[PATCH 0/3] object-file: fix ".keep" lockfile ownership in ODB transactions

[PATCH 0/3] object-file: fix ".keep" lockfile ownership in ODB transactions

[Justin Tobler]

Greetings,

This series addresses a bug report[1] regarding a git-receive-pack(1)
process deleting ".keep" lockfiles of with the same name but from
another concurrent git-receive-pack(1) process receiving an identical
packfile. Deleting the ".keep" lockfile for another process leaves its
newly written packfile vulnerable to deletion by another concurrent
housekeeping process.

The root of the problem here is that the temporary ".keep" lockfiles are
generated and registered in the object quarantine prior to being written
in the primary ODB. This means that even if the ODB transaction never
commits and exits early, the tempfile atexit(3) handler still runs and
tries to delete the lockfile at its expected final location. This
blindly deletes any file matching regardless of whether is was created
by the process.

The first two patches are some preparatory changes so that
`odb_transaction_files_write_pack()` is able to start explicitly
creating the lockfiles without relying on git-index-pack(1). Patch three
migrates lockfile generation to happen during ODB transaction commit to
ensure the lockfile is actually able to be written before registering it
as a tempfile.

Thanks,
-Justin

[1]: <932e8e425aecfbd33c1e5caf66c80a0226abacba.1789385483.git.gitgitgadget@gmail.com>

Justin Tobler (3):
  pack-write: lift and expose index-pack report parsing
  object-file: explicitly write ".keep" lockfile
  object-file: write ".keep" lockfiles during transaction commit

 object-file.c              | 117 +++++++++++++++++++++++++++++--------
 pack-write.c               |  69 ++++++++++++++++------
 pack.h                     |  10 ++++
 t/t5547-push-quarantine.sh |  29 ++++++++-
 4 files changed, 181 insertions(+), 44 deletions(-)


base-commit: f57c9a73884400380bddd172acc88b569834f909
-- 
2.55.0.424.g13c7afec21

[PATCH 1/3] pack-write: lift and expose index-pack report parsing

[Justin Tobler]

Once git-index-pack(1) finishes writing a packfile, it reports the
hexadecimal name of that packfile on stdout prefixed with either "pack"
or "keep" depending on whether a ".keep" file was also created alongside
it. Callers that are required to manage the lifecycle of ".keep" files
parse this report via `index_pack_lockfile()`, which provides the path
to the temporary ".keep" file.

In a subsequent commit, the "files" ODB transaction begins staging
packfiles in a transaction without eagerly creating a ".keep" file and
must instead read git-index-pack(1) "pack" output to get the packfile
name. Extract the generic parsing logic for the git-index-pack(1) output
into `read_index_pack_report()` in preparation accordingly.

Signed-off-by: Justin Tobler <jltobler@gmail.com>
---
 pack-write.c | 69 +++++++++++++++++++++++++++++++++++++++-------------
 pack.h       | 10 ++++++++
 2 files changed, 62 insertions(+), 17 deletions(-)

diff --git a/pack-write.c b/pack-write.c
index 85674e4b72..5f5bb894a5 100644
--- a/pack-write.c
+++ b/pack-write.c
@@ -469,33 +469,68 @@ void fixup_pack_header_footer(const struct git_hash_algo *hash_algo,
 	fsync_component_or_die(FSYNC_COMPONENT_PACK, pack_fd, pack_name);
 }
 
-char *index_pack_lockfile(struct odb_source *source, int ip_out,
-			  int *is_well_formed)
+int read_index_pack_report(struct repository *repo, int ip_out,
+			   struct strbuf *pack_name, int *is_keep)
 {
+	const int len = repo->hash_algo->hexsz + 6;
 	char packname[GIT_MAX_HEXSZ + 6];
-	const int len = source->odb->repo->hash_algo->hexsz + 6;
+	const char *name;
+	int keep;
 
 	/*
-	 * The first thing we expect from index-pack's output
-	 * is "pack\t%40s\n" or "keep\t%40s\n" (46 bytes) where
-	 * %40s is the newly created pack SHA1 name.  In the "keep"
-	 * case, we need it to remove the corresponding .keep file
-	 * later on.  If we don't get that then tough luck with it.
+	 * The first thing we expect from index-pack's output is
+	 * "pack\t<name>\n" or "keep\t<name>\n", where <name> is the hexadecimal
+	 * name of the packfile it has just written. The line is therefore five
+	 * bytes of prefix, `hexsz` bytes of name, and a trailing newline. A
+	 * "keep" prefix tells us that index-pack has also created the
+	 * corresponding ".keep" file.
 	 */
-	if (read_in_full(ip_out, packname, len) == len && packname[len-1] == '\n') {
-		const char *name;
+	if (read_in_full(ip_out, packname, len) != len || packname[len-1] != '\n')
+		return -1;
+
+	packname[len-1] = 0;
+	if (skip_prefix(packname, "pack\t", &name))
+		keep = 0;
+	else if (skip_prefix(packname, "keep\t", &name))
+		keep = 1;
+	else
+		return -1;
+
+	strbuf_addstr(pack_name, name);
+	if (is_keep)
+		*is_keep = keep;
+
+	return 0;
+}
+
+char *index_pack_lockfile(struct odb_source *source, int ip_out,
+			  int *is_well_formed)
+{
+	struct strbuf pack_name = STRBUF_INIT;
+	char *lockfile = NULL;
+	int is_keep = 0;
 
+	if (read_index_pack_report(source->odb->repo, ip_out, &pack_name, &is_keep)) {
 		if (is_well_formed)
-			*is_well_formed = 1;
-		packname[len-1] = 0;
-		if (skip_prefix(packname, "keep\t", &name))
-			return xstrfmt("%s/pack/pack-%s.keep",
-				       source->path, name);
+			*is_well_formed = 0;
 		return NULL;
 	}
+
 	if (is_well_formed)
-		*is_well_formed = 0;
-	return NULL;
+		*is_well_formed = 1;
+
+	/*
+	 * If there is no ".keep" file, then there is no lockfile to return.
+	 */
+	if (!is_keep) {
+		strbuf_release(&pack_name);
+		return NULL;
+	}
+
+	lockfile = xstrfmt("%s/pack/pack-%s.keep", source->path, pack_name.buf);
+
+	strbuf_release(&pack_name);
+	return lockfile;
 }
 
 /*
diff --git a/pack.h b/pack.h
index ada506b5c5..ddf2d52934 100644
--- a/pack.h
+++ b/pack.h
@@ -106,6 +106,16 @@ off_t write_pack_header(struct hashfile *f, uint32_t);
 void fixup_pack_header_footer(const struct git_hash_algo *, int,
 			      unsigned char *, const char *, uint32_t,
 			      unsigned char *, off_t);
+
+/*
+ * Read the report that git-index-pack(1) writes to its standard output after
+ * the packfile is written. The hexadecimal name of the packfile is appended to
+ * `pack_name` and, if `is_keep` is non-NULL, set it to indicate whether a
+ * ".keep" file was written. Return 0 on success and a negative error code on
+ * failure.
+ */
+int read_index_pack_report(struct repository *repo, int ip_out,
+			   struct strbuf *pack_name, int *is_keep);
 char *index_pack_lockfile(struct odb_source *source, int fd,
 			  int *is_well_formed);
 
-- 
2.55.0.424.g13c7afec21

[PATCH 2/3] object-file: explicitly write ".keep" lockfile

[Justin Tobler]

The `odb_transaction_write_pack()` interface for the "files" backend
relies on git-index-pack(1) to write the specified packfile into an ODB
transaction as a packfile. To prevent a concurrent housekeeping
operation from deleting the newly written packfile prior to the
references being updated, a ".keep" lockfile is also generated by
git-index-pack(1) via the `--keep` option. Consequently, the ".keep"
lockfile is written into the ODB transaction alongside the packfile and
later migrated together to the primary ODB during transaction commit.

In a subsequent commit, writing the lockfile is deferred until
transaction commit and thus cannot rely on git-index-pack(1) to write
it. In preparation for this, update `odb_transaction_files_write_pack()`
to explicitly generate the same ".keep" lockfile itself.

Signed-off-by: Justin Tobler <jltobler@gmail.com>
---
 object-file.c | 99 ++++++++++++++++++++++++++++++++++++++++-----------
 1 file changed, 79 insertions(+), 20 deletions(-)

diff --git a/object-file.c b/object-file.c
index 210984f825..ebe46d35de 100644
--- a/object-file.c
+++ b/object-file.c
@@ -1256,6 +1256,43 @@ int read_loose_object(struct repository *repo,
 	return ret;
 }
 
+static int write_pack_lock(struct odb_source *source, const char *pack_name,
+			   const char *keep_msg)
+{
+	struct strbuf lock_path = STRBUF_INIT;
+	struct strbuf contents = STRBUF_INIT;
+	int ret = -1;
+	int fd;
+
+	strbuf_addf(&lock_path, "%s/pack/pack-%s.keep",
+		    source->path, pack_name);
+	strbuf_addf(&contents, "%s\n", keep_msg);
+
+	fd = safe_create_file_with_leading_directories(source->odb->repo,
+						       lock_path.buf);
+	if (fd < 0)
+		goto out;
+
+	if (write_in_full(fd, contents.buf, contents.len) < 0) {
+		close(fd);
+		unlink(lock_path.buf);
+		goto out;
+	}
+
+	if (close(fd)) {
+		unlink(lock_path.buf);
+		goto out;
+	}
+
+	ret = 0;
+
+out:
+	strbuf_release(&lock_path);
+	strbuf_release(&contents);
+
+	return ret;
+}
+
 static int odb_transaction_files_commit(struct odb_transaction *base)
 {
 	struct odb_transaction_files *transaction =
@@ -1389,18 +1426,17 @@ static int odb_transaction_files_write_pack(struct odb_transaction *base,
 			return -1;
 		}
 	} else {
+		struct strbuf final_path = STRBUF_INIT;
+		struct strbuf pack_name = STRBUF_INIT;
+		struct strbuf keep_msg = STRBUF_INIT;
 		char hostname[HOST_NAME_MAX + 1];
-		char *lockfile;
+		int ret;
 
 		strvec_pushl(&child.args, "index-pack", "--stdin", NULL);
 		push_header_arg(&child.args, &hdr);
 
 		if (xgethostname(hostname, sizeof(hostname)))
 			xsnprintf(hostname, sizeof(hostname), "localhost");
-		strvec_pushf(&child.args,
-			     "--keep=receive-pack %"PRIuMAX" on %s",
-			     (uintmax_t)getpid(),
-			     hostname);
 
 		if (!opts->quiet && err_fd)
 			strvec_push(&child.args, "--show-resolving-progress");
@@ -1424,31 +1460,54 @@ static int odb_transaction_files_write_pack(struct odb_transaction *base,
 			return -1;
 		}
 
-		/*
-		 * The lockfile filepath is expected to be the final location of
-		 * the ".keep" file after being migrated to the main ODB source.
-		 * This ensures the lockfile can be found and removed later
-		 * after the ODB transaction has been committed.
-		 */
-		lockfile = index_pack_lockfile(base->source, child.out, NULL);
-		if (lockfile) {
-			ALLOC_GROW(transaction->pack_lockfiles,
-				   transaction->pack_lockfiles_nr + 1,
-				   transaction->pack_lockfiles_alloc);
-			transaction->pack_lockfiles[transaction->pack_lockfiles_nr++] =
-				register_tempfile(lockfile);
-			free(lockfile);
-		}
+		ret = read_index_pack_report(repo, child.out, &pack_name, NULL);
+
 		close(child.out);
 
 		status = finish_command(&child);
 		if (status) {
+			strbuf_release(&pack_name);
 			strbuf_addstr(err_msg, "index-pack abnormal exit");
 			return -1;
 		}
 
+		if (ret) {
+			strbuf_release(&pack_name);
+			strbuf_addstr(err_msg, "index-pack failed to report pack name");
+			return -1;
+		}
+
+		strbuf_addf(&keep_msg, "receive-pack %"PRIuMAX" on %s",
+			    (uintmax_t)getpid(), hostname);
+
+		if (write_pack_lock(transaction->quarantine, pack_name.buf, keep_msg.buf)) {
+			strbuf_release(&pack_name);
+			strbuf_release(&keep_msg);
+			strbuf_addstr(err_msg, "lockfile write failed");
+			return -1;
+		}
+
+		ALLOC_GROW(transaction->pack_lockfiles,
+			   transaction->pack_lockfiles_nr + 1,
+			   transaction->pack_lockfiles_alloc);
+
+		/*
+		 * The lockfile filepath is expected to be the final location of
+		 * the ".keep" file after being migrated to the main ODB source.
+		 * This ensures the lockfile can be found and removed later
+		 * after the ODB transaction has been committed.
+		 */
+		strbuf_addf(&final_path, "%s/pack/pack-%s.keep",
+			    base->source->path, pack_name.buf);
+		transaction->pack_lockfiles[transaction->pack_lockfiles_nr++] =
+			register_tempfile(final_path.buf);
+
 		odb_source_prepare(transaction->quarantine,
 				   ODB_PREPARE_FLUSH_CACHES);
+
+		strbuf_release(&final_path);
+		strbuf_release(&pack_name);
+		strbuf_release(&keep_msg);
 	}
 
 	return 0;
-- 
2.55.0.424.g13c7afec21

[PATCH 3/3] object-file: write ".keep" lockfiles during transaction commit

[Justin Tobler]

The "files" ODB backend writes a ".keep" lockfile into the transaction
quarantine directory via `odb_transaction_files_write_pack()`. When the
ODB transaction is committed, the ".keep" lockfile is migrated to the
primary ODB alongside its packfile. Writing the temporary lockfile in
this manner has a couple of downsides:

  - Failing to migrate the lockfile during transaction commit due to a
    concurrent process already acquiring the same lock still allows the
    rest of the transaction to be migrated into the primary ODB. This
    dirties the primary ODB with new objects even though no references
    were updated and the operation failed.

  - Even more concerning, the ".keep" tempfile atexit(3) handler tries
    to remove any lockfile of the same name regardless of whether it was
    created by the same process. If the ".keep" lockfile of the same
    name for a concurrent process gets deleted prematurely, its packfile
    becomes at risk of being deleted by concurrent housekeeping.

Stop creating/registering the temporary lockfiles when writing the
packfile to the transaction and instead do so when committing the
transaction. This ensures the lockfile is properly acquired prior to
migrating objects and that only the owning process can delete its own
lockfiles.

Reported-by: Qin ShiCheng <qeesung@live.com>
Signed-off-by: Justin Tobler <jltobler@gmail.com>
---
 object-file.c              | 76 +++++++++++++++++++++-----------------
 t/t5547-push-quarantine.sh | 29 +++++++++++++--
 2 files changed, 69 insertions(+), 36 deletions(-)

diff --git a/object-file.c b/object-file.c
index ebe46d35de..39e6299be1 100644
--- a/object-file.c
+++ b/object-file.c
@@ -492,6 +492,7 @@ struct odb_transaction_files {
 	struct transaction_packfile packfile;
 	const char *prefix;
 
+	struct strvec pack_names;
 	struct tempfile **pack_lockfiles;
 	size_t pack_lockfiles_nr;
 	size_t pack_lockfiles_alloc;
@@ -1257,7 +1258,7 @@ int read_loose_object(struct repository *repo,
 }
 
 static int write_pack_lock(struct odb_source *source, const char *pack_name,
-			   const char *keep_msg)
+			   const char *keep_msg, struct tempfile **out)
 {
 	struct strbuf lock_path = STRBUF_INIT;
 	struct strbuf contents = STRBUF_INIT;
@@ -1284,6 +1285,7 @@ static int write_pack_lock(struct odb_source *source, const char *pack_name,
 		goto out;
 	}
 
+	*out = register_tempfile(lock_path.buf);
 	ret = 0;
 
 out:
@@ -1293,6 +1295,40 @@ static int write_pack_lock(struct odb_source *source, const char *pack_name,
 	return ret;
 }
 
+static int write_pack_locks(struct odb_transaction_files *transaction)
+{
+	struct strbuf keep_msg = STRBUF_INIT;
+	char hostname[HOST_NAME_MAX + 1];
+	int ret = 0;
+
+	if (!transaction->pack_names.nr)
+		return 0;
+
+	if (xgethostname(hostname, sizeof(hostname)))
+		xsnprintf(hostname, sizeof(hostname), "localhost");
+	strbuf_addf(&keep_msg, "receive-pack %"PRIuMAX" on %s",
+		    (uintmax_t)getpid(), hostname);
+
+	for (size_t i = 0; i < transaction->pack_names.nr; i++) {
+		struct tempfile *lock;
+
+		ret = write_pack_lock(transaction->base.source,
+				      transaction->pack_names.v[i],
+				      keep_msg.buf, &lock);
+		if (ret < 0)
+			break;
+
+		ALLOC_GROW(transaction->pack_lockfiles,
+			   transaction->pack_lockfiles_nr + 1,
+			   transaction->pack_lockfiles_alloc);
+		transaction->pack_lockfiles[transaction->pack_lockfiles_nr++] = lock;
+	}
+
+	strbuf_release(&keep_msg);
+
+	return ret;
+}
+
 static int odb_transaction_files_commit(struct odb_transaction *base)
 {
 	struct odb_transaction_files *transaction =
@@ -1321,6 +1357,9 @@ static int odb_transaction_files_commit(struct odb_transaction *base)
 		delete_tempfile(&temp);
 		strbuf_release(&temp_path);
 
+		if (write_pack_locks(transaction))
+			return error(_("unable to lock packfiles"));
+
 		/*
 		 * Make the object files visible in the primary ODB after their data is
 		 * fully durable.
@@ -1426,18 +1465,12 @@ static int odb_transaction_files_write_pack(struct odb_transaction *base,
 			return -1;
 		}
 	} else {
-		struct strbuf final_path = STRBUF_INIT;
 		struct strbuf pack_name = STRBUF_INIT;
-		struct strbuf keep_msg = STRBUF_INIT;
-		char hostname[HOST_NAME_MAX + 1];
 		int ret;
 
 		strvec_pushl(&child.args, "index-pack", "--stdin", NULL);
 		push_header_arg(&child.args, &hdr);
 
-		if (xgethostname(hostname, sizeof(hostname)))
-			xsnprintf(hostname, sizeof(hostname), "localhost");
-
 		if (!opts->quiet && err_fd)
 			strvec_push(&child.args, "--show-resolving-progress");
 		if (err_fd)
@@ -1477,37 +1510,12 @@ static int odb_transaction_files_write_pack(struct odb_transaction *base,
 			return -1;
 		}
 
-		strbuf_addf(&keep_msg, "receive-pack %"PRIuMAX" on %s",
-			    (uintmax_t)getpid(), hostname);
-
-		if (write_pack_lock(transaction->quarantine, pack_name.buf, keep_msg.buf)) {
-			strbuf_release(&pack_name);
-			strbuf_release(&keep_msg);
-			strbuf_addstr(err_msg, "lockfile write failed");
-			return -1;
-		}
-
-		ALLOC_GROW(transaction->pack_lockfiles,
-			   transaction->pack_lockfiles_nr + 1,
-			   transaction->pack_lockfiles_alloc);
-
-		/*
-		 * The lockfile filepath is expected to be the final location of
-		 * the ".keep" file after being migrated to the main ODB source.
-		 * This ensures the lockfile can be found and removed later
-		 * after the ODB transaction has been committed.
-		 */
-		strbuf_addf(&final_path, "%s/pack/pack-%s.keep",
-			    base->source->path, pack_name.buf);
-		transaction->pack_lockfiles[transaction->pack_lockfiles_nr++] =
-			register_tempfile(final_path.buf);
+		strvec_push(&transaction->pack_names, pack_name.buf);
 
 		odb_source_prepare(transaction->quarantine,
 				   ODB_PREPARE_FLUSH_CACHES);
 
-		strbuf_release(&final_path);
 		strbuf_release(&pack_name);
-		strbuf_release(&keep_msg);
 	}
 
 	return 0;
@@ -1523,6 +1531,7 @@ static int odb_transaction_files_finalize(struct odb_transaction *base)
 		ret |= delete_tempfile(&transaction->pack_lockfiles[i]);
 
 	free(transaction->pack_lockfiles);
+	strvec_clear(&transaction->pack_names);
 
 	return ret;
 }
@@ -1555,6 +1564,7 @@ int odb_transaction_files_begin(struct odb_source *source,
 	transaction->base.write_pack = odb_transaction_files_write_pack;
 	transaction->base.env = odb_transaction_files_env;
 	transaction->flags = flags;
+	strvec_init(&transaction->pack_names);
 
 	transaction->prefix = "bulk-fsync";
 	if (flags & ODB_TRANSACTION_RECEIVE) {
diff --git a/t/t5547-push-quarantine.sh b/t/t5547-push-quarantine.sh
index 1b7097179e..cf349eee2c 100755
--- a/t/t5547-push-quarantine.sh
+++ b/t/t5547-push-quarantine.sh
@@ -77,10 +77,9 @@ test_expect_success '.keep file is removed after push' '
 	git -C keep.git config set receive.unpackLimit 0 &&
 
 	# While incoming objects are still quarantined, validate that the
-	# ".keep" lockfile is present in the quarantine directory.
+	# ".keep" lockfile is not present in the quarantine directory.
 	test_hook -C keep.git pre-receive <<-\EOF &&
-	keep="$(ls "$GIT_QUARANTINE_PATH"/pack/pack-*.keep)" &&
-	test -f "$keep"
+	test ! -e "$GIT_QUARANTINE_PATH"/pack/pack-*.keep
 	EOF
 
 	# After quarantined objects are migrated, validate that the ".keep"
@@ -101,4 +100,28 @@ test_expect_success '.keep file is removed after push' '
 	test_path_is_missing "$keep"
 '
 
+test_expect_success '.keep file created outside of process is not removed' '
+	test_when_finished rm -rf keep.git &&
+	git init --bare keep.git &&
+
+	git -C keep.git config set receive.unpackLimit 0 &&
+
+	# Create a ".keep" lockfile matching the packfile to simulate a
+	# concurrent git-receive-pack(1) that was already written its lockfile.
+	test_hook -C keep.git pre-receive <<-\EOF &&
+	pack="$(ls "$GIT_QUARANTINE_PATH"/pack/pack-*.pack)" &&
+	pack="${pack##*/}" &&
+	echo "baz" >"objects/pack/${pack%.pack}.keep"
+	EOF
+
+	test_commit bar &&
+	test_must_fail git push keep.git HEAD &&
+
+	# The ".keep" lockfile of the concurrent git-receive-pack(1) should be
+	# left in its place.
+	keep="$(ls keep.git/objects/pack/pack-*.keep)" &&
+	test_path_is_file "$keep" &&
+	test_grep "baz" "$keep"
+'
+
 test_done
-- 
2.55.0.424.g13c7afec21