A couple fixes in add_rule() for XFLG_ABS_IF_SLASH:
[rsync/rsync.git] / exclude.c
index 16ab02d..4c7a847 100644 (file)
--- a/exclude.c
+++ b/exclude.c
@@ -1,12 +1,14 @@
-/* -*- c-file-style: "linux" -*-
+/*
+ * The filter include/exclude routines.
  *
- * Copyright (C) 1996-2001 by Andrew Tridgell <tridge@samba.org>
- * Copyright (C) 1996 by Paul Mackerras
- * Copyright (C) 2002 by Martin Pool
+ * Copyright (C) 1996-2001 Andrew Tridgell <tridge@samba.org>
+ * Copyright (C) 1996 Paul Mackerras
+ * Copyright (C) 2002 Martin Pool
+ * Copyright (C) 2003-2008 Wayne Davison
  *
  * 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
+ * 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,
  * 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.
+ * You should have received a copy of the GNU General Public License along
+ * with this program; if not, visit the http://fsf.org website.
  */
 
-/* a lot of this stuff was originally derived from GNU tar, although
-   it has now changed so much that it is hard to tell :) */
-
-/* include/exclude cluestick added by Martin Pool <mbp@samba.org> */
-
 #include "rsync.h"
 
 extern int verbose;
+extern int am_server;
+extern int am_sender;
+extern int eol_nulls;
+extern int io_error;
+extern int local_server;
+extern int prune_empty_dirs;
+extern int ignore_perishable;
+extern int delete_mode;
+extern int delete_excluded;
+extern int cvs_exclude;
+extern int sanitize_paths;
+extern int protocol_version;
+extern int module_id;
+
+extern char curr_dir[];
+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]" };
+
+/* Need room enough for ":MODS " prefix plus some room to grow. */
+#define MAX_RULE_PREFIX (16)
+
+#define MODIFIERS_MERGE_FILE "-+Cenw"
+#define MODIFIERS_INCL_EXCL "/!Crsp"
+#define MODIFIERS_HIDE_PROTECT "/!p"
+
+/* The dirbuf is set by push_local_filters() to the current subdirectory
+ * relative to curr_dir that is being processed.  The path always has a
+ * trailing slash appended, and the variable dirbuf_len contains the length
+ * of this path prefix.  The path is always absolute. */
+static char dirbuf[MAXPATHLEN+1];
+static unsigned int dirbuf_len = 0;
+static int dirbuf_depth;
+
+/* This is True when we're scanning parent dirs for per-dir merge-files. */
+static BOOL parent_dirscan = False;
+
+/* This array contains a list of all the currently active per-dir merge
+ * files.  This makes it easier to save the appropriate values when we
+ * "push" down into each subdirectory. */
+static struct filter_struct **mergelist_parents;
+static int mergelist_cnt = 0;
+static int mergelist_size = 0;
+
+/* Each filter_list_struct describes a singly-linked list by keeping track
+ * of both the head and tail pointers.  The list is slightly unusual in that
+ * a parent-dir's content can be appended to the end of the local list in a
+ * special way:  the last item in the local list has its "next" pointer set
+ * to point to the inherited list, but the local list's tail pointer points
+ * at the end of the local list.  Thus, if the local list is empty, the head
+ * will be pointing at the inherited content but the tail will be NULL.  To
+ * help you visualize this, here are the possible list arrangements:
+ *
+ * Completely Empty                     Local Content Only
+ * ==================================   ====================================
+ * head -> NULL                         head -> Local1 -> Local2 -> NULL
+ * tail -> NULL                         tail -------------^
+ *
+ * Inherited Content Only               Both Local and Inherited Content
+ * ==================================   ====================================
+ * head -> Parent1 -> Parent2 -> NULL   head -> L1 -> L2 -> P1 -> P2 -> NULL
+ * tail -> NULL                         tail ---------^
+ *
+ * This means that anyone wanting to traverse the whole list to use it just
+ * needs to start at the head and use the "next" pointers until it goes
+ * NULL.  To add new local content, we insert the item after the tail item
+ * and update the tail (obviously, if "tail" was NULL, we insert it at the
+ * head).  To clear the local list, WE MUST NOT FREE THE INHERITED CONTENT
+ * because it is shared between the current list and our parent list(s).
+ * The easiest way to handle this is to simply truncate the list after the
+ * tail item and then free the local list from the head.  When inheriting
+ * the list for a new local dir, we just save off the filter_list_struct
+ * values (so we can pop back to them later) and set the tail to NULL.
+ */
 
-struct exclude_struct **exclude_list;
-struct exclude_struct **local_exclude_list;
-struct exclude_struct **server_exclude_list;
-char *exclude_path_prefix = NULL;
+static void free_filter(struct filter_struct *ex)
+{
+       if (ex->match_flags & MATCHFLG_PERDIR_MERGE) {
+               free(ex->u.mergelist->debug_type);
+               free(ex->u.mergelist);
+               mergelist_cnt--;
+       }
+       free(ex->pattern);
+       free(ex);
+}
 
-/** Build an exclude structure given a exclude pattern */
-static struct exclude_struct *make_exclude(const char *pattern, int pat_len,
-                                          int include)
+/* 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,
+                    unsigned int pat_len, uint32 mflags, int xflags)
 {
-       struct exclude_struct *ret;
+       struct filter_struct *ret;
        const char *cp;
-       int ex_len;
+       unsigned int pre_len, slash_cnt = 0;
+
+       if (verbose > 2) {
+               rprintf(FINFO, "[%s] add_rule(%s%.*s%s)%s\n",
+                       who_am_i(), get_rule_prefix(mflags, pat, 0, NULL),
+                       (int)pat_len, pat,
+                       (mflags & MATCHFLG_DIRECTORY) ? "/" : "",
+                       listp->debug_type);
+       }
+
+       /* These flags also indicate that we're reading a list that
+        * needs to be filtered now, not post-filtered later. */
+       if (xflags & (XFLG_ANCHORED2ABS|XFLG_ABS_IF_SLASH)) {
+               uint32 mf = mflags & (MATCHFLG_RECEIVER_SIDE|MATCHFLG_SENDER_SIDE);
+               if (am_sender) {
+                       if (mf == MATCHFLG_RECEIVER_SIDE)
+                               return;
+               } else {
+                       if (mf == MATCHFLG_SENDER_SIDE)
+                               return;
+               }
+       }
 
-       ret = new(struct exclude_struct);
-       if (!ret)
-               out_of_memory("make_exclude");
+       if (!(ret = new0(struct filter_struct)))
+               out_of_memory("add_rule");
 
-       memset(ret, 0, sizeof ret[0]);
-       ret->include = include;
+       if (pat_len > 1 && pat[pat_len-1] == '/') {
+               pat_len--;
+               mflags |= MATCHFLG_DIRECTORY;
+       }
 
-       if (exclude_path_prefix)
-               ret->match_flags |= MATCHFLG_ABS_PATH;
-       if (exclude_path_prefix && *pattern == '/')
-               ex_len = strlen(exclude_path_prefix);
-       else
-               ex_len = 0;
-       ret->pattern = new_array(char, ex_len + pat_len + 1);
-       if (!ret->pattern)
-               out_of_memory("make_exclude");
-       if (ex_len)
-               memcpy(ret->pattern, exclude_path_prefix, ex_len);
-       strlcpy(ret->pattern + ex_len, pattern, pat_len + 1);
-       pat_len += ex_len;
+       for (cp = pat; cp < pat + pat_len; cp++) {
+               if (*cp == '/')
+                       slash_cnt++;
+       }
+
+       if (!(mflags & (MATCHFLG_ABS_PATH | MATCHFLG_MERGE_FILE))
+        && ((xflags & (XFLG_ANCHORED2ABS|XFLG_ABS_IF_SLASH) && *pat == '/')
+         || (xflags & XFLG_ABS_IF_SLASH && slash_cnt))) {
+               mflags |= MATCHFLG_ABS_PATH;
+               if (*pat == '/')
+                       pre_len = dirbuf_len - module_dirlen - 1;
+               else
+                       pre_len = 0;
+       } else
+               pre_len = 0;
+
+       if (!(ret->pattern = new_array(char, pre_len + pat_len + 1)))
+               out_of_memory("add_rule");
+       if (pre_len) {
+               memcpy(ret->pattern, dirbuf + module_dirlen, pre_len);
+               for (cp = ret->pattern; cp < ret->pattern + pre_len; cp++) {
+                       if (*cp == '/')
+                               slash_cnt++;
+               }
+       }
+       strlcpy(ret->pattern + pre_len, pat, pat_len + 1);
+       pat_len += pre_len;
 
        if (strpbrk(ret->pattern, "*[?")) {
-               ret->match_flags |= MATCHFLG_WILD;
+               mflags |= MATCHFLG_WILD;
                if ((cp = strstr(ret->pattern, "**")) != NULL) {
-                       ret->match_flags |= MATCHFLG_WILD2;
+                       mflags |= MATCHFLG_WILD2;
                        /* If the pattern starts with **, note that. */
                        if (cp == ret->pattern)
-                               ret->match_flags |= MATCHFLG_WILD2_PREFIX;
+                               mflags |= MATCHFLG_WILD2_PREFIX;
+                       /* If the pattern ends with ***, note that. */
+                       if (pat_len >= 3
+                        && ret->pattern[pat_len-3] == '*'
+                        && ret->pattern[pat_len-2] == '*'
+                        && ret->pattern[pat_len-1] == '*')
+                               mflags |= MATCHFLG_WILD3_SUFFIX;
+               }
+       }
+
+       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
+                * add it again. */
+               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) {
+                               free_filter(ret);
+                               return;
+                       }
+               }
+
+               if (!(lp = new_array(struct filter_list_struct, 1)))
+                       out_of_memory("add_rule");
+               lp->head = lp->tail = NULL;
+               if (asprintf(&lp->debug_type, " [per-dir %s]", cp) < 0)
+                       out_of_memory("add_rule");
+               ret->u.mergelist = lp;
+
+               if (mergelist_cnt == mergelist_size) {
+                       mergelist_size += 5;
+                       mergelist_parents = realloc_array(mergelist_parents,
+                                               struct filter_struct *,
+                                               mergelist_size);
+                       if (!mergelist_parents)
+                               out_of_memory("add_rule");
+               }
+               mergelist_parents[mergelist_cnt++] = ret;
+       } else
+               ret->u.slash_cnt = slash_cnt;
+
+       ret->match_flags = mflags;
+
+       if (!listp->tail) {
+               ret->next = listp->head;
+               listp->head = listp->tail = ret;
+       } else {
+               ret->next = listp->tail->next;
+               listp->tail->next = ret;
+               listp->tail = ret;
+       }
+}
+
+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);
+               }
+       }
+
+       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. */
+static char *parse_merge_name(const char *merge_file, unsigned int *len_ptr,
+                             unsigned int prefix_skip)
+{
+       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;
+               /* null-terminate the name if it isn't already */
+               if (len_ptr && merge_file[*len_ptr]) {
+                       char *to = fn == buf ? tmpbuf : buf;
+                       strlcpy(to, merge_file, *len_ptr + 1);
+                       merge_file = to;
                }
+               if (!sanitize_path(fn, merge_file, r, dirbuf_depth, SP_DEFAULT)) {
+                       rprintf(FERROR, "merge-file name overflows: %s\n",
+                               merge_file);
+                       return NULL;
+               }
+               fn_len = strlen(fn);
+       } else {
+               strlcpy(fn, merge_file, len_ptr ? *len_ptr + 1 : MAXPATHLEN);
+               fn_len = clean_fname(fn, CFN_COLLAPSE_DOT_DOT_DIRS);
        }
 
-       if (pat_len > 1 && ret->pattern[pat_len-1] == '/') {
-               ret->pattern[pat_len-1] = 0;
-               ret->directory = 1;
+       /* If the name isn't in buf yet, it's wasn't absolute. */
+       if (fn != buf) {
+               if (dirbuf_len + fn_len >= MAXPATHLEN) {
+                       rprintf(FERROR, "merge-file name overflows: %s\n", fn);
+                       return NULL;
+               }
+               memcpy(buf, dirbuf + prefix_skip, dirbuf_len - prefix_skip);
+               memcpy(buf + dirbuf_len - prefix_skip, fn, fn_len + 1);
+               fn_len = clean_fname(buf, CFN_COLLAPSE_DOT_DOT_DIRS);
        }
 
-       for (cp = ret->pattern; (cp = strchr(cp, '/')) != NULL; cp++)
-               ret->slash_cnt++;
+       if (len_ptr)
+               *len_ptr = fn_len;
+       return buf;
+}
 
-       return ret;
+/* Sets the dirbuf and dirbuf_len values. */
+void set_filter_dir(const char *dir, unsigned int dirlen)
+{
+       unsigned int len;
+       if (*dir != '/') {
+               memcpy(dirbuf, curr_dir, curr_dir_len);
+               dirbuf[curr_dir_len] = '/';
+               len = curr_dir_len + 1;
+               if (len + dirlen >= MAXPATHLEN)
+                       dirlen = 0;
+       } else
+               len = 0;
+       memcpy(dirbuf + len, dir, dirlen);
+       dirbuf[dirlen + len] = '\0';
+       dirbuf_len = clean_fname(dirbuf, CFN_COLLAPSE_DOT_DOT_DIRS);
+       if (dirbuf_len > 1 && dirbuf[dirbuf_len-1] == '.'
+           && dirbuf[dirbuf_len-2] == '/')
+               dirbuf_len -= 2;
+       if (dirbuf_len != 1)
+               dirbuf[dirbuf_len++] = '/';
+       dirbuf[dirbuf_len] = '\0';
+       if (sanitize_paths)
+               dirbuf_depth = count_dir_elements(dirbuf + module_dirlen);
 }
 
-static void free_exclude(struct exclude_struct *ex)
+/* This routine takes a per-dir merge-file entry and finishes its setup.
+ * If the name has a path portion then we check to see if it refers to a
+ * 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,
+                            struct filter_list_struct *lp)
 {
-       free(ex->pattern);
-       memset(ex, 0, sizeof ex[0]);
-       free(ex);
+       char buf[MAXPATHLEN];
+       char *x, *y, *pat = ex->pattern;
+       unsigned int len;
+
+       if (!(x = parse_merge_name(pat, NULL, 0)) || *x != '/')
+               return 0;
+
+       y = strrchr(x, '/');
+       *y = '\0';
+       ex->pattern = strdup(y+1);
+       if (!*x)
+               x = "/";
+       if (*x == '/')
+               strlcpy(buf, x, MAXPATHLEN);
+       else
+               pathjoin(buf, MAXPATHLEN, dirbuf, x);
+
+       len = clean_fname(buf, CFN_COLLAPSE_DOT_DOT_DIRS);
+       if (len != 1 && len < MAXPATHLEN-1) {
+               buf[len++] = '/';
+               buf[len] = '\0';
+       }
+       /* This ensures that the specified dir is a parent of the transfer. */
+       for (x = buf, y = dirbuf; *x && *x == *y; x++, y++) {}
+       if (*x)
+               y += strlen(y); /* nope -- skip the scan */
+
+       parent_dirscan = True;
+       while (*y) {
+               char save[MAXPATHLEN];
+               strlcpy(save, y, MAXPATHLEN);
+               *y = '\0';
+               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)
+                       lp->head = NULL;
+               lp->tail = NULL;
+               strlcpy(y, save, MAXPATHLEN);
+               while ((*x++ = *y++) != '/') {}
+       }
+       parent_dirscan = False;
+       free(pat);
+       return 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;
+       int i;
+
+       set_filter_dir(dir, dirlen);
+
+       if (!mergelist_cnt)
+               return NULL;
+
+       push = new_array(struct filter_list_struct, mergelist_cnt);
+       if (!push)
+               out_of_memory("push_local_filters");
+
+       for (i = 0, ap = push; i < mergelist_cnt; i++) {
+               memcpy(ap++, mergelist_parents[i]->u.mergelist,
+                      sizeof (struct filter_list_struct));
+       }
+
+       /* Note: parse_filter_file() might increase mergelist_cnt, so keep
+        * this loop separate from the above loop. */
+       for (i = 0; i < mergelist_cnt; i++) {
+               struct filter_struct *ex = mergelist_parents[i];
+               struct filter_list_struct *lp = ex->u.mergelist;
+
+               if (verbose > 2) {
+                       rprintf(FINFO, "[%s] pushing filter list%s\n",
+                               who_am_i(), lp->debug_type);
+               }
+
+               lp->tail = NULL; /* Switch any local rules to inherited. */
+               if (ex->match_flags & MATCHFLG_NO_INHERIT)
+                       lp->head = NULL;
+
+               if (ex->match_flags & MATCHFLG_FINISH_SETUP) {
+                       ex->match_flags &= ~MATCHFLG_FINISH_SETUP;
+                       if (setup_merge_file(ex, lp))
+                               set_filter_dir(dir, dirlen);
+               }
+
+               if (strlcpy(dirbuf + dirbuf_len, ex->pattern,
+                   MAXPATHLEN - dirbuf_len) < MAXPATHLEN - dirbuf_len) {
+                       parse_filter_file(lp, dirbuf, ex->match_flags,
+                                         XFLG_ANCHORED2ABS);
+               } else {
+                       io_error |= IOERR_GENERAL;
+                       rprintf(FINFO,
+                           "cannot add local filter rules in long-named directory: %s\n",
+                           full_fname(dirbuf));
+               }
+               dirbuf[dirbuf_len] = '\0';
+       }
+
+       return (void*)push;
+}
 
-void free_exclude_list(struct exclude_struct ***listp)
+void pop_local_filters(void *mem)
 {
-       struct exclude_struct **list = *listp;
+       struct filter_list_struct *ap, *pop = (struct filter_list_struct*)mem;
+       int i;
 
-       if (verbose > 2)
-               rprintf(FINFO, "[%s] clearing exclude list\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 (!list)
+               if (verbose > 2) {
+                       rprintf(FINFO, "[%s] popping filter list%s\n",
+                               who_am_i(), lp->debug_type);
+               }
+
+               clear_filter_list(lp);
+       }
+
+       if (!pop)
                return;
 
-       while (*list)
-               free_exclude(*list++);
+       for (i = 0, ap = pop; i < mergelist_cnt; i++) {
+               memcpy(mergelist_parents[i]->u.mergelist, ap++,
+                      sizeof (struct filter_list_struct));
+       }
 
-       free(*listp);
-       *listp = NULL;
+       free(pop);
 }
 
-static int check_one_exclude(char *name, struct exclude_struct *ex,
-                             int name_is_dir)
+void change_local_filter_dir(const char *dname, int dlen, int dir_depth)
 {
-       char *p;
-       int match_start = 0;
-       char *pattern = ex->pattern;
-
-       /* If the pattern does not have any slashes AND it does not have
-        * a "**" (which could match a slash), then we just match the
-        * name portion of the path. */
-       if (!ex->slash_cnt && !(ex->match_flags & MATCHFLG_WILD2)) {
-               if ((p = strrchr(name,'/')) != NULL)
-                       name = p+1;
+       static int cur_depth = -1;
+       static void *filt_array[MAXPATHLEN/2+1];
+
+       if (!dname) {
+               for ( ; cur_depth >= 0; cur_depth--) {
+                       if (filt_array[cur_depth]) {
+                               pop_local_filters(filt_array[cur_depth]);
+                               filt_array[cur_depth] = NULL;
+                       }
+               }
+               return;
        }
-       else if ((ex->match_flags & MATCHFLG_ABS_PATH) && *name != '/') {
-               static char full_name[MAXPATHLEN];
-               extern char curr_dir[];
-               int plus = curr_dir[1] == '\0'? 1 : 0;
-               pathjoin(full_name, sizeof full_name, curr_dir+plus, name);
-               name = full_name;
+
+       assert(dir_depth < MAXPATHLEN/2+1);
+
+       for ( ; cur_depth >= dir_depth; cur_depth--) {
+               if (filt_array[cur_depth]) {
+                       pop_local_filters(filt_array[cur_depth]);
+                       filt_array[cur_depth] = NULL;
+               }
        }
 
-       if (!name[0]) return 0;
+       cur_depth = dir_depth;
+       filt_array[cur_depth] = push_local_filters(dname, dlen);
+}
 
-       if (ex->directory && !name_is_dir) return 0;
+static int rule_matches(const char *fname, struct filter_struct *ex, int name_is_dir)
+{
+       int slash_handling, str_cnt = 0, anchored_match = 0;
+       int ret_match = ex->match_flags & MATCHFLG_NEGATE ? 0 : 1;
+       char *p, *pattern = ex->pattern;
+       const char *strings[16]; /* more than enough */
+       const char *name = fname + (*fname == '/');
+
+       if (!*name)
+               return 0;
+
+       if (!ex->u.slash_cnt && !(ex->match_flags & MATCHFLG_WILD2)) {
+               /* If the pattern does not have any slashes AND it does
+                * not have a "**" (which could match a slash), then we
+                * just match the name portion of the path. */
+               if ((p = strrchr(name,'/')) != NULL)
+                       name = p+1;
+       } else if (ex->match_flags & MATCHFLG_ABS_PATH && *fname != '/'
+           && curr_dir_len > module_dirlen + 1) {
+               /* If we're matching against an absolute-path pattern,
+                * we need to prepend our full path info. */
+               strings[str_cnt++] = curr_dir + module_dirlen + 1;
+               strings[str_cnt++] = "/";
+       } else if (ex->match_flags & MATCHFLG_WILD2_PREFIX && *fname != '/') {
+               /* Allow "**"+"/" to match at the start of the string. */
+               strings[str_cnt++] = "/";
+       }
+       strings[str_cnt++] = name;
+       if (name_is_dir) {
+               /* Allow a trailing "/"+"***" to match the directory. */
+               if (ex->match_flags & MATCHFLG_WILD3_SUFFIX)
+                       strings[str_cnt++] = "/";
+       } else if (ex->match_flags & MATCHFLG_DIRECTORY)
+               return !ret_match;
+       strings[str_cnt] = NULL;
 
        if (*pattern == '/') {
-               match_start = 1;
+               anchored_match = 1;
                pattern++;
-               if (*name == '/')
-                       name++;
        }
 
-       if (ex->match_flags & MATCHFLG_WILD) {
+       if (!anchored_match && ex->u.slash_cnt
+           && !(ex->match_flags & MATCHFLG_WILD2)) {
                /* A non-anchored match with an infix slash and no "**"
                 * needs to match the last slash_cnt+1 name elements. */
-               if (!match_start && ex->slash_cnt &&
-                   !(ex->match_flags & MATCHFLG_WILD2)) {
-                       int cnt = ex->slash_cnt + 1;
-                       for (p = name + strlen(name) - 1; p >= name; p--) {
-                               if (*p == '/' && !--cnt)
-                                       break;
-                       }
-                       name = p+1;
-               }
-               if (wildmatch(pattern, name))
-                       return 1;
-               if (ex->match_flags & MATCHFLG_WILD2_PREFIX) {
-                       /* If the **-prefixed pattern has a '/' as the next
-                        * character, then try to match the rest of the
-                        * pattern at the root. */
-                       if (pattern[2] == '/' && wildmatch(pattern+3, name))
-                               return 1;
-               }
-               else if (!match_start && ex->match_flags & MATCHFLG_WILD2) {
-                       /* A non-anchored match with an infix or trailing "**"
-                        * (but not a prefixed "**") needs to try matching
-                        * after every slash. */
-                       while ((name = strchr(name, '/')) != NULL) {
-                               name++;
-                               if (wildmatch(pattern, name))
-                                       return 1;
-                       }
-               }
-       } else if (match_start) {
-               if (strcmp(name,pattern) == 0)
-                       return 1;
+               slash_handling = ex->u.slash_cnt + 1;
+       } else if (!anchored_match && !(ex->match_flags & MATCHFLG_WILD2_PREFIX)
+                                  && ex->match_flags & MATCHFLG_WILD2) {
+               /* A non-anchored match with an infix or trailing "**" (but not
+                * a prefixed "**") needs to try matching after every slash. */
+               slash_handling = -1;
+       } else {
+               /* The pattern matches only at the start of the path or name. */
+               slash_handling = 0;
+       }
+
+       if (ex->match_flags & MATCHFLG_WILD) {
+               if (wildmatch_array(pattern, strings, slash_handling))
+                       return ret_match;
+       } else if (str_cnt > 1) {
+               if (litmatch_array(pattern, strings, slash_handling))
+                       return ret_match;
+       } else if (anchored_match) {
+               if (strcmp(name, pattern) == 0)
+                       return ret_match;
        } else {
                int l1 = strlen(name);
                int l2 = strlen(pattern);
                if (l2 <= l1 &&
                    strcmp(name+(l1-l2),pattern) == 0 &&
                    (l1==l2 || name[l1-(l2+1)] == '/')) {
-                       return 1;
+                       return ret_match;
                }
        }
 
-       return 0;
+       return !ret_match;
 }
 
 
-static void report_exclude_result(char const *name,
-                                  struct exclude_struct const *ent,
-                                  int name_is_dir)
+static void report_filter_result(char const *name,
+                                 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 make_exclude.  So as a special
+        * then it is stripped out by add_rule().  So as a special
         * case we add it back in here. */
 
        if (verbose >= 2) {
-               rprintf(FINFO, "[%s] %scluding %s %s because of pattern %s%s\n",
-                       who_am_i(), ent->include ? "in" : "ex",
-                       name_is_dir ? "directory" : "file", name, ent->pattern,
-                       ent->directory ? "/" : "");
+               static char *actions[2][2]
+                   = { {"show", "hid"}, {"risk", "protect"} };
+               const char *w = who_am_i();
+               rprintf(FINFO, "[%s] %sing %s %s because of pattern %s%s%s\n",
+                   w, actions[*w!='s'][!(ent->match_flags&MATCHFLG_INCLUDE)],
+                   name_is_dir ? "directory" : "file", name, ent->pattern,
+                   ent->match_flags & MATCHFLG_DIRECTORY ? "/" : "", type);
        }
 }
 
 
 /*
- * Return true if file NAME is defined to be excluded by either
- * LOCAL_EXCLUDE_LIST or the globals EXCLUDE_LIST.
+ * Return -1 if file "name" is defined to be excluded by the specified
+ * exclude list, 1 if it is included, and 0 if it was not matched.
  */
-int check_exclude(struct exclude_struct **list, char *name, int name_is_dir)
+int check_filter(struct filter_list_struct *listp, const char *name, int name_is_dir)
 {
-       struct exclude_struct *ent;
+       struct filter_struct *ent;
 
-       while ((ent = *list++) != NULL) {
-               if (check_one_exclude(name, ent, name_is_dir)) {
-                       report_exclude_result(name, ent, name_is_dir);
-                       return !ent->include;
+       for (ent = listp->head; ent; ent = ent->next) {
+               if (ignore_perishable && ent->match_flags & MATCHFLG_PERISHABLE)
+                       continue;
+               if (ent->match_flags & MATCHFLG_PERDIR_MERGE) {
+                       int rc = check_filter(ent->u.mergelist, name,
+                                             name_is_dir);
+                       if (rc)
+                               return rc;
+                       continue;
+               }
+               if (ent->match_flags & MATCHFLG_CVS_IGNORE) {
+                       int rc = check_filter(&cvs_filter_list, name,
+                                             name_is_dir);
+                       if (rc)
+                               return rc;
+                       continue;
+               }
+               if (rule_matches(name, ent, name_is_dir)) {
+                       report_filter_result(name, ent, name_is_dir,
+                                             listp->debug_type);
+                       return ent->match_flags & MATCHFLG_INCLUDE ? 1 : -1;
                }
        }
 
        return 0;
 }
 
+#define RULE_STRCMP(s,r) rule_strcmp((s), (r), sizeof (r) - 1)
+
+static const uchar *rule_strcmp(const uchar *str, const char *rule, int rule_len)
+{
+       if (strncmp((char*)str, rule, rule_len) != 0)
+               return NULL;
+       if (isspace(str[rule_len]) || str[rule_len] == '_' || !str[rule_len])
+               return str + rule_len - 1;
+       if (str[rule_len] == ',')
+               return str + rule_len;
+       return NULL;
+}
 
 /* Get the next include/exclude arg from the string.  The token will not
  * be '\0' terminated, so use the returned length to limit the string.
  * Also, be sure to add this length to the returned pointer before passing
- * it back to ask for the next token.  This routine will not split off a
- * prefix of "+ " or "- " unless xflags contains XFLG_NO_PREFIXES.  The
- * *incl_ptr value will be 1 for an include, 0 for an exclude, and -1 for
- * the list-clearing "!" token.
- */
-static const char *get_exclude_tok(const char *p, int *len_ptr, int *incl_ptr,
-                                  int xflags)
+ * it back to ask for the next token.  This routine parses the "!" (list-
+ * clearing) token and (depending on the mflags) the various prefixes.
+ * The *mflags_ptr value will be set on exit to the new MATCHFLG_* bits
+ * for the current token. */
+static const char *parse_rule_tok(const char *p, uint32 mflags, int xflags,
+                                 unsigned int *len_ptr, uint32 *mflags_ptr)
 {
-       const unsigned char *s = (const unsigned char *)p;
-       int len;
+       const uchar *s = (const uchar *)p;
+       uint32 new_mflags;
+       unsigned int len;
 
-       if (xflags & XFLG_WORD_SPLIT) {
+       if (mflags & MATCHFLG_WORD_SPLIT) {
                /* Skip over any initial whitespace. */
                while (isspace(*s))
                        s++;
-               /* Update for "!" check. */
+               /* Update to point to real start of rule. */
                p = (const char *)s;
        }
+       if (!*s)
+               return NULL;
+
+       new_mflags = mflags & MATCHFLGS_FROM_CONTAINER;
+
+       /* Figure out what kind of a filter rule "s" is pointing at.  Note
+        * that if MATCHFLG_NO_PREFIXES is set, the rule is either an include
+        * or an exclude based on the inheritance of the MATCHFLG_INCLUDE
+        * flag (above).  XFLG_OLD_PREFIXES indicates a compatibility mode
+        * for old include/exclude patterns where just "+ " and "- " are
+        * allowed as optional prefixes.  */
+       if (mflags & MATCHFLG_NO_PREFIXES) {
+               if (*s == '!' && mflags & MATCHFLG_CVS_IGNORE)
+                       new_mflags |= MATCHFLG_CLEAR_LIST; /* Tentative! */
+       } else if (xflags & XFLG_OLD_PREFIXES) {
+               if (*s == '-' && s[1] == ' ') {
+                       new_mflags &= ~MATCHFLG_INCLUDE;
+                       s += 2;
+               } else if (*s == '+' && s[1] == ' ') {
+                       new_mflags |= MATCHFLG_INCLUDE;
+                       s += 2;
+               } else if (*s == '!')
+                       new_mflags |= MATCHFLG_CLEAR_LIST; /* Tentative! */
+       } else {
+               char ch = 0, *mods = "";
+               switch (*s) {
+               case 'c':
+                       if ((s = RULE_STRCMP(s, "clear")) != NULL)
+                               ch = '!';
+                       break;
+               case 'd':
+                       if ((s = RULE_STRCMP(s, "dir-merge")) != NULL)
+                               ch = ':';
+                       break;
+               case 'e':
+                       if ((s = RULE_STRCMP(s, "exclude")) != NULL)
+                               ch = '-';
+                       break;
+               case 'h':
+                       if ((s = RULE_STRCMP(s, "hide")) != NULL)
+                               ch = 'H';
+                       break;
+               case 'i':
+                       if ((s = RULE_STRCMP(s, "include")) != NULL)
+                               ch = '+';
+                       break;
+               case 'm':
+                       if ((s = RULE_STRCMP(s, "merge")) != NULL)
+                               ch = '.';
+                       break;
+               case 'p':
+                       if ((s = RULE_STRCMP(s, "protect")) != NULL)
+                               ch = 'P';
+                       break;
+               case 'r':
+                       if ((s = RULE_STRCMP(s, "risk")) != NULL)
+                               ch = 'R';
+                       break;
+               case 's':
+                       if ((s = RULE_STRCMP(s, "show")) != NULL)
+                               ch = 'S';
+                       break;
+               default:
+                       ch = *s;
+                       if (s[1] == ',')
+                               s++;
+                       break;
+               }
+               switch (ch) {
+               case ':':
+                       new_mflags |= MATCHFLG_PERDIR_MERGE
+                                   | MATCHFLG_FINISH_SETUP;
+                       /* FALL THROUGH */
+               case '.':
+                       new_mflags |= MATCHFLG_MERGE_FILE;
+                       mods = MODIFIERS_INCL_EXCL MODIFIERS_MERGE_FILE;
+                       break;
+               case '+':
+                       new_mflags |= MATCHFLG_INCLUDE;
+                       /* FALL THROUGH */
+               case '-':
+                       mods = MODIFIERS_INCL_EXCL;
+                       break;
+               case 'S':
+                       new_mflags |= MATCHFLG_INCLUDE;
+                       /* FALL THROUGH */
+               case 'H':
+                       new_mflags |= MATCHFLG_SENDER_SIDE;
+                       mods = MODIFIERS_HIDE_PROTECT;
+                       break;
+               case 'R':
+                       new_mflags |= MATCHFLG_INCLUDE;
+                       /* FALL THROUGH */
+               case 'P':
+                       new_mflags |= MATCHFLG_RECEIVER_SIDE;
+                       mods = MODIFIERS_HIDE_PROTECT;
+                       break;
+               case '!':
+                       new_mflags |= MATCHFLG_CLEAR_LIST;
+                       mods = NULL;
+                       break;
+               default:
+                       rprintf(FERROR, "Unknown filter rule: `%s'\n", p);
+                       exit_cleanup(RERR_SYNTAX);
+               }
+               while (mods && *++s && *s != ' ' && *s != '_') {
+                       if (strchr(mods, *s) == NULL) {
+                               if (mflags & MATCHFLG_WORD_SPLIT && isspace(*s)) {
+                                       s--;
+                                       break;
+                               }
+                           invalid:
+                               rprintf(FERROR,
+                                       "invalid modifier sequence at '%c' in filter rule: %s\n",
+                                       *s, p);
+                               exit_cleanup(RERR_SYNTAX);
+                       }
+                       switch (*s) {
+                       case '-':
+                               if (new_mflags & MATCHFLG_NO_PREFIXES)
+                                   goto invalid;
+                               new_mflags |= MATCHFLG_NO_PREFIXES;
+                               break;
+                       case '+':
+                               if (new_mflags & MATCHFLG_NO_PREFIXES)
+                                   goto invalid;
+                               new_mflags |= MATCHFLG_NO_PREFIXES
+                                           | MATCHFLG_INCLUDE;
+                               break;
+                       case '/':
+                               new_mflags |= MATCHFLG_ABS_PATH;
+                               break;
+                       case '!':
+                               new_mflags |= MATCHFLG_NEGATE;
+                               break;
+                       case 'C':
+                               if (new_mflags & MATCHFLG_NO_PREFIXES)
+                                   goto invalid;
+                               new_mflags |= MATCHFLG_NO_PREFIXES
+                                           | MATCHFLG_WORD_SPLIT
+                                           | MATCHFLG_NO_INHERIT
+                                           | MATCHFLG_CVS_IGNORE;
+                               break;
+                       case 'e':
+                               new_mflags |= MATCHFLG_EXCLUDE_SELF;
+                               break;
+                       case 'n':
+                               new_mflags |= MATCHFLG_NO_INHERIT;
+                               break;
+                       case 'p':
+                               new_mflags |= MATCHFLG_PERISHABLE;
+                               break;
+                       case 'r':
+                               new_mflags |= MATCHFLG_RECEIVER_SIDE;
+                               break;
+                       case 's':
+                               new_mflags |= MATCHFLG_SENDER_SIDE;
+                               break;
+                       case 'w':
+                               new_mflags |= MATCHFLG_WORD_SPLIT;
+                               break;
+                       }
+               }
+               if (*s)
+                       s++;
+       }
 
-       /* Is this a '+' or '-' followed by a space (not whitespace)? */
-       if (!(xflags & XFLG_NO_PREFIXES)
-           && (*s == '-' || *s == '+') && s[1] == ' ') {
-               *incl_ptr = *s == '+';
-               s += 2;
-       } else
-               *incl_ptr = xflags & XFLG_DEF_INCLUDE;
-
-       if (xflags & XFLG_WORD_SPLIT) {
-               const unsigned char *cp = s;
+       if (mflags & MATCHFLG_WORD_SPLIT) {
+               const uchar *cp = s;
                /* Token ends at whitespace or the end of the string. */
                while (!isspace(*cp) && *cp != '\0')
                        cp++;
                len = cp - s;
        } else
-               len = strlen(s);
+               len = strlen((char*)s);
+
+       if (new_mflags & MATCHFLG_CLEAR_LIST) {
+               if (!(mflags & MATCHFLG_NO_PREFIXES)
+                && !(xflags & XFLG_OLD_PREFIXES) && len) {
+                       rprintf(FERROR,
+                               "'!' rule has trailing characters: %s\n", p);
+                       exit_cleanup(RERR_SYNTAX);
+               }
+               if (len > 1)
+                       new_mflags &= ~MATCHFLG_CLEAR_LIST;
+       } else if (!len && !(new_mflags & MATCHFLG_CVS_IGNORE)) {
+               rprintf(FERROR, "unexpected end of filter rule: %s\n", p);
+               exit_cleanup(RERR_SYNTAX);
+       }
 
-       if (*p == '!' && len == 1 && !(xflags & XFLG_NO_PREFIXES))
-               *incl_ptr = -1;
+       /* --delete-excluded turns an un-modified include/exclude into a
+        * sender-side rule.  We also affect per-dir merge files that take
+        * no prefixes as a simple optimization. */
+       if (delete_excluded
+        && !(new_mflags & (MATCHFLG_RECEIVER_SIDE|MATCHFLG_SENDER_SIDE))
+        && (!(new_mflags & MATCHFLG_PERDIR_MERGE)
+         || new_mflags & MATCHFLG_NO_PREFIXES))
+               new_mflags |= MATCHFLG_SENDER_SIDE;
 
        *len_ptr = len;
+       *mflags_ptr = new_mflags;
        return (const char *)s;
 }
 
 
-void add_exclude(struct exclude_struct ***listp, const char *pattern, int xflags)
+static char default_cvsignore[] =
+       /* These default ignored items come from the CVS manual. */
+       "RCS SCCS CVS CVS.adm RCSLOG cvslog.* tags TAGS"
+       " .make.state .nse_depinfo *~ #* .#* ,* _$* *$"
+       " *.old *.bak *.BAK *.orig *.rej .del-*"
+       " *.a *.olb *.o *.obj *.so *.exe"
+       " *.Z *.elc *.ln core"
+       /* The rest we added to suit ourself. */
+       " .svn/ .git/ .bzr/";
+
+static void get_cvs_excludes(uint32 mflags)
 {
-       struct exclude_struct **list = *listp;
-       int pat_len, list_len = 0;
-       int incl, add_cnt = 1;
-       const char *cp;
+       static int initialized = 0;
+       char *p, fname[MAXPATHLEN];
 
-       if (!pattern)
+       if (initialized)
                return;
+       initialized = 1;
 
-       if (xflags & XFLG_WORD_SPLIT) {
-               int add = 0;
-               /* Count maximum extra tokens we might encounter. */
-               for (cp = pattern; *cp; cp++) {
-                       if (isspace(*(unsigned char *)cp)) {
-                               add_cnt += add;
-                               add = 0;
-                       } else
-                               add = 1;
-               }
-       }
+       parse_rule(&cvs_filter_list, default_cvsignore,
+                  mflags | (protocol_version >= 30 ? MATCHFLG_PERISHABLE : 0),
+                  0);
+
+       p = module_id >= 0 && lp_use_chroot(module_id) ? "/" : getenv("HOME");
+       if (p && pathjoin(fname, MAXPATHLEN, p, ".cvsignore") < MAXPATHLEN)
+               parse_filter_file(&cvs_filter_list, fname, mflags, 0);
 
-       cp = get_exclude_tok(pattern, &pat_len, &incl, xflags);
-       if (!pat_len)
+       parse_rule(&cvs_filter_list, getenv("CVSIGNORE"), mflags, 0);
+}
+
+
+void parse_rule(struct filter_list_struct *listp, const char *pattern,
+               uint32 mflags, int xflags)
+{
+       unsigned int pat_len;
+       uint32 new_mflags;
+       const char *cp, *p;
+
+       if (!pattern)
                return;
 
-       /* Check for the special "!" token that clears the list.  Yes, we
-        * only honor it at the start of a XFLG_WORD_SPLIT string. */
-       if (incl < 0) {
-               free_exclude_list(listp);
-               if (!--add_cnt)
-                       return;
-               cp = get_exclude_tok(cp + pat_len, &pat_len, &incl, xflags);
-               if (!pat_len)
-                       return;
-       }
+       while (1) {
+               /* Remember that the returned string is NOT '\0' terminated! */
+               cp = parse_rule_tok(pattern, mflags, xflags,
+                                   &pat_len, &new_mflags);
+               if (!cp)
+                       break;
 
-       if (list)
-               for ( ; list[list_len]; list_len++) {}
+               pattern = cp + pat_len;
 
-       list = *listp = realloc_array(list, struct exclude_struct *,
-                                     list_len + add_cnt + 1);
-       if (!list)
-               out_of_memory("add_exclude");
+               if (pat_len >= MAXPATHLEN) {
+                       rprintf(FERROR, "discarding over-long filter: %.*s\n",
+                               (int)pat_len, cp);
+                       continue;
+               }
 
-       while (pat_len) {
-               list[list_len++] = make_exclude(cp, pat_len, incl);
+               if (new_mflags & MATCHFLG_CLEAR_LIST) {
+                       if (verbose > 2) {
+                               rprintf(FINFO,
+                                       "[%s] clearing filter list%s\n",
+                                       who_am_i(), listp->debug_type);
+                       }
+                       clear_filter_list(listp);
+                       continue;
+               }
 
-               if (verbose > 2) {
-                       rprintf(FINFO, "[%s] add_exclude(%s,%s)\n",
-                               who_am_i(), cp,
-                               incl ? "include" : "exclude");
+               if (new_mflags & MATCHFLG_MERGE_FILE) {
+                       unsigned int len;
+                       if (!pat_len) {
+                               cp = ".cvsignore";
+                               pat_len = 10;
+                       }
+                       len = pat_len;
+                       if (new_mflags & MATCHFLG_EXCLUDE_SELF) {
+                               const char *name = cp + len;
+                               while (name > cp && name[-1] != '/') name--;
+                               len -= name - cp;
+                               add_rule(listp, name, len, 0, 0);
+                               new_mflags &= ~MATCHFLG_EXCLUDE_SELF;
+                               len = pat_len;
+                       }
+                       if (new_mflags & MATCHFLG_PERDIR_MERGE) {
+                               if (parent_dirscan) {
+                                       if (!(p = parse_merge_name(cp, &len,
+                                                               module_dirlen)))
+                                               continue;
+                                       add_rule(listp, p, len, new_mflags, 0);
+                                       continue;
+                               }
+                       } else {
+                               if (!(p = parse_merge_name(cp, &len, 0)))
+                                       continue;
+                               parse_filter_file(listp, p, new_mflags,
+                                                 XFLG_FATAL_ERRORS);
+                               continue;
+                       }
                }
-               cp = get_exclude_tok(cp + pat_len, &pat_len, &incl, xflags);
-       }
 
-       list[list_len] = NULL;
+               add_rule(listp, cp, pat_len, new_mflags, xflags);
+
+               if (new_mflags & MATCHFLG_CVS_IGNORE
+                   && !(new_mflags & MATCHFLG_MERGE_FILE))
+                       get_cvs_excludes(new_mflags);
+       }
 }
 
 
-void add_exclude_file(struct exclude_struct ***listp, const char *fname,
-                     int xflags)
+void parse_filter_file(struct filter_list_struct *listp, const char *fname,
+                      uint32 mflags, int xflags)
 {
        FILE *fp;
-       char line[MAXPATHLEN];
-       char *eob = line + MAXPATHLEN - 1;
-       extern int eol_nulls;
+       char line[BIGPATHBUFLEN];
+       char *eob = line + sizeof line - 1;
+       int word_split = mflags & MATCHFLG_WORD_SPLIT;
 
        if (!fname || !*fname)
                return;
 
-       if (*fname != '-' || fname[1])
-               fp = fopen(fname, "rb");
-       else
+       if (*fname != '-' || fname[1] || am_server) {
+               if (daemon_filter_list.head) {
+                       strlcpy(line, fname, sizeof line);
+                       clean_fname(line, CFN_COLLAPSE_DOT_DOT_DIRS);
+                       if (check_filter(&daemon_filter_list, line, 0) < 0)
+                               fp = NULL;
+                       else
+                               fp = fopen(line, "rb");
+               } else
+                       fp = fopen(fname, "rb");
+       } else
                fp = stdin;
+
+       if (verbose > 2) {
+               rprintf(FINFO, "[%s] parse_filter_file(%s,%x,%x)%s\n",
+                       who_am_i(), fname, mflags, xflags,
+                       fp ? "" : " [not found]");
+       }
+
        if (!fp) {
                if (xflags & XFLG_FATAL_ERRORS) {
                        rsyserr(FERROR, errno,
-                               "failed to open %s file %s",
-                               xflags & XFLG_DEF_INCLUDE ? "include" : "exclude",
+                               "failed to open %sclude file %s",
+                               mflags & MATCHFLG_INCLUDE ? "in" : "ex",
                                fname);
                        exit_cleanup(RERR_FILEIO);
                }
                return;
        }
+       dirbuf[dirbuf_len] = '\0';
 
        while (1) {
                char *s = line;
-               int ch;
+               int ch, overflow = 0;
                while (1) {
                        if ((ch = getc(fp)) == EOF) {
-                               if (ferror(fp) && errno == EINTR)
+                               if (ferror(fp) && errno == EINTR) {
+                                       clearerr(fp);
                                        continue;
+                               }
                                break;
                        }
+                       if (word_split && isspace(ch))
+                               break;
                        if (eol_nulls? !ch : (ch == '\n' || ch == '\r'))
                                break;
                        if (s < eob)
                                *s++ = ch;
+                       else
+                               overflow = 1;
                }
-               *s = '\0';
-               if (*line && *line != ';' && *line != '#') {
-                       /* Skip lines starting with semicolon or pound.
-                        * It probably wouldn't cause any harm to not skip
-                        * them but there's no need to save them. */
-                       add_exclude(listp, line, xflags);
+               if (overflow) {
+                       rprintf(FERROR, "discarding over-long filter: %s...\n", line);
+                       s = line;
                }
+               *s = '\0';
+               /* Skip an empty token and (when line parsing) comments. */
+               if (*line && (word_split || (*line != ';' && *line != '#')))
+                       parse_rule(listp, line, mflags, xflags);
                if (ch == EOF)
                        break;
        }
        fclose(fp);
 }
 
-
-void send_exclude_list(int f)
+/* If the "for_xfer" flag is set, the prefix is made compatible with the
+ * current protocol_version (if possible) or a NULL is returned (if not
+ * possible). */
+char *get_rule_prefix(int match_flags, const char *pat, int for_xfer,
+                     unsigned int *plen_ptr)
 {
-       int i;
-       extern int list_only, recurse;
-
-       /* This is a complete hack - blame Rusty.
-        *
-        * FIXME: This pattern shows up in the output of
-        * report_exclude_result(), which is not ideal. */
-       if (list_only && !recurse)
-               add_exclude(&exclude_list, "/*/*", 0);
-
-       if (!exclude_list) {
-               write_int(f, 0);
-               return;
+       static char buf[MAX_RULE_PREFIX+1];
+       char *op = buf;
+       int legal_len = for_xfer && protocol_version < 29 ? 1 : MAX_RULE_PREFIX-1;
+
+       if (match_flags & MATCHFLG_PERDIR_MERGE) {
+               if (legal_len == 1)
+                       return NULL;
+               *op++ = ':';
+       } else if (match_flags & MATCHFLG_INCLUDE)
+               *op++ = '+';
+       else if (legal_len != 1
+           || ((*pat == '-' || *pat == '+') && pat[1] == ' '))
+               *op++ = '-';
+       else
+               legal_len = 0;
+
+       if (match_flags & MATCHFLG_NEGATE)
+               *op++ = '!';
+       if (match_flags & MATCHFLG_CVS_IGNORE)
+               *op++ = 'C';
+       else {
+               if (match_flags & MATCHFLG_NO_INHERIT)
+                       *op++ = 'n';
+               if (match_flags & MATCHFLG_WORD_SPLIT)
+                       *op++ = 'w';
+               if (match_flags & MATCHFLG_NO_PREFIXES) {
+                       if (match_flags & MATCHFLG_INCLUDE)
+                               *op++ = '+';
+                       else
+                               *op++ = '-';
+               }
        }
+       if (match_flags & MATCHFLG_EXCLUDE_SELF)
+               *op++ = 'e';
+       if (match_flags & MATCHFLG_SENDER_SIDE
+           && (!for_xfer || protocol_version >= 29))
+               *op++ = 's';
+       if (match_flags & MATCHFLG_RECEIVER_SIDE
+           && (!for_xfer || protocol_version >= 29
+            || (delete_excluded && am_sender)))
+               *op++ = 'r';
+       if (match_flags & MATCHFLG_PERISHABLE) {
+               if (!for_xfer || protocol_version >= 30)
+                       *op++ = 'p';
+               else if (am_sender)
+                       return NULL;
+       }
+       if (op - buf > legal_len)
+               return NULL;
+       if (legal_len)
+               *op++ = ' ';
+       *op = '\0';
+       if (plen_ptr)
+               *plen_ptr = op - buf;
+       return buf;
+}
 
-       for (i = 0; exclude_list[i]; i++) {
-               unsigned int l;
-               char p[MAXPATHLEN+1];
-
-               l = strlcpy(p, exclude_list[i]->pattern, sizeof p);
-               if (l == 0 || l >= MAXPATHLEN)
+static void send_rules(int f_out, struct filter_list_struct *flp)
+{
+       struct filter_struct *ent, *prev = NULL;
+
+       for (ent = flp->head; ent; ent = ent->next) {
+               unsigned int len, plen, dlen;
+               int elide = 0;
+               char *p;
+
+               /* Note we need to check delete_excluded here in addition to
+                * the code in parse_rule_tok() because some rules may have
+                * been added before we found the --delete-excluded option.
+                * We must also elide any CVS merge-file rules to avoid a
+                * backward compatibility problem, and we elide any no-prefix
+                * merge files as an optimization (since they can only have
+                * include/exclude rules). */
+               if (ent->match_flags & MATCHFLG_SENDER_SIDE)
+                       elide = am_sender ? 1 : -1;
+               if (ent->match_flags & MATCHFLG_RECEIVER_SIDE)
+                       elide = elide ? 0 : am_sender ? -1 : 1;
+               else if (delete_excluded && !elide
+                && (!(ent->match_flags & MATCHFLG_PERDIR_MERGE)
+                 || ent->match_flags & MATCHFLG_NO_PREFIXES))
+                       elide = am_sender ? 1 : -1;
+               if (elide < 0) {
+                       if (prev)
+                               prev->next = ent->next;
+                       else
+                               flp->head = ent->next;
+               } else
+                       prev = ent;
+               if (elide > 0)
                        continue;
-               if (exclude_list[i]->directory) {
-                       p[l++] = '/';
-                       p[l] = '\0';
+               if (ent->match_flags & MATCHFLG_CVS_IGNORE
+                   && !(ent->match_flags & MATCHFLG_MERGE_FILE)) {
+                       int f = am_sender || protocol_version < 29 ? f_out : -2;
+                       send_rules(f, &cvs_filter_list);
+                       if (f == f_out)
+                               continue;
                }
-
-               if (exclude_list[i]->include) {
-                       write_int(f, l + 2);
-                       write_buf(f, "+ ", 2);
-               } else if ((*p == '-' || *p == '+') && p[1] == ' ') {
-                       write_int(f, l + 2);
-                       write_buf(f, "- ", 2);
-               } else
-                       write_int(f, l);
-               write_buf(f, p, l);
+               p = get_rule_prefix(ent->match_flags, ent->pattern, 1, &plen);
+               if (!p) {
+                       rprintf(FERROR,
+                               "filter rules are too modern for remote rsync.\n");
+                       exit_cleanup(RERR_PROTOCOL);
+               }
+               if (f_out < 0)
+                       continue;
+               len = strlen(ent->pattern);
+               dlen = ent->match_flags & MATCHFLG_DIRECTORY ? 1 : 0;
+               if (!(plen + len + dlen))
+                       continue;
+               write_int(f_out, plen + len + dlen);
+               if (plen)
+                       write_buf(f_out, p, plen);
+               write_buf(f_out, ent->pattern, len);
+               if (dlen)
+                       write_byte(f_out, '/');
        }
-
-       write_int(f, 0);
+       flp->tail = prev;
 }
 
-
-void recv_exclude_list(int f)
+/* This is only called by the client. */
+void send_filter_list(int f_out)
 {
-       char line[MAXPATHLEN+1]; /* Allows a trailing slash on a max-len dir */
-       unsigned int l;
-
-       while ((l = read_int(f)) != 0) {
-               if (l >= sizeof line)
-                       overflow("recv_exclude_list");
-               read_sbuf(f, line, l);
-               add_exclude(&exclude_list, line, 0);
+       int receiver_wants_list = prune_empty_dirs
+           || (delete_mode && (!delete_excluded || protocol_version >= 29));
+
+       if (local_server || (am_sender && !receiver_wants_list))
+               f_out = -1;
+       if (cvs_exclude && am_sender) {
+               if (protocol_version >= 29)
+                       parse_rule(&filter_list, ":C", 0, 0);
+               parse_rule(&filter_list, "-C", 0, 0);
        }
-}
 
+       send_rules(f_out, &filter_list);
 
-static char default_cvsignore[] = 
-       /* These default ignored items come from the CVS manual. */
-       "RCS SCCS CVS CVS.adm RCSLOG cvslog.* tags TAGS"
-       " .make.state .nse_depinfo *~ #* .#* ,* _$* *$"
-       " *.old *.bak *.BAK *.orig *.rej .del-*"
-       " *.a *.olb *.o *.obj *.so *.exe"
-       " *.Z *.elc *.ln core"
-       /* The rest we added to suit ourself. */
-       " .svn/";
+       if (f_out >= 0)
+               write_int(f_out, 0);
 
-void add_cvs_excludes(void)
-{
-       char fname[MAXPATHLEN];
-       char *p;
+       if (cvs_exclude) {
+               if (!am_sender || protocol_version < 29)
+                       parse_rule(&filter_list, ":C", 0, 0);
+               if (!am_sender)
+                       parse_rule(&filter_list, "-C", 0, 0);
+       }
+}
 
-       add_exclude(&exclude_list, default_cvsignore,
-                   XFLG_WORD_SPLIT | XFLG_NO_PREFIXES);
+/* This is only called by the server. */
+void recv_filter_list(int f_in)
+{
+       char line[BIGPATHBUFLEN];
+       int xflags = protocol_version >= 29 ? 0 : XFLG_OLD_PREFIXES;
+       int receiver_wants_list = prune_empty_dirs
+           || (delete_mode
+            && (!delete_excluded || protocol_version >= 29));
+       unsigned int len;
+
+       if (!local_server && (am_sender || receiver_wants_list)) {
+               while ((len = read_int(f_in)) != 0) {
+                       if (len >= sizeof line)
+                               overflow_exit("recv_rules");
+                       read_sbuf(f_in, line, len);
+                       parse_rule(&filter_list, line, 0, xflags);
+               }
+       }
 
-       if ((p = getenv("HOME"))
-           && pathjoin(fname, sizeof fname, p, ".cvsignore") < sizeof fname) {
-               add_exclude_file(&exclude_list, fname,
-                                XFLG_WORD_SPLIT | XFLG_NO_PREFIXES);
+       if (cvs_exclude) {
+               if (local_server || am_sender || protocol_version < 29)
+                       parse_rule(&filter_list, ":C", 0, 0);
+               if (local_server || am_sender)
+                       parse_rule(&filter_list, "-C", 0, 0);
        }
 
-       add_exclude(&exclude_list, getenv("CVSIGNORE"),
-                   XFLG_WORD_SPLIT | XFLG_NO_PREFIXES);
+       if (local_server) /* filter out any rules that aren't for us. */
+               send_rules(-1, &filter_list);
 }