+#ifdef HAVE_MMAP
+ data = mmap(0, rrd_file->file_len, rrd_file->mm_prot, rrd_file->mm_flags,
+ rrd_file->fd, offset);
+
+ /* lets see if the first read worked */
+ if (data == MAP_FAILED) {
+ rrd_set_error("mmaping file '%s': %s", file_name,
+ rrd_strerror(errno));
+ goto out_close;
+ }
+ rrd_file->file_start = data;
+ if (rdwr & RRD_CREAT) {
+ memset(data, DNAN, newfile_size - 1);
+ goto out_done;
+ }
+#endif
+ if (rdwr & RRD_CREAT)
+ goto out_done;
+#ifdef USE_MADVISE
+ if (rdwr & RRD_COPY) {
+ /* We will read everything in a moment (copying) */
+ madvise(data, rrd_file->file_len, MADV_WILLNEED | MADV_SEQUENTIAL);
+ } else {
+ /* We do not need to read anything in for the moment */
+ madvise(data, rrd_file->file_len, MADV_RANDOM);
+ /* the stat_head will be needed soonish, so hint accordingly */
+ madvise(data, sizeof(stat_head_t), MADV_WILLNEED | MADV_RANDOM);
+ }
+#endif
+
+ __rrd_read(rrd->stat_head, stat_head_t,
+ 1);
+
+ /* lets do some test if we are on track ... */
+ if (memcmp(rrd->stat_head->cookie, RRD_COOKIE, sizeof(RRD_COOKIE)) != 0) {
+ rrd_set_error("'%s' is not an RRD file", file_name);
+ goto out_nullify_head;
+ }
+
+ if (rrd->stat_head->float_cookie != FLOAT_COOKIE) {
+ rrd_set_error("This RRD was created on another architecture");
+ goto out_nullify_head;
+ }
+
+ version = atoi(rrd->stat_head->version);
+
+ if (version > atoi(RRD_VERSION)) {
+ rrd_set_error("can't handle RRD file version %s",
+ rrd->stat_head->version);
+ goto out_nullify_head;
+ }
+#if defined USE_MADVISE
+ /* the ds_def will be needed soonish, so hint accordingly */
+ madvise(data + PAGE_START(offset),
+ sizeof(ds_def_t) * rrd->stat_head->ds_cnt, MADV_WILLNEED);
+#endif
+ __rrd_read(rrd->ds_def, ds_def_t,
+ rrd->stat_head->ds_cnt);
+
+#if defined USE_MADVISE
+ /* the rra_def will be needed soonish, so hint accordingly */
+ madvise(data + PAGE_START(offset),
+ sizeof(rra_def_t) * rrd->stat_head->rra_cnt, MADV_WILLNEED);
+#endif
+ __rrd_read(rrd->rra_def, rra_def_t,
+ rrd->stat_head->rra_cnt);
+
+ /* handle different format for the live_head */
+ if (version < 3) {
+ rrd->live_head = (live_head_t *) malloc(sizeof(live_head_t));
+ if (rrd->live_head == NULL) {
+ rrd_set_error("live_head_t malloc");
+ goto out_close;
+ }
+#if defined USE_MADVISE
+ /* the live_head will be needed soonish, so hint accordingly */
+ madvise(data + PAGE_START(offset), sizeof(time_t), MADV_WILLNEED);
+#endif
+ __rrd_read(rrd->legacy_last_up, time_t,
+ 1);
+
+ rrd->live_head->last_up = *rrd->legacy_last_up;
+ rrd->live_head->last_up_usec = 0;
+ } else {
+#if defined USE_MADVISE
+ /* the live_head will be needed soonish, so hint accordingly */
+ madvise(data + PAGE_START(offset),
+ sizeof(live_head_t), MADV_WILLNEED);
+#endif
+ __rrd_read(rrd->live_head, live_head_t,
+ 1);
+ }
+ __rrd_read(rrd->pdp_prep, pdp_prep_t,
+ rrd->stat_head->ds_cnt);
+ __rrd_read(rrd->cdp_prep, cdp_prep_t,
+ rrd->stat_head->rra_cnt * rrd->stat_head->ds_cnt);
+ __rrd_read(rrd->rra_ptr, rra_ptr_t,
+ rrd->stat_head->rra_cnt);
+
+ rrd_file->header_len = offset;
+ rrd_file->pos = offset;
+
+ {
+ unsigned long row_cnt = 0;
+ unsigned long i;
+
+ for (i=0; i<rrd->stat_head->rra_cnt; i++)
+ row_cnt += rrd->rra_def[i].row_cnt;
+
+ off_t correct_len = rrd_file->header_len +
+ sizeof(rrd_value_t) * row_cnt * rrd->stat_head->ds_cnt;
+
+ if (correct_len > rrd_file->file_len)
+ {
+ rrd_set_error("'%s' is too small (should be %ld bytes)",
+ file_name, (long long) correct_len);
+ goto out_nullify_head;
+ }
+ }
+
+ out_done:
+ return (rrd_file);
+ out_nullify_head:
+ rrd->stat_head = NULL;
+ out_close:
+#ifdef HAVE_MMAP
+ if (data != MAP_FAILED)
+ munmap(data, rrd_file->file_len);
+#endif
+ close(rrd_file->fd);
+ out_free:
+ free(rrd_file);
+ return NULL;
+}
+
+
+#if defined DEBUG && DEBUG > 1
+/* Print list of in-core pages of a the current rrd_file. */
+static
+void mincore_print(
+ rrd_file_t *rrd_file,
+ char *mark)
+{
+#ifdef HAVE_MMAP
+ /* pretty print blocks in core */
+ off_t off;
+ unsigned char *vec;
+ ssize_t _page_size = sysconf(_SC_PAGESIZE);
+
+ off = rrd_file->file_len +
+ ((rrd_file->file_len + _page_size - 1) / _page_size);
+ vec = malloc(off);
+ if (vec != NULL) {
+ memset(vec, 0, off);
+ if (mincore(rrd_file->file_start, rrd_file->file_len, vec) == 0) {
+ int prev;
+ unsigned is_in = 0, was_in = 0;
+
+ for (off = 0, prev = 0; off < rrd_file->file_len; ++off) {
+ is_in = vec[off] & 1; /* if lsb set then is core resident */
+ if (off == 0)
+ was_in = is_in;
+ if (was_in != is_in) {
+ fprintf(stderr, "%s: %sin core: %p len %ld\n", mark,
+ was_in ? "" : "not ", vec + prev, off - prev);
+ was_in = is_in;
+ prev = off;
+ }
+ }
+ fprintf(stderr,
+ "%s: %sin core: %p len %ld\n", mark,
+ was_in ? "" : "not ", vec + prev, off - prev);
+ } else
+ fprintf(stderr, "mincore: %s", rrd_strerror(errno));
+ }
+#else
+ fprintf(stderr, "sorry mincore only works with mmap");
+#endif
+}
+#endif /* defined DEBUG && DEBUG > 1 */
+
+/*
+ * get exclusive lock to whole file.
+ * lock gets removed when we close the file
+ *
+ * returns 0 on success
+ */
+int rrd_lock(
+ rrd_file_t *file)
+{
+ int rcstat;
+
+ {
+#if defined(_WIN32) && !defined(__CYGWIN__) && !defined(__CYGWIN32__)
+ struct _stat st;
+
+ if (_fstat(file->fd, &st) == 0) {
+ rcstat = _locking(file->fd, _LK_NBLCK, st.st_size);
+ } else {
+ rcstat = -1;
+ }
+#else
+ struct flock lock;
+
+ lock.l_type = F_WRLCK; /* exclusive write lock */
+ lock.l_len = 0; /* whole file */
+ lock.l_start = 0; /* start of file */
+ lock.l_whence = SEEK_SET; /* end of file */
+
+ rcstat = fcntl(file->fd, F_SETLK, &lock);
+#endif
+ }
+
+ return (rcstat);
+}
+
+
+/* drop cache except for the header and the active pages */
+void rrd_dontneed(
+ rrd_file_t *rrd_file,
+ rrd_t *rrd)
+{
+#if defined USE_MADVISE || defined HAVE_POSIX_FADVISE
+ off_t dontneed_start;
+ off_t rra_start;
+ off_t active_block;
+ unsigned long i;
+ ssize_t _page_size = sysconf(_SC_PAGESIZE);
+
+ if (rrd_file == NULL) {
+#if defined DEBUG && DEBUG
+ fprintf (stderr, "rrd_dontneed: Argument 'rrd_file' is NULL.\n");
+#endif
+ return;
+ }
+
+#if defined DEBUG && DEBUG > 1
+ mincore_print(rrd_file, "before");
+#endif
+
+ /* ignoring errors from RRDs that are smaller then the file_len+rounding */
+ rra_start = rrd_file->header_len;
+ dontneed_start = PAGE_START(rra_start) + _page_size;
+ for (i = 0; i < rrd->stat_head->rra_cnt; ++i) {
+ active_block =
+ PAGE_START(rra_start
+ + rrd->rra_ptr[i].cur_row
+ * rrd->stat_head->ds_cnt * sizeof(rrd_value_t));
+ if (active_block > dontneed_start) {
+#ifdef USE_MADVISE
+ madvise(rrd_file->file_start + dontneed_start,
+ active_block - dontneed_start - 1, MADV_DONTNEED);
+#endif
+/* in linux at least only fadvise DONTNEED seems to purge pages from cache */
+#ifdef HAVE_POSIX_FADVISE
+ posix_fadvise(rrd_file->fd, dontneed_start,
+ active_block - dontneed_start - 1,
+ POSIX_FADV_DONTNEED);
+#endif
+ }
+ dontneed_start = active_block;
+ /* do not release 'hot' block if update for this RAA will occur
+ * within 10 minutes */
+ if (rrd->stat_head->pdp_step * rrd->rra_def[i].pdp_cnt -
+ rrd->live_head->last_up % (rrd->stat_head->pdp_step *
+ rrd->rra_def[i].pdp_cnt) < 10 * 60) {
+ dontneed_start += _page_size;
+ }
+ rra_start +=
+ rrd->rra_def[i].row_cnt * rrd->stat_head->ds_cnt *
+ sizeof(rrd_value_t);
+ }
+
+ if (dontneed_start < rrd_file->file_len) {
+#ifdef USE_MADVISE
+ madvise(rrd_file->file_start + dontneed_start,
+ rrd_file->file_len - dontneed_start, MADV_DONTNEED);
+#endif
+#ifdef HAVE_POSIX_FADVISE
+ posix_fadvise(rrd_file->fd, dontneed_start,
+ rrd_file->file_len - dontneed_start,
+ POSIX_FADV_DONTNEED);
+#endif
+ }