X-Git-Url: https://mattmccutchen.net/rsync/rsync.git/blobdiff_plain/951e826b75c4a4e6bc066e248d7489fb6eba6fde..refs/heads/wip/perdir-merge-2.1:/exclude.c diff --git a/exclude.c b/exclude.c index 080aa492..f46b3e75 100644 --- a/exclude.c +++ b/exclude.c @@ -36,13 +36,13 @@ extern int sanitize_paths; extern int protocol_version; extern int module_id; -extern char curr_dir[]; +extern char curr_dir[MAXPATHLEN]; extern unsigned int curr_dir_len; extern unsigned int module_dirlen; -struct filter_list_struct filter_list = { 0, 0, "" }; -struct filter_list_struct cvs_filter_list = { 0, 0, " [global CVS]" }; -struct filter_list_struct daemon_filter_list = { 0, 0, " [daemon]" }; +struct filter_list_struct filter_list = { .debug_type = "" }; +struct filter_list_struct cvs_filter_list = { .debug_type = " [global CVS]" }; +struct filter_list_struct daemon_filter_list = { .debug_type = " [daemon]" }; /* Need room enough for ":MODS " prefix plus some room to grow. */ #define MAX_RULE_PREFIX (16) @@ -102,17 +102,63 @@ static int mergelist_size = 0; * values (so we can pop back to them later) and set the tail to NULL. */ -static void free_filter(struct filter_struct *ex) +static void teardown_mergelist(struct filter_struct *ex) { - if (ex->match_flags & MATCHFLG_PERDIR_MERGE) { - free(ex->u.mergelist->debug_type); - free(ex->u.mergelist); - mergelist_cnt--; + if (DEBUG_GTE(FILTER, 2)) { + rprintf(FINFO, "[%s] deactivating mergelist #%d%s\n", + who_am_i(), mergelist_cnt - 1, + ex->u.mergelist->debug_type); } + + /* We should deactivate mergelists in LIFO order. */ + assert(mergelist_cnt > 0); + assert(ex == mergelist_parents[mergelist_cnt - 1]); + + /* The parent_dirscan filters should have been freed. */ + assert(ex->u.mergelist->parent_dirscan_head == NULL); + + free(ex->u.mergelist->debug_type); + free(ex->u.mergelist); + mergelist_cnt--; +} + +static void free_filter(struct filter_struct *ex) +{ + if (ex->match_flags & MATCHFLG_PERDIR_MERGE) + teardown_mergelist(ex); free(ex->pattern); free(ex); } +static void free_filters(struct filter_struct *head) +{ + struct filter_struct *rev_head = NULL; + + /* Reverse the list so we deactivate mergelists in the proper LIFO + * order. */ + while (head) { + struct filter_struct *next = head->next; + head->next = rev_head; + rev_head = head; + head = next; + } + + while (rev_head) { + struct filter_struct *prev = rev_head->next; + free_filter(rev_head); + rev_head = prev; + } +} + +/* Helper for identifying duplicate per-dir merge rules. */ +static const char *perdir_merge_relative_part(const char *merge_file) +{ + if (merge_file[0] == '/' || strncmp(merge_file, "../", 3) == 0) + return strrchr(merge_file, '/') + 1; /* There is a slash. */ + else + return merge_file; +} + /* Build a filter structure given a filter pattern. The value in "pat" * is not null-terminated. */ static void add_rule(struct filter_list_struct *listp, const char *pat, @@ -210,26 +256,22 @@ static void add_rule(struct filter_list_struct *listp, const char *pat, if (mflags & MATCHFLG_PERDIR_MERGE) { struct filter_list_struct *lp; - unsigned int len; int i; - if ((cp = strrchr(ret->pattern, '/')) != NULL) - cp++; - else - cp = ret->pattern; - - /* If the local merge file was already mentioned, don't + clean_fname(ret->pattern, CFN_COLLAPSE_DOT_DOT_DIRS); + /* Throw out a corner case so that we are left with either a + * ..-free path (relative) or an absolute or ../-starting path + * (parent_dirscanning).. */ + if (strcmp(ret->pattern, "..") == 0) { + rprintf(FERROR, "Dir-merging .. is not allowed.\n"); + exit_cleanup(RERR_UNSUPPORTED); + } + /* If this per-dir merge file was already mentioned, don't * add it again. */ + cp = perdir_merge_relative_part(ret->pattern); for (i = 0; i < mergelist_cnt; i++) { struct filter_struct *ex = mergelist_parents[i]; - const char *s = strrchr(ex->pattern, '/'); - if (s) - s++; - else - s = ex->pattern; - len = strlen(s); - if (len == pat_len - (cp - ret->pattern) - && memcmp(s, cp, len) == 0) { + if (strcmp(cp, perdir_merge_relative_part(ex->pattern)) == 0) { free_filter(ret); return; } @@ -237,7 +279,7 @@ static void add_rule(struct filter_list_struct *listp, const char *pat, if (!(lp = new_array(struct filter_list_struct, 1))) out_of_memory("add_rule"); - lp->head = lp->tail = NULL; + lp->head = lp->tail = lp->parent_dirscan_head = NULL; if (asprintf(&lp->debug_type, " [per-dir %s]", cp) < 0) out_of_memory("add_rule"); ret->u.mergelist = lp; @@ -250,6 +292,10 @@ static void add_rule(struct filter_list_struct *listp, const char *pat, if (!mergelist_parents) out_of_memory("add_rule"); } + if (DEBUG_GTE(FILTER, 2)) { + rprintf(FINFO, "[%s] activating mergelist #%d%s\n", + who_am_i(), mergelist_cnt, lp->debug_type); + } mergelist_parents[mergelist_cnt++] = ret; } else ret->u.slash_cnt = slash_cnt; @@ -269,46 +315,62 @@ static void add_rule(struct filter_list_struct *listp, const char *pat, static void clear_filter_list(struct filter_list_struct *listp) { if (listp->tail) { - struct filter_struct *ent, *next; /* Truncate any inherited items from the local list. */ listp->tail->next = NULL; /* Now free everything that is left. */ - for (ent = listp->head; ent; ent = next) { - next = ent->next; - free_filter(ent); - } + free_filters(listp->head); } listp->head = listp->tail = NULL; } -/* This returns an expanded (absolute) filename for the merge-file name if - * the name has any slashes in it OR if the parent_dirscan var is True; - * otherwise it returns the original merge_file name. If the len_ptr value - * is non-NULL the merge_file name is limited by the referenced length - * value and will be updated with the length of the resulting name. We - * always return a name that is null terminated, even if the merge_file - * name was not. */ +/* parse_merge_name is called to process merge-file names in three cases: + * + * 1. When an non-per-dir merge rule is added. + * - Sanitize path using NULL (module_dir) and dirbuf_depth (which is 0 as + * desired if the rule came from initial filter processing rather than a + * per-dir merge file). If non-sanitizing, just cancel .. components. + * - If the path is relative, prepend dirbuf. During initial filter + * processing, dirbuf is empty, so this step has no effect and we will + * follow the path relative to the current directory, as desired. From a + * per-dir merge file, this step makes the path absolute so that we + * interpret it correctly. + * 2. When a dir-merge rule is set up (setup_merge_file), with dirbuf containing + * the topmost dir to which it will apply. + * - Sanitize path using NULL (module_dir) and dirbuf_depth. If non- + * sanitizing, just cancel .. components. + * - If the path begins with ../, then prepend the dirbuf, yielding an + * absolute path. + * - A path that is absolute at this point will activate the parent_dirscan + * in setup_merge_file, while a relative path (which contains no .. due to + * sanitization/cleaning) will be stored for following from each dir. + * 3. When a dir-merge rule is added during another parent_dirscan. + * - We want to interpret the rule relative to the ancestor dir in which it + * was defined, even though we don't actually push into that dir to set up + * the rule. Thus, we apply the same processing as in #2 while we have the + * desired dirbuf value, leaving an absolute or ..-free relative path that + * #2 won't modify further, except... + * - #2 will prepend the module_dir to an absolute path, so we are passed + * prefix_skip == module_dirlen to avoid prepending a duplicate module_dir + * now. + * + * FIXME: A per-dir merge rule added during a parent-dirscan needs to load + * files in the same ancestor dirs, so it should be made absolute + * unconditionally. This might make a relative rule parent-dirscanning, and + * it will mess up multicomponent relative rules. Quite possibly I should + * introduce a better syntax that allows separate specification of the relative + * and parent-dirscanning parts of the path, like ./ for --relative. + * + * When this is called on a rule addition (#1 or #3), merge_file is not null- + * terminated, but len_ptr points to its length and is updated with the length + * of the result. In all cases, the result is null-terminated. */ static char *parse_merge_name(const char *merge_file, unsigned int *len_ptr, - unsigned int prefix_skip) + unsigned int prefix_skip, BOOL is_perdir) { static char buf[MAXPATHLEN]; char *fn, tmpbuf[MAXPATHLEN]; unsigned int fn_len; - if (!parent_dirscan && *merge_file != '/') { - /* Return the name unchanged it doesn't have any slashes. */ - if (len_ptr) { - const char *p = merge_file + *len_ptr; - while (--p > merge_file && *p != '/') {} - if (p == merge_file) { - strlcpy(buf, merge_file, *len_ptr + 1); - return buf; - } - } else if (strchr(merge_file, '/') == NULL) - return (char *)merge_file; - } - fn = *merge_file == '/' ? buf : tmpbuf; if (sanitize_paths) { const char *r = prefix_skip ? "/" : NULL; @@ -329,8 +391,10 @@ static char *parse_merge_name(const char *merge_file, unsigned int *len_ptr, fn_len = clean_fname(fn, CFN_COLLAPSE_DOT_DOT_DIRS); } - /* If the name isn't in buf yet, it's wasn't absolute. */ - if (fn != buf) { + /* If the name isn't in buf yet, it's relative. In case #1, we want to + * prepend the dirbuf unconditionally. Otherwise, we want to make + * parent_dirscan-activating paths absolute but leave others alone. */ + if (fn != buf && (!is_perdir || strncmp(fn, "../", 3) == 0)) { int d_len = dirbuf_len - prefix_skip; if (d_len + fn_len >= MAXPATHLEN) { rprintf(FERROR, "merge-file name overflows: %s\n", fn); @@ -338,12 +402,13 @@ static char *parse_merge_name(const char *merge_file, unsigned int *len_ptr, } memcpy(buf, dirbuf + prefix_skip, d_len); memcpy(buf + d_len, fn, fn_len + 1); + fn = buf; fn_len = clean_fname(buf, CFN_COLLAPSE_DOT_DOT_DIRS); } if (len_ptr) *len_ptr = fn_len; - return buf; + return fn; } /* Sets the dirbuf and dirbuf_len values. */ @@ -376,16 +441,22 @@ void set_filter_dir(const char *dir, unsigned int dirlen) * parent directory of the first transfer dir. If it does, we scan all the * dirs from that point through the parent dir of the transfer dir looking * for the per-dir merge-file in each one. */ -static BOOL setup_merge_file(struct filter_struct *ex, +static BOOL setup_merge_file(int mergelist_num, struct filter_struct *ex, struct filter_list_struct *lp) { char buf[MAXPATHLEN]; char *x, *y, *pat = ex->pattern; unsigned int len; - if (!(x = parse_merge_name(pat, NULL, 0)) || *x != '/') + /* Case #2 in parse_merge_name comment: make a parent_dirscanning path + * absolute. */ + if (!(x = parse_merge_name(pat, NULL, 0, True)) || *x != '/') return 0; + if (DEBUG_GTE(FILTER, 2)) { + rprintf(FINFO, "[%s] performing parent_dirscan for mergelist #%d%s\n", + who_am_i(), mergelist_num, lp->debug_type); + } y = strrchr(x, '/'); *y = '\0'; ex->pattern = strdup(y+1); @@ -414,37 +485,63 @@ static BOOL setup_merge_file(struct filter_struct *ex, dirbuf_len = y - dirbuf; strlcpy(x, ex->pattern, MAXPATHLEN - (x - buf)); parse_filter_file(lp, buf, ex->match_flags, XFLG_ANCHORED2ABS); - if (ex->match_flags & MATCHFLG_NO_INHERIT) + if (ex->match_flags & MATCHFLG_NO_INHERIT) { + /* Free the undesired rules to clean up any per-dir + * mergelists they defined. Otherwise pop_local_filters + * may crash trying to restore nonexistent state for + * those mergelists. */ + free_filters(lp->head); lp->head = NULL; + } lp->tail = NULL; strlcpy(y, save, MAXPATHLEN); while ((*x++ = *y++) != '/') {} } + /* Save current head for freeing when the mergelist becomes inactive. */ + lp->parent_dirscan_head = lp->head; parent_dirscan = False; + if (DEBUG_GTE(FILTER, 2)) { + rprintf(FINFO, "[%s] completed parent_dirscan for mergelist #%d%s\n", + who_am_i(), mergelist_num, lp->debug_type); + } free(pat); return 1; } +struct local_filter_state { + int mergelist_cnt; + struct filter_list_struct mergelists[1]; +}; + /* Each time rsync changes to a new directory it call this function to * handle all the per-dir merge-files. The "dir" value is the current path * relative to curr_dir (which might not be null-terminated). We copy it * into dirbuf so that we can easily append a file name on the end. */ void *push_local_filters(const char *dir, unsigned int dirlen) { - struct filter_list_struct *ap, *push; + struct local_filter_state *push; int i; set_filter_dir(dir, dirlen); + if (DEBUG_GTE(FILTER, 2)) { + rprintf(FINFO, "[%s] pushing local filters for %s\n", + who_am_i(), dirbuf); + } - if (!mergelist_cnt) + if (!mergelist_cnt) { + /* No old state to save and no new merge files to push. */ return NULL; + } - push = new_array(struct filter_list_struct, mergelist_cnt); + push = (struct local_filter_state *)new_array(char, + sizeof (struct local_filter_state) + + (mergelist_cnt-1) * sizeof (struct filter_list_struct)); if (!push) out_of_memory("push_local_filters"); - for (i = 0, ap = push; i < mergelist_cnt; i++) { - memcpy(ap++, mergelist_parents[i]->u.mergelist, + push->mergelist_cnt = mergelist_cnt; + for (i = 0; i < mergelist_cnt; i++) { + memcpy(&push->mergelists[i], mergelist_parents[i]->u.mergelist, sizeof (struct filter_list_struct)); } @@ -455,8 +552,8 @@ void *push_local_filters(const char *dir, unsigned int dirlen) struct filter_list_struct *lp = ex->u.mergelist; if (DEBUG_GTE(FILTER, 2)) { - rprintf(FINFO, "[%s] pushing filter list%s\n", - who_am_i(), lp->debug_type); + rprintf(FINFO, "[%s] pushing mergelist #%d%s\n", + who_am_i(), i, lp->debug_type); } lp->tail = NULL; /* Switch any local rules to inherited. */ @@ -465,7 +562,7 @@ void *push_local_filters(const char *dir, unsigned int dirlen) if (ex->match_flags & MATCHFLG_FINISH_SETUP) { ex->match_flags &= ~MATCHFLG_FINISH_SETUP; - if (setup_merge_file(ex, lp)) + if (setup_merge_file(i, ex, lp)) set_filter_dir(dir, dirlen); } @@ -475,7 +572,7 @@ void *push_local_filters(const char *dir, unsigned int dirlen) XFLG_ANCHORED2ABS); } else { io_error |= IOERR_GENERAL; - rprintf(FINFO, + rprintf(FERROR, "cannot add local filter rules in long-named directory: %s\n", full_fname(dirbuf)); } @@ -487,26 +584,52 @@ void *push_local_filters(const char *dir, unsigned int dirlen) void pop_local_filters(void *mem) { - struct filter_list_struct *ap, *pop = (struct filter_list_struct*)mem; + struct local_filter_state *pop = (struct local_filter_state *)mem; int i; + int old_mergelist_cnt = pop ? pop->mergelist_cnt : 0; + + if (DEBUG_GTE(FILTER, 2)) + rprintf(FINFO, "[%s] popping local filters\n", who_am_i()); for (i = mergelist_cnt; i-- > 0; ) { struct filter_struct *ex = mergelist_parents[i]; struct filter_list_struct *lp = ex->u.mergelist; if (DEBUG_GTE(FILTER, 2)) { - rprintf(FINFO, "[%s] popping filter list%s\n", - who_am_i(), lp->debug_type); + rprintf(FINFO, "[%s] popping mergelist #%d%s\n", + who_am_i(), i, lp->debug_type); } clear_filter_list(lp); + + if (i >= old_mergelist_cnt) { + /* This mergelist does not exist in the state to be + * restored. Free its parent_dirscan list to clean up + * any per-dir mergelists defined there so we don't + * crash trying to restore nonexistent state for them + * below. (Counterpart to setup_merge_file call in + * push_local_filters. Must be done here, not in + * free_filter, for LIFO order.) */ + if (DEBUG_GTE(FILTER, 2)) { + rprintf(FINFO, "[%s] freeing parent_dirscan filters of mergelist #%d%s\n", + who_am_i(), i, ex->u.mergelist->debug_type); + } + free_filters(lp->parent_dirscan_head); + lp->parent_dirscan_head = NULL; + } } - if (!pop) + /* If we cleaned things up properly, the only still-active mergelists + * should be those with a state to be restored. */ + assert(mergelist_cnt == old_mergelist_cnt); + + if (!pop) { + /* No state to restore. */ return; + } - for (i = 0, ap = pop; i < mergelist_cnt; i++) { - memcpy(mergelist_parents[i]->u.mergelist, ap++, + for (i = 0; i < mergelist_cnt; i++) { + memcpy(mergelist_parents[i]->u.mergelist, &pop->mergelists[i], sizeof (struct filter_list_struct)); } @@ -621,8 +744,8 @@ static int rule_matches(const char *fname, struct filter_struct *ex, int name_is static void report_filter_result(enum logcode code, char const *name, - struct filter_struct const *ent, - int name_is_dir, const char *type) + struct filter_struct const *ent, + int name_is_dir, const char *type) { /* If a trailing slash is present to match only directories, * then it is stripped out by add_rule(). So as a special @@ -998,14 +1121,21 @@ void parse_rule(struct filter_list_struct *listp, const char *pattern, } if (new_mflags & MATCHFLG_PERDIR_MERGE) { if (parent_dirscan) { + /* Case #3 in parse_merge_name comment: + * record correct starting dir for a + * parent_dirscanning per-dir merge rule + * found during another parent_dirscan. */ if (!(p = parse_merge_name(cp, &len, - module_dirlen))) + module_dirlen, True))) continue; add_rule(listp, p, len, new_mflags, 0); continue; } } else { - if (!(p = parse_merge_name(cp, &len, 0))) + /* Case #1 in parse_merge_name comment: + * interpret the path w.r.t. the correct + * directory. */ + if (!(p = parse_merge_name(cp, &len, 0, False))) continue; parse_filter_file(listp, p, new_mflags, XFLG_FATAL_ERRORS);