X-Git-Url: https://mattmccutchen.net/rsync/rsync.git/blobdiff_plain/ebdd24d6d08bc4ffd632a64b3781f48cc8a46eb0..3ed8eafccb0f8880be295e67486347e5ffd93320:/util.c diff --git a/util.c b/util.c index 96b7ac4f..d37ab61e 100644 --- a/util.c +++ b/util.c @@ -31,8 +31,10 @@ extern int verbose; extern int dry_run; extern int module_id; extern int modify_window; +extern int relative_paths; +extern int human_readable; extern char *partial_dir; -extern struct exclude_list_struct server_exclude_list; +extern struct filter_list_struct server_filter_list; int sanitize_paths = 0; @@ -79,7 +81,7 @@ int fd_pair(int fd[2]) { int ret; -#if HAVE_SOCKETPAIR +#ifdef HAVE_SOCKETPAIR ret = socketpair(AF_UNIX, SOCK_STREAM, 0, fd); #else ret = pipe(fd); @@ -120,7 +122,7 @@ void out_of_memory(char *str) exit_cleanup(RERR_MALLOC); } -void overflow(char *str) +void overflow_exit(char *str) { rprintf(FERROR, "ERROR: buffer overflow in %s\n", str); exit_cleanup(RERR_MALLOC); @@ -128,10 +130,12 @@ void overflow(char *str) -int set_modtime(char *fname, time_t modtime) +int set_modtime(char *fname, time_t modtime, mode_t mode) { - if (dry_run) - return 0; +#if !defined HAVE_LUTIMES || !defined HAVE_UTIMES + if (S_ISLNK(mode)) + return 1; +#endif if (verbose > 2) { rprintf(FINFO, "set modtime of %s to (%ld) %s", @@ -139,24 +143,33 @@ int set_modtime(char *fname, time_t modtime) asctime(localtime(&modtime))); } + if (dry_run) + return 0; + { -#ifdef HAVE_UTIMBUF +#ifdef HAVE_UTIMES + struct timeval t[2]; + t[0].tv_sec = time(NULL); + t[0].tv_usec = 0; + t[1].tv_sec = modtime; + t[1].tv_usec = 0; +# ifdef HAVE_LUTIMES + if (S_ISLNK(mode)) + return lutimes(fname, t); +# endif + return utimes(fname, t); +#elif defined HAVE_UTIMBUF struct utimbuf tbuf; tbuf.actime = time(NULL); tbuf.modtime = modtime; return utime(fname,&tbuf); -#elif defined(HAVE_UTIME) +#elif defined HAVE_UTIME time_t t[2]; t[0] = time(NULL); t[1] = modtime; return utime(fname,t); #else - struct timeval t[2]; - t[0].tv_sec = time(NULL); - t[0].tv_usec = 0; - t[1].tv_sec = modtime; - t[1].tv_usec = 0; - return utimes(fname,t); +#error No file-time-modification routine found! #endif } } @@ -197,7 +210,7 @@ int create_directory_path(char *fname, int base_umask) * * Derived from GNU C's cccp.c. */ -static int full_write(int desc, char *ptr, size_t len) +int full_write(int desc, char *ptr, size_t len) { int total_written; @@ -245,8 +258,9 @@ static int safe_read(int desc, char *ptr, size_t len) /** Copy a file. * - * This is used in conjunction with the --temp-dir option */ -int copy_file(char *source, char *dest, mode_t mode) + * This is used in conjunction with the --temp-dir, --backup, and + * --copy-dest options. */ +int copy_file(const char *source, const char *dest, mode_t mode) { int ifd; int ofd; @@ -315,7 +329,7 @@ int copy_file(char *source, char *dest, mode_t mode) * --delete trying to remove old .rsyncNNN files, hence it renames it * each time. **/ -int robust_unlink(char *fname) +int robust_unlink(const char *fname) { #ifndef ETXTBSY return do_unlink(fname); @@ -362,9 +376,12 @@ int robust_unlink(char *fname) #endif } -/* Returns 0 on success, -1 on most errors, and -2 if we got an error - * trying to copy the file across file systems. */ -int robust_rename(char *from, char *to, int mode) +/* Returns 0 on successful rename, 1 if we successfully copied the file + * across filesystems, -2 if copy_file() failed, and -1 on other errors. + * If partialptr is not NULL and we need to do a copy, copy the file into + * the active partial-dir instead of over the destination file. */ +int robust_rename(char *from, char *to, char *partialptr, + int mode) { int tries = 4; @@ -380,10 +397,15 @@ int robust_rename(char *from, char *to, int mode) break; #endif case EXDEV: + if (partialptr) { + if (!handle_partial_dir(partialptr,PDIR_CREATE)) + return -1; + to = partialptr; + } if (copy_file(from, to, mode) != 0) return -2; do_unlink(from); - return 0; + return 1; default: return -1; } @@ -479,14 +501,14 @@ int lock_range(int fd, int offset, int len) return fcntl(fd,F_SETLK,&lock) == 0; } -static int exclude_server_path(char *arg) +static int filter_server_path(char *arg) { char *s; - if (server_exclude_list.head) { + if (server_filter_list.head) { for (s = arg; (s = strchr(s, '/')) != NULL; ) { *s = '\0'; - if (check_exclude(&server_exclude_list, arg, 1) < 0) { + if (check_filter(&server_filter_list, arg, 1) < 0) { /* We must leave arg truncated! */ return 1; } @@ -502,7 +524,7 @@ static void glob_expand_one(char *s, char ***argv_ptr, int *argc_ptr, char **argv = *argv_ptr; int argc = *argc_ptr; int maxargs = *maxargs_ptr; -#if !(defined(HAVE_GLOB) && defined(HAVE_GLOB_H)) +#if !defined HAVE_GLOB || !defined HAVE_GLOB_H if (argc == maxargs) { maxargs += MAX_ARGS; if (!(argv = realloc_array(argv, char *, maxargs))) @@ -513,10 +535,9 @@ static void glob_expand_one(char *s, char ***argv_ptr, int *argc_ptr, if (!*s) s = "."; s = argv[argc++] = strdup(s); - exclude_server_path(s); + filter_server_path(s); #else glob_t globbuf; - int i; if (maxargs <= argc) return; @@ -524,12 +545,12 @@ static void glob_expand_one(char *s, char ***argv_ptr, int *argc_ptr, s = "."; if (sanitize_paths) - s = sanitize_path(NULL, s, NULL); + s = sanitize_path(NULL, s, "", 0); else s = strdup(s); memset(&globbuf, 0, sizeof globbuf); - if (!exclude_server_path(s)) + if (!filter_server_path(s)) glob(s, 0, NULL, &globbuf); if (MAX((int)globbuf.gl_pathc, 1) > maxargs - argc) { maxargs += globbuf.gl_pathc + MAX_ARGS; @@ -541,9 +562,9 @@ static void glob_expand_one(char *s, char ***argv_ptr, int *argc_ptr, if (globbuf.gl_pathc == 0) argv[argc++] = s; else { - int j = globbuf.gl_pathc; + int i; free(s); - for (i = 0; i < j; i++) { + for (i = 0; i < (int)globbuf.gl_pathc; i++) { if (!(argv[argc++] = strdup(globbuf.gl_pathv[i]))) out_of_memory("glob_expand_one"); } @@ -650,13 +671,31 @@ size_t stringjoin(char *dest, size_t destsize, ...) return ret; } -void clean_fname(char *name) +int count_dir_elements(const char *p) +{ + int cnt = 0, new_component = 1; + while (*p) { + if (*p++ == '/') + new_component = 1; + else if (new_component) { + new_component = 0; + cnt++; + } + } + return cnt; +} + +/* Turns multiple adjacent slashes into a single slash, gets rid of "./" + * elements (but not a trailing dot dir), removes a trailing slash, and + * optionally collapses ".." elements (except for those at the start of the + * string). If the resulting name would be empty, change it into a ".". */ +unsigned int clean_fname(char *name, BOOL collapse_dot_dot) { - char *limit = name, *t = name, *f = name; + char *limit = name - 1, *t = name, *f = name; int anchored; if (!name) - return; + return 0; if ((anchored = *f == '/') != 0) *t++ = *f++; @@ -669,25 +708,24 @@ void clean_fname(char *name) if (*f == '.') { /* discard "." dirs (but NOT a trailing '.'!) */ if (f[1] == '/') { - f++; /* not += 2! */ + f += 2; continue; } /* collapse ".." dirs */ - if (f[1] == '.' && (f[2] == '/' || !f[2])) { + if (collapse_dot_dot + && f[1] == '.' && (f[2] == '/' || !f[2])) { char *s = t - 1; if (s == name && anchored) { f += 2; continue; } while (s > limit && *--s != '/') {} - if (s != t - 1 && *s == '/') { + if (s != t - 1 && (s < name || *s == '/')) { t = s + 1; f += 2; continue; } - *t++ = *f++; - *t++ = *f++; - limit = t; + limit = t + 2; } } while (*f && (*t++ = *f++) != '/') {} @@ -698,43 +736,44 @@ void clean_fname(char *name) if (t == name) *t++ = '.'; *t = '\0'; + + return t - name; } /* Make path appear as if a chroot had occurred. This handles a leading * "/" (either removing it or expanding it) and any leading or embedded * ".." components that attempt to escape past the module's top dir. * - * If dest is NULL, a buffer is allocated to hold the result. If dest is - * the same buffer as p (the path) OR if reldir is NULL, a leading slash - * is dropped instead of being expanded to be the module's top dir. + * If dest is NULL, a buffer is allocated to hold the result. It is legal + * to call with the dest and the path (p) pointing to the same buffer, but + * rootdir will be ignored to avoid expansion of the string. * - * If reldir is non-NULL (and non-empty), it is a sanitized directory that - * the path will be relative to, so allow as many '..'s 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. + * The rootdir string contains a value to use in place of a leading slash. + * Specify NULL to get the default of lp_path(module_id). * - * 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 depth is >= 0, it is a count of how many '..'s to allow at the start + * of the path. Use -1 to allow unlimited depth. * - * If the resulting path would be empty, change it into ".". - */ -char *sanitize_path(char *dest, const char *p, const char *reldir) + * We also clean the path in a manner similar to clean_fname() but with a + * few differences: + * + * Turns multiple adjacent slashes into a single slash, gets rid of "." dir + * elements (INCLUDING a trailing dot dir), PRESERVES a trailing slash, and + * ALWAYS collapses ".." elements (except for those at the start of the + * string up to "depth" deep). If the resulting name would be empty, + * change it into a ".". */ +char *sanitize_path(char *dest, const char *p, const char *rootdir, int depth) { char *start, *sanp; - int depth = 0; - int allowdotdot = 0; - int rlen = 0; + int rlen = 0, leave_one_dotdir = relative_paths; if (dest != p) { int plen = strlen(p); - if (*p == '/' && reldir) { - rlen = strlen(lp_path(module_id)); - reldir = NULL; + if (*p == '/') { + if (!rootdir) + rootdir = lp_path(module_id); + rlen = strlen(rootdir); + depth = 0; p++; } if (dest) { @@ -743,53 +782,36 @@ char *sanitize_path(char *dest, const char *p, const char *reldir) } else if (!(dest = new_array(char, rlen + plen + 1))) out_of_memory("sanitize_path"); if (rlen) { - memcpy(dest, lp_path(module_id), rlen); + memcpy(dest, rootdir, rlen); if (rlen > 1) dest[rlen++] = '/'; } } - if (reldir) { - int new_component = 1; - while (*reldir) { - if (*reldir++ == '/') - new_component = 1; - else if (new_component) { - new_component = 0; - depth++; - } - } - } - start = sanp = dest + rlen; - while (*p == '/') { - /* remove leading slashes */ - p++; - } while (*p != '\0') { + /* discard leading or extra slashes */ + if (*p == '/') { + p++; + continue; + } /* this loop iterates once per filename component in p. * 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')) { - /* skip "." component */ - while (*++p == '/') { - /* skip following slashes */ - ; + if (leave_one_dotdir && p[1]) + leave_one_dotdir = 0; + else { + /* skip "." component */ + p++; + continue; } - continue; } - allowdotdot = 0; if (*p == '.' && p[1] == '.' && (p[2] == '/' || p[2] == '\0')) { /* ".." component followed by slash or end */ - if (depth > 0 && sanp == start) { - /* allow depth levels of .. at the beginning */ - --depth; - allowdotdot = 1; - } else { + if (depth <= 0 || sanp != start) { p += 2; - if (*p == '/') - p++; if (sanp != start) { /* back up sanp one level */ --sanp; /* now pointing at slash */ @@ -800,22 +822,13 @@ char *sanitize_path(char *dest, const char *p, const char *reldir) } continue; } - } - while (1) { - /* copy one component through next slash */ - *sanp++ = *p++; - if (*p == '\0' || p[-1] == '/') { - while (*p == '/') { - /* skip multiple slashes */ - p++; - } - break; - } - } - if (allowdotdot) { + /* allow depth levels of .. at the beginning */ + depth--; /* move the virtual beginning to leave the .. alone */ - start = sanp; + start = sanp + 3; } + /* copy one component through next slash */ + while (*p && (*sanp++ = *p++) != '/') {} } if (sanp == dest) { /* ended up with nothing, so put in "." component */ @@ -867,7 +880,7 @@ int push_dir(char *dir) curr_dir_len += len; } - clean_fname(curr_dir); + curr_dir_len = clean_fname(curr_dir, 1); return 1; } @@ -888,30 +901,6 @@ int pop_dir(char *dir) return 1; } -/** - * Return the filename, turning any newlines into '?'s. This ensures that - * outputting it on a line of its own cannot generate an empty line. This - * function can handle only 2 names at a time! - **/ -const char *safe_fname(const char *fname) -{ - static char fbuf1[MAXPATHLEN], fbuf2[MAXPATHLEN]; - static char *fbuf = fbuf2; - char *nl = strchr(fname, '\n'); - - if (!nl) - return fname; - - fbuf = fbuf == fbuf1 ? fbuf2 : fbuf1; - strlcpy(fbuf, fname, MAXPATHLEN); - nl = fbuf + (nl - (char *)fname); - do { - *nl = '?'; - } while ((nl = strchr(nl+1, '\n')) != NULL); - - return fbuf; -} - /** * Return a quoted string with the full pathname of the indicated filename. * The string " (in MODNAME)" may also be appended. The returned pointer @@ -926,30 +915,25 @@ char *full_fname(const char *fn) if (result) free(result); - fn = safe_fname(fn); if (*fn == '/') p1 = p2 = ""; else { p1 = curr_dir; - p2 = "/"; + for (p2 = p1; *p2 == '/'; p2++) {} + if (*p2) + p2 = "/"; } if (module_id >= 0) { m1 = " (in "; m2 = lp_name(module_id); m3 = ")"; - if (*p1) { + if (p1 == curr_dir) { if (!lp_use_chroot(module_id)) { char *p = lp_path(module_id); if (*p != '/' || p[1]) p1 += strlen(p); } - if (!*p1) - p2++; - else - p1++; } - else - fn++; } else m1 = m2 = m3 = ""; @@ -978,9 +962,17 @@ char *partial_dir_fname(const char *fname) fn = fname; if ((int)pathjoin(t, sz, partial_dir, fn) >= sz) return NULL; - if (server_exclude_list.head - && check_exclude(&server_exclude_list, partial_fname, 0) < 0) - return NULL; + if (server_filter_list.head) { + static int len; + if (!len) + len = strlen(partial_dir); + t[len] = '\0'; + if (check_filter(&server_filter_list, partial_fname, 1) < 0) + return NULL; + t[len] = '/'; + if (check_filter(&server_filter_list, partial_fname, 0) < 0) + return NULL; + } return partial_fname; } @@ -1002,11 +994,7 @@ int handle_partial_dir(const char *fname, int create) dir = partial_fname; if (create) { STRUCT_STAT st; -#if SUPPORT_LINKS int statret = do_lstat(dir, &st); -#else - int statret = do_stat(dir, &st); -#endif if (statret == 0 && !S_ISDIR(st.st_mode)) { if (do_unlink(dir) < 0) return 0; @@ -1021,22 +1009,6 @@ int handle_partial_dir(const char *fname, int create) return 1; } -/** 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) -{ - const uchar *s1 = (const uchar *)cs1; - const uchar *s2 = (const uchar *)cs2; - - while (*s1 && *s2 && (*s1 == *s2)) { - s1++; s2++; - } - - return (int)*s1 - (int)*s2; -} - - - /** * Determine if a symlink points outside the current directory tree. * This is considered "unsafe" because e.g. when mirroring somebody @@ -1101,6 +1073,63 @@ int unsafe_symlink(const char *dest, const char *src) return (depth < 0); } +/* Return the int64 number as a string. If the --human-readable option was + * specified, we may output the number in K, M, or G units. We can return + * up to 4 buffers at a time. */ +char *human_num(int64 num) +{ + static char bufs[4][128]; /* more than enough room */ + static unsigned int n; + char *s; + + n = (n + 1) % (sizeof bufs / sizeof bufs[0]); + + if (human_readable) { + char units = '\0'; + int mult = human_readable == 1 ? 1024 : 1000; + double dnum = 0; + if (num > mult*mult*mult) { + dnum = (double)num / (mult*mult*mult); + units = 'G'; + } else if (num > mult*mult) { + dnum = (double)num / (mult*mult); + units = 'M'; + } else if (num > mult) { + dnum = (double)num / mult; + units = 'K'; + } + if (units) { + sprintf(bufs[n], "%.2f%c", dnum, units); + return bufs[n]; + } + } + + s = bufs[n] + sizeof bufs[0] - 1; + *s = '\0'; + + if (!num) + *--s = '0'; + while (num) { + *--s = (num % 10) + '0'; + num /= 10; + } + return s; +} + +/* Return the double number as a string. If the --human-readable option was + * specified, we may output the number in K, M, or G units. We use a buffer + * from human_num() to return our result. */ +char *human_dnum(double dnum, int decimal_digits) +{ + char *buf = human_num(dnum); + int len = strlen(buf); + if (isdigit(*(uchar*)(buf+len-1))) { + /* There's extra room in buf prior to the start of the num. */ + buf -= decimal_digits + 1; + snprintf(buf, len + decimal_digits + 2, "%.*f", decimal_digits, dnum); + } + return buf; +} /** * Return the date and time as a string @@ -1229,3 +1258,110 @@ void *_realloc_array(void *ptr, unsigned int size, unsigned long num) return malloc(size * num); return realloc(ptr, size * num); } + +/* Take a filename and filename length and return the most significant + * filename suffix we can find. This ignores suffixes such as "~", + * ".bak", ".orig", ".~1~", etc. */ +const char *find_filename_suffix(const char *fn, int fn_len, int *len_ptr) +{ + const char *suf, *s; + BOOL had_tilde; + int s_len; + + /* One or more dots at the start aren't a suffix. */ + while (fn_len && *fn == '.') fn++, fn_len--; + + /* Ignore the ~ in a "foo~" filename. */ + if (fn_len > 1 && fn[fn_len-1] == '~') + fn_len--, had_tilde = True; + else + had_tilde = False; + + /* Assume we don't find an suffix. */ + suf = ""; + *len_ptr = 0; + + /* Find the last significant suffix. */ + for (s = fn + fn_len; fn_len > 1; ) { + while (*--s != '.' && s != fn) {} + if (s == fn) + break; + s_len = fn_len - (s - fn); + fn_len = s - fn; + if (s_len == 4) { + if (strcmp(s+1, "bak") == 0 + || strcmp(s+1, "old") == 0) + continue; + } else if (s_len == 5) { + if (strcmp(s+1, "orig") == 0) + continue; + } else if (s_len > 2 && had_tilde + && s[1] == '~' && isdigit(*(uchar*)(s+2))) + continue; + *len_ptr = s_len; + suf = s; + if (s_len == 1) + break; + /* Determine if the suffix is all digits. */ + for (s++, s_len--; s_len > 0; s++, s_len--) { + if (!isdigit(*(uchar*)s)) + return suf; + } + /* An all-digit suffix may not be that signficant. */ + s = suf; + } + + return suf; +} + +/* This is an implementation of the Levenshtein distance algorithm. It + * was implemented to avoid needing a two-dimensional matrix (to save + * memory). It was also tweaked to try to factor in the ASCII distance + * between changed characters as a minor distance quantity. The normal + * Levenshtein units of distance (each signifying a single change between + * the two strings) are defined as a "UNIT". */ + +#define UNIT (1 << 16) + +uint32 fuzzy_distance(const char *s1, int len1, const char *s2, int len2) +{ + uint32 a[MAXPATHLEN], diag, above, left, diag_inc, above_inc, left_inc; + int32 cost; + int i1, i2; + + if (!len1 || !len2) { + if (!len1) { + s1 = s2; + len1 = len2; + } + for (i1 = 0, cost = 0; i1 < len1; i1++) + cost += s1[i1]; + return (int32)len1 * UNIT + cost; + } + + for (i2 = 0; i2 < len2; i2++) + a[i2] = (i2+1) * UNIT; + + for (i1 = 0; i1 < len1; i1++) { + diag = i1 * UNIT; + above = (i1+1) * UNIT; + for (i2 = 0; i2 < len2; i2++) { + left = a[i2]; + if ((cost = *((uchar*)s1+i1) - *((uchar*)s2+i2)) != 0) { + if (cost < 0) + cost = UNIT - cost; + else + cost = UNIT + cost; + } + diag_inc = diag + cost; + left_inc = left + UNIT + *((uchar*)s1+i1); + above_inc = above + UNIT + *((uchar*)s2+i2); + a[i2] = above = left < above + ? (left_inc < diag_inc ? left_inc : diag_inc) + : (above_inc < diag_inc ? above_inc : diag_inc); + diag = left; + } + } + + return a[len2-1]; +}