X-Git-Url: https://mattmccutchen.net/rsync/rsync.git/blobdiff_plain/5a788adec19043d4d93c4fa2713f9f633bfd35ab..32f761755e2cfee243a1513c9ea9871586463fa6:/util.c diff --git a/util.c b/util.c index e04d0d78..63af7dab 100644 --- a/util.c +++ b/util.c @@ -1,162 +1,114 @@ -/* - Copyright (C) Andrew Tridgell 1996 - Copyright (C) Paul Mackerras 1996 - - 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 2 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, write to the Free Software - Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA. -*/ - -/* - Utilities used in rsync - - tridge, June 1996 - */ +/* -*- c-file-style: "linux" -*- + * + * Copyright (C) 1996-2000 by Andrew Tridgell + * Copyright (C) Paul Mackerras 1996 + * Copyright (C) 2001, 2002 by Martin Pool + * + * 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 2 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, write to the Free Software + * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA. + */ + +/** + * @file + * + * Utilities used in rsync + **/ + #include "rsync.h" -/**************************************************************************** -Set a fd into nonblocking mode. Uses POSIX O_NONBLOCK if available, -else -if SYSV use O_NDELAY -if BSD use FNDELAY -****************************************************************************/ -int set_nonblocking(int fd) +extern int verbose; + +int sanitize_paths = 0; + + + +/** + * Set a fd into nonblocking mode + **/ +void set_nonblocking(int fd) { int val; -#ifdef O_NONBLOCK -#define FLAG_TO_SET O_NONBLOCK -#else -#ifdef SYSV -#define FLAG_TO_SET O_NDELAY -#else /* BSD */ -#define FLAG_TO_SET FNDELAY -#endif -#endif - - if((val = fcntl(fd, F_GETFL, 0)) == -1) - return -1; - val |= FLAG_TO_SET; - return fcntl( fd, F_SETFL, val); -#undef FLAG_TO_SET -} - - -/* this is taken from CVS */ -int piped_child(char **command,int *f_in,int *f_out) -{ - int pid; - int to_child_pipe[2]; - int from_child_pipe[2]; - - if (pipe(to_child_pipe) < 0 || - pipe(from_child_pipe) < 0) { - rprintf(FERROR,"pipe: %s\n",strerror(errno)); - exit_cleanup(RERR_IPC); - } - - - pid = do_fork(); - if (pid < 0) { - rprintf(FERROR,"fork: %s\n",strerror(errno)); - exit_cleanup(RERR_IPC); - } - - if (pid == 0) - { - extern int orig_umask; - if (dup2(to_child_pipe[0], STDIN_FILENO) < 0 || - close(to_child_pipe[1]) < 0 || - close(from_child_pipe[0]) < 0 || - dup2(from_child_pipe[1], STDOUT_FILENO) < 0) { - rprintf(FERROR,"Failed to dup/close : %s\n",strerror(errno)); - exit_cleanup(RERR_IPC); - } - if (to_child_pipe[0] != STDIN_FILENO) close(to_child_pipe[0]); - if (from_child_pipe[1] != STDOUT_FILENO) close(from_child_pipe[1]); - umask(orig_umask); - execvp(command[0], command); - rprintf(FERROR,"Failed to exec %s : %s\n", - command[0],strerror(errno)); - exit_cleanup(RERR_IPC); - } - - if (close(from_child_pipe[1]) < 0 || - close(to_child_pipe[0]) < 0) { - rprintf(FERROR,"Failed to close : %s\n",strerror(errno)); - exit_cleanup(RERR_IPC); - } - - *f_in = from_child_pipe[0]; - *f_out = to_child_pipe[1]; - - set_nonblocking(*f_in); - set_nonblocking(*f_out); - - return pid; + + if ((val = fcntl(fd, F_GETFL, 0)) == -1) + return; + if (!(val & NONBLOCK_FLAG)) { + val |= NONBLOCK_FLAG; + fcntl(fd, F_SETFL, val); + } } -int local_child(int argc, char **argv,int *f_in,int *f_out) +/** + * Set a fd into blocking mode + **/ +void set_blocking(int fd) { - int pid; - int to_child_pipe[2]; - int from_child_pipe[2]; + int val; - if (pipe(to_child_pipe) < 0 || - pipe(from_child_pipe) < 0) { - rprintf(FERROR,"pipe: %s\n",strerror(errno)); - exit_cleanup(RERR_IPC); + if ((val = fcntl(fd, F_GETFL, 0)) == -1) + return; + if (val & NONBLOCK_FLAG) { + val &= ~NONBLOCK_FLAG; + fcntl(fd, F_SETFL, val); } +} + +/** + * Create a file descriptor pair - like pipe() but use socketpair if + * possible (because of blocking issues on pipes). + * + * Always set non-blocking. + */ +int fd_pair(int fd[2]) +{ + int ret; + +#if HAVE_SOCKETPAIR + ret = socketpair(AF_UNIX, SOCK_STREAM, 0, fd); +#else + ret = pipe(fd); +#endif - pid = do_fork(); - if (pid < 0) { - rprintf(FERROR,"fork: %s\n",strerror(errno)); - exit_cleanup(RERR_IPC); + if (ret == 0) { + set_nonblocking(fd[0]); + set_nonblocking(fd[1]); } - if (pid == 0) { - extern int am_sender; - extern int am_server; + return ret; +} - am_sender = !am_sender; - am_server = 1; - if (dup2(to_child_pipe[0], STDIN_FILENO) < 0 || - close(to_child_pipe[1]) < 0 || - close(from_child_pipe[0]) < 0 || - dup2(from_child_pipe[1], STDOUT_FILENO) < 0) { - rprintf(FERROR,"Failed to dup/close : %s\n",strerror(errno)); - exit_cleanup(RERR_IPC); +void print_child_argv(char **cmd) +{ + rprintf(FINFO, "opening connection using "); + for (; *cmd; cmd++) { + /* Look for characters that ought to be quoted. This + * is not a great quoting algorithm, but it's + * sufficient for a log message. */ + if (strspn(*cmd, "abcdefghijklmnopqrstuvwxyz" + "ABCDEFGHIJKLMNOPQRSTUVWXYZ" + "0123456789" + ",.-_=+@/") != strlen(*cmd)) { + rprintf(FINFO, "\"%s\" ", *cmd); + } else { + rprintf(FINFO, "%s ", *cmd); } - if (to_child_pipe[0] != STDIN_FILENO) close(to_child_pipe[0]); - if (from_child_pipe[1] != STDOUT_FILENO) close(from_child_pipe[1]); - start_server(STDIN_FILENO, STDOUT_FILENO, argc, argv); } - - if (close(from_child_pipe[1]) < 0 || - close(to_child_pipe[0]) < 0) { - rprintf(FERROR,"Failed to close : %s\n",strerror(errno)); - exit_cleanup(RERR_IPC); - } - - *f_in = from_child_pipe[0]; - *f_out = to_child_pipe[1]; - - return pid; + rprintf(FINFO, "\n"); } - void out_of_memory(char *str) { rprintf(FERROR,"ERROR: out of memory in %s\n",str); @@ -171,10 +123,18 @@ void overflow(char *str) -int set_modtime(char *fname,time_t modtime) +int set_modtime(char *fname, time_t modtime) { extern int dry_run; - if (dry_run) return 0; + if (dry_run) + return 0; + + if (verbose > 2) { + rprintf(FINFO, "set modtime of %s to (%ld) %s", + fname, (long) modtime, + asctime(localtime(&modtime))); + } + { #ifdef HAVE_UTIMBUF struct utimbuf tbuf; @@ -198,14 +158,13 @@ int set_modtime(char *fname,time_t modtime) } -/**************************************************************************** -create any necessary directories in fname. Unfortunately we don't know -what perms to give the directory when this is called so we need to rely -on the umask -****************************************************************************/ -int create_directory_path(char *fname) +/** + Create any necessary directories in fname. Unfortunately we don't know + what perms to give the directory when this is called so we need to rely + on the umask +**/ +int create_directory_path(char *fname, int base_umask) { - extern int orig_umask; char *p; while (*fname == '/') fname++; @@ -214,7 +173,7 @@ int create_directory_path(char *fname) p = fname; while ((p=strchr(p,'/'))) { *p = 0; - do_mkdir(fname,0777 & ~orig_umask); + do_mkdir(fname, 0777 & ~base_umask); *p = '/'; p++; } @@ -222,12 +181,17 @@ int create_directory_path(char *fname) } -/* Write LEN bytes at PTR to descriptor DESC, retrying if interrupted. - Return LEN upon success, write's (negative) error code otherwise. - - derived from GNU C's cccp.c. -*/ -static int full_write(int desc, char *ptr, int len) +/** + * Write @p len bytes at @p ptr to descriptor @p desc, retrying if + * interrupted. + * + * @retval len upon success + * + * @retval <0 write's (negative) error code + * + * Derived from GNU C's cccp.c. + */ +static int full_write(int desc, char *ptr, size_t len) { int total_written; @@ -248,16 +212,23 @@ static int full_write(int desc, char *ptr, int len) return total_written; } -/* Read LEN bytes at PTR from descriptor DESC, retrying if interrupted. - Return the actual number of bytes read, zero for EOF, or negative - for an error. - derived from GNU C's cccp.c. */ -static int safe_read(int desc, char *ptr, int len) +/** + * Read @p len bytes at @p ptr from descriptor @p desc, retrying if + * interrupted. + * + * @retval >0 the actual number of bytes read + * + * @retval 0 for EOF + * + * @retval <0 for an error. + * + * Derived from GNU C's cccp.c. */ +static int safe_read(int desc, char *ptr, size_t len) { int n_chars; - if (len <= 0) + if (len == 0) return len; #ifdef EINTR @@ -272,7 +243,9 @@ static int safe_read(int desc, char *ptr, int len) } -/* copy a file - this is used in conjunction with the --temp-dir option */ +/** Copy a file. + * + * This is used in conjunction with the --temp-dir option */ int copy_file(char *source, char *dest, mode_t mode) { int ifd; @@ -280,14 +253,14 @@ int copy_file(char *source, char *dest, mode_t mode) char buf[1024 * 8]; int len; /* Number of bytes read into `buf'. */ - ifd = open(source, O_RDONLY); + ifd = do_open(source, O_RDONLY, 0); if (ifd == -1) { rprintf(FERROR,"open %s: %s\n", source,strerror(errno)); return -1; } - if (do_unlink(dest) && errno != ENOENT) { + if (robust_unlink(dest) && errno != ENOENT) { rprintf(FERROR,"unlink %s: %s\n", dest,strerror(errno)); return -1; @@ -323,42 +296,128 @@ int copy_file(char *source, char *dest, mode_t mode) return 0; } -/* sleep for a while via select */ -void u_sleep(int usec) +/* MAX_RENAMES should be 10**MAX_RENAMES_DIGITS */ +#define MAX_RENAMES_DIGITS 3 +#define MAX_RENAMES 1000 + +/** + * Robust unlink: some OS'es (HPUX) refuse to unlink busy files, so + * rename to /.rsyncNNN instead. + * + * Note that successive rsync runs will shuffle the filenames around a + * bit as long as the file is still busy; this is because this function + * does not know if the unlink call is due to a new file coming in, or + * --delete trying to remove old .rsyncNNN files, hence it renames it + * each time. + **/ +int robust_unlink(char *fname) { - struct timeval tv; +#ifndef ETXTBSY + return do_unlink(fname); +#else + static int counter = 1; + int rc, pos, start; + char path[MAXPATHLEN]; + + rc = do_unlink(fname); + if ((rc == 0) || (errno != ETXTBSY)) + return rc; + + strlcpy(path, fname, MAXPATHLEN); - tv.tv_sec = 0; - tv.tv_usec = usec; - select(0, NULL, NULL, NULL, &tv); + pos = strlen(path); + while((path[--pos] != '/') && (pos >= 0)) + ; + ++pos; + strlcpy(&path[pos], ".rsync", MAXPATHLEN-pos); + pos += sizeof(".rsync")-1; + + if (pos > (MAXPATHLEN-MAX_RENAMES_DIGITS-1)) { + errno = ETXTBSY; + return -1; + } + + /* start where the last one left off to reduce chance of clashes */ + start = counter; + do { + sprintf(&path[pos], "%03d", counter); + if (++counter >= MAX_RENAMES) + counter = 1; + } while (((rc = access(path, 0)) == 0) && (counter != start)); + + if (verbose > 0) + rprintf(FINFO,"renaming %s to %s because of text busy\n", + fname, path); + + /* maybe we should return rename()'s exit status? Nah. */ + if (do_rename(fname, path) != 0) { + errno = ETXTBSY; + return -1; + } + return 0; +#endif +} + +int robust_rename(char *from, char *to) +{ +#ifndef ETXTBSY + return do_rename(from, to); +#else + int rc = do_rename(from, to); + if ((rc == 0) || (errno != ETXTBSY)) + return rc; + if (robust_unlink(to) != 0) + return -1; + return do_rename(from, to); +#endif } static pid_t all_pids[10]; static int num_pids; -/* fork and record the pid of the child */ +/** Fork and record the pid of the child. **/ pid_t do_fork(void) { pid_t newpid = fork(); - if (newpid) { + if (newpid != 0 && newpid != -1) { all_pids[num_pids++] = newpid; } return newpid; } -/* kill all children */ +/** + * Kill all children. + * + * @todo It would be kind of nice to make sure that they are actually + * all our children before we kill them, because their pids may have + * been recycled by some other process. Perhaps when we wait for a + * child, we should remove it from this array. Alternatively we could + * perhaps use process groups, but I think that would not work on + * ancient Unix versions that don't support them. + **/ void kill_all(int sig) { int i; - for (i=0;i= n || ret < 0) { - str[n-1] = 0; - return -1; - } - str[ret] = 0; - return ret; -} - - -/* like snprintf but always null terminates */ -int slprintf(char *str, int n, char *format, ...) -{ - va_list ap; - int ret; - - va_start(ap, format); - ret = vslprintf(str,n,format,ap); - va_end(ap); - return ret; -} - - void *Realloc(void *p, int size) { if (!p) return (void *)malloc(size); @@ -553,31 +587,55 @@ void clean_fname(char *name) } } -/* +/** * Make path appear as if a chroot had occurred: - * 1. remove leading "/" (or replace with "." if at end) - * 2. remove leading ".." components - * 3. delete any other "/.." (recursively) + * + * @li 1. remove leading "/" (or replace with "." if at end) + * + * @li 2. remove leading ".." components (except those allowed by @p reldir) + * + * @li 3. delete any other "/.." (recursively) + * + * Can only shrink paths, so sanitizes in place. + * * While we're at it, remove double slashes and "." components like - * clean_fname does(), but DON'T remove a trailing slash because that + * clean_fname() does, but DON'T remove a trailing slash because that * is sometimes significant on command line arguments. - * Return a malloc'ed copy. + * + * If @p reldir is non-null, it is a sanitized directory that the path will be + * relative to, so allow as many ".." at the beginning of the path as + * there are components in reldir. This is used for symbolic link targets. + * If reldir is non-null and the path began with "/", to be completely like + * a chroot we should add in depth levels of ".." at the beginning of the + * path, but that would blow the assumption that the path doesn't grow and + * it is not likely to end up being a valid symlink anyway, so just do + * the normal removal of the leading "/" instead. + * * Contributed by Dave Dykstra */ - -char *sanitize_path(char *p) +void sanitize_path(char *p, char *reldir) { - char *copy, *copyp; + char *start, *sanp; + int depth = 0; + int allowdotdot = 0; - copy = (char *) malloc(strlen(p)+1); - copyp = copy; + if (reldir) { + depth++; + while (*reldir) { + if (*reldir++ == '/') { + depth++; + } + } + } + start = p; + sanp = p; while (*p == '/') { /* remove leading slashes */ p++; } while (*p != '\0') { /* this loop iterates once per filename component in p. - * both p (and copyp if the original had a slash) should + * both p (and sanp if the original had a slash) should * always be left pointing after a slash */ if ((*p == '.') && ((*(p+1) == '/') || (*(p+1) == '\0'))) { @@ -586,47 +644,67 @@ char *sanitize_path(char *p) /* skip following slashes */ ; } - } else if ((*p == '.') && (*(p+1) == '.') && + continue; + } + allowdotdot = 0; + if ((*p == '.') && (*(p+1) == '.') && ((*(p+2) == '/') || (*(p+2) == '\0'))) { - /* skip ".." component followed by slash or end */ - p += 2; - if (*p == '/') - p++; - if (copyp != copy) { - /* back up the copy one level */ - --copyp; /* now pointing at slash */ - while ((copyp > copy) && (*(copyp - 1) != '/')) { - /* skip back up to slash */ - copyp--; + /* ".." component followed by slash or end */ + if ((depth > 0) && (sanp == start)) { + /* allow depth levels of .. at the beginning */ + --depth; + allowdotdot = 1; + } else { + p += 2; + if (*p == '/') + p++; + if (sanp != start) { + /* back up sanp one level */ + --sanp; /* now pointing at slash */ + while ((sanp > start) && (*(sanp - 1) != '/')) { + /* skip back up to slash */ + sanp--; + } } + continue; } - } else { - while (1) { - /* copy one component through next slash */ - *copyp++ = *p++; - if ((*p == '\0') || (*(p-1) == '/')) { - while (*p == '/') { - /* skip multiple slashes */ - p++; - } - break; + } + while (1) { + /* copy one component through next slash */ + *sanp++ = *p++; + if ((*p == '\0') || (*(p-1) == '/')) { + while (*p == '/') { + /* skip multiple slashes */ + p++; } + break; } } + if (allowdotdot) { + /* move the virtual beginning to leave the .. alone */ + start = sanp; + } } - if (copyp == copy) { + if ((sanp == start) && !allowdotdot) { /* ended up with nothing, so put in "." component */ - *copyp++ = '.'; + /* + * note that the !allowdotdot doesn't prevent this from + * happening in all allowed ".." situations, but I didn't + * think it was worth putting in an extra variable to ensure + * it since an extra "." won't hurt in those situations. + */ + *sanp++ = '.'; } - *copyp = '\0'; - return(copy); + *sanp = '\0'; } static char curr_dir[MAXPATHLEN]; -/* like chdir() but can be reversed with pop_dir() if save is set. It - is also much faster as it remembers where we have been */ +/** + * Like chdir() but can be reversed with pop_dir() if @p save is set. + * It is also much faster as it remembers where we have been. + **/ char *push_dir(char *dir, int save) { char *ret = curr_dir; @@ -637,6 +715,8 @@ char *push_dir(char *dir, int save) getcwd(curr_dir, sizeof(curr_dir)-1); } + if (!dir) return NULL; /* this call was probably just to initialize */ + if (chdir(dir)) return NULL; if (save) { @@ -655,7 +735,7 @@ char *push_dir(char *dir, int save) return ret; } -/* reverse a push_dir call */ +/** Reverse a push_dir() call */ int pop_dir(char *dir) { int ret; @@ -673,7 +753,7 @@ int pop_dir(char *dir) return 0; } -/* we need to supply our own strcmp function for file list comparisons +/** We need to supply our own strcmp function for file list comparisons to ensure that signed/unsigned usage is consistent between machines. */ int u_strcmp(const char *cs1, const char *cs2) { @@ -687,32 +767,32 @@ int u_strcmp(const char *cs1, const char *cs2) return (int)*s1 - (int)*s2; } -static OFF_T last_ofs; -void end_progress(void) -{ - extern int do_progress, am_server; - - if (do_progress && !am_server) { - rprintf(FINFO,"\n"); - } - last_ofs = 0; -} -void show_progress(OFF_T ofs, OFF_T size) -{ - extern int do_progress, am_server; - - if (do_progress && !am_server) { - if (ofs > last_ofs + 1000) { - int pct = (int)((100.0*ofs)/size); - rprintf(FINFO,"%.0f (%d%%)\r", (double)ofs, pct); - last_ofs = ofs; - } - } -} - -/* determine if a symlink points outside the current directory tree */ +/** + * Determine if a symlink points outside the current directory tree. + * This is considered "unsafe" because e.g. when mirroring somebody + * else's machine it might allow them to establish a symlink to + * /etc/passwd, and then read it through a web server. + * + * Null symlinks and absolute symlinks are always unsafe. + * + * Basically here we are concerned with symlinks whose target contains + * "..", because this might cause us to walk back up out of the + * transferred directory. We are not allowed to go back up and + * reenter. + * + * @param dest Target of the symlink in question. + * + * @param src Top source directory currently applicable. Basically this + * is the first parameter to rsync in a simple invocation, but it's + * modified by flist.c in slightly complex ways. + * + * @retval True if unsafe + * @retval False is unsafe + * + * @sa t_unsafe.c + **/ int unsafe_symlink(char *dest, char *src) { char *tok; @@ -760,9 +840,9 @@ int unsafe_symlink(char *dest, char *src) } -/**************************************************************************** - return the date and time as a string -****************************************************************************/ +/** + * Return the date and time as a string + **/ char *timestring(time_t t) { static char TimeBuf[200]; @@ -781,3 +861,90 @@ char *timestring(time_t t) return(TimeBuf); } + +/** + * Sleep for a specified number of milliseconds. + * + * Always returns TRUE. (In the future it might return FALSE if + * interrupted.) + **/ +int msleep(int t) +{ + int tdiff=0; + struct timeval tval,t1,t2; + + gettimeofday(&t1, NULL); + gettimeofday(&t2, NULL); + + while (tdiff < t) { + tval.tv_sec = (t-tdiff)/1000; + tval.tv_usec = 1000*((t-tdiff)%1000); + + errno = 0; + select(0,NULL,NULL, NULL, &tval); + + gettimeofday(&t2, NULL); + tdiff = (t2.tv_sec - t1.tv_sec)*1000 + + (t2.tv_usec - t1.tv_usec)/1000; + } + + return True; +} + + +/** + * Determine if two file modification times are equivalent (either + * exact or in the modification timestamp window established by + * --modify-window). + * + * @retval 0 if the times should be treated as the same + * + * @retval +1 if the first is later + * + * @retval -1 if the 2nd is later + **/ +int cmp_modtime(time_t file1, time_t file2) +{ + extern int modify_window; + + if (file2 > file1) { + if (file2 - file1 <= modify_window) return 0; + return -1; + } + if (file1 - file2 <= modify_window) return 0; + return 1; +} + + +#ifdef __INSURE__XX +#include + +/** + This routine is a trick to immediately catch errors when debugging + with insure. A xterm with a gdb is popped up when insure catches + a error. It is Linux specific. +**/ +int _Insure_trap_error(int a1, int a2, int a3, int a4, int a5, int a6) +{ + static int (*fn)(); + int ret; + char *cmd; + + asprintf(&cmd, "/usr/X11R6/bin/xterm -display :0 -T Panic -n Panic -e /bin/sh -c 'cat /tmp/ierrs.*.%d ; gdb /proc/%d/exe %d'", + getpid(), getpid(), getpid()); + + if (!fn) { + static void *h; + h = dlopen("/usr/local/parasoft/insure++lite/lib.linux2/libinsure.so", RTLD_LAZY); + fn = dlsym(h, "_Insure_trap_error"); + } + + ret = fn(a1, a2, a3, a4, a5, a6); + + system(cmd); + + free(cmd); + + return ret; +} +#endif