatus { SafeChdirOK, SafeChdirFailSymlink, SafeChdirFailNotDir, SafeChdirFailStat, SafeChdirFailWouldBeUnableToReturn, SafeChdirFailChdirFailed, SafeChdirFailNonexistent, SafeChdirFailDestUnreadable }; /* Safely perform a change in directory. We do this by calling * lstat() on the subdirectory, using chdir() to move into it, and * then lstat()ing ".". We compare the results of the two stat calls * to see if they are consistent. If not, we sound the alarm. * * If following_links() is true, we do follow symbolic links. */ static enum SafeChdirStatus safely_chdir_lstat (const char *dest, enum TraversalDirection direction, struct stat *statbuf_dest, enum ChdirSymlinkHandling symlink_follow_option, bool *did_stat) { struct stat statbuf_arrived; int rv, dotfd=-1; int saved_errno; /* specific_dirname() changes errno. */ bool rv_set = false; bool statflag = false; int tries = 0; enum WdSanityCheckFatality isfatal = RETRY_IF_SANITY_CHECK_FAILS; saved_errno = errno = 0; dotfd = open_cloexec (".", O_RDONLY #if defined O_LARGEFILE |O_LARGEFILE #endif ); /* We jump back to here if wd_sanity_check() * recoverably triggers an alert. */ retry: ++tries; if (dotfd >= 0) { /* Stat the directory we're going to. */ set_stat_placeholders (statbuf_dest); if (0 == options.xstat (dest, statbuf_dest)) { statflag = true; #ifdef S_ISLNK /* symlink_follow_option might be set to SymlinkFollowOk, which * would allow us to chdir() into a symbolic link. This is * only useful for the case where the directory we're * chdir()ing into is the basename of a command line * argument, for example where "foo/bar/baz" is specified on * the command line. When -P is in effect (the default), * baz will not be followed if it is a symlink, but if bar * is a symlink, it _should_ be followed. Hence we need the * ability to override the policy set by following_links(). */ if (!following_links () && S_ISLNK(statbuf_dest->st_mode)) { /* We're not supposed to be following links, but this is * a link. Check symlink_follow_option to see if we should * make a special exception. */ if (symlink_follow_option == SymlinkFollowOk) { /* We need to re-stat() the file so that the * sanity check can pass. */ if (0 != stat (dest, statbuf_dest)) { rv = SafeChdirFailNonexistent; rv_set = true; saved_errno = errno; goto fail; } statflag = true; } else { /* Not following symlinks, so the attempt to * chdir() into a symlink should be prevented. */ rv = SafeChdirFailSymlink; rv_set = true; saved_errno = 0; /* silence the error message */ goto fail; } } #endif #ifdef S_ISDIR /* Although the immediately following chdir() would detect * the fact that this is not a directory for us, this would * result in an extra system call that fails. Anybody * examining the system-call trace should ideally not be * concerned that something is actually failing. */ if (!S_ISDIR(statbuf_dest->st_mode)) { rv = SafeChdirFailNotDir; rv_set = true; saved_errno = 0; /* silence the error message */ goto fail; } #endif if (options.debug_options & DebugSearch) fprintf (stderr, "safely_chdir(): chdir(\"%s\")\n", dest); if (0 == chdir (dest)) { /* check we ended up where we wanted to go */ bool changed = false; if (!wd_sanity_check (".", program_name, ".", statbuf_dest->st_dev, statbuf_dest->st_ino, &statbuf_arrived, 0, __LINE__, direction, isfatal, &changed)) { /* Only allow one failure. */ if (RETRY_IF_SANITY_CHECK_FAILS == isfatal) { if (0 == fchdir (dotfd)) { isfatal = FATAL_IF_SANITY_CHECK_FAILS; goto retry; } else { /* Failed to return to original directory, * but we know that the current working * directory is not the one that we intend * to be in. Since fchdir() failed, we * can't recover from this and so this error * is fatal. */ error (EXIT_FAILURE, errno, _("failed to return to parent directory")); } } else { /* XXX: not sure what to use as an excuse here. */ rv = SafeChdirFailNonexistent; rv_set = true; saved_errno = 0; goto fail; } } close (dotfd); return SafeChdirOK; } else { saved_errno = errno; if (ENOENT == saved_errno) { rv = SafeChdirFailNonexistent; rv_set = true; if (options.ignore_readdir_race) errno = 0; /* don't issue err msg */ } else if (ENOTDIR == saved_errno) { /* This can happen if the we stat a directory, * and then file system activity changes it into * a non-directory. */ saved_errno = 0; /* don't issue err msg */ rv = SafeChdirFailNotDir; rv_set = true; } else { rv = SafeChdirFailChdirFailed; rv_set = true; } goto fail; } } else { saved_errno = errno; rv = SafeChdirFailStat; rv_set = true; if ( (ENOENT == saved_errno) || (0 == state.curdepth)) saved_errno = 0; /* don't issue err msg */ goto fail; } } else { /* We do not have read permissions on "." */ rv = SafeChdirFailWouldBeUnableToReturn; rv_set = true; goto fail; } /* This is the success path, so we clear errno. The caller probably * won't be calling error() anyway. */ saved_errno = 0; /* We use the same exit path for success or failure. * which has occurred is recorded in RV. */ fail: /* We do not call error() as this would result in a duplicate error * message when the caller does the same thing. */ if (saved_errno) errno = saved_errno; if (dotfd >= 0) { close (dotfd); dotfd = -1; } *did_stat = statflag; assert (rv_set); return rv; } /* Safely change working directory to the specified subdirectory. If * we are not allowed to follow symbolic links, we use open() with * O_NOFOLLOW, followed by fchdir(). This ensures that we don't * follow symbolic links (of course, we do follow them if the -L * option is in effect). */ static enum SafeChdirStatus safely_chdir_nofollow (const char *dest, enum TraversalDirection direction, struct stat *statbuf_dest, enum ChdirSymlinkHandling symlink_follow_option, bool *did_stat) { int extraflags, fd; (void) direction; (void) statbuf_dest; extraflags = 0; *did_stat = false; switch (symlink_follow_option) { case SymlinkFollowOk: extraflags = 0; break; case SymlinkHandleDefault: if (following_links ()) extraflags = 0; else extraflags = O_NOFOLLOW; /* ... which may still be 0. */ break; } errno = 0; fd = open (dest, O_RDONLY #if defined O_LARGEFILE |O_LARGEFILE #endif #if defined O_CLOEXEC |O_CLOEXEC #endif |extraflags); if (fd < 0) { switch (errno) { case ELOOP: return SafeChdirFailSymlink; /* This is why we use O_NOFOLLOW */ case ENOENT: return SafeChdirFailNonexistent; default: return SafeChdirFailDestUnreadable; } } errno = 0; if (0 == fchdir (fd)) { close (fd); return SafeChdirOK; } else { int saved_errno = errno; close (fd); errno = saved_errno; switch (errno) { case ENOTDIR: return SafeChdirFailNotDir; case EACCES: case EBADF: /* Shouldn't happen */ case EINTR: case EIO: default: return SafeChdirFailChdirFailed; } } } static enum SafeChdirStatus safely_chdir (const char *dest, enum TraversalDirection direction, struct stat *statbuf_dest, enum ChdirSymlinkHandling symlink_follow_option, bool *did_stat) { enum SafeChdirStatus result; /* We're about to leave a directory. If there are any -execdir * argument lists which have been built but have not yet been * processed, do them now because they must be done in the same * directory. */ complete_pending_execdirs (); /* gnulib defines O_NOFOLLOW to 0 if the OS doesn't have it. */ options.open_nofollow_available = !!O_NOFOLLOW; if (options.open_nofollow_available) { result = safely_chdir_nofollow (dest, direction, statbuf_dest, symlink_follow_option, did_stat); if (SafeChdirFailDestUnreadable != result) { return result; } else { /* Savannah bug #15384: fall through to use safely_chdir_lstat * if the directory is not readable. */ /* Do nothing. */ } } /* Even if O_NOFOLLOW is available, we may need to use the alternative * method, since parent of the start point may be executable but not * readable. */ return safely_chdir_lstat (dest, direction, statbuf_dest, symlink_follow_option, did_stat); } /* Safely go back to the starting directory. */ static void chdir_back (void) { if (options.debug_options & DebugSearch) fprintf (stderr, "chdir_back(): chdir to start point\n"); restore_cwd (initial_wd); } /* Move to the parent of a given directory and then call a function, * restoring the cwd. Don't bother changing directory if the * specified directory is a child of "." or is the root directory. */ static void at_top (char *pathname, mode_t mode, ino_t inum, struct stat *pstat, void (*action)(char *pathname, char *basename, int mode, ino_t inum, struct stat *pstat)) { int dirchange; char *parent_dir = dir_name (pathname); char *base = last_component (pathname); state.curdepth = 0; state.starting_path_length = strlen (pathname); if (0 == *base || 0 == strcmp (parent_dir, ".")) { dirchange = 0; base = pathname; } else { enum TraversalDirection direction; enum SafeChdirStatus chdir_status; struct stat st; bool did_stat = false; dirchange = 1; if (0 == strcmp (base, "..")) direction = TraversingUp; else direction = TraversingDown; /* We pass SymlinkFollowOk to safely_chdir(), which allows it to * chdir() into a symbolic link. This is only useful for the * case where the directory we're chdir()ing into is the * basename of a command line argument, for example where * "foo/bar/baz" is specified on the command line. When -P is * in effect (the default), baz will not be followed if it is a * symlink, but if bar is a symlink, it _should_ be followed. * Hence we need the ability to override the policy set by * following_links(). */ chdir_status = safely_chdir (parent_dir, direction, &st, SymlinkFollowOk, &did_stat); if (SafeChdirOK != chdir_status) { const char *what = (SafeChdirFailWouldBeUnableToReturn == chdir_status) ? "." : parent_dir; if (errno) error (0, errno, "%s", safely_quote_err_filename (0, what)); else error (0, 0, _("Failed to safely change directory into %s"), safely_quote_err_filename (0, parent_dir)); /* We can't process this command-line argument. */ state.exit_status = 1; return; } } free (parent_dir); parent_dir = NULL; action (pathname, base, mode, inum, pstat); if (dirchange) { chdir_back (); } } static void do_process_top_dir (char *pathname, char *base, int mode, ino_t inum, struct stat *pstat) { (void) pstat; process_path (pathname, base, false, ".", mode, inum); complete_pending_execdirs (); } static void do_process_predicate (char *pathname, char *base, int mode, ino_t inum, struct stat *pstat) { (void) mode; (void) inum; state.rel_pathname = base; /* cwd_dir_fd was already set by safely_chdir */ apply_predicate (pathname, pstat, get_eval_tree ()); } /* Descend PATHNAME, which is a command-line argument. Actions like -execdir assume that we are in the parent directory of the file we're examining, and on entry to this function our working directory is whatever it was when find was invoked. Therefore If PATHNAME is "." we just leave things as they are. Otherwise, we figure out what the parent directory is, and move to that. */ static void process_top_path (char *pathname, mode_t mode, ino_t inum) { at_top (pathname, mode, inum, NULL, do_process_top_dir); } /* Info on each directory in the current tree branch, to avoid getting stuck in symbolic link loops. */ static struct dir_id *dir_ids = NULL; /* Entries allocated in `dir_ids'. */ static int dir_alloc = 0; /* Index in `dir_ids' of directory currently being searched. This is always the last valid entry. */ static int dir_curr = -1; /* (Arbitrary) number of entries to grow `dir_ids' by. */ #define DIR_ALLOC_STEP 32 /* We've detected a file system loop. This is caused by one of * two things: * * 1. Option -L is in effect and we've hit a symbolic link that * points to an ancestor. This is harmless. We won't traverse the * symbolic link. * * 2. We have hit a real cycle in the directory hierarchy. In this * case, we issue a diagnostic message (POSIX requires this) and we * skip that directory entry. */ static void issue_loop_warning (const char *name, const char *pathname, int level) { struct stat stbuf_link; if (lstat (name, &stbuf_link) != 0) stbuf_link.st_mode = S_IFREG; if (S_ISLNK(stbuf_link.st_mode)) { error (0, 0, _("Symbolic link %s is part of a loop in the directory hierarchy; we have already visited the directory to which it points."), safely_quote_err_filename (0, pathname)); /* XXX: POSIX appears to require that the exit status be non-zero if a * diagnostic is issued. */ } else { int distance = 1 + (dir_curr-level); /* We have found an infinite loop. POSIX requires us to * issue a diagnostic. Usually we won't get to here * because when the leaf optimisation is on, it will cause * the subdirectory to be skipped. If /a/b/c/d is a hard * link to /a/b, then the link count of /a/b/c is 2, * because the ".." entry of /b/b/c/d points to /a, not * to /a/b/c. */ error (0, 0, ngettext ( "Filesystem loop detected; %s has the same device number and inode as " "a directory which is %d level higher in the file system hierarchy", "Filesystem loop detected; %s has the same device number and inode as " "a directory which is %d levels higher in the file system hierarchy", (long)distance), safely_quote_err_filename (0, pathname), distance); } } /* Recursively descend path PATHNAME, applying the predicates. LEAF is true if PATHNAME is known to be in a directory that has no more unexamined subdirectories, and therefore it is not a directory. Knowing this allows us to avoid calling stat as long as possible for leaf files. NAME is PATHNAME relative to the current directory. We access NAME but print PATHNAME. PARENT is the path of the parent of NAME, relative to find's starting directory. Return nonzero iff PATHNAME is a directory. */ static int process_path (char *pathname, char *name, bool leaf, char *parent, mode_t mode, ino_t inum) { struct stat stat_buf; static dev_t root_dev; /* Device ID of current argument pathname. */ int i; struct predicate *eval_tree; eval_tree = get_eval_tree (); /* Assume it is a non-directory initially. */ stat_buf.st_mode = 0; /* The caller usually knows the inode number, either from readdir or * a *stat call. We use that value (the caller passes 0 to indicate * ignorance of the inode number). */ stat_buf.st_ino = inum; state.rel_pathname = name; state.type = 0; state.have_stat = false; state.have_type = false; state.already_issued_stat_error_msg = false; if (!digest_mode (&mode, pathname, name, &stat_buf, leaf)) return 0; if (!S_ISDIR (state.type)) { if (state.curdepth >= options.mindepth) apply_predicate (pathname, &stat_buf, eval_tree); return 0; } /* From here on, we're working on a directory. */ /* Now we really need to stat the directory, even if we know the * type, because we need information like struct stat.st_rdev. */ if (get_statinfo (pathname, name, &stat_buf) != 0) return 0; state.have_stat = true; mode = state.type = stat_buf.st_mode; /* use full info now that we have it. */ state.stop_at_current_level = options.maxdepth >= 0 && state.curdepth >= options.maxdepth; /* If we've already seen this directory on this branch, don't descend it again. */ for (i = 0; i <= dir_curr; i++) if (stat_buf.st_ino == dir_ids[i].ino && stat_buf.st_dev == dir_ids[i].dev) { state.stop_at_current_level = true; issue_loop_warning (name, pathname, i); } if (dir_alloc <= ++dir_curr) { dir_alloc += DIR_ALLOC_STEP; dir_ids = (struct dir_id *) xrealloc ((char *) dir_ids, dir_alloc * sizeof (struct dir_id)); } dir_ids[dir_curr].ino = stat_buf.st_ino; dir_ids[dir_curr].dev = stat_buf.st_dev; if (options.stay_on_filesystem) { if (state.curdepth == 0) root_dev = stat_buf.st_dev; else if (stat_buf.st_dev != root_dev) state.stop_at_current_level = true; } if (options.do_dir_first && state.curdepth >= options.mindepth) apply_predicate (pathname, &stat_buf, eval_tree); if (options.debug_options & DebugSearch) fprintf (stderr, "pathname = %s, stop_at_current_level = %d\n", pathname, state.stop_at_current_level); if (state.stop_at_current_level == false) { /* Scan directory on disk. */ process_dir (pathname, name, strlen (pathname), &stat_buf, parent); } if (options.do_dir_first == false && state.curdepth >= options.mindepth) { /* The fields in 'state' are now out of date. Correct them. */ if (!digest_mode (&mode, pathname, name, &stat_buf, leaf)) return 0; if (0 == dir_curr) { at_top (pathname, mode, stat_buf.st_ino, &stat_buf, do_process_predicate); } else { do_process_predicate (pathname, name, mode, stat_buf.st_ino, &stat_buf); } } dir_curr--; return 1; } /* Scan directory PATHNAME and recurse through process_path for each entry. PATHLEN is the length of PATHNAME. NAME is PATHNAME relative to the current directory. STATP is the results of *options.xstat on it. PARENT is the path of the parent of NAME, relative to find's starting directory. */ static void process_dir (char *pathname, char *name, int pathlen, const struct stat *statp, char *parent) { int subdirs_left; /* Number of unexamined subdirs in PATHNAME. */ bool subdirs_unreliable; /* if true, cannot use dir link count as subdir limif (if false, it may STILL be unreliable) */ struct stat stat_buf; size_t dircount = 0u; DIR *dirp; if (statp->st_nlink < 2) { subdirs_unreliable = true; subdirs_left = 0; } else { subdirs_unreliable = false; /* not necessarily right */ subdirs_left = statp->st_nlink - 2; /* Account for name and ".". */ } errno = 0; dirp = opendir_safer (name); if (dirp == NULL) { assert (errno != 0); error (0, errno, "%s", safely_quote_err_filename (0, pathname)); state.exit_status = 1; } else { char *cur_path; /* Full path of each file to process. */ char *cur_name; /* Base name of each file to process. */ unsigned cur_path_size; /* Bytes allocated for `cur_path'. */ register unsigned file_len; /* Length of each path to process. */ register unsigned pathname_len; /* PATHLEN plus trailing '/'. */ bool did_stat = false; if (pathname[pathlen - 1] == '/') pathname_len = pathlen + 1; /* For '\0'; already have '/'. */ else pathname_len = pathlen + 2; /* For '/' and '\0'. */ cur_path_size = 0; cur_path = NULL; /* We're about to leave the directory. If there are any * -execdir argument lists which have been built but have not * yet been processed, do them now because they must be done in * the same directory. */ complete_pending_execdirs (); if (strcmp (name, ".")) { enum SafeChdirStatus status = safely_chdir (name, TraversingDown, &stat_buf, SymlinkHandleDefault, &did_stat); switch (status) { case SafeChdirOK: /* If there had been a change but wd_sanity_check() * accepted it, we need to accept that on the * way back up as well, so modify our record * of what we think we should see later. * If there was no change, the assignments are a no-op. * * However, before performing the assignment, we need to * check that we have the stat information. If O_NOFOLLOW * is available, safely_chdir() will not have needed to use * stat(), and so stat_buf will just contain random data. */ if (!did_stat) { /* If there is a link we need to follow it. Hence * the direct call to stat() not through (options.xstat) */ set_stat_placeholders (&stat_buf); if (0 != stat (".", &stat_buf)) break; /* skip the assignment. */ } dir_ids[dir_curr].dev = stat_buf.st_dev; dir_ids[dir_curr].ino = stat_buf.st_ino; break; case SafeChdirFailWouldBeUnableToReturn: error (0, errno, "."); state.exit_status = 1; break; case SafeChdirFailNonexistent: case SafeChdirFailDestUnreadable: case SafeChdirFailStat: case SafeChdirFailNotDir: case SafeChdirFailChdirFailed: error (0, errno, "%s", safely_quote_err_filename (0, pathname)); state.exit_status = 1; return; case SafeChdirFailSymlink: error (0, 0, _("warning: not following the symbolic link %s"), safely_quote_err_filename (0, pathname)); state.exit_status = 1; return; } } while (1) { const char *namep; mode_t mode = 0; const struct dirent *dp; /* We reset errno here to distinguish between end-of-directory and an error */ errno = 0; dp = readdir (dirp); if (NULL == dp) { if (errno) { /* an error occurred, but we are not yet at the end of the directory stream. */ error (0, errno, "%s", safely_quote_err_filename (0, pathname)); continue; } else { break; /* End of the directory stream. */ } } else { namep = dp->d_name; /* Skip "", ".", and "..". "" is returned by at least one buggy implementation: Solaris 2.4 readdir on NFS file systems. */ if (!namep[0] || (namep[0] == '.' && (namep[1] == 0 || (namep[1] == '.' && namep[2] == 0)))) continue; } #if defined HAVE_STRUCT_DIRENT_D_TYPE if (dp->d_type != DT_UNKNOWN) mode = type_to_mode (dp->d_type); #endif /* Append this directory entry's name to the path being searched. */ file_len = pathname_len + strlen (namep); if (file_len > cur_path_size) { while (file_len > cur_path_size) cur_path_size += 1024; free (cur_path); cur_path = xmalloc (cur_path_size); strcpy (cur_path, pathname); cur_path[pathname_len - 2] = '/'; } cur_name = cur_path + pathname_len - 1; strcpy (cur_name, namep); state.curdepth++; if (!options.no_leaf_check && !subdirs_unreliable) { if (mode && S_ISDIR(mode) && (subdirs_left == 0)) { /* This is a subdirectory, but the number of directories we * have found now exceeds the number we would expect given * the hard link count on the parent. This is likely to be * a bug in the file system driver (e.g. Linux's * /proc file system) or may just be a fact that the OS * doesn't really handle hard links with Unix semantics. * In the latter case, -noleaf should be used routinely. */ error (0, 0, _("WARNING: Hard link count is wrong for %s (saw only st_nlink=%" PRIuMAX " but we already saw %" PRIuMAX " subdirectories): this may be a bug in your file system driver. Automatically turning on find's -noleaf option. Earlier results may have failed to include directories that should have been searched."), safely_quote_err_filename(0, pathname), (uintmax_t) statp->st_nlink, (uintmax_t) dircount); state.exit_status = 1; /* We know the result is wrong, now */ options.no_leaf_check = true; /* Don't make same mistake again */ subdirs_unreliable = 1; subdirs_left = 1; /* band-aid for this iteration. */ } /* Normal case optimization. On normal Unix file systems, a directory that has no subdirectories has two links: its name, and ".". Any additional links are to the ".." entries of its subdirectories. Once we have processed as many subdirectories as there are additional links, we know that the rest of the entries are non-directories -- in other words, leaf files. */ { int count; count = process_path (cur_path, cur_name, subdirs_left == 0, pathname, mode, D_INO(dp)); subdirs_left -= count; dircount += count; } } else { /* There might be weird (e.g., CD-ROM or MS-DOS) file systems mounted, which don't have Unix-like directory link counts. */ process_path (cur_path, cur_name, false, pathname, mode, D_INO(dp)); } state.curdepth--; } /* We're about to leave the directory. If there are any * -execdir argument lists which have been built but have not * yet been processed, do them now because they must be done in * the same directory. */ complete_pending_execdirs (); if (strcmp (name, ".")) { enum SafeChdirStatus status; /* We could go back and do the next command-line arg instead, maybe using longjmp. */ char const *dir; bool deref = following_links () ? true : false; if ( (state.curdepth>0) && !deref) dir = ".."; else { chdir_back (); dir = parent; } did_stat = false; status = safely_chdir (dir, TraversingUp, &stat_buf, SymlinkHandleDefault, &did_stat); switch (status) { case SafeChdirOK: break; case SafeChdirFailWouldBeUnableToReturn: error (EXIT_FAILURE, errno, "."); return; case SafeChdirFailNonexistent: case SafeChdirFailDestUnreadable: case SafeChdirFailStat: case SafeChdirFailSymlink: case SafeChdirFailNotDir: case SafeChdirFailChdirFailed: error (EXIT_FAILURE, errno, "%s", safely_quote_err_filename (0, pathname)); return; } } free (cur_path); CLOSEDIR (dirp); } if (subdirs_unreliable) { /* Make sure we hasn't used the variable subdirs_left if we knew * we shouldn't do so. */ assert (0 == subdirs_left || options.no_leaf_check); } }