rst: Evaluate per-task clone mask early

When we've read all pstree-items and their ids we
can get the desired clone-flags early and avoid all
these dances with flag calculations in fork_with_pid
and company.

Signed-off-by: Pavel Emelyanov <xemul@parallels.com>

cr-restore.c

@@ -799,19 +799,17 @@ static void write_pidfile(char *pfname, int pid)
 	close(fd);
 }
 
-static inline int fork_with_pid(struct pstree_item *item, unsigned long ns_clone_flags)
+static inline int fork_with_pid(struct pstree_item *item)
 {
 	int ret = -1;
 	struct cr_clone_arg ca;
 	pid_t pid = item->pid.virt;
 
-	pr_info("Forking task with %d pid (flags 0x%lx)\n", pid, ns_clone_flags);
 
 	ca.item = item;
-	ca.clone_flags = ns_clone_flags;
+	ca.clone_flags = item->rst->clone_flags;
 
-	if (shared_fdtable(item))
+	pr_info("Forking task with %d pid (flags 0x%lx)\n", pid, ca.clone_flags);
-		ca.clone_flags |= CLONE_FILES;
 
 	if (!(ca.clone_flags & CLONE_NEWPID)) {
 		char buf[32];
@@ -1070,7 +1068,7 @@ static int restore_task_with_children(void *_arg)
 
 		BUG_ON(child->born_sid != -1 && getsid(getpid()) != child->born_sid);
 
-		ret = fork_with_pid(child, 0);
+		ret = fork_with_pid(child);
 		if (ret < 0)
 			exit(1);
 	}
@@ -1081,7 +1079,7 @@ static int restore_task_with_children(void *_arg)
 	list_for_each_entry(child, &current->children, sibling) {
 		if (restore_before_setsid(child))
 			continue;
-		ret = fork_with_pid(child, 0);
+		ret = fork_with_pid(child);
 		if (ret < 0)
 			exit(1);
 	}
@@ -1176,7 +1174,7 @@ static int restore_root_task(struct pstree_item *init, struct cr_options *opts)
 
 	futex_set(&task_entries->nr_in_progress, stage_participants(CR_STATE_FORKING));
 
-	ret = fork_with_pid(init, current_ns_mask);
+	ret = fork_with_pid(init);
 	if (ret < 0)
 		return -1;
 

include/crtools.h

@@ -254,9 +254,11 @@ struct rst_info {
 
 	void			*premmapped_addr;
 	unsigned long		premmapped_len;
+	unsigned long		clone_flags;
 
 	int service_fd_id;
 	struct fdt		*fdt;
+
 };
 
 static inline int in_vma_area(struct vma_area *vma, unsigned long addr)

include/namespaces.h

@@ -22,6 +22,5 @@ struct pstree_item;
 int dump_task_ns_ids(struct pstree_item *);
 
 extern unsigned long current_ns_mask;
-int check_ns_ids(struct pstree_item *);
 
 #endif /* __CR_NS_H__ */

include/syscall-types.h

@@ -55,6 +55,8 @@ struct robust_list_head;
 #define CLONE_NEWNET	0x40000000
 #endif
 
+#define CLONE_ALLNS	(CLONE_NEWPID | CLONE_NEWNET | CLONE_NEWIPC | CLONE_NEWUTS | CLONE_NEWNS)
+
 #define setns	sys_setns
 
 struct rlimit;

namespaces.c

@@ -260,56 +260,6 @@ int dump_namespaces(struct pid *ns_pid, unsigned int ns_flags)
 	return 0;
 }
 
-static unsigned long get_clone_mask(TaskKobjIdsEntry *i,
-		TaskKobjIdsEntry *p)
-{
-	unsigned long mask = 0;
-
-	if (i->pid_ns_id != p->pid_ns_id)
-		mask |= CLONE_NEWPID;
-	if (i->net_ns_id != p->net_ns_id)
-		mask |= CLONE_NEWNET;
-	if (i->ipc_ns_id != p->ipc_ns_id)
-		mask |= CLONE_NEWIPC;
-	if (i->uts_ns_id != p->uts_ns_id)
-		mask |= CLONE_NEWUTS;
-	if (i->mnt_ns_id != p->mnt_ns_id)
-		mask |= CLONE_NEWNS;
-
-	return mask;
-}
-
-int check_ns_ids(struct pstree_item *item)
-{
-	struct pstree_item *p = item->parent;
-
-	if (!p) {
-		if (!root_ids) {
-			current_ns_mask = opts.rst_namespaces_flags;
-			return 0;
-		}
-
-		current_ns_mask = get_clone_mask(item->ids, root_ids);
-		pr_info("Will restore in %lx namespaces\n", current_ns_mask);
-		return 0;
-	}
-
-	if (!item->ids)
-		return 0;
-
-	while (!p->ids) {
-		p = p->parent;
-		BUG_ON(!p); /* must meet the root_item */
-	}
-
-	if (get_clone_mask(item->ids, p->ids)) {
-		pr_err("Task in sub namespace\n");
-		return -1;
-	}
-
-	return 0;
-}
-
 int prepare_namespace(int pid, unsigned long clone_flags)
 {
 	pr_info("Restoring namespaces %d flags 0x%lx\n",

pstree.c

@@ -304,7 +304,7 @@ static int prepare_pstree_ids(void)
 		if (item->sid == root_item->sid || item->sid == item->pid.virt)
 			continue;
 
-		helper = alloc_pstree_item();
+		helper = alloc_pstree_item_with_rst();
 		if (helper == NULL)
 			return -1;
 		helper->sid = item->sid;
@@ -422,7 +422,7 @@ static int prepare_pstree_ids(void)
 		if (current_pgid == item->pgid)
 			continue;
 
-		helper = alloc_pstree_item();
+		helper = alloc_pstree_item_with_rst();
 		if (helper == NULL)
 			return -1;
 		helper->sid = item->sid;
@@ -440,6 +440,27 @@ static int prepare_pstree_ids(void)
 	return 0;
 }
 
+static unsigned long get_clone_mask(TaskKobjIdsEntry *i,
+		TaskKobjIdsEntry *p)
+{
+	unsigned long mask = 0;
+
+	if (i->files_id == p->files_id)
+		mask |= CLONE_FILES;
+	if (i->pid_ns_id != p->pid_ns_id)
+		mask |= CLONE_NEWPID;
+	if (i->net_ns_id != p->net_ns_id)
+		mask |= CLONE_NEWNET;
+	if (i->ipc_ns_id != p->ipc_ns_id)
+		mask |= CLONE_NEWIPC;
+	if (i->uts_ns_id != p->uts_ns_id)
+		mask |= CLONE_NEWUTS;
+	if (i->mnt_ns_id != p->mnt_ns_id)
+		mask |= CLONE_NEWNS;
+
+	return mask;
+}
+
 static int prepare_pstree_kobj_ids(void)
 {
 	struct pstree_item *item;
@@ -447,23 +468,49 @@ static int prepare_pstree_kobj_ids(void)
 	/* Find a process with minimal pid for shared fd tables */
 	for_each_pstree_item(item) {
 		struct pstree_item *parent = item->parent;
+		TaskKobjIdsEntry *ids;
+		unsigned long cflags;
+
+		if (!item->ids) {
+			if (item == root_item) {
+				cflags = opts.rst_namespaces_flags;
+				goto set_mask;
+			}
 
-		if (item->state == TASK_HELPER)
 			continue;
+		}
 
-		if (check_ns_ids(item))
+		if (parent)
-			return -1;
+			ids = parent->ids;
+		else
+			ids = root_ids;
 
-		if (parent == NULL)
+		cflags = get_clone_mask(item->ids, ids);
-			continue;
 
-		if (shared_fdtable(item)) {
+		if (cflags & CLONE_FILES) {
 			int ret;
 
 			ret = shared_fdt_prepare(item);
 			if (ret)
 				return ret;
 		}
+
+set_mask:
+		item->rst->clone_flags = cflags;
+
+		/*
+		 * Workaround for current namespaces model --
+		 * all tasks should be in one namespace. And
+		 * this namespace is either inherited from the
+		 * crtools or is created for the init task (only)
+		 */
+		if (item == root_item) {
+			pr_info("Will restore in %lx namespaces\n", cflags);
+			current_ns_mask = cflags & CLONE_ALLNS;
+		} else if (cflags & CLONE_ALLNS) {
+			pr_err("Can't restore sub-task in NS\n");
+			return -1;
+		}
 	}
 
 	return 0;