sssd/0001-KCM-fix-memory-leak.patch
2025-02-10 16:12:49 +01:00

114 lines
3.9 KiB
Diff

From 50f703f25914254d2a545f52f504dfa5a6f65546 Mon Sep 17 00:00:00 2001
From: Alexey Tikhonov <atikhono@redhat.com>
Date: Tue, 4 Feb 2025 18:59:36 +0100
Subject: [PATCH] KCM: fix memory leak
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit
The copy of 'secret' argument - `secret_val.data` - was left hanging
on `sss_sec_ctx`, effectively resulting in a memory leak.
But this copy isn't actually required as this data isn't modified in
below operations.
Skipping alloc+memcpy+erase is also beneficial performance wise.
:fixes:'sssd_kcm' memory leak was fixed.
Reviewed-by: Alejandro López <allopez@redhat.com>
Reviewed-by: Justin Stephenson <jstephen@redhat.com>
(cherry picked from commit 7f1b7c9689827df92e8b2166423d4e80688dbacb)
---
src/responder/kcm/secrets/secrets.c | 34 ++++++++++-------------------
1 file changed, 12 insertions(+), 22 deletions(-)
diff --git a/src/responder/kcm/secrets/secrets.c b/src/responder/kcm/secrets/secrets.c
index 625a09f39..fe7410cb3 100644
--- a/src/responder/kcm/secrets/secrets.c
+++ b/src/responder/kcm/secrets/secrets.c
@@ -979,7 +979,7 @@ errno_t sss_sec_put(struct sss_sec_req *req,
size_t secret_len)
{
struct ldb_message *msg;
- struct ldb_val secret_val = { .data = NULL };
+ const struct ldb_val secret_val = { .length = secret_len, .data = secret };
bool erase_msg = false;
int ret;
@@ -1029,13 +1029,11 @@ errno_t sss_sec_put(struct sss_sec_req *req,
goto done;
}
- secret_val.length = secret_len;
- secret_val.data = talloc_memdup(req->sctx, secret, secret_len);
- if (!secret_val.data) {
- ret = ENOMEM;
- goto done;
- }
-
+ /* `ldb_msg_add_value()` does NOT make a copy of secret_val::*data
+ * but rather copies a pointer under the hood.
+ * This is fine since no operations modifying this data are performed
+ * below and 'msg' is freed before function returns.
+ */
ret = ldb_msg_add_value(msg, SEC_ATTR_SECRET, &secret_val, NULL);
if (ret != EOK) {
DEBUG(SSSDBG_OP_FAILURE,
@@ -1069,9 +1067,6 @@ errno_t sss_sec_put(struct sss_sec_req *req,
ret = EOK;
done:
- if (secret_val.data != NULL) {
- sss_erase_mem_securely(secret_val.data, secret_val.length);
- }
if (erase_msg) {
db_result_erase_message_securely(msg, SEC_ATTR_SECRET);
}
@@ -1084,7 +1079,7 @@ errno_t sss_sec_update(struct sss_sec_req *req,
size_t secret_len)
{
struct ldb_message *msg;
- struct ldb_val secret_val = { .data = NULL };
+ const struct ldb_val secret_val = { .length = secret_len, .data = secret };
bool erase_msg = false;
int ret;
@@ -1134,13 +1129,6 @@ errno_t sss_sec_update(struct sss_sec_req *req,
goto done;
}
- secret_val.length = secret_len;
- secret_val.data = talloc_memdup(req->sctx, secret, secret_len);
- if (!secret_val.data) {
- ret = ENOMEM;
- goto done;
- }
-
/* FIXME - should we have a lastUpdate timestamp? */
ret = ldb_msg_add_empty(msg, SEC_ATTR_SECRET, LDB_FLAG_MOD_REPLACE, NULL);
if (ret != LDB_SUCCESS) {
@@ -1150,6 +1138,11 @@ errno_t sss_sec_update(struct sss_sec_req *req,
goto done;
}
+ /* `ldb_msg_add_value()` does NOT make a copy of secret_val::*data
+ * but rather copies a pointer under the hood.
+ * This is fine since no operations modifying this data are performed
+ * below and 'msg' is freed before function returns.
+ */
ret = ldb_msg_add_value(msg, SEC_ATTR_SECRET, &secret_val, NULL);
if (ret != LDB_SUCCESS) {
DEBUG(SSSDBG_MINOR_FAILURE,
@@ -1174,9 +1167,6 @@ errno_t sss_sec_update(struct sss_sec_req *req,
ret = EOK;
done:
- if (secret_val.data != NULL) {
- sss_erase_mem_securely(secret_val.data, secret_val.length);
- }
if (erase_msg) {
db_result_erase_message_securely(msg, SEC_ATTR_SECRET);
}
--
2.47.0