fsmonitor: fix khash memory leak in do_handle_client

From: Paul Tarjan <[email protected]>

The do_handle_client() function allocates a khash table to de-duplicate pathnames when responding to client requests. Two issues existed:

1. kh_release_str() was used instead of kh_destroy_str(). The release function only frees internal arrays (flags, keys, vals) but not the struct itself (allocated by kh_init_str via xcalloc). This caused a 40-byte leak per request.

2. The khash was freed mid-function rather than in the cleanup section, so if the worker thread was interrupted before reaching that point during daemon shutdown, the memory would leak.

Fix both issues by:
- Initializing shown = NULL at declaration
- Using kh_destroy_str() which handles NULL and frees both internal arrays and the struct itself
- Moving the cleanup to the cleanup section so it runs on all exit paths

---
fsmonitor: fix khash memory leak in do_handle_client

The do_handle_client() function allocates a khash table to de-duplicate pathnames when responding to client requests. Two issues existed:

1. kh_release_str() was used instead of kh_destroy_str(). The release function only frees internal arrays (flags, keys, vals) but not the struct itself (allocated by kh_init_str via xcalloc). This caused a 40-byte leak per request.

2. The khash was freed mid-function rather than in the cleanup section, so if the worker thread was interrupted before reaching that point during daemon shutdown, the memory would leak.

Fix both issues by:

* Initializing shown = NULL at declaration
* Using kh_destroy_str() which handles NULL and frees both internal arrays and the struct itself
* Moving the cleanup to the cleanup section so it runs on all exit paths

Published-As: https://github.com/gitgitgadget/git/releases/tag/pr-git-2148%2Fptarjan%2Fclaude%2Ffix-fsmonitor-hashmap-leak-gfDCU-v1
Fetch-It-Via: git fetch https://github.com/gitgitgadget/git pr-git-2148/ptarjan/claude/fix-fsmonitor-hashmap-leak-gfDCU-v1
Pull-Request: https://github.com/git/git/pull/2148

 builtin/fsmonitor--daemon.c | 5 ++---
 1 file changed, 2 insertions(+), 3 deletions(-)

index 242c594646..bc4571938c 100644
@@ -671,7 +671,7 @@ static int do_handle_client(struct fsmonitor_daemon_state *state,
 	const struct fsmonitor_batch *batch;
 	struct fsmonitor_batch *remainder = NULL;
 	intmax_t count = 0, duplicates = 0;
-	kh_str_t *shown;
+	kh_str_t *shown = NULL;
 	int hash_ret;
 	int do_trivial = 0;
 	int do_flush = 0;

@@ -909,8 +909,6 @@ static int do_handle_client(struct fsmonitor_daemon_state *state,
 		total_response_len += payload.len;
 	}
 
-	kh_release_str(shown);
-
 	pthread_mutex_lock(&state->main_lock);
 
 	if (token_data->client_ref_count > 0)

@@ -954,6 +952,7 @@ static int do_handle_client(struct fsmonitor_daemon_state *state,
 	trace2_data_intmax("fsmonitor", the_repository, "response/count/duplicates", duplicates);
 
 cleanup:
+	kh_destroy_str(shown);
 	strbuf_release(&response_token);
 	strbuf_release(&requested_token_id);
 	strbuf_release(&payload);

base-commit: 7c7698a654a7a0031f65b0ab0c1c4e438e95df60

gitgitgadget

Makes sense.

Really? Would an interrupted thread even reach its cleanup section?

From: Paul Tarjan <[email protected]>

The do_handle_client() function allocates a khash table to de-duplicate pathnames when responding to client requests. However, kh_release_str() was used instead of kh_destroy_str(). The release function only frees internal arrays (flags, keys, vals) but not the struct itself, which is allocated by kh_init_str() via xcalloc. This caused a 40-byte leak per client request.

Fix by using kh_destroy_str() which properly frees both internal arrays and the struct itself. Also move the cleanup to the cleanup section and initialize shown to NULL so that kh_destroy_str() is safe to call on all exit paths.

---
fsmonitor: fix khash memory leak in do_handle_client

The do_handle_client() function allocates a khash table to de-duplicate pathnames when responding to client requests. However, kh_release_str() was used instead of kh_destroy_str(). The release function only frees internal arrays (flags, keys, vals) but not the struct itself, which is allocated by kh_init_str() via xcalloc. This caused a 40-byte leak per client request.

Fix by using kh_destroy_str() which properly frees both internal arrays and the struct itself. Also move the cleanup to the cleanup section and initialize shown to NULL so that kh_destroy_str() is safe to call on all exit paths.

Changes since v1:

* Removed incorrect claim about interrupted threads reaching cleanup
* Simplified commit message to focus on the real bug (kh_release vs kh_destroy)

Published-As: https://github.com/gitgitgadget/git/releases/tag/pr-git-2148%2Fptarjan%2Fclaude%2Ffix-fsmonitor-hashmap-leak-gfDCU-v2
Fetch-It-Via: git fetch https://github.com/gitgitgadget/git pr-git-2148/ptarjan/claude/fix-fsmonitor-hashmap-leak-gfDCU-v2
Pull-Request: https://github.com/git/git/pull/2148

Range-diff vs v1:

1: 18670d3230 ! 1: 0a754c7f09 fsmonitor: fix khash memory leak in do_handle_client
@@ Commit message
fsmonitor: fix khash memory leak in do_handle_client

The do_handle_client() function allocates a khash table to de-duplicate
- pathnames when responding to client requests. Two issues existed:
+ pathnames when responding to client requests. However, kh_release_str()
+ was used instead of kh_destroy_str(). The release function only frees
+ internal arrays (flags, keys, vals) but not the struct itself, which is
+ allocated by kh_init_str() via xcalloc. This caused a 40-byte leak per
+ client request.

- 1. kh_release_str() was used instead of kh_destroy_str(). The release
- function only frees internal arrays (flags, keys, vals) but not the
- struct itself (allocated by kh_init_str via xcalloc). This caused a
- 40-byte leak per request.
+ Fix by using kh_destroy_str() which properly frees both internal arrays
+ and the struct itself. Also move the cleanup to the cleanup section and
+ initialize shown to NULL so that kh_destroy_str() is safe to call on all
+ exit paths.

- 2. The khash was freed mid-function rather than in the cleanup section,
- so if the worker thread was interrupted before reaching that point
- during daemon shutdown, the memory would leak.
-
- Fix both issues by:
- - Initializing shown = NULL at declaration
- - Using kh_destroy_str() which handles NULL and frees both internal
- arrays and the struct itself
- - Moving the cleanup to the cleanup section so it runs on all exit paths
-
- Signed-off-by: Claude <[email protected]>
+ Signed-off-by: Paul Tarjan <[email protected]>

## builtin/fsmonitor--daemon.c ##
@@ builtin/fsmonitor--daemon.c: static int do_handle_client(struct fsmonitor_daemon_state *state,

 builtin/fsmonitor--daemon.c | 5 ++---
 1 file changed, 2 insertions(+), 3 deletions(-)

index 242c594646..bc4571938c 100644
@@ -671,7 +671,7 @@ static int do_handle_client(struct fsmonitor_daemon_state *state,
 	const struct fsmonitor_batch *batch;
 	struct fsmonitor_batch *remainder = NULL;
 	intmax_t count = 0, duplicates = 0;
-	kh_str_t *shown;
+	kh_str_t *shown = NULL;
 	int hash_ret;
 	int do_trivial = 0;
 	int do_flush = 0;

@@ -909,8 +909,6 @@ static int do_handle_client(struct fsmonitor_daemon_state *state,
 		total_response_len += payload.len;
 	}
 
-	kh_release_str(shown);
-
 	pthread_mutex_lock(&state->main_lock);
 
 	if (token_data->client_ref_count > 0)

@@ -954,6 +952,7 @@ static int do_handle_client(struct fsmonitor_daemon_state *state,
 	trace2_data_intmax("fsmonitor", the_repository, "response/count/duplicates", duplicates);
 
 cleanup:
+	kh_destroy_str(shown);
 	strbuf_release(&response_token);
 	strbuf_release(&requested_token_id);
 	strbuf_release(&payload);

base-commit: 7c7698a654a7a0031f65b0ab0c1c4e438e95df60

gitgitgadget

This is already in v4 of the other larger fsmonitor-linux patch, right?

Thanks.

Correct. I sent it separately since it is currently a bug in existing code and you might want to merge it sooner.

Thanks.