Add file:advise/4 - a wrapper to the POSIX syscall posix_fadvise

Useful for informing the Operating System about the access pattern
for a file's data, so that it can adapt the caching strategy to
maximize disk IO performance.

7 files changed, 103 insertions, 1 deletions

diff --git a/erts/configure.in b/erts/configure.in
index f11aba1b15..63bf548c89 100644
--- a/erts/configure.in
+++ b/erts/configure.in
@@ -1054,6 +1054,7 @@ fi
 
 AC_SUBST(ERTS_BUILD_SMP_EMU)
 
+AC_CHECK_FUNCS([posix_fadvise])
 
 
 #
diff --git a/erts/emulator/drivers/common/efile_drv.c b/erts/emulator/drivers/common/efile_drv.c
index b64d27de68..60ae4cb108 100644
--- a/erts/emulator/drivers/common/efile_drv.c
+++ b/erts/emulator/drivers/common/efile_drv.c
@@ -54,6 +54,7 @@
 #define FILE_ALTNAME            28
 #define FILE_READ_LINE          29
 #define FILE_FDATASYNC          30
+#define FILE_FADVISE            31
 
 /* Return codes */
 
@@ -358,6 +359,11 @@ struct t_data
 	    struct t_readdir_buf *first_buf;
 	    struct t_readdir_buf *last_buf;
 	} read_dir;
+	struct {
+	    Sint64 offset;
+	    Sint64 length;
+	    int advise;
+	} fadvise;
     } c;
     char b[1];
 };
@@ -1647,6 +1653,18 @@ static void invoke_open(void *data)
     d->result_ok = status;
 }
 
+static void invoke_fadvise(void *data)
+{
+    struct t_data *d = (struct t_data *) data;
+    int fd = (int) d->fd;
+    off_t offset = (off_t) d->c.fadvise.offset;
+    off_t length = (off_t) d->c.fadvise.length;
+    int advise = (int) d->c.fadvise.advise;
+
+    d->again = 0;
+    d->result_ok = efile_fadvise(&d->errInfo, fd, offset, length, advise);
+}
+
 static void free_readdir(void *data)
 {
     struct t_data *d = (struct t_data *) data;
@@ -1936,6 +1954,7 @@ file_async_ready(ErlDrvData e, ErlDrvThreadData data)
       case FILE_SYMLINK:
       case FILE_RENAME:
       case FILE_WRITE_INFO:
+      case FILE_FADVISE:
 	reply(desc, d->result_ok, &d->errInfo);
 	free_data(data);
 	break;
@@ -2355,6 +2374,21 @@ file_output(ErlDrvData e, char* buf, int count)
 	    goto done;
 	}
 
+    case FILE_FADVISE:
+    {
+        d = EF_SAFE_ALLOC(sizeof(struct t_data));
+
+        d->fd = fd;
+        d->command = command;
+        d->invoke = invoke_fadvise;
+        d->free = free_data;
+        d->level = 2;
+        d->c.fadvise.offset = get_int64((uchar*) buf);
+        d->c.fadvise.length = get_int64(((uchar*) buf) + sizeof(Sint64));
+        d->c.fadvise.advise = get_int32(((uchar*) buf) + 2 * sizeof(Sint64));
+        goto done;
+    }
+
     }
 
     /*
diff --git a/erts/emulator/drivers/common/erl_efile.h b/erts/emulator/drivers/common/erl_efile.h
index 6821a0e2ee..61684af6c9 100644
--- a/erts/emulator/drivers/common/erl_efile.h
+++ b/erts/emulator/drivers/common/erl_efile.h
@@ -151,3 +151,5 @@ int efile_altname(Efile_error* errInfo, char *name,
 int efile_link(Efile_error* errInfo, char* old, char* new);
 int efile_symlink(Efile_error* errInfo, char* old, char* new);
 int efile_may_openfile(Efile_error* errInfo, char *name);
+int efile_fadvise(Efile_error* errInfo, int fd, Sint64 offset, Sint64 length,
+		  int advise);
diff --git a/erts/emulator/drivers/common/ram_file_drv.c b/erts/emulator/drivers/common/ram_file_drv.c
index d4e547ade6..abedcc933a 100644
--- a/erts/emulator/drivers/common/ram_file_drv.c
+++ b/erts/emulator/drivers/common/ram_file_drv.c
@@ -46,6 +46,8 @@
 #define RAM_FILE_UUENCODE      35  /* uuencode file */
 #define RAM_FILE_UUDECODE      36  /* uudecode file */
 #define RAM_FILE_SIZE          37  /* get file size */
+#define RAM_FILE_ADVISE        38  /* predeclare the access
+                                    * pattern for file data */
 /* possible new operations include:
    DES_ENCRYPT
    DES_DECRYPT
@@ -693,6 +695,13 @@ static void rfile_command(ErlDrvData e, char* buf, int count)
     case RAM_FILE_UUDECODE:   /* uudecode file */
 	ram_file_uudecode(f);
 	break;
+
+    case RAM_FILE_ADVISE:
+	if (f->flags == 0)
+	    error_reply(f, EBADF);
+	else
+	    reply(f, 1, 0);
+	break;
     }
     /*
      * Ignore anything else -- let the caller hang.
diff --git a/erts/emulator/drivers/unix/unix_efile.c b/erts/emulator/drivers/unix/unix_efile.c
index 0cece76221..ea016526ef 100644
--- a/erts/emulator/drivers/unix/unix_efile.c
+++ b/erts/emulator/drivers/unix/unix_efile.c
@@ -1448,3 +1448,14 @@ efile_symlink(Efile_error* errInfo, char* old, char* new)
     return check_error(symlink(old, new), errInfo);
 #endif
 }
+
+int
+efile_fadvise(Efile_error* errInfo, int fd, Sint64 offset,
+		  Sint64 length, int advise)
+{
+#ifdef HAVE_POSIX_FADVISE
+    return check_error(posix_fadvise(fd, offset, length, advise), errInfo);
+#else
+    return check_error(0, errInfo);
+#endif
+}
diff --git a/erts/emulator/drivers/win32/win_efile.c b/erts/emulator/drivers/win32/win_efile.c
index cdf3bb424c..06850b4945 100644
--- a/erts/emulator/drivers/win32/win_efile.c
+++ b/erts/emulator/drivers/win32/win_efile.c
@@ -1433,3 +1433,12 @@ efile_symlink(Efile_error* errInfo, char* old, char* new)
     errno = ENOTSUP;
     return check_error(-1, errInfo);
 }
+
+int
+efile_fadvise(Efile_error* errInfo, int fd, Sint64 offset,
+	    Sint64 length, int advise)
+{
+    /* posix_fadvise is not available on Windows, do nothing */
+    errno = ERROR_SUCCESS;
+    return check_error(0, errInfo);
+}
diff --git a/erts/preloaded/src/prim_file.erl b/erts/preloaded/src/prim_file.erl
index 53c741e66e..e9df5752ab 100644
--- a/erts/preloaded/src/prim_file.erl
+++ b/erts/preloaded/src/prim_file.erl
@@ -25,7 +25,7 @@
 %%% Interface towards a single file's contents. Uses ?FD_DRV.
 
 %% Generic file contents operations
--export([open/2, close/1, datasync/1, sync/1, position/2, truncate/1,
+-export([open/2, close/1, datasync/1, sync/1, advise/4, position/2, truncate/1,
 	 write/2, pwrite/2, pwrite/3, read/2, read_line/1, pread/2, pread/3, copy/3]).
 
 %% Specialized file operations
@@ -97,6 +97,7 @@
 -define(FILE_ALTNAME,          28).
 -define(FILE_READ_LINE,        29).
 -define(FILE_FDATASYNC,        30).
+-define(FILE_ADVISE,           31).
 
 %% Driver responses
 -define(FILE_RESP_OK,          0).
@@ -131,6 +132,13 @@
 %% IPREAD variants
 -define(IPREAD_S32BU_P32BU, 0).
 
+%% POSIX file advises
+-define(POSIX_FADV_NORMAL,     0).
+-define(POSIX_FADV_RANDOM,     1).
+-define(POSIX_FADV_SEQUENTIAL, 2).
+-define(POSIX_FADV_WILLNEED,   3).
+-define(POSIX_FADV_DONTNEED,   4).
+-define(POSIX_FADV_NOREUSE,    5).
 
 
 %%%-----------------------------------------------------------------
@@ -221,7 +229,35 @@ close(#file_descriptor{module = ?MODULE, data = {Port, _}}) ->
 close(Port) when is_port(Port) ->
     drv_close(Port).
 
+-define(ADVISE(Offs, Len, Adv),
+	<<?FILE_ADVISE, Offs:64/signed, Len:64/signed,
+	  Adv:32/signed>>).
 
+%% Returns {error, Reason} | ok.
+advise(#file_descriptor{module = ?MODULE, data = {Port, _}},
+       Offset, Length, Advise) ->
+    case Advise of
+	normal ->
+	    Cmd = ?ADVISE(Offset, Length, ?POSIX_FADV_NORMAL),
+	    drv_command(Port, Cmd);
+	random ->
+	    Cmd = ?ADVISE(Offset, Length, ?POSIX_FADV_RANDOM),
+	    drv_command(Port, Cmd);
+	sequential ->
+	    Cmd = ?ADVISE(Offset, Length, ?POSIX_FADV_SEQUENTIAL),
+	    drv_command(Port, Cmd);
+	will_need ->
+	    Cmd = ?ADVISE(Offset, Length, ?POSIX_FADV_WILLNEED),
+	    drv_command(Port, Cmd);
+	dont_need ->
+	    Cmd = ?ADVISE(Offset, Length, ?POSIX_FADV_DONTNEED),
+	    drv_command(Port, Cmd);
+	no_reuse ->
+	    Cmd = ?ADVISE(Offset, Length, ?POSIX_FADV_NOREUSE),
+	    drv_command(Port, Cmd);
+	_ ->
+	    {error, einval}
+    end.
 
 %% Returns {error, Reason} | ok.
 write(#file_descriptor{module = ?MODULE, data = {Port, _}}, Bytes) ->