You are viewing a plain text version of this content. The canonical link for it is here.
Posted to common-commits@hadoop.apache.org by ch...@apache.org on 2026/09/04 07:23:49 UTC
(hadoop) branch branch-3.4 updated: YARN-11987. container-executor delete-as-user fails to rmdir private filecache directories. (#8700)
This is an automated email from the ASF dual-hosted git repository.
pan3793 pushed a commit to branch branch-3.4
in repository https://gitbox.apache.org/repos/asf/hadoop.git
The following commit(s) were added to refs/heads/branch-3.4 by this push:
new 041e1abb203 YARN-11987. container-executor delete-as-user fails to rmdir private filecache directories. (#8700)
041e1abb203 is described below
commit 041e1abb2034dcfcd27e30aa0f48fdc5d1b8c30f
Author: Yubi Lee <eu...@gmail.com>
AuthorDate: Fri Sep 4 16:22:06 2026 +0900
YARN-11987. container-executor delete-as-user fails to rmdir private filecache directories. (#8700)
Authored-by: Yubi Lee <eu...@naver.com>
Reviewed-by: Shilun Fan <sl...@apache.org>
Signed-off-by: Cheng Pan <ch...@apache.org>
---
.../container-executor/impl/container-executor.c | 33 +++++++------
.../test/test-container-executor.c | 57 ++++++++++++++++++++++
2 files changed, 76 insertions(+), 14 deletions(-)
diff --git a/hadoop-yarn-project/hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-nodemanager/src/main/native/container-executor/impl/container-executor.c b/hadoop-yarn-project/hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-nodemanager/src/main/native/container-executor/impl/container-executor.c
index ab03d3991dc..19b057187aa 100644
--- a/hadoop-yarn-project/hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-nodemanager/src/main/native/container-executor/impl/container-executor.c
+++ b/hadoop-yarn-project/hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-nodemanager/src/main/native/container-executor/impl/container-executor.c
@@ -2737,22 +2737,27 @@ static int delete_path(const char *full_path,
return -1;
}
- /*
- * If required, do the final rmdir as root on the top level.
- * That handles the case where the top level directory is in a directory
- * owned by the node manager.
- */
- if (needs_tt_user) {
- return rmdir_as_nm(full_path);
- }
- /* Otherwise rmdir the top level as the current user. */
+ /* rmdir the top level as the current user. */
if (rmdir(full_path) != 0) {
ret = errno;
- if (ret != ENOENT) {
- fprintf(LOGFILE, "Couldn't delete directory %s - %s\n",
- full_path, strerror(ret));
- return -1;
- }
+ if (ret == ENOENT) {
+ return 0;
+ }
+ /*
+ * If required, retry the final rmdir as the node manager user.
+ * That handles the case where the top level directory is in a directory
+ * owned by the node manager, e.g. usercache/<user>. The first attempt
+ * as the current user handles the opposite case where the top level
+ * directory is in a directory owned by the run-as user but not writable
+ * by the node manager user, e.g. the private file cache
+ * usercache/<user>/filecache/<id>.
+ */
+ if (needs_tt_user && (ret == EACCES || ret == EPERM)) {
+ return rmdir_as_nm(full_path);
+ }
+ fprintf(LOGFILE, "Couldn't delete directory %s - %s\n",
+ full_path, strerror(ret));
+ return -1;
}
return 0;
}
diff --git a/hadoop-yarn-project/hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-nodemanager/src/main/native/container-executor/test/test-container-executor.c b/hadoop-yarn-project/hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-nodemanager/src/main/native/container-executor/test/test-container-executor.c
index f209ea53a00..090718269b2 100644
--- a/hadoop-yarn-project/hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-nodemanager/src/main/native/container-executor/test/test-container-executor.c
+++ b/hadoop-yarn-project/hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-nodemanager/src/main/native/container-executor/test/test-container-executor.c
@@ -640,6 +640,62 @@ void test_delete_user() {
free(app_dir);
}
+/**
+ * Test that delete_as_user with an empty relative path can remove a top
+ * level directory whose parent is owned by the run-as user and is not
+ * writable by the node manager user, as happens when the private file
+ * cache (usercache/<user>/filecache/<id>) is evicted. Run this test as
+ * root with two different users (test-container-executor <nm-user>
+ * <run-as-user>) to exercise the failure mode.
+ */
+void test_delete_dir_in_user_owned_parent() {
+ printf("\nTesting delete_dir_in_user_owned_parent\n");
+ if (seteuid(0) != 0) {
+ printf("Ignoring test_delete_dir_in_user_owned_parent; not running as root\n");
+ return;
+ }
+ char buffer[100000];
+ // the private filecache is created by the ContainerLocalizer with perms
+ // 0710 and is owned by the run-as user, so the node manager user cannot
+ // write to it.
+ sprintf(buffer, "mkdir -p " TEST_ROOT "/local-1/usercache/%s/filecache/10",
+ yarn_username);
+ run(buffer);
+ sprintf(buffer, "chown -R %s " TEST_ROOT "/local-1/usercache/%s",
+ yarn_username, yarn_username);
+ run(buffer);
+ sprintf(buffer, "chmod 710 " TEST_ROOT "/local-1/usercache/%s/filecache",
+ yarn_username);
+ run(buffer);
+
+ if (set_user(yarn_username) != 0) {
+ printf("FAIL: failed to set user to %s\n", yarn_username);
+ exit(1);
+ }
+ char child[PATH_MAX];
+ sprintf(child, TEST_ROOT "/local-1/usercache/%s/filecache/10",
+ yarn_username);
+ char * dirs[] = {child, 0};
+ int ret = delete_as_user(yarn_username, "", dirs);
+ if (seteuid(0) != 0) {
+ printf("FAIL: could not become root again\n");
+ exit(1);
+ }
+ if (ret != 0) {
+ printf("FAIL: return code from delete_as_user is %d\n", ret);
+ exit(1);
+ }
+ if (access(child, F_OK) == 0) {
+ printf("FAIL: failed to delete the directory - %s\n", child);
+ exit(1);
+ }
+ sprintf(buffer, TEST_ROOT "/local-1/usercache/%s/filecache", yarn_username);
+ if (access(buffer, F_OK) != 0) {
+ printf("FAIL: accidently deleted the parent directory - %s\n", buffer);
+ exit(1);
+ }
+}
+
/**
* Read a file and tokenize it on newlines. Place up to max lines into lines.
* The max+1st element of lines will be set to NULL.
@@ -1779,6 +1835,7 @@ int main(int argc, char **argv) {
ret++;
// test_delete_user must run as root since that's how we use the delete_as_user
test_delete_user();
+ test_delete_dir_in_user_owned_parent();
free_executor_configurations();
printf("\nTrying banned default user()\n");
---------------------------------------------------------------------
To unsubscribe, e-mail: common-commits-unsubscribe@hadoop.apache.org
For additional commands, e-mail: common-commits-help@hadoop.apache.org