Merge branch 'checkpoint-restore:criu-dev' into assorted-fixes

criu/cgroup.c

@@ -639,8 +639,8 @@ static int open_cgroupfs(struct cg_ctl *cc)
 		return -1;
 	}
 
-	if (mount("none", prefix, fstype, 0, mopts) < 0) {
-		pr_perror("Unable to mount %s", mopts);
+	if (mount("none", prefix, fstype, 0, mopts[0] ? mopts : NULL) < 0) {
+		pr_perror("Unable to mount %s %s", fstype, mopts);
 		rmdir(prefix);
 		return -1;
 	}

criu/cr-restore.c

@@ -2971,7 +2971,7 @@ static int prepare_mm(pid_t pid, struct task_restore_args *args)
 
 	args->fd_exe_link = exe_fd;
 
-	args->has_thp_enabled = rsti(current)->has_thp_enabled;
+	args->thp_disabled = mm->has_thp_disabled && mm->thp_disabled;
 
 	ret = 0;
 out:

criu/include/restorer.h

@@ -144,7 +144,7 @@ struct task_restore_args {
 	struct timeval logstart;
 
 	int uffd;
-	bool has_thp_enabled;
+	bool thp_disabled;
 
 	/* threads restoration */
 	int nr_threads;				 /* number of threads */

criu/include/rst_info.h

@@ -73,8 +73,6 @@ struct rst_info {
 	 */
 	bool has_old_seccomp_filter;
 
-	bool has_thp_enabled;
-
 	struct rst_rseq *rseqe;
 
 	void *breakpoint;

criu/kerndat.c

@@ -1324,6 +1324,8 @@ int kerndat_has_thp_disable(void)
 
 			parse_vmflags(str, &flags, &madv, &io_pf);
 			kdat.has_thp_disable = !(madv & (1 << MADV_NOHUGEPAGE));
+			if (!kdat.has_thp_disable)
+				pr_warn("prctl PR_SET_THP_DISABLE sets MADV_NOHUGEPAGE");
 			break;
 		}
 	}

criu/mem.c

@@ -1217,8 +1217,6 @@ static int restore_priv_vma_content(struct pstree_item *t, struct page_read *pr)
 
 static int maybe_disable_thp(struct pstree_item *t, struct page_read *pr)
 {
-	MmEntry *mm = rsti(t)->mm;
-
 	/*
 	 * There is no need to disable it if the page read doesn't
 	 * have parent. In this case VMA will be empty until
@@ -1241,8 +1239,6 @@ static int maybe_disable_thp(struct pstree_item *t, struct page_read *pr)
 		pr_perror("Cannot disable THP");
 		return -1;
 	}
-	if (!(mm->has_thp_disabled && mm->thp_disabled))
-		rsti(t)->has_thp_enabled = true;
 
 	return 0;
 }

criu/memfd.c

@@ -93,8 +93,17 @@ static int dump_memfd_inode(int fd, struct memfd_dump_inode *inode, const char *
 	}
 
 	mie.seals = fcntl(fd, F_GET_SEALS);
-	if (mie.seals == -1)
-		goto out;
+	if (mie.seals == -1) {
+		if (errno != EINVAL || ~mie.hugetlb_flag & MFD_HUGETLB) {
+			pr_perror("fcntl(F_GET_SEALS)");
+			goto out;
+		}
+		/* Kernels before 4.16 don't allow MFD_HUGETLB |
+		 * MFD_ALLOW_SEALING and return EINVAL for
+		 * fcntl(MFD_HUGETLB-enabled fd).
+		 */
+		mie.seals = F_SEAL_SEAL;
+	}
 
 	if (pb_write_one(img_from_set(glob_imgset, CR_FD_MEMFD_INODE), &mie, PB_MEMFD_INODE))
 		goto out;

criu/pie/restorer.c

@@ -1635,17 +1635,13 @@ long __export_restore_task(struct task_restore_args *args)
 			goto core_restore_end;
 	}
 
-	if (args->uffd > -1) {
-		/* re-enable THP if we disabled it previously */
-		if (args->has_thp_enabled) {
-			int ret;
-			ret = sys_prctl(PR_SET_THP_DISABLE, 0, 0, 0, 0);
-			if (ret) {
-				pr_err("Cannot re-enable THP: %d\n", ret);
-				goto core_restore_end;
-			}
-		}
+	ret = sys_prctl(PR_SET_THP_DISABLE, args->thp_disabled, 0, 0, 0);
+	if (ret) {
+		pr_err("Cannot restore THP_DISABLE=%d flag: %ld\n", args->thp_disabled, ret);
+		goto core_restore_end;
+	}
 
+	if (args->uffd > -1) {
 		pr_debug("lazy-pages: closing uffd %d\n", args->uffd);
 		/*
 		 * All userfaultfd configuration has finished at this point.

test/zdtm/static/thp_disable.c

@@ -17,6 +17,7 @@ int main(int argc, char **argv)
 	unsigned long orig_flags = 0, new_flags = 0;
 	unsigned long orig_madv = 0, new_madv = 0;
 	void *area;
+	int ret;
 
 	test_init(argc, argv);
 
@@ -35,9 +36,46 @@ int main(int argc, char **argv)
 		return -1;
 	}
 
+	ret = prctl(PR_GET_THP_DISABLE, 0, 0, 0, 0);
+	if (ret < 0) {
+		pr_perror("Getting THP-disabled flag failed");
+		return -1;
+	}
+	if (ret != 1) {
+		errno = 0;
+		fail("prctl(GET_THP_DISABLE) returned unexpected value: %d != 1\n", ret);
+		return -1;
+	}
+
+	test_msg("Fetch pre-migration flags/adv\n");
+	if (get_smaps_bits((unsigned long)area, &new_flags, &new_madv))
+		return -1;
+
+	errno = 0;
+	if (orig_flags != new_flags) {
+		fail("Flags changed %lx -> %lx\n", orig_flags, new_flags);
+		return -1;
+	}
+
+	if (orig_madv != new_madv) {
+		fail("Madvs changed %lx -> %lx\n", orig_madv, new_madv);
+		return -1;
+	}
+
 	test_daemon();
 	test_waitsig();
 
+	ret = prctl(PR_GET_THP_DISABLE, 0, 0, 0, 0);
+	if (ret < 0) {
+		pr_perror("Getting post-migration THP-disabled flag failed");
+		return -1;
+	}
+	if (ret != 1) {
+		errno = 0;
+		fail("post-migration prctl(GET_THP_DISABLE) returned unexpected value: %d != 1\n", ret);
+		return -1;
+	}
+
 	if (prctl(PR_SET_THP_DISABLE, 0, 0, 0, 0)) {
 		pr_perror("Enabling THP failed");
 		return -1;
@@ -47,15 +85,14 @@ int main(int argc, char **argv)
 	if (get_smaps_bits((unsigned long)area, &new_flags, &new_madv))
 		return -1;
 
+	errno = 0;
 	if (orig_flags != new_flags) {
-		pr_err("Flags are changed %lx -> %lx\n", orig_flags, new_flags);
-		fail();
+		fail("Flags changed %lx -> %lx\n", orig_flags, new_flags);
 		return -1;
 	}
 
 	if (orig_madv != new_madv) {
-		pr_err("Madvs are changed %lx -> %lx\n", orig_madv, new_madv);
-		fail();
+		fail("Madvs changed %lx -> %lx\n", orig_madv, new_madv);
 		return -1;
 	}