This builds on the checksum-reading patch and adds the ability to create and/or update the .rsyncsums files using extended mode args to the --sumfiles=MODE option and the "checksum files = MODE" daemon parameter. CAUTION: This patch is only lightly tested. If you're interested in using it, please help out. To use this patch, run these commands for a successful build: patch -p1 to_redo) + static struct csum_cache { struct file_list *flist; + const char *dirname; + int checksum_matches; + int checksum_updates; } *csum_cache = NULL; static void flist_sort_and_clean(struct file_list *flist, int flags); @@ -352,7 +361,79 @@ static void flist_done_allocating(struct file_list *flist) flist->pool_boundary = ptr; } -void reset_checksum_cache() +static void checksum_filename(int slot, const char *dirname, char *fbuf) +{ + if (dirname && *dirname) { + unsigned int len; + if (slot) { + len = strlcpy(fbuf, basis_dir[slot-1], MAXPATHLEN); + if (len >= MAXPATHLEN) + return; + } else + len = 0; + if (pathjoin(fbuf+len, MAXPATHLEN-len, dirname, RSYNCSUMS_FILE) >= MAXPATHLEN-len) + return; + } else + strlcpy(fbuf, RSYNCSUMS_FILE, MAXPATHLEN); +} + +static void write_checksums(int slot, struct file_list *flist, int whole_dir) +{ + int i; + FILE *out_fp; + char fbuf[MAXPATHLEN]; + int new_entries = csum_cache[slot].checksum_updates != 0; + int counts_match = flist->used == csum_cache[slot].checksum_matches; + int no_skipped = whole_dir && REGULAR_SKIPPED(flist) == 0; + const char *dirname = csum_cache[slot].dirname; + + flist_sort_and_clean(flist, 0); + + if (dry_run && !(checksum_files & CSF_AFFECT_DRYRUN)) + return; + + checksum_filename(slot, dirname, fbuf); + + if (flist->high - flist->low < 0 && no_skipped) { + unlink(fbuf); + return; + } + + if (!new_entries && (counts_match || !whole_dir)) + return; + + if (!(out_fp = fopen(fbuf, "w"))) + return; + + for (i = flist->low; i <= flist->high; i++) { + struct file_struct *file = flist->sorted[i]; + const char *cp = F_SUM(file); + const char *end = cp + checksum_len; + const char *alt_sum = file->basename + strlen(file->basename) + 1; + if (whole_dir && !(file->flags & FLAG_SUM_KEEP)) + continue; + if (protocol_version >= 30) + fprintf(out_fp, "%s ", alt_sum); + if (file->flags & FLAG_SUM_MISSING) { + do { + fputs("==", out_fp); + } while (++cp != end); + } else { + do { + fprintf(out_fp, "%02x", (int)CVAL(cp, 0)); + } while (++cp != end); + } + if (protocol_version < 30) + fprintf(out_fp, " %s", alt_sum); + fprintf(out_fp, " %10.0f %10.0f %10lu %10lu %s\n", + (double)F_LENGTH(file), (double)file->modtime, + (long)F_CTIME(file), (long)F_INODE(file), file->basename); + } + + fclose(out_fp); +} + +void reset_checksum_cache(int whole_dir) { int slot, slots = am_sender ? 1 : basis_dir_cnt + 1; @@ -366,6 +447,9 @@ void reset_checksum_cache() struct file_list *flist = csum_cache[slot].flist; if (flist) { + if (checksum_files & CSF_UPDATE && flist->next) + write_checksums(slot, flist, whole_dir); + /* Reset the pool memory and empty the file-list array. */ pool_free_old(flist->file_pool, pool_boundary(flist->file_pool, 0)); @@ -376,6 +460,10 @@ void reset_checksum_cache() flist->low = 0; flist->high = -1; flist->next = NULL; + + csum_cache[slot].checksum_matches = 0; + csum_cache[slot].checksum_updates = 0; + REGULAR_SKIPPED(flist) = 0; } } @@ -383,7 +471,7 @@ void reset_checksum_cache() static int add_checksum(struct file_list *flist, const char *dirname, const char *basename, int basename_len, OFF_T file_length, time_t mtime, uint32 ctime, uint32 inode, - const char *sum) + const char *sum, const char *alt_sum, int flags) { struct file_struct *file; int alloc_len, extra_len; @@ -400,7 +488,7 @@ static int add_checksum(struct file_list *flist, const char *dirname, if (extra_len & (EXTRA_ROUNDING * EXTRA_LEN)) extra_len = (extra_len | (EXTRA_ROUNDING * EXTRA_LEN)) + EXTRA_LEN; #endif - alloc_len = FILE_STRUCT_LEN + extra_len + basename_len; + alloc_len = FILE_STRUCT_LEN + extra_len + basename_len + checksum_len*2 + 1; bp = pool_alloc(flist->file_pool, alloc_len, "add_checksum"); memset(bp, 0, extra_len + FILE_STRUCT_LEN); @@ -409,7 +497,14 @@ static int add_checksum(struct file_list *flist, const char *dirname, bp += FILE_STRUCT_LEN; memcpy(bp, basename, basename_len); + if (alt_sum) + strlcpy(bp+basename_len, alt_sum, checksum_len*2 + 1); + else { + memset(bp+basename_len, '=', checksum_len*2); + bp[basename_len+checksum_len*2] = '\0'; + } + file->flags = flags; file->mode = S_IFREG; file->modtime = mtime; file->len32 = (uint32)file_length; @@ -438,10 +533,11 @@ static void read_checksums(int slot, struct file_list *flist, const char *dirnam char line[MAXPATHLEN+1024], fbuf[MAXPATHLEN], sum[MAX_DIGEST_LEN]; FILE *fp; char *cp; - int len, i; time_t mtime; + int len, i, flags; OFF_T file_length; uint32 ctime, inode; + const char *alt_sum = NULL; int dlen = dirname ? strlcpy(fbuf, dirname, sizeof fbuf) : 0; if (dlen >= (int)(sizeof fbuf - 1 - RSYNCSUMS_LEN)) @@ -462,7 +558,7 @@ static void read_checksums(int slot, struct file_list *flist, const char *dirnam while (fgets(line, sizeof line, fp)) { cp = line; if (protocol_version >= 30) { - char *alt_sum = cp; + alt_sum = cp; if (*cp == '=') while (*++cp == '=') {} else @@ -473,7 +569,14 @@ static void read_checksums(int slot, struct file_list *flist, const char *dirnam } if (*cp == '=') { - continue; + for (i = 0; i < checksum_len*2; i++, cp++) { + if (*cp != '=') { + cp = ""; + break; + } + } + memset(sum, 0, checksum_len); + flags = FLAG_SUM_MISSING; } else { for (i = 0; i < checksum_len*2; i++, cp++) { int x; @@ -491,13 +594,14 @@ static void read_checksums(int slot, struct file_list *flist, const char *dirnam else sum[i/2] = x << 4; } + flags = 0; } if (*cp != ' ') break; while (*++cp == ' ') {} if (protocol_version < 30) { - char *alt_sum = cp; + alt_sum = cp; if (*cp == '=') while (*++cp == '=') {} else @@ -547,24 +651,112 @@ static void read_checksums(int slot, struct file_list *flist, const char *dirnam continue; strlcpy(fbuf+dlen, cp, sizeof fbuf - dlen); + if (is_excluded(fbuf, 0, ALL_FILTERS)) { + flags |= FLAG_SUM_KEEP; + csum_cache[slot].checksum_matches++; + } add_checksum(flist, dirname, cp, len, file_length, mtime, ctime, inode, - sum); + sum, alt_sum, flags); } fclose(fp); flist_sort_and_clean(flist, CLEAN_KEEP_LAST); } +void set_cached_checksum(struct file_list *file_flist, struct file_struct *file) +{ + int j; + FILE *out_fp; + STRUCT_STAT st; + char fbuf[MAXPATHLEN]; + const char *fn = f_name(file, NULL); + struct file_list *flist = csum_cache[0].flist; + + if (dry_run && !(checksum_files & CSF_AFFECT_DRYRUN)) + return; + + if (stat(fn, &st) < 0) + return; + + checksum_filename(0, file->dirname, fbuf); + + if (file_flist != flist->next) { + const char *cp = F_SUM(file); + const char *end = cp + checksum_len; + + if (!(out_fp = fopen(fbuf, "a"))) + return; + + if (protocol_version >= 30) { + for (j = 0; j < checksum_len; j++) + fputs("==", out_fp); + fputc(' ', out_fp); + } + do { + fprintf(out_fp, "%02x", (int)CVAL(cp, 0)); + } while (++cp != end); + if (protocol_version < 30) { + fputc(' ', out_fp); + for (j = 0; j < checksum_len; j++) + fputs("==", out_fp); + } + fprintf(out_fp, " %10.0f %10.0f %10lu %10lu %s\n", + (double)st.st_size, (double)st.st_mtime, + (long)(uint32)st.st_ctime, (long)(uint32)st.st_ino, + file->basename); + + fclose(out_fp); + return; + } + + if ((j = flist_find(flist, file)) >= 0) { + struct file_struct *fp = flist->sorted[j]; + int inc = 0; + if (F_LENGTH(fp) != st.st_size) { + fp->len32 = (uint32)st.st_size; + if (st.st_size > 0xFFFFFFFFu) { + OPT_EXTRA(fp, 0)->unum = (uint32)(st.st_size >> 32); + fp->flags |= FLAG_LENGTH64; + } else + fp->flags &= FLAG_LENGTH64; + inc = 1; + } + if (fp->modtime != st.st_mtime) { + fp->modtime = st.st_mtime; + inc = 1; + } + if (F_CTIME(fp) != (uint32)st.st_ctime) { + F_CTIME(fp) = (uint32)st.st_ctime; + inc = 1; + } + if (F_INODE(fp) != (uint32)st.st_ino) { + F_INODE(fp) = (uint32)st.st_ino; + inc = 1; + } + memcpy(F_SUM(fp), F_SUM(file), MAX_DIGEST_LEN); + csum_cache[0].checksum_updates += inc; + fp->flags &= ~FLAG_SUM_MISSING; + fp->flags |= FLAG_SUM_KEEP; + return; + } + + csum_cache[0].checksum_updates += + add_checksum(flist, file->dirname, file->basename, strlen(file->basename) + 1, + st.st_size, (uint32)st.st_mtime, (uint32)st.st_ctime, + st.st_ino, F_SUM(file), NULL, FLAG_SUM_KEEP); +} + void get_cached_checksum(int slot, const char *fname, struct file_struct *file, - STRUCT_STAT *stp, char *sum_buf) + int basename_len, STRUCT_STAT *stp, char *sum_buf) { struct file_list *flist = csum_cache[slot].flist; int j; if (!flist->next) { flist->next = cur_flist; /* next points from checksum flist to file flist */ + csum_cache[slot].dirname = file->dirname; read_checksums(slot, flist, file->dirname); } @@ -576,12 +768,31 @@ void get_cached_checksum(int slot, const char *fname, struct file_struct *file, && (checksum_files & CSF_LAX || (F_CTIME(fp) == (uint32)stp->st_ctime && F_INODE(fp) == (uint32)stp->st_ino))) { - memcpy(sum_buf, F_SUM(fp), MAX_DIGEST_LEN); + if (fp->flags & FLAG_SUM_MISSING) { + fp->flags &= ~FLAG_SUM_MISSING; + csum_cache[slot].checksum_updates++; + file_checksum(fname, stp->st_size, sum_buf); + memcpy(F_SUM(fp), sum_buf, MAX_DIGEST_LEN); + } else { + csum_cache[slot].checksum_matches++; + memcpy(sum_buf, F_SUM(fp), MAX_DIGEST_LEN); + } + fp->flags |= FLAG_SUM_KEEP; return; } + clear_file(fp); } file_checksum(fname, stp->st_size, sum_buf); + + if (checksum_files & CSF_UPDATE) { + if (basename_len < 0) + basename_len = strlen(file->basename) + 1; + csum_cache[slot].checksum_updates += + add_checksum(flist, file->dirname, file->basename, basename_len, + stp->st_size, stp->st_mtime, (uint32)stp->st_ctime, + (uint32)stp->st_ino, sum_buf, NULL, FLAG_SUM_KEEP); + } } /* Call this with EITHER (1) "file, NULL, 0" to chdir() to the file's @@ -1407,6 +1618,8 @@ struct file_struct *make_file(const char *fname, struct file_list *flist, if (is_excluded(thisname, S_ISDIR(st.st_mode) != 0, filter_level)) { if (ignore_perishable) non_perishable_cnt++; + if (S_ISREG(st.st_mode)) + REGULAR_SKIPPED(flist)++; return NULL; } @@ -1453,13 +1666,13 @@ struct file_struct *make_file(const char *fname, struct file_list *flist, lastdir[len] = '\0'; lastdir_len = len; if (checksum_files && am_sender && flist) - reset_checksum_cache(); + reset_checksum_cache(0); } } else { basename = thisname; if (checksum_files && am_sender && flist && lastdir_len == -2) { lastdir_len = -1; - reset_checksum_cache(); + reset_checksum_cache(0); } } basename_len = strlen(basename) + 1; /* count the '\0' */ @@ -1545,7 +1758,7 @@ struct file_struct *make_file(const char *fname, struct file_list *flist, if (always_checksum && am_sender && S_ISREG(st.st_mode)) { if (flist && checksum_files) - get_cached_checksum(0, thisname, file, &st, tmp_sum); + get_cached_checksum(0, thisname, file, basename_len, &st, tmp_sum); else file_checksum(thisname, st.st_size, tmp_sum); } @@ -1898,6 +2111,9 @@ static void send_directory(int f, struct file_list *flist, char *fbuf, int len, closedir(d); + if (checksum_files & CSF_UPDATE && am_sender && f >= 0) + reset_checksum_cache(1); + if (f >= 0 && recurse && !divert_dirs) { int i, end = flist->used - 1; /* send_if_directory() bumps flist->used, so use "end". */ @@ -2495,6 +2711,9 @@ struct file_list *send_file_list(int f, int argc, char *argv[]) } else flist_eof = 1; + if (checksum_files & CSF_UPDATE && flist_eof) + reset_checksum_cache(0); /* writes any last updates */ + return flist; } diff --git a/generator.c b/generator.c --- a/generator.c +++ b/generator.c @@ -114,6 +114,7 @@ static int dir_tweaking; static int symlink_timeset_failed_flags; static int need_retouch_dir_times; static int need_retouch_dir_perms; +static int started_whole_dir, upcoming_whole_dir; static const char *solo_file = NULL; /* For calling delete_item() and delete_dir_contents(). */ @@ -726,7 +727,7 @@ int unchanged_file(char *fn, struct file_struct *file, STRUCT_STAT *st, int slot if (always_checksum > 0 && S_ISREG(st->st_mode)) { char sum[MAX_DIGEST_LEN]; if (checksum_files && slot >= 0) - get_cached_checksum(slot, fn, file, st, sum); + get_cached_checksum(slot, fn, file, -1, st, sum); else file_checksum(fn, st->st_size, sum); return memcmp(sum, F_SUM(file), checksum_len) == 0; @@ -1369,7 +1370,8 @@ static void recv_generator(char *fname, struct file_struct *file, int ndx, fuzzy_dirlist = get_dirlist(fnamecmpbuf, -1, 1); } if (checksum_files) { - reset_checksum_cache(); + reset_checksum_cache(started_whole_dir); + started_whole_dir = upcoming_whole_dir; } need_new_dirscan = 0; } @@ -1519,6 +1521,7 @@ static void recv_generator(char *fname, struct file_struct *file, int ndx, else change_local_filter_dir(fname, strlen(fname), F_DEPTH(file)); } + upcoming_whole_dir = file->flags & FLAG_CONTENT_DIR && f_out != -1 ? 1 : 0; goto cleanup; } @@ -1811,6 +1814,8 @@ static void recv_generator(char *fname, struct file_struct *file, int ndx, handle_partial_dir(partialptr, PDIR_DELETE); } set_file_attrs(fname, file, &sx, NULL, maybe_ATTRS_REPORT); + if (checksum_files & CSF_UPDATE) + set_cached_checksum(cur_flist, file); if (itemizing) itemize(fnamecmp, file, ndx, statret, &sx, 0, 0, NULL); #ifdef SUPPORT_HARD_LINKS @@ -2252,6 +2257,7 @@ void generate_files(int f_out, const char *local_name) } else change_local_filter_dir(fbuf, strlen(fbuf), F_DEPTH(fp)); } + upcoming_whole_dir = fp->flags & FLAG_CONTENT_DIR ? 1 : 0; } for (i = cur_flist->low; i <= cur_flist->high; i++) { struct file_struct *file = cur_flist->sorted[i]; @@ -2338,6 +2344,9 @@ void generate_files(int f_out, const char *local_name) wait_for_receiver(); } + if (checksum_files) + reset_checksum_cache(started_whole_dir); + do_progress = save_do_progress; if (delete_during == 2) do_delayed_deletions(fbuf); diff --git a/io.c b/io.c --- a/io.c +++ b/io.c @@ -50,6 +50,7 @@ extern int read_batch; extern int csum_length; extern int protect_args; extern int checksum_seed; +extern int checksum_files; extern int protocol_version; extern int remove_source_files; extern int preserve_hard_links; @@ -161,6 +162,9 @@ static void got_flist_entry_status(enum festatus status, const char *buf) flist_ndx_push(&hlink_list, ndx); flist->in_progress++; } + } else if (checksum_files & CSF_UPDATE) { + struct file_struct *file = flist->files[ndx - flist->ndx_start]; + set_cached_checksum(flist, file); } break; case FES_REDO: diff --git a/loadparm.c b/loadparm.c --- a/loadparm.c +++ b/loadparm.c @@ -300,6 +300,10 @@ static struct enum_list enum_csum_modes[] = { { CSF_IGNORE_FILES, "none" }, { CSF_LAX_MODE, "lax" }, { CSF_STRICT_MODE, "strict" }, + { CSF_LAX_MODE|CSF_UPDATE, "+lax" }, + { CSF_STRICT_MODE|CSF_UPDATE, "+strict" }, + { CSF_LAX_MODE|CSF_UPDATE|CSF_AFFECT_DRYRUN, "++lax" }, + { CSF_STRICT_MODE|CSF_UPDATE|CSF_AFFECT_DRYRUN, "++strict" }, { -1, NULL } }; diff --git a/options.c b/options.c --- a/options.c +++ b/options.c @@ -1234,7 +1234,15 @@ int parse_arguments(int *argc_p, const char ***argv_p) case OPT_SUMFILES: arg = poptGetOptArg(pc); - checksum_files = 0; + if (*arg == '+') { + arg++; + checksum_files = CSF_UPDATE; + if (*arg == '+') { + arg++; + checksum_files |= CSF_AFFECT_DRYRUN; + } + } else + checksum_files = 0; if (strcmp(arg, "lax") == 0) checksum_files |= CSF_LAX_MODE; else if (strcmp(arg, "strict") == 0) diff --git a/receiver.c b/receiver.c --- a/receiver.c +++ b/receiver.c @@ -47,6 +47,7 @@ extern int append_mode; extern int sparse_files; extern int keep_partial; extern int checksum_seed; +extern int checksum_files; extern int inplace; extern int delay_updates; extern mode_t orig_umask; @@ -343,7 +344,7 @@ static void handle_delayed_updates(char *local_name) "rename failed for %s (from %s)", full_fname(fname), partialptr); } else { - if (remove_source_files + if (remove_source_files || checksum_files & CSF_UPDATE || (preserve_hard_links && F_IS_HLINKED(file))) send_msg_int(MSG_SUCCESS, ndx); handle_partial_dir(partialptr, PDIR_DELETE); @@ -758,7 +759,7 @@ int recv_files(int f_in, char *local_name) case 2: break; case 1: - if (remove_source_files || inc_recurse + if (remove_source_files || inc_recurse || checksum_files & CSF_UPDATE || (preserve_hard_links && F_IS_HLINKED(file))) send_msg_int(MSG_SUCCESS, ndx); break; diff --git a/rsync.h b/rsync.h --- a/rsync.h +++ b/rsync.h @@ -885,6 +885,8 @@ typedef struct { #define CSF_ENABLE (1<<1) #define CSF_LAX (1<<2) +#define CSF_UPDATE (1<<3) +#define CSF_AFFECT_DRYRUN (1<<4) #define CSF_IGNORE_FILES 0 #define CSF_LAX_MODE (CSF_ENABLE|CSF_LAX) diff --git a/rsync.yo b/rsync.yo --- a/rsync.yo +++ b/rsync.yo @@ -548,9 +548,13 @@ computed just as it would be if bf(--sumfiles) was not specified. The MODE value is either "lax", for relaxed checking (which compares size and mtime), "strict" (which also compares ctime and inode), or "none" to -ignore any .rsyncsums files ("none" is the default). Rsync does not create -or update these files, but there is a perl script in the support directory -named "rsyncsums" that can be used for that. +ignore any .rsyncsums files ("none" is the default). +If you want rsync to create and/or update these files, specify a prefixed +plus ("+lax" or "+strict"). +Adding a second prefixed '+' causes the checksum-file updates to happen +even when the transfer is in bf(--dry-run) mode ("++lax" or "++strict"). +There is also a perl script in the support directory named "rsyncsums" +that can be used to update the .rsyncsums files. This option has no effect unless bf(--checksum, -c) was also specified. It also only affects the current side of the transfer, so if you want the diff --git a/rsyncd.conf.yo b/rsyncd.conf.yo --- a/rsyncd.conf.yo +++ b/rsyncd.conf.yo @@ -284,13 +284,15 @@ The default is tt(/var/run/rsyncd.lock). dit(bf(checksum files)) This parameter tells rsync to make use of any cached checksum information it finds in per-directory .rsyncsums files when the current transfer is using the bf(--checksum) option. The value can be set -to either "lax", "strict", or "none" -- see the client's bf(--sumfiles) -option for what these choices do. +to either "lax", "strict", "+lax", "+strict", "++lax", "++strict", or +"none". See the client's bf(--sumfiles) option for what these choices do. Note also that the client's command-line option, bf(--sumfiles), has no effect on a daemon. A daemon will only access checksum files if this -config option tells it to. See also the bf(exclude) directive for a way -to hide the .rsyncsums files from the user. +config option tells it to. You can configure updating of the .rsyncsums +files even if the module itself is configured to be read-only. See also +the bf(exclude) directive for a way to hide the .rsyncsums files from the +user. dit(bf(read only)) This parameter determines whether clients will be able to upload files or not. If "read only" is true then any