Merge pull request #17 from WukLab/integration/tfork-perf-tune-filecow-pidns

Integration/tfork perf tune filecow pidns

Author: dothyt

README.md

@@ -178,7 +178,10 @@ check [agent-s README](agents/agent-s/README.md)
 - btrfs only
 - rootful only
 - amd64 only
-- linux page-cache CoW currently has a memory leak that will be fixed
+- Filecow layers are reclaimed with their cached inode/mapping owners rather
+  than eagerly when a child is deleted. Repeated unchanged forks reuse the
+  current layer, so retained layers are bounded by changed generations instead
+  of the total fork count.
 
 ## License
 

criu/build.sh

@@ -93,12 +93,25 @@ for mod in "${ACTIVE_MODULES[@]}"; do
     fi
 
     if lsmod | awk '{print $1}' | grep -qx "${mod}"; then
-        echo "  -- ${mod}: already loaded, rmmod then insmod"
-        rmmod "${mod}" 2>/dev/null || true
-    else
-        echo "  -- ${mod}: insmod"
+        echo "  -- ${mod}: already loaded, trying reload"
+        if ! rmmod "${mod}"; then
+            echo "build.sh: failed to unload loaded module ${mod}; aborting to avoid stale module" >&2
+            echo "build.sh: stop running tfork/podman containers that may still hold ${mod}, then retry" >&2
+            exit 1
+        fi
+        if lsmod | awk '{print $1}' | grep -qx "${mod}"; then
+            echo "build.sh: module ${mod} is still loaded after rmmod; aborting to avoid stale module" >&2
+            echo "build.sh: stop running tfork/podman containers that may still hold ${mod}, then retry" >&2
+            exit 1
+        fi
+    fi
+
+    echo "  -- ${mod}: insmod"
+    if ! insmod "${ko_path}"; then
+        echo "build.sh: failed to insert rebuilt module ${mod} from ${ko_path}" >&2
+        echo "build.sh: if the old module is still active, stop running tfork/podman containers and rerun this script" >&2
+        exit 1
     fi
-    insmod "${ko_path}"
 done
 
 echo "  -- verify all modules loaded"

criu/criu/action-scripts.c

@@ -32,6 +32,9 @@ static const char *action_names[ACT_MAX] = {
 	[ACT_ORPHAN_PTS_MASTER] = "orphan-pts-master",
 	[ACT_STATUS_READY] = "status-ready",
 	[ACT_QUERY_EXT_FILES] = "query-ext-files",
+	[ACT_POST_TFORK_FREEZE] = "post-tfork-freeze",
+	[ACT_PRE_TFORK_RESTORE] = "pre-tfork-restore",
+	[ACT_TFORK_SOURCE_DETACHED] = "tfork-source-detached",
 };
 
 struct script {

criu/criu/clone-noasan.c

@@ -78,7 +78,10 @@ int clone3_with_pid_noasan(int (*fn)(void *), void *arg, int flags, int exit_sig
 	c_args.flags = flags;
 	c_args.set_tid = ptr_to_u64(&pid);
 	c_args.set_tid_size = 1;
+	pr_info("clone3 set_tid pid=%d flags=0x%x size=1\n", pid, flags);
 	pid = syscall(__NR_clone3, &c_args, sizeof(c_args));
+	if (pid < 0)
+		pr_perror("clone3 set_tid failed flags=0x%x size=1", flags);
 	if (pid == 0)
 		exit(fn(arg));
 	return pid;
@@ -99,6 +102,12 @@ int clone3_with_nested_pid_noasan(int (*fn)(void *), void *arg, int flags, int e
 	BUG_ON(pid->ns_level > MAX_PID_NS_LEVEL || pid->ns_level <= 1);
 	for (i = 0; i < pid->ns_level; i++)
 		tids[i] = pid->ns[i].ns_pid;
+	pr_info("clone3 nested set_tid flags=0x%x size=%d tids=%d/%d/%d/%d\n",
+		flags, pid->ns_level,
+		tids[0],
+		pid->ns_level > 1 ? tids[1] : -1,
+		pid->ns_level > 2 ? tids[2] : -1,
+		pid->ns_level > 3 ? tids[3] : -1);
 
 	if (!(flags & CLONE_PARENT)) {
 		if (exit_signal != SIGCHLD) {
@@ -112,6 +121,13 @@ int clone3_with_nested_pid_noasan(int (*fn)(void *), void *arg, int flags, int e
 	c_args.set_tid = ptr_to_u64(tids);
 	c_args.set_tid_size = pid->ns_level;
 	pid_ret = syscall(__NR_clone3, &c_args, sizeof(c_args));
+	if (pid_ret < 0)
+		pr_perror("clone3 nested set_tid failed flags=0x%x size=%d tids=%d/%d/%d/%d",
+			  flags, pid->ns_level,
+			  tids[0],
+			  pid->ns_level > 1 ? tids[1] : -1,
+			  pid->ns_level > 2 ? tids[2] : -1,
+			  pid->ns_level > 3 ? tids[3] : -1);
 	if (pid_ret == 0)
 		exit(fn(arg));
 	return pid_ret;