X-Git-Url: https://mattmccutchen.net/rsync/rsync.git/blobdiff_plain/79452d46930e5e3ec5b14d4c8380ff1e0329b9a8..fb4c98c2c8a618855ebe4d11c36f8a120fab2aa1:/util.c diff --git a/util.c b/util.c index 2d1989d4..9b54c05a 100644 --- a/util.c +++ b/util.c @@ -1,6 +1,8 @@ -/* - Copyright (C) Andrew Tridgell 1996 - Copyright (C) Paul Mackerras 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 @@ -17,111 +19,183 @@ Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA. */ -/* - Utilities used in rsync +/** + * + * @file util.c + * + * Utilities used in rsync + **/ - tridge, June 1996 - */ #include "rsync.h" extern int verbose; -/**************************************************************************** -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) + +/** + Set a fd into nonblocking mode +**/ +void set_nonblocking(int fd) +{ + int val; + + if((val = fcntl(fd, F_GETFL, 0)) == -1) + return; + if (!(val & NONBLOCK_FLAG)) { + val |= NONBLOCK_FLAG; + fcntl(fd, F_SETFL, val); + } +} + +/** +Set a fd into blocking mode +*/ +void set_blocking(int fd) { int val; -#ifdef O_NONBLOCK -#define FLAG_TO_SET O_NONBLOCK + + 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 -#ifdef SYSV -#define FLAG_TO_SET O_NDELAY -#else /* BSD */ -#define FLAG_TO_SET FNDELAY -#endif + ret = pipe(fd); #endif + + if (ret == 0) { + set_nonblocking(fd[0]); + set_nonblocking(fd[1]); + } - 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); - } + return ret; +} - 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]; +static 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); + } + } + rprintf(FINFO, "\n"); +} - set_nonblocking(*f_in); - set_nonblocking(*f_out); - - return pid; + +/** + * Create a child connected to use on stdin/stdout. + * + * This is derived from CVS code + * + * Note that in the child STDIN is set to blocking and STDOUT + * is set to non-blocking. This is necessary as rsh relies on stdin being blocking + * and ssh relies on stdout being non-blocking + * + * If blocking_io is set then use blocking io on both fds. That can be + * used to cope with badly broken rsh implementations like the one on + * Solaris. + **/ +pid_t piped_child(char **command, int *f_in, int *f_out) +{ + pid_t pid; + int to_child_pipe[2]; + int from_child_pipe[2]; + extern int blocking_io; + + if (verbose >= 2) { + print_child_argv(command); + } + + if (fd_pair(to_child_pipe) < 0 || fd_pair(from_child_pipe) < 0) { + rprintf(FERROR, "pipe: %s\n", strerror(errno)); + exit_cleanup(RERR_IPC); + } + + + pid = do_fork(); + if (pid == -1) { + 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); + set_blocking(STDIN_FILENO); + if (blocking_io) { + set_blocking(STDOUT_FILENO); + } + 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]; + + return pid; } -int local_child(int argc, char **argv,int *f_in,int *f_out) +pid_t local_child(int argc, char **argv,int *f_in,int *f_out, + int (*child_main)(int, char **)) { - int pid; + pid_t pid; int to_child_pipe[2]; int from_child_pipe[2]; + extern int read_batch; /* dw */ - if (pipe(to_child_pipe) < 0 || - pipe(from_child_pipe) < 0) { + if (fd_pair(to_child_pipe) < 0 || + fd_pair(from_child_pipe) < 0) { rprintf(FERROR,"pipe: %s\n",strerror(errno)); exit_cleanup(RERR_IPC); } pid = do_fork(); - if (pid < 0) { + if (pid == -1) { rprintf(FERROR,"fork: %s\n",strerror(errno)); exit_cleanup(RERR_IPC); } @@ -130,7 +204,7 @@ int local_child(int argc, char **argv,int *f_in,int *f_out) extern int am_sender; extern int am_server; - am_sender = !am_sender; + am_sender = read_batch ? 0 : !am_sender; am_server = 1; if (dup2(to_child_pipe[0], STDIN_FILENO) < 0 || @@ -142,7 +216,7 @@ int local_child(int argc, char **argv,int *f_in,int *f_out) } 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); + child_main(argc, argv); } if (close(from_child_pipe[1]) < 0 || @@ -173,10 +247,18 @@ 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; @@ -200,11 +282,11 @@ 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 -****************************************************************************/ +/** + 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) { extern int orig_umask; @@ -224,12 +306,17 @@ 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; @@ -250,16 +337,23 @@ 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 @@ -274,7 +368,9 @@ 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; @@ -282,7 +378,7 @@ 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)); @@ -325,18 +421,21 @@ int copy_file(char *source, char *dest, mode_t mode) return 0; } -/* - 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. -*/ /* 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) { #ifndef ETXTBSY @@ -397,45 +496,54 @@ int robust_rename(char *from, char *to) return -1; return do_rename(from, to); #endif - } - - -/* sleep for a while via select */ -void u_sleep(int usec) -{ - struct timeval tv; - - tv.tv_sec = 0; - tv.tv_usec = usec; - select(0, NULL, NULL, NULL, &tv); } 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); @@ -634,15 +712,19 @@ 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 (except those allowed by "reldir") * 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 * is sometimes significant on command line arguments. + * * If "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. @@ -651,9 +733,9 @@ void clean_fname(char *name) * 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 */ - void sanitize_path(char *p, char *reldir) { char *start, *sanp; @@ -742,7 +824,7 @@ void sanitize_path(char *p, char *reldir) static char curr_dir[MAXPATHLEN]; -/* like chdir() but can be reversed with pop_dir() if save is set. It +/** 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 */ char *push_dir(char *dir, int save) { @@ -774,7 +856,7 @@ 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; @@ -792,7 +874,7 @@ 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) { @@ -806,32 +888,117 @@ int u_strcmp(const char *cs1, const char *cs2) return (int)*s1 - (int)*s2; } -static OFF_T last_ofs; +static OFF_T last_ofs; +static struct timeval print_time; +static struct timeval start_time; +static OFF_T start_ofs; + +static unsigned long msdiff(struct timeval *t1, struct timeval *t2) +{ + return (t2->tv_sec - t1->tv_sec) * 1000 + + (t2->tv_usec - t1->tv_usec) / 1000; +} + + +/** + * @param ofs Current position in file + * @param size Total size of file + * @param is_last True if this is the last time progress will be + * printed for this file, so we should output a newline. (Not + * necessarily the same as all bytes being received.) + **/ +static void rprint_progress(OFF_T ofs, OFF_T size, struct timeval *now, + int is_last) +{ + int pct = (ofs == size) ? 100 : (int)((100.0*ofs)/size); + unsigned long diff = msdiff(&start_time, now); + double rate = diff ? (double) (ofs-start_ofs) * 1000.0 / diff / 1024.0 : 0; + const char *units; + /* If we've finished transferring this file, show the time taken; + * otherwise show expected time to complete. That's kind of + * inconsistent, but people can probably cope. Hopefully we'll + * get more consistent and complete progress reporting soon. -- + * mbp */ + double remain = is_last + ? (double) diff / 1000.0 + : rate ? (double) (size-ofs) / rate / 1000.0 : 0.0; + int remain_h, remain_m, remain_s; + + if (rate > 1024*1024) { + rate /= 1024.0 * 1024.0; + units = "GB/s"; + } else if (rate > 1024) { + rate /= 1024.0; + units = "MB/s"; + } else { + units = "kB/s"; + } + + remain_s = (int) remain % 60; + remain_m = (int) (remain / 60.0) % 60; + remain_h = (int) (remain / 3600.0); + + rprintf(FINFO, "%12.0f %3d%% %7.2f%s %4d:%02d:%02d%s", + (double) ofs, pct, rate, units, + remain_h, remain_m, remain_s, + is_last ? "\n" : "\r"); +} -void end_progress(void) +void end_progress(OFF_T size) { extern int do_progress, am_server; if (do_progress && !am_server) { - rprintf(FINFO,"\n"); + struct timeval now; + gettimeofday(&now, NULL); + rprint_progress(size, size, &now, True); } - last_ofs = 0; + last_ofs = 0; + start_ofs = 0; + print_time.tv_sec = print_time.tv_usec = 0; + start_time.tv_sec = start_time.tv_usec = 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; - } + struct timeval now; + + gettimeofday(&now, NULL); + + if (!start_time.tv_sec && !start_time.tv_usec) { + start_time.tv_sec = now.tv_sec; + start_time.tv_usec = now.tv_usec; + start_ofs = ofs; + } + + if (do_progress + && !am_server + && ofs > last_ofs + 1000 + && msdiff(&print_time, &now) > 250) { + rprint_progress(ofs, size, &now, False); + last_ofs = ofs; + print_time.tv_sec = now.tv_sec; + print_time.tv_usec = now.tv_usec; } } -/* 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. + * + * @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 + **/ int unsafe_symlink(char *dest, char *src) { char *tok; @@ -879,9 +1046,9 @@ 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]; @@ -900,3 +1067,90 @@ 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