s3:smbd: restructure kernel oplocks code

This converts the irix oplocks code to use a fd event
and removes the last special case for file descriptors
for the main sys_select().

metze

6 files changed, 117 insertions, 136 deletions

diff --git a/source3/smbd/globals.c b/source3/smbd/globals.c
index 6fcf18da09..c5681223f9 100644
--- a/source3/smbd/globals.c
+++ b/source3/smbd/globals.c
@@ -175,20 +175,13 @@ char *LastDir = NULL;
 #if HAVE_KERNEL_OPLOCKS_LINUX
 SIG_ATOMIC_T oplock_signals_received = 0;
 SIG_ATOMIC_T fd_pending_array[FD_PENDING_SIZE];
-struct kernel_oplocks linux_koplocks;
-#endif
-
-#if HAVE_KERNEL_OPLOCKS_IRIX
-int oplock_pipe_write = -1;
-int oplock_pipe_read = -1;
-struct kernel_oplocks irix_koplocks;
 #endif
 
 /* Current number of oplocks we have outstanding. */
 int32_t exclusive_oplocks_open = 0;
 int32_t level_II_oplocks_open = 0;
 bool global_client_failed_oplock_break = false;
-const struct kernel_oplocks *koplocks = NULL;
+struct kernel_oplocks *koplocks = NULL;
 
 struct notify_mid_map *notify_changes_by_mid = NULL;
 
@@ -215,9 +208,5 @@ void smbd_init_globals(void)
 
 #if HAVE_KERNEL_OPLOCKS_LINUX
 	ZERO_STRUCT(fd_pending_array);
-	ZERO_STRUCT(linux_koplocks);
-#endif
-#if HAVE_KERNEL_OPLOCKS_IRIX
-	ZERO_STRUCT(irix_koplocks);
 #endif
 }
diff --git a/source3/smbd/globals.h b/source3/smbd/globals.h
index 157089f37c..2c4f8b5821 100644
--- a/source3/smbd/globals.h
+++ b/source3/smbd/globals.h
@@ -188,20 +188,13 @@ extern char *LastDir;
 extern SIG_ATOMIC_T oplock_signals_received;
 #define FD_PENDING_SIZE 100
 extern SIG_ATOMIC_T fd_pending_array[FD_PENDING_SIZE];
-extern struct kernel_oplocks linux_koplocks;
-#endif
-
-#if HAVE_KERNEL_OPLOCKS_IRIX
-extern int oplock_pipe_write;
-extern int oplock_pipe_read;
-extern struct kernel_oplocks irix_koplocks;
 #endif
 
 /* Current number of oplocks we have outstanding. */
 extern int32_t exclusive_oplocks_open;
 extern int32_t level_II_oplocks_open;
 extern bool global_client_failed_oplock_break;
-extern const struct kernel_oplocks *koplocks;
+extern struct kernel_oplocks *koplocks;
 
 extern struct notify_mid_map *notify_changes_by_mid;
 
diff --git a/source3/smbd/oplock.c b/source3/smbd/oplock.c
index 3fd5afef22..e4b5016538 100644
--- a/source3/smbd/oplock.c
+++ b/source3/smbd/oplock.c
@@ -36,9 +36,9 @@ int32 get_number_of_exclusive_open_oplocks(void)
  Return True if an oplock message is pending.
 ****************************************************************************/
 
-bool oplock_message_waiting(fd_set *fds)
+bool oplock_message_waiting(void)
 {
-	if (koplocks && koplocks->msg_waiting(fds)) {
+	if (koplocks && koplocks->ops->msg_waiting(koplocks)) {
 		return True;
 	}
 
@@ -52,7 +52,7 @@ bool oplock_message_waiting(fd_set *fds)
  we're calling this in a shutting down state.
 ****************************************************************************/
 
-void process_kernel_oplocks(struct messaging_context *msg_ctx, fd_set *pfds)
+void process_kernel_oplocks(struct messaging_context *msg_ctx)
 {
 	/*
 	 * We need to check for kernel oplocks before going into the select
@@ -64,11 +64,11 @@ void process_kernel_oplocks(struct messaging_context *msg_ctx, fd_set *pfds)
 		return;
 	}
 
-	while (koplocks->msg_waiting(pfds)) { 
+	while (koplocks->ops->msg_waiting(koplocks)) {
 		files_struct *fsp;
 		char msg[MSG_SMB_KERNEL_BREAK_SIZE];
 
-		fsp = koplocks->receive_message(pfds);
+		fsp = koplocks->ops->receive_message(koplocks);
 
 		if (fsp == NULL) {
 			DEBUG(3, ("Kernel oplock message announced, but none "
@@ -99,7 +99,7 @@ bool set_file_oplock(files_struct *fsp, int oplock_type)
 	if ((fsp->oplock_type != NO_OPLOCK) &&
 	    (fsp->oplock_type != FAKE_LEVEL_II_OPLOCK) &&
 	    koplocks &&
-	    !koplocks->set_oplock(fsp, oplock_type)) {
+	    !koplocks->ops->set_oplock(koplocks, fsp, oplock_type)) {
 		return False;
 	}
 
@@ -129,7 +129,7 @@ void release_file_oplock(files_struct *fsp)
 	if ((fsp->oplock_type != NO_OPLOCK) &&
 	    (fsp->oplock_type != FAKE_LEVEL_II_OPLOCK) &&
 	    koplocks) {
-		koplocks->release_oplock(fsp);
+		koplocks->ops->release_oplock(koplocks, fsp);
 	}
 
 	if (fsp->oplock_type == LEVEL_II_OPLOCK) {
@@ -161,7 +161,7 @@ void release_file_oplock(files_struct *fsp)
 static void downgrade_file_oplock(files_struct *fsp)
 {
 	if (koplocks) {
-		koplocks->release_oplock(fsp);
+		koplocks->ops->release_oplock(koplocks, fsp);
 	}
 	fsp->oplock_type = LEVEL_II_OPLOCK;
 	exclusive_oplocks_open--;
@@ -227,19 +227,6 @@ bool downgrade_oplock(files_struct *fsp)
 }
 
 /****************************************************************************
- Return the fd (if any) used for receiving oplock notifications.
-****************************************************************************/
-
-int oplock_notify_fd(void)
-{
-	if (koplocks) {
-		return koplocks->notification_fd;
-	}
-
-	return -1;
-}
-
-/****************************************************************************
  Set up an oplock break message.
 ****************************************************************************/
 
@@ -914,9 +901,9 @@ bool init_oplocks(struct messaging_context *msg_ctx)
 
 	if (lp_kernel_oplocks()) {
 #if HAVE_KERNEL_OPLOCKS_IRIX
-		koplocks = irix_init_kernel_oplocks();
+		koplocks = irix_init_kernel_oplocks(talloc_autofree_context());
 #elif HAVE_KERNEL_OPLOCKS_LINUX
-		koplocks = linux_init_kernel_oplocks();
+		koplocks = linux_init_kernel_oplocks(talloc_autofree_context());
 #endif
 	}
 
diff --git a/source3/smbd/oplock_irix.c b/source3/smbd/oplock_irix.c
index 496b0b72ea..0cfa960425 100644
--- a/source3/smbd/oplock_irix.c
+++ b/source3/smbd/oplock_irix.c
@@ -23,6 +23,14 @@
 
 #if HAVE_KERNEL_OPLOCKS_IRIX
 
+struct irix_oplocks_context {
+	struct kernel_oplocks *ctx;
+	int write_fd;
+	int read_fd;
+	struct fd_event *read_fde;
+	bool pending;
+};
+
 /****************************************************************************
  Test to see if IRIX kernel oplocks work.
 ****************************************************************************/
@@ -100,22 +108,27 @@ static bool irix_oplocks_available(void)
  * oplock break protocol.
 ****************************************************************************/
 
-static files_struct *irix_oplock_receive_message(fd_set *fds)
+static files_struct *irix_oplock_receive_message(struct kernel_oplocks *_ctx)
 {
+	struct irix_oplocks_context *ctx = talloc_get_type(_ctx->private_data,
+					   struct irix_oplocks_context);
 	oplock_stat_t os;
 	char dummy;
 	struct file_id fileid;
 	files_struct *fsp;
 
-	/* Ensure we only get one call per select fd set. */
-	FD_CLR(oplock_pipe_read, fds);
+	/*
+	 * TODO: is it correct to assume we only get one
+	 * oplock break, for each byte we read from the pipe?
+	 */
+	ctx->pending = false;
 
 	/*
 	 * Read one byte of zero to clear the
 	 * kernel break notify message.
 	 */
 
-	if(read(oplock_pipe_read, &dummy, 1) != 1) {
+	if(read(ctx->read_fd, &dummy, 1) != 1) {
 		DEBUG(0,("irix_oplock_receive_message: read of kernel "
 			 "notification failed. Error was %s.\n",
 			 strerror(errno) ));
@@ -128,7 +141,7 @@ static files_struct *irix_oplock_receive_message(fd_set *fds)
 	 * request outstanding.
 	 */
 
-	if(sys_fcntl_ptr(oplock_pipe_read, F_OPLKSTAT, &os) < 0) {
+	if(sys_fcntl_ptr(ctx->read_fd, F_OPLKSTAT, &os) < 0) {
 		DEBUG(0,("irix_oplock_receive_message: fcntl of kernel "
 			 "notification failed. Error was %s.\n",
 			 strerror(errno) ));
@@ -170,9 +183,13 @@ static files_struct *irix_oplock_receive_message(fd_set *fds)
  Attempt to set an kernel oplock on a file.
 ****************************************************************************/
 
-static bool irix_set_kernel_oplock(files_struct *fsp, int oplock_type)
+static bool irix_set_kernel_oplock(struct kernel_oplocks *_ctx,
+				   files_struct *fsp, int oplock_type)
 {
-	if (sys_fcntl_long(fsp->fh->fd, F_OPLKREG, oplock_pipe_write) == -1) {
+	struct irix_oplocks_context *ctx = talloc_get_type(_ctx->private_data,
+					   struct irix_oplocks_context);
+
+	if (sys_fcntl_long(fsp->fh->fd, F_OPLKREG, ctx->write_fd) == -1) {
 		if(errno != EAGAIN) {
 			DEBUG(0,("irix_set_kernel_oplock: Unable to get "
 				 "kernel oplock on file %s, file_id %s "
@@ -204,7 +221,8 @@ static bool irix_set_kernel_oplock(files_struct *fsp, int oplock_type)
  Release a kernel oplock on a file.
 ****************************************************************************/
 
-static void irix_release_kernel_oplock(files_struct *fsp)
+static void irix_release_kernel_oplock(struct kernel_oplocks *_ctx,
+				       files_struct *fsp)
 {
 	if (DEBUGLVL(10)) {
 		/*
@@ -234,63 +252,77 @@ static void irix_release_kernel_oplock(files_struct *fsp)
 	}
 }
 
-/****************************************************************************
- See if there is a message waiting in this fd set.
- Note that fds MAY BE NULL ! If so we must do our own select.
-****************************************************************************/
-
-static bool irix_oplock_msg_waiting(fd_set *fds)
+static bool irix_oplock_msg_waiting(struct kernel_oplocks *_ctx)
 {
-	int selrtn;
-	fd_set myfds;
-	struct timeval to;
-
-	if (oplock_pipe_read == -1)
-		return False;
-
-	if (fds) {
-		return FD_ISSET(oplock_pipe_read, fds);
-	}
-
-	/* Do a zero-time select. We just need to find out if there
-	 * are any outstanding messages. We use sys_select_intr as
-	 * we need to ignore any signals. */
+	struct irix_oplocks_context *ctx = talloc_get_type(_ctx->private_data,
+					   struct irix_oplocks_context);
+	return ctx->pending;
+}
 
-	FD_ZERO(&myfds);
-	FD_SET(oplock_pipe_read, &myfds);
+static void irix_oplocks_read_fde_handler(struct event_context *ev,
+					  struct fd_event *fde,
+					  uint16_t flags,
+					  void *private_data)
+{
+	struct irix_oplocks_context *ctx = talloc_get_type(private_data,
+					   struct irix_oplocks_context);
 
-	to = timeval_set(0, 0);
-	selrtn = sys_select_intr(oplock_pipe_read+1,&myfds,NULL,NULL,&to);
-	return (selrtn == 1) ? True : False;
+	ctx->pending = true;
+	process_kernel_oplocks(smbd_messaging_context());
+	ctx->pending = false;
 }
 
 /****************************************************************************
  Setup kernel oplocks.
 ****************************************************************************/
 
-struct kernel_oplocks *irix_init_kernel_oplocks(void) 
+static const struct kernel_oplocks_ops irix_koplocks = {
+	.receive_message	= irix_oplock_receive_message,
+	.set_oplock		= irix_set_kernel_oplock,
+	.release_oplock		= irix_release_kernel_oplock,
+	.msg_waiting		= irix_oplock_msg_waiting
+};
+
+struct kernel_oplocks *irix_init_kernel_oplocks(TALLOC_CTX *mem_ctx)
 {
+	struct kernel_oplocks *_ctx;
+	struct irix_oplocks_context *ctx;
 	int pfd[2];
 
 	if (!irix_oplocks_available())
 		return NULL;
 
+	_ctx = talloc_zero(mem_ctx, struct kernel_oplocks);
+	if (!_ctx) {
+		return NULL;
+	}
+
+	ctx = talloc_zero(_ctx, struct irix_oplocks_context);
+	if (!ctx) {
+		talloc_free(_ctx);
+		return NULL;
+	}
+	_ctx->ops = &irix_koplocks;
+	_ctx->private_data = ctx;
+	ctx->ctx = _ctx;
+
 	if(pipe(pfd) != 0) {
+		talloc_free(_ctx);
 		DEBUG(0,("setup_kernel_oplock_pipe: Unable to create pipe. "
 			 "Error was %s\n", strerror(errno) ));
 		return False;
 	}
 
-	oplock_pipe_read = pfd[0];
-	oplock_pipe_write = pfd[1];
-
-	irix_koplocks.receive_message = irix_oplock_receive_message;
-	irix_koplocks.set_oplock = irix_set_kernel_oplock;
-	irix_koplocks.release_oplock = irix_release_kernel_oplock;
-	irix_koplocks.msg_waiting = irix_oplock_msg_waiting;
-	irix_koplocks.notification_fd = oplock_pipe_read;
+	ctx->read_fd = pfd[0];
+	ctx->write_fd = pfd[1];
 
-	return &irix_koplocks;
+	ctx->read_fde = event_add_fd(smbd_event_context(),
+				     ctx,
+				     ctx->read_fd,
+				     EVENT_FD_READ,
+				     irix_oplocks_read_fde_handler,
+				     ctx);
+	return _ctx;
 }
 #else
  void oplock_irix_dummy(void);
diff --git a/source3/smbd/oplock_linux.c b/source3/smbd/oplock_linux.c
index cb37a81fb9..8087167ff4 100644
--- a/source3/smbd/oplock_linux.c
+++ b/source3/smbd/oplock_linux.c
@@ -101,7 +101,7 @@ int linux_setlease(int fd, int leasetype)
  * oplock break protocol.
 ****************************************************************************/
 
-static files_struct *linux_oplock_receive_message(fd_set *fds)
+static files_struct *linux_oplock_receive_message(struct kernel_oplocks *ctx)
 {
 	int fd;
 	files_struct *fsp;
@@ -125,7 +125,8 @@ static files_struct *linux_oplock_receive_message(fd_set *fds)
  Attempt to set an kernel oplock on a file.
 ****************************************************************************/
 
-static bool linux_set_kernel_oplock(files_struct *fsp, int oplock_type)
+static bool linux_set_kernel_oplock(struct kernel_oplocks *ctx,
+				    files_struct *fsp, int oplock_type)
 {
 	if ( SMB_VFS_LINUX_SETLEASE(fsp, F_WRLCK) == -1) {
 		DEBUG(3,("linux_set_kernel_oplock: Refused oplock on file %s, "
@@ -148,7 +149,8 @@ static bool linux_set_kernel_oplock(files_struct *fsp, int oplock_type)
  Release a kernel oplock on a file.
 ****************************************************************************/
 
-static void linux_release_kernel_oplock(files_struct *fsp)
+static void linux_release_kernel_oplock(struct kernel_oplocks *ctx,
+					files_struct *fsp)
 {
 	if (DEBUGLVL(10)) {
 		/*
@@ -181,7 +183,7 @@ static void linux_release_kernel_oplock(files_struct *fsp)
  See if a oplock message is waiting.
 ****************************************************************************/
 
-static bool linux_oplock_msg_waiting(fd_set *fds)
+static bool linux_oplock_msg_waiting(struct kernel_oplocks *ctx)
 {
 	return oplock_signals_received != 0;
 }
@@ -205,15 +207,31 @@ static bool linux_oplocks_available(void)
  Setup kernel oplocks.
 ****************************************************************************/
 
-struct kernel_oplocks *linux_init_kernel_oplocks(void) 
+static const struct kernel_oplocks_ops linux_koplocks = {
+	.receive_message	= linux_oplock_receive_message,
+	.set_oplock		= linux_set_kernel_oplock,
+	.release_oplock		= linux_release_kernel_oplock,
+	.msg_waiting		= linux_oplock_msg_waiting
+};
+
+struct kernel_oplocks *linux_init_kernel_oplocks(TALLOC_CTX *mem_ctx)
 {
         struct sigaction act;
+	struct kernel_oplocks *ctx;
 
 	if (!linux_oplocks_available()) {
 		DEBUG(3,("Linux kernel oplocks not available\n"));
 		return NULL;
 	}
 
+	ctx = talloc_zero(mem_ctx, struct kernel_oplocks);
+	if (!ctx) {
+		DEBUG(0,("Linux Kernel oplocks talloc_Zero failed\n"));
+		return NULL;
+	}
+
+	ctx->ops = &linux_koplocks;
+
 	ZERO_STRUCT(act);
 
 	act.sa_handler = NULL;
@@ -225,18 +243,12 @@ struct kernel_oplocks *linux_init_kernel_oplocks(void)
 		return NULL;
 	}
 
-	linux_koplocks.receive_message = linux_oplock_receive_message;
-	linux_koplocks.set_oplock = linux_set_kernel_oplock;
-	linux_koplocks.release_oplock = linux_release_kernel_oplock;
-	linux_koplocks.msg_waiting = linux_oplock_msg_waiting;
-	linux_koplocks.notification_fd = -1;
-
 	/* the signal can start off blocked due to a bug in bash */
 	BlockSignals(False, RT_SIGNAL_LEASE);
 
 	DEBUG(3,("Linux kernel oplocks enabled\n"));
 
-	return &linux_koplocks;
+	return ctx;
 }
 #else
  void oplock_linux_dummy(void);
diff --git a/source3/smbd/process.c b/source3/smbd/process.c
index 7847505f7c..78e66e4620 100644
--- a/source3/smbd/process.c
+++ b/source3/smbd/process.c
@@ -700,13 +700,13 @@ struct idle_event *event_add_idle(struct event_context *event_ctx,
  notify events etc.
 ****************************************************************************/
 
-static void async_processing(fd_set *pfds)
+static void async_processing(void)
 {
 	DEBUG(10,("async_processing: Doing async processing.\n"));
 
 	process_aio_queue();
 
-	process_kernel_oplocks(smbd_messaging_context(), pfds);
+	process_kernel_oplocks(smbd_messaging_context());
 
 	/* Do the aio check again after receive_local_message as it does a
 	   select and may have eaten our signal. */
@@ -727,20 +727,6 @@ static void async_processing(fd_set *pfds)
 }
 
 /****************************************************************************
- Add a fd to the set we will be select(2)ing on.
-****************************************************************************/
-
-static int select_on_fd(int fd, int maxfd, fd_set *fds)
-{
-	if (fd != -1) {
-		FD_SET(fd, fds);
-		maxfd = MAX(maxfd, fd);
-	}
-
-	return maxfd;
-}
-
-/****************************************************************************
   Do a select on an two fd's - with timeout. 
 
   If a local udp message has been pushed onto the
@@ -794,9 +780,9 @@ static NTSTATUS smbd_server_connection_loop_once(struct smbd_server_connection *
 	 * This is hideously complex - *MUST* be simplified for 3.0 ! JRA.
 	 */
 
-	if (oplock_message_waiting(&r_fds)) {
+	if (oplock_message_waiting()) {
 		DEBUG(10,("receive_message_or_smb: oplock_message is waiting.\n"));
-		async_processing(&r_fds);
+		async_processing();
 		/*
 		 * After async processing we must go and do the select again, as
 		 * the state of the flag in fds for the server file descriptor is
@@ -827,8 +813,6 @@ static NTSTATUS smbd_server_connection_loop_once(struct smbd_server_connection *
 		int sav;
 		START_PROFILE(smbd_idle);
 
-		maxfd = select_on_fd(oplock_notify_fd(), maxfd, &r_fds);
-
 		selrtn = sys_select(maxfd+1,&r_fds,&w_fds,NULL,&to);
 		sav = errno;
 
@@ -845,7 +829,7 @@ static NTSTATUS smbd_server_connection_loop_once(struct smbd_server_connection *
 	   is the best we can do until the oplock code knows more about
 	   signals */
 	if (selrtn == -1 && errno == EINTR) {
-		async_processing(&r_fds);
+		async_processing();
 		/*
 		 * After async processing we must go and do the select again, as
 		 * the state of the flag in fds for the server file descriptor is
@@ -866,22 +850,6 @@ static NTSTATUS smbd_server_connection_loop_once(struct smbd_server_connection *
 	}
 
 	/*
-	 * Ensure we process oplock break messages by preference.
-	 * This is IMPORTANT ! Otherwise we can starve other processes
-	 * sending us an oplock break message. JRA.
-	 */
-
-	if (oplock_message_waiting(&r_fds)) {
-		async_processing(&r_fds);
-		/*
-		 * After async processing we must go and do the select again, as
-		 * the state of the flag in fds for the server file descriptor is
-		 * indeterminate - we may have done I/O on it in the oplock processing. JRA.
-		 */
-		return NT_STATUS_RETRY;
-	}
-
-	/*
 	 * We've just woken up from a protentially long select sleep.
 	 * Ensure we process local messages as we need to synchronously
 	 * process any messages from other smbd's to avoid file rename race
@@ -935,7 +903,7 @@ void respond_to_all_remaining_local_messages(void)
 		return;
 	}
 
-	process_kernel_oplocks(smbd_messaging_context(), NULL);
+	process_kernel_oplocks(smbd_messaging_context());
 
 	return;
 }