Proposal for support of dmabuf with CRIU

criu/cr-dump.c

@@ -2228,6 +2228,9 @@ int cr_dump_tasks(pid_t pid)
 			goto err;
 	}
 
+	if(run_plugins(DUMP_DEVICE_LATE, pid))
+		goto err;
+
 	if (parent_ie) {
 		inventory_entry__free_unpacked(parent_ie, NULL);
 		parent_ie = NULL;

criu/cr-restore.c

@@ -3470,6 +3470,7 @@ static int sigreturn_restore(pid_t pid, struct task_restore_args *task_args, uns
 	close_image_dir();
 	close_proc();
 	close_service_fd(TRANSPORT_FD_OFF);
+	close_service_fd(DMABUF_FD_OFF);
 	close_service_fd(CR_PROC_FD_OFF);
 	close_service_fd(ROOT_FD_OFF);
 	close_service_fd(USERNSD_SK);

criu/files-ext.c

@@ -45,10 +45,11 @@ static int open_fd(struct file_desc *d, int *new_fd)
 {
 	struct ext_file_info *xfi;
 	int fd;
+	bool retry_needed;
 
 	xfi = container_of(d, struct ext_file_info, d);
 
-	fd = run_plugins(RESTORE_EXT_FILE, xfi->xfe->id);
+	fd = run_plugins(RESTORE_EXT_FILE, xfi->xfe->id, &retry_needed);
 	if (fd < 0) {
 		pr_err("Unable to restore %#x\n", xfi->xfe->id);
 		return -1;
@@ -57,8 +58,11 @@ static int open_fd(struct file_desc *d, int *new_fd)
 	if (restore_fown(fd, xfi->xfe->fown))
 		return -1;
 
-	*new_fd = fd;
-	return 0;
+	if (!retry_needed)
+		*new_fd = fd;
+	else
+		*new_fd = -1;
+	return retry_needed;
 }
 
 static struct file_desc_ops ext_desc_ops = {

criu/files.c

@@ -62,6 +62,8 @@
 static struct hlist_head file_desc_hash[FDESC_HASH_SIZE];
 /* file_desc's, which fle is not owned by a process, that is able to open them */
 static LIST_HEAD(fake_master_head);
+/* processes that have a file from a plugin that can use shared dmabuf_fds */
+static LIST_HEAD(dmabuf_processes);
 
 static u32 max_file_desc_id = 0;
 
@@ -831,10 +833,35 @@
 	}
 }
 
+static int add_pid_to_dmabuf_list(int pid)
+{
+	struct fdinfo_list_entry *le;
+
+	list_for_each_entry(le, &dmabuf_processes, desc_list)
+		if (le->pid == pid)
+			return 0;
+
+	le = alloc_fle(pid, NULL);
+
+	if (!le)
+		return -ENOMEM;
+
+	list_add(&le->desc_list, &dmabuf_processes);
+
+	return 0;
+}
+
 struct fdinfo_list_entry *collect_fd_to(int pid, FdinfoEntry *e, struct rst_info *rst_info, struct file_desc *fdesc,
 					bool fake, bool force_master)
 {
 	struct fdinfo_list_entry *new_le;
+	int ret;
+
+	if (fdesc->ops->type == FD_TYPES__EXT) {
+		ret = add_pid_to_dmabuf_list(pid);
+		if (ret)
+			return NULL;
+	}
 
 	new_le = alloc_fle(pid, e);
 	if (new_le) {
@@ -983,6 +1010,14 @@
 	*addr->sun_path = '\0';
 }
 
+static void dmabuf_socket_name_gen(struct sockaddr_un *addr, int *len, int pid)
+{
+	addr->sun_family = AF_UNIX;
+	snprintf(addr->sun_path, UNIX_PATH_MAX, "x/crtools-dmabuf-%d-%" PRIx64, pid, criu_run_id);
+	*len = SUN_LEN(addr);
+	*addr->sun_path = '\0';
+}
+
 static bool task_fle(struct pstree_item *task, struct fdinfo_list_entry *fle)
 {
 	struct fdinfo_list_entry *tmp;
@@ -1028,6 +1063,45 @@
 	return 0;
 }
 
+static int recv_dmabuf_fds(void)
+{
+	int fd, newfd, ret, tsock, handle;
+
+	tsock = get_service_fd(DMABUF_FD_OFF);
+
+	while (true) {
+		ret = __recv_fds(tsock, &fd, 1, (void *)&handle, sizeof(handle), MSG_DONTWAIT);
+
+		if (ret == -EAGAIN || ret == -EWOULDBLOCK)
+			return 1;
+		else if (ret)
+			return -1;
+
+		newfd = get_unused_high_fd();
+
+		reopen_fd_as(newfd, fd);
+
+		run_plugins(DMABUF_FD, handle, newfd);
+	}
+
+	return 0;
+}
+
+static int send_dmabuf_fd_to_peer(int handle, int fd, int pid)
+{
+	struct sockaddr_un saddr;
+	int len, sock, ret;
+
+	sock = get_service_fd(DMABUF_FD_OFF);
+
+	dmabuf_socket_name_gen(&saddr, &len, pid);
+	pr_info("\t\tSend dmabuf fd %d for handle %d to %s\n", fd, handle, saddr.sun_path + 1);
+	ret = send_fds(sock, &saddr, len, &fd, 1, (void *)&handle, sizeof(handle));
+	if (ret < 0)
+		return -1;
+	return set_fds_event(pid);
+}
+
 static int send_fd_to_peer(int fd, struct fdinfo_list_entry *fle)
 {
 	struct sockaddr_un saddr;
@@ -1132,6 +1206,25 @@
 	return 0;
 }
 
+int serve_out_dmabuf_fd(int handle, int fd)
+{
+	int ret;
+	struct fdinfo_list_entry *fle;
+
+	list_for_each_entry(fle, &dmabuf_processes, desc_list) {
+		ret = send_dmabuf_fd_to_peer(handle, fd, fle->pid);
+
+		if (ret) {
+			pr_err("Can't sent fd %d to %d\n", fd, fle->pid);
+			goto out;
+		}
+	}
+
+	ret = 0;
+out:
+	return ret;
+}
+
 static int open_fd(struct fdinfo_list_entry *fle)
 {
 	struct file_desc *d = fle->desc;
@@ -1212,6 +1305,7 @@
 	do {
 		progress = again = false;
 		clear_fds_event();
+		recv_dmabuf_fds();
 
 		list_for_each_entry_safe(fle, tmp, list, ps_list) {
 			st = fle->stage;
@@ -1707,6 +1801,25 @@
 
 	if (install_service_fd(TRANSPORT_FD_OFF, sock) < 0)
 		goto out;
+
+
+	sock = socket(PF_UNIX, SOCK_DGRAM | SOCK_CLOEXEC, 0);
+	if (sock < 0) {
+		pr_perror("Can't create socket");
+		goto out;
+	}
+
+	dmabuf_socket_name_gen(&saddr, &slen, pid);
+	if (bind(sock, (struct sockaddr *)&saddr, slen) < 0) {
+		pr_perror("Can't bind dmabuf socket %s", saddr.sun_path + 1);
+		close(sock);
+		goto out;
+	}
+
+	if (install_service_fd(DMABUF_FD_OFF, sock) < 0)
+		goto out;
+
+
 	ret = 0;
 out:
 	return ret;

criu/include/criu-plugin.h

@@ -60,6 +60,10 @@ enum {
 
 	CR_PLUGIN_HOOK__CHECKPOINT_DEVICES = 11,
 
+	CR_PLUGIN_HOOK__DUMP_DEVICE_LATE = 12,
+
+	CR_PLUGIN_HOOK__DMABUF_FD = 13,
+
 	CR_PLUGIN_HOOK__MAX
 };
 
@@ -68,7 +72,7 @@ enum {
 DECLARE_PLUGIN_HOOK_ARGS(CR_PLUGIN_HOOK__DUMP_UNIX_SK, int fd, int id);
 DECLARE_PLUGIN_HOOK_ARGS(CR_PLUGIN_HOOK__RESTORE_UNIX_SK, int id);
 DECLARE_PLUGIN_HOOK_ARGS(CR_PLUGIN_HOOK__DUMP_EXT_FILE, int fd, int id);
-DECLARE_PLUGIN_HOOK_ARGS(CR_PLUGIN_HOOK__RESTORE_EXT_FILE, int id);
+DECLARE_PLUGIN_HOOK_ARGS(CR_PLUGIN_HOOK__RESTORE_EXT_FILE, int id, bool *retry_needed);
 DECLARE_PLUGIN_HOOK_ARGS(CR_PLUGIN_HOOK__DUMP_EXT_MOUNT, char *mountpoint, int id);
 DECLARE_PLUGIN_HOOK_ARGS(CR_PLUGIN_HOOK__RESTORE_EXT_MOUNT, int id, char *mountpoint, char *old_root, int *is_file);
 DECLARE_PLUGIN_HOOK_ARGS(CR_PLUGIN_HOOK__DUMP_EXT_LINK, int index, int type, char *kind);
@@ -78,6 +82,8 @@ DECLARE_PLUGIN_HOOK_ARGS(CR_PLUGIN_HOOK__UPDATE_VMA_MAP, const char *path, const
 DECLARE_PLUGIN_HOOK_ARGS(CR_PLUGIN_HOOK__RESUME_DEVICES_LATE, int pid);
 DECLARE_PLUGIN_HOOK_ARGS(CR_PLUGIN_HOOK__PAUSE_DEVICES, int pid);
 DECLARE_PLUGIN_HOOK_ARGS(CR_PLUGIN_HOOK__CHECKPOINT_DEVICES, int pid);
+DECLARE_PLUGIN_HOOK_ARGS(CR_PLUGIN_HOOK__DUMP_DEVICE_LATE, int id);
+DECLARE_PLUGIN_HOOK_ARGS(CR_PLUGIN_HOOK__DMABUF_FD, int handle, int fd);
 
 enum {
 	CR_PLUGIN_STAGE__DUMP,

criu/include/files.h

@@ -195,5 +195,7 @@ extern int open_transport_socket(void);
 extern int set_fds_event(pid_t virt);
 extern void wait_fds_event(void);
 
+int serve_out_dmabuf_fd(int handle, int fd);
+
 int find_unused_fd_pid(pid_t pid);
 #endif /* __CR_FILES_H__ */

criu/include/servicefd.h

@@ -28,6 +28,7 @@ enum sfd_type {
 	USERNSD_SK,	  /* Socket for usernsd */
 	NS_FD_OFF,	  /* Node's net namespace fd */
 	TRANSPORT_FD_OFF, /* to transfer file descriptors */
+	DMABUF_FD_OFF,
 	RPC_SK_OFF,
 	FDSTORE_SK_OFF,
 
@@ -47,5 +48,6 @@ extern int install_service_fd(enum sfd_type type, int fd);
 extern int close_service_fd(enum sfd_type type);
 extern void __close_service_fd(enum sfd_type type);
 extern int clone_service_fd(struct pstree_item *me);
+extern int get_unused_high_fd(void);
 
 #endif /* __CR_SERVICE_FD_H__ */

criu/pie/restorer.c

@@ -1920,6 +1920,10 @@ __visible long __export_restore_task(struct task_restore_args *args)
 
 		for (m = 0; m < sizeof(vma_entry->madv) * 8; m++) {
 			if (vma_entry->madv & (1ul << m)) {
+
+				if (!(vma_entry_is(vma_entry, VMA_AREA_REGULAR)))
+					continue;
+
 				ret = sys_madvise(vma_entry->start, vma_entry_len(vma_entry), m);
 				if (ret) {
 					pr_err("madvise(%" PRIx64 ", %" PRIu64 ", %ld) "

criu/plugin.c

@@ -59,6 +59,7 @@ static cr_plugin_desc_t *cr_gen_plugin_desc(void *h, char *path)
 	__assign_hook(RESUME_DEVICES_LATE, "cr_plugin_resume_devices_late");
 	__assign_hook(PAUSE_DEVICES, "cr_plugin_pause_devices");
 	__assign_hook(CHECKPOINT_DEVICES, "cr_plugin_checkpoint_devices");
+	__assign_hook(DUMP_DEVICE_LATE, "cr_plugin_dump_device_late");
 
 #undef __assign_hook
 

criu/servicefd.c

@@ -25,6 +25,7 @@ int service_fd_rlim_cur;
 
 /* Base of current process service fds set */
 static int service_fd_base;
+static int next_high_fd;
 
 /* Id of current process in shared fdt */
 static int service_fd_id = 0;
@@ -53,6 +54,7 @@ const char *sfd_type_name(enum sfd_type type)
 		[USERNSD_SK] = __stringify_1(USERNSD_SK),
 		[NS_FD_OFF] = __stringify_1(NS_FD_OFF),
 		[TRANSPORT_FD_OFF] = __stringify_1(TRANSPORT_FD_OFF),
+		[DMABUF_FD_OFF] = __stringify_1(DMABUF_FD_OFF),
 		[RPC_SK_OFF] = __stringify_1(RPC_SK_OFF),
 		[FDSTORE_SK_OFF] = __stringify_1(FDSTORE_SK_OFF),
 		[SERVICE_FD_MAX] = __stringify_1(SERVICE_FD_MAX),
@@ -312,5 +314,15 @@ int clone_service_fd(struct pstree_item *me)
 	service_fd_id = id;
 	ret = 0;
 
+	next_high_fd = service_fd_base + 1024;
+
 	return ret;
 }
+
+int get_unused_high_fd(void)
+{
+	if (next_high_fd > service_fd_rlim_cur)
+		return -1;
+	next_high_fd += 1;
+	return next_high_fd - 1;
+}