rrd_notify_row patch:
[rrdtool.git] / src / rrd_open.c
index 4caebd3..654005f 100644 (file)
@@ -1,5 +1,5 @@
 /*****************************************************************************
- * RRDtool 1.3rc9  Copyright by Tobi Oetiker, 1997-2008
+ * RRDtool 1.3.2  Copyright by Tobi Oetiker, 1997-2008
  *****************************************************************************
  * rrd_open.c  Open an RRD File
  *****************************************************************************
 /* the cast to void* is there to avoid this warning seen on ia64 with certain
    versions of gcc: 'cast increases required alignment of target type'
 */
-#define __rrd_read(dst, dst_t, cnt) \
+#define __rrd_read(dst, dst_t, cnt) { \
+       size_t wanted = sizeof(dst_t)*(cnt); \
+       if (offset + wanted > rrd_file->file_len) { \
+               rrd_set_error("reached EOF while loading header " #dst); \
+               goto out_nullify_head; \
+       } \
        (dst) = (dst_t*)(void*) (data + offset); \
-       offset += sizeof(dst_t) * (cnt)
+       offset += wanted; \
+    }
 #else
-#define __rrd_read(dst, dst_t, cnt) \
-       if ((dst = malloc(sizeof(dst_t)*(cnt))) == NULL) { \
+#define __rrd_read(dst, dst_t, cnt) { \
+       size_t wanted = sizeof(dst_t)*(cnt); \
+        size_t got; \
+       if ((dst = malloc(wanted)) == NULL) { \
                rrd_set_error(#dst " malloc"); \
                goto out_nullify_head; \
        } \
-       offset += read (rrd_file->fd, dst, sizeof(dst_t)*(cnt))
+        got = read (rrd_simple_file->fd, dst, wanted); \
+       if (got != wanted) { \
+               rrd_set_error("short read while reading header " #dst); \
+                goto out_nullify_head; \
+       } \
+       offset += got; \
+    }
 #endif
 
 /* get the address of the start of this page */
 #endif
 #endif
 
+long int  rra_random_row(
+    rra_def_t *);
+
+
 /* Open a database file, return its header and an open filehandle,
  * positioned to the first cdp in the first rra.
  * In the error path of rrd_open, only rrd_free(&rrd) has to be called
  * before returning an error. Do not call rrd_close upon failure of rrd_open.
+ * If creating a new file, the parameter rrd must be initialised with
+ * details of the file content.
+ * If opening an existing file, then use rrd must be initialised by
+ * rrd_init(rrd) prior to invoking rrd_open
  */
 
 rrd_file_t *rrd_open(
@@ -51,27 +73,43 @@ rrd_file_t *rrd_open(
     rrd_t *rrd,
     unsigned rdwr)
 {
+    int i;
     int       flags = 0;
-    mode_t    mode = S_IRUSR;
     int       version;
 
 #ifdef HAVE_MMAP
     ssize_t   _page_size = sysconf(_SC_PAGESIZE);
-    int       mm_prot = PROT_READ, mm_flags = 0;
-    char     *data;
+    char     *data = MAP_FAILED;
 #endif
     off_t     offset = 0;
     struct stat statb;
     rrd_file_t *rrd_file = NULL;
+    rrd_simple_file_t *rrd_simple_file = NULL;
     off_t     newfile_size = 0;
-
-    if (rdwr & RRD_CREAT) {
-        /* yes bad inline signaling alert, we are using the
-           floatcookie to pass the size in ... only used in resize */
-        newfile_size = (off_t) rrd->stat_head->float_cookie;
-        free(rrd->stat_head);
+    off_t header_len, value_cnt, data_len;
+
+    /* Are we creating a new file? */
+    if((rdwr & RRD_CREAT) && (rrd->stat_head != NULL))
+    {
+        header_len = \
+          sizeof(stat_head_t) + \
+          sizeof(ds_def_t) * rrd->stat_head->ds_cnt + \
+          sizeof(rra_def_t) * rrd->stat_head->rra_cnt + \
+          sizeof(time_t) + \
+          sizeof(live_head_t) + \
+          sizeof(pdp_prep_t) * rrd->stat_head->ds_cnt + \
+          sizeof(cdp_prep_t) * rrd->stat_head->ds_cnt * rrd->stat_head->rra_cnt + \
+          sizeof(rra_ptr_t) * rrd->stat_head->rra_cnt;
+
+        value_cnt = 0;
+        for (i = 0; i < rrd->stat_head->rra_cnt; i++)
+            value_cnt += rrd->stat_head->ds_cnt * rrd->rra_def[i].row_cnt;
+
+        data_len = sizeof(rrd_value_t) * value_cnt;
+
+        newfile_size = header_len + data_len;
     }
-    rrd_init(rrd);
+    
     rrd_file = malloc(sizeof(rrd_file_t));
     if (rrd_file == NULL) {
         rrd_set_error("allocating rrd_file descriptor for '%s'", file_name);
@@ -79,6 +117,14 @@ rrd_file_t *rrd_open(
     }
     memset(rrd_file, 0, sizeof(rrd_file_t));
 
+    rrd_file->pvt = malloc(sizeof(rrd_simple_file_t));
+    if(rrd_file->pvt == NULL) {
+        rrd_set_error("allocating rrd_simple_file for '%s'", file_name);
+        return NULL;
+    }
+    memset(rrd_file->pvt, 0, sizeof(rrd_simple_file_t));
+    rrd_simple_file = (rrd_simple_file_t *)rrd_file->pvt;
+
 #ifdef DEBUG
     if ((rdwr & (RRD_READONLY | RRD_READWRITE)) ==
         (RRD_READONLY | RRD_READWRITE)) {
@@ -87,21 +133,26 @@ rrd_file_t *rrd_open(
         exit(-1);
     }
 #endif
+
+#ifdef HAVE_MMAP
+    rrd_simple_file->mm_prot = PROT_READ;
+    rrd_simple_file->mm_flags = 0;
+#endif
+
     if (rdwr & RRD_READONLY) {
         flags |= O_RDONLY;
 #ifdef HAVE_MMAP
-        mm_flags = MAP_PRIVATE;
+        rrd_simple_file->mm_flags = MAP_PRIVATE;
 # ifdef MAP_NORESERVE
-        mm_flags |= MAP_NORESERVE;  /* readonly, so no swap backing needed */
+        rrd_simple_file->mm_flags |= MAP_NORESERVE;  /* readonly, so no swap backing needed */
 # endif
 #endif
     } else {
         if (rdwr & RRD_READWRITE) {
-            mode |= S_IWUSR;
             flags |= O_RDWR;
 #ifdef HAVE_MMAP
-            mm_flags = MAP_SHARED;
-            mm_prot |= PROT_WRITE;
+            rrd_simple_file->mm_flags = MAP_SHARED;
+            rrd_simple_file->mm_prot |= PROT_WRITE;
 #endif
         }
         if (rdwr & RRD_CREAT) {
@@ -110,21 +161,24 @@ rrd_file_t *rrd_open(
     }
     if (rdwr & RRD_READAHEAD) {
 #ifdef MAP_POPULATE
-        mm_flags |= MAP_POPULATE;   /* populate ptes and data */
+        rrd_simple_file->mm_flags |= MAP_POPULATE;   /* populate ptes and data */
 #endif
 #if defined MAP_NONBLOCK
-        mm_flags |= MAP_NONBLOCK;   /* just populate ptes */
+        rrd_simple_file->mm_flags |= MAP_NONBLOCK;   /* just populate ptes */
 #endif
     }
+#if defined(_WIN32) && !defined(__CYGWIN__) && !defined(__CYGWIN32__)
+    flags |= O_BINARY;
+#endif
 
-    if ((rrd_file->fd = open(file_name, flags, mode)) < 0) {
+    if ((rrd_simple_file->fd = open(file_name, flags, 0666)) < 0) {
         rrd_set_error("opening '%s': %s", file_name, rrd_strerror(errno));
         goto out_free;
     }
 
     /* Better try to avoid seeks as much as possible. stat may be heavy but
      * many concurrent seeks are even worse.  */
-    if (newfile_size == 0 && ((fstat(rrd_file->fd, &statb)) < 0)) {
+    if (newfile_size == 0 && ((fstat(rrd_simple_file->fd, &statb)) < 0)) {
         rrd_set_error("fstat '%s': %s", file_name, rrd_strerror(errno));
         goto out_close;
     }
@@ -132,30 +186,32 @@ rrd_file_t *rrd_open(
         rrd_file->file_len = statb.st_size;
     } else {
         rrd_file->file_len = newfile_size;
-        lseek(rrd_file->fd, newfile_size - 1, SEEK_SET);
-        write(rrd_file->fd, "\0", 1);   /* poke */
-        lseek(rrd_file->fd, 0, SEEK_SET);
+        lseek(rrd_simple_file->fd, newfile_size - 1, SEEK_SET);
+        write(rrd_simple_file->fd, "\0", 1);   /* poke */
+        lseek(rrd_simple_file->fd, 0, SEEK_SET);
     }
 #ifdef HAVE_POSIX_FADVISE
     /* In general we need no read-ahead when dealing with rrd_files.
        When we stop reading, it is highly unlikely that we start up again.
        In this manner we actually save time and diskaccess (and buffer cache).
        Thanks to Dave Plonka for the Idea of using POSIX_FADV_RANDOM here. */
-    posix_fadvise(rrd_file->fd, 0, 0, POSIX_FADV_RANDOM);
+    posix_fadvise(rrd_simple_file->fd, 0, 0, POSIX_FADV_RANDOM);
 #endif
 
 /*
         if (rdwr & RRD_READWRITE)
         {
-           if (setvbuf((rrd_file->fd),NULL,_IONBF,2)) {
+           if (setvbuf((rrd_simple_file->fd),NULL,_IONBF,2)) {
                   rrd_set_error("failed to disable the stream buffer\n");
                   return (-1);
            }
         }
 */
+
 #ifdef HAVE_MMAP
-    data = mmap(0, rrd_file->file_len, mm_prot, mm_flags,
-                rrd_file->fd, offset);
+    data = mmap(0, rrd_file->file_len, 
+        rrd_simple_file->mm_prot, rrd_simple_file->mm_flags,
+        rrd_simple_file->fd, offset);
 
     /* lets see if the first read worked */
     if (data == MAP_FAILED) {
@@ -163,7 +219,7 @@ rrd_file_t *rrd_open(
                       rrd_strerror(errno));
         goto out_close;
     }
-    rrd_file->file_start = data;
+    rrd_simple_file->file_start = data;
     if (rdwr & RRD_CREAT) {
         memset(data, DNAN, newfile_size - 1);
         goto out_done;
@@ -254,13 +310,37 @@ rrd_file_t *rrd_open(
 
     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:
-    close(rrd_file->fd);
+#ifdef HAVE_MMAP
+    if (data != MAP_FAILED)
+      munmap(data, rrd_file->file_len);
+#endif
+    close(rrd_simple_file->fd);
   out_free:
+    free(rrd_file->pvt);
     free(rrd_file);
     return NULL;
 }
@@ -273,6 +353,8 @@ void mincore_print(
     rrd_file_t *rrd_file,
     char *mark)
 {
+    rrd_simple_file_t *rrd_simple_file;
+    rrd_simple_file = (rrd_simple_file_t *)rrd_file->pvt;
 #ifdef HAVE_MMAP
     /* pretty print blocks in core */
     off_t     off;
@@ -284,7 +366,7 @@ void mincore_print(
     vec = malloc(off);
     if (vec != NULL) {
         memset(vec, 0, off);
-        if (mincore(rrd_file->file_start, rrd_file->file_len, vec) == 0) {
+        if (mincore(rrd_simple_file->file_start, rrd_file->file_len, vec) == 0) {
             int       prev;
             unsigned  is_in = 0, was_in = 0;
 
@@ -311,19 +393,64 @@ void mincore_print(
 }
 #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 *rrd_file)
+{
+    int       rcstat;
+    rrd_simple_file_t *rrd_simple_file;
+    rrd_simple_file = (rrd_simple_file_t *)rrd_file->pvt;
+
+    {
+#if defined(_WIN32) && !defined(__CYGWIN__) && !defined(__CYGWIN32__)
+        struct _stat st;
+
+        if (_fstat(rrd_simple_file->fd, &st) == 0) {
+            rcstat = _locking(rrd_simple_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(rrd_simple_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)
 {
+    rrd_simple_file_t *rrd_simple_file = (rrd_simple_file_t *)rrd_file->pvt;
 #if defined USE_MADVISE || defined HAVE_POSIX_FADVISE
-    unsigned long dontneed_start;
-    unsigned long rra_start;
-    unsigned long active_block;
+    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
@@ -338,12 +465,12 @@ void rrd_dontneed(
                        * rrd->stat_head->ds_cnt * sizeof(rrd_value_t));
         if (active_block > dontneed_start) {
 #ifdef USE_MADVISE
-            madvise(rrd_file->file_start + dontneed_start,
+            madvise(rrd_simple_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,
+            posix_fadvise(rrd_simple_file->fd, dontneed_start,
                           active_block - dontneed_start - 1,
                           POSIX_FADV_DONTNEED);
 #endif
@@ -360,14 +487,19 @@ void rrd_dontneed(
             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);
+           madvise(rrd_simple_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);
+           posix_fadvise(rrd_simple_file->fd, dontneed_start,
+                         rrd_file->file_len - dontneed_start,
+                         POSIX_FADV_DONTNEED);
 #endif
+    }
+
 #if defined DEBUG && DEBUG > 1
     mincore_print(rrd_file, "after");
 #endif
@@ -381,19 +513,22 @@ void rrd_dontneed(
 int rrd_close(
     rrd_file_t *rrd_file)
 {
+    rrd_simple_file_t *rrd_simple_file;
+    rrd_simple_file = (rrd_simple_file_t *)rrd_file->pvt;
     int       ret;
 
 #ifdef HAVE_MMAP
-    ret = msync(rrd_file->file_start, rrd_file->file_len, MS_ASYNC);
+    ret = msync(rrd_simple_file->file_start, rrd_file->file_len, MS_ASYNC);
     if (ret != 0)
         rrd_set_error("msync rrd_file: %s", rrd_strerror(errno));
-    ret = munmap(rrd_file->file_start, rrd_file->file_len);
+    ret = munmap(rrd_simple_file->file_start, rrd_file->file_len);
     if (ret != 0)
         rrd_set_error("munmap rrd_file: %s", rrd_strerror(errno));
 #endif
-    ret = close(rrd_file->fd);
+    ret = close(rrd_simple_file->fd);
     if (ret != 0)
         rrd_set_error("closing file: %s", rrd_strerror(errno));
+    free(rrd_file->pvt);
     free(rrd_file);
     rrd_file = NULL;
     return ret;
@@ -408,6 +543,8 @@ off_t rrd_seek(
     int whence)
 {
     off_t     ret = 0;
+    rrd_simple_file_t *rrd_simple_file;
+    rrd_simple_file = (rrd_simple_file_t *)rrd_file->pvt;
 
 #ifdef HAVE_MMAP
     if (whence == SEEK_SET)
@@ -417,7 +554,7 @@ off_t rrd_seek(
     else if (whence == SEEK_END)
         rrd_file->pos = rrd_file->file_len + off;
 #else
-    ret = lseek(rrd_file->fd, off, whence);
+    ret = lseek(rrd_simple_file->fd, off, whence);
     if (ret < 0)
         rrd_set_error("lseek: %s", rrd_strerror(errno));
     rrd_file->pos = ret;
@@ -429,7 +566,7 @@ off_t rrd_seek(
 
 /* Get current position in rrd_file.  */
 
-inline off_t rrd_tell(
+off_t rrd_tell(
     rrd_file_t *rrd_file)
 {
     return rrd_file->pos;
@@ -439,11 +576,12 @@ inline off_t rrd_tell(
 /* Read count bytes into buffer buf, starting at rrd_file->pos.
  * Returns the number of bytes read or <0 on error.  */
 
-inline ssize_t rrd_read(
+ssize_t rrd_read(
     rrd_file_t *rrd_file,
     void *buf,
     size_t count)
 {
+    rrd_simple_file_t *rrd_simple_file = (rrd_simple_file_t *)rrd_file->pvt;
 #ifdef HAVE_MMAP
     size_t    _cnt = count;
     ssize_t   _surplus;
@@ -458,14 +596,14 @@ inline ssize_t rrd_read(
     }
     if (_cnt == 0)
         return 0;       /* EOF */
-    buf = memcpy(buf, rrd_file->file_start + rrd_file->pos, _cnt);
+    buf = memcpy(buf, rrd_simple_file->file_start + rrd_file->pos, _cnt);
 
     rrd_file->pos += _cnt;  /* mimmic read() semantics */
     return _cnt;
 #else
     ssize_t   ret;
 
-    ret = read(rrd_file->fd, buf, count);
+    ret = read(rrd_simple_file->fd, buf, count);
     if (ret > 0)
         rrd_file->pos += ret;   /* mimmic read() semantics */
     return ret;
@@ -474,24 +612,32 @@ inline ssize_t rrd_read(
 
 
 /* Write count bytes from buffer buf to the current position
- * rrd_file->pos of rrd_file->fd.
+ * rrd_file->pos of rrd_simple_file->fd.
  * Returns the number of bytes written or <0 on error.  */
 
-inline ssize_t rrd_write(
+ssize_t rrd_write(
     rrd_file_t *rrd_file,
     const void *buf,
     size_t count)
 {
+    rrd_simple_file_t *rrd_simple_file = (rrd_simple_file_t *)rrd_file->pvt;
 #ifdef HAVE_MMAP
+    int old_size = rrd_file->file_len;
     if (count == 0)
         return 0;
     if (buf == NULL)
         return -1;      /* EINVAL */
-    memcpy(rrd_file->file_start + rrd_file->pos, buf, count);
+    
+    if((rrd_file->pos + count) > old_size)
+    {
+        rrd_set_error("attempting to write beyond end of file");
+        return -1;
+    }
+    memcpy(rrd_simple_file->file_start + rrd_file->pos, buf, count);
     rrd_file->pos += count;
     return count;       /* mimmic write() semantics */
 #else
-    ssize_t   _sz = write(rrd_file->fd, buf, count);
+    ssize_t   _sz = write(rrd_simple_file->fd, buf, count);
 
     if (_sz > 0)
         rrd_file->pos += _sz;
@@ -502,11 +648,13 @@ inline ssize_t rrd_write(
 
 /* flush all data pending to be written to FD.  */
 
-inline void rrd_flush(
+void rrd_flush(
     rrd_file_t *rrd_file)
 {
-    if (fdatasync(rrd_file->fd) != 0) {
-        rrd_set_error("flushing fd %d: %s", rrd_file->fd,
+    rrd_simple_file_t *rrd_simple_file;
+    rrd_simple_file = (rrd_simple_file_t *)rrd_file->pvt;
+    if (fdatasync(rrd_simple_file->fd) != 0) {
+        rrd_set_error("flushing fd %d: %s", rrd_simple_file->fd,
                       rrd_strerror(errno));
     }
 }
@@ -563,3 +711,46 @@ void rrd_freemem(
 {
     free(mem);
 }
+
+/*
+ * rra_update informs us about the RRAs being updated
+ * The low level storage API may use this information for
+ * aligning RRAs within stripes, or other performance enhancements
+ */
+void rrd_notify_row(
+    rrd_file_t *rrd_file,
+    int rra_idx,
+    unsigned long rra_row,
+    time_t rra_time)
+{
+}
+
+/*
+ * This function is called when creating a new RRD
+ * The storage implementation can use this opportunity to select
+ * a sensible starting row within the file.
+ * The default implementation is random, to ensure that all RRAs
+ * don't change to a new disk block at the same time
+ */
+unsigned long rrd_select_initial_row(
+    rrd_file_t *rrd_file,
+    int rra_idx,
+    rra_def_t *rra
+    )
+{
+    return rra_random_row(rra);
+}
+
+static int rand_init = 0;
+
+long int rra_random_row(
+    rra_def_t *rra)
+{
+    if (!rand_init) {
+        srandom((unsigned int) time(NULL) + (unsigned int) getpid());
+        rand_init++;
+    }
+
+    return random() % rra->row_cnt;
+}
+