[PATCH v3 0/8] sequencer API & users: fix widespread leaks

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

 



This series fixes various widespread leaks in the sequencer and its
users (rebase, revert, cherry-pick). As a result 18 tests become
leak-free in their entirety.

See the v1 for a longer general summary:
https://lore.kernel.org/git/cover-00.10-00000000000-20221230T071741Z-avarab@xxxxxxxxx/

Changes since v2:

 * Reword/amend commit messages for some of the functional changes in
   v1..v2.
 * Remove leftover "opts->xopts_nr = 0" now that we don't
   FREE_AND_NULL() anymore.
 * Drop the "squash_onto_name" refactoring to a "to_free"
 * Instead add a new "keep_base_onto_name" in the next commit. I'd
   missed that if both options were provided we'd die(), so that
   free() wasn't needed before re-assignment, as Phillip pointed
   out...

CI & branch for this available at:
https://github.com/avar/git/tree/avar/leak-fixes-sequencer-rebase-3

Ævar Arnfjörð Bjarmason (8):
  rebase: use "cleanup" pattern in do_interactive_rebase()
  sequencer.c: split up sequencer_remove_state()
  rebase & sequencer API: fix get_replay_opts() leak in "rebase"
  builtin/revert.c: move free-ing of "revs" to replay_opts_release()
  builtin/rebase.c: fix "options.onto_name" leak
  sequencer.c: always free() the "msgbuf" in do_pick_commit()
  builtin/rebase.c: free() "options.strategy_opts"
  commit.c: free() revs.commit in get_fork_point()

 builtin/rebase.c                       | 22 ++++++++------
 builtin/revert.c                       |  8 ++---
 commit.c                               |  1 +
 sequencer.c                            | 42 ++++++++++++++++----------
 sequencer.h                            |  1 +
 t/t3405-rebase-malformed.sh            |  1 +
 t/t3412-rebase-root.sh                 |  1 +
 t/t3416-rebase-onto-threedots.sh       |  1 +
 t/t3419-rebase-patch-id.sh             |  1 +
 t/t3423-rebase-reword.sh               |  1 +
 t/t3425-rebase-topology-merges.sh      |  2 ++
 t/t3431-rebase-fork-point.sh           |  1 +
 t/t3432-rebase-fast-forward.sh         |  1 +
 t/t3437-rebase-fixup-options.sh        |  1 +
 t/t3438-rebase-broken-files.sh         |  2 ++
 t/t3501-revert-cherry-pick.sh          |  1 +
 t/t3502-cherry-pick-merge.sh           |  1 +
 t/t3503-cherry-pick-root.sh            |  1 +
 t/t3506-cherry-pick-ff.sh              |  1 +
 t/t3511-cherry-pick-x.sh               |  1 +
 t/t7402-submodule-rebase.sh            |  1 +
 t/t9106-git-svn-commit-diff-clobber.sh |  1 -
 t/t9164-git-svn-dcommit-concurrent.sh  |  1 -
 23 files changed, 61 insertions(+), 33 deletions(-)

Range-diff against v2:
 1:  d0a0524f3d4 =  1:  b223429df33 rebase: use "cleanup" pattern in do_interactive_rebase()
 2:  c4eaa8dfef4 =  2:  00c7f04363f sequencer.c: split up sequencer_remove_state()
 3:  f06f565ceaf !  3:  e4a96898a68 rebase & sequencer API: fix get_replay_opts() leak in "rebase"
    @@ Commit message
         get_replay_opts() function in "builtin/rebase.c". See [1] for the
         initial addition of get_replay_opts().
     
    -    To safely call our new replay_opts_release() we'll need to change all
    -    the free() to a FREE_AND_NULL(), and set "xopts_nr" to "0" after we
    -    loop over it and free() it (the free() in the loop doesn't need to be
    -    a FREE_AND_NULL()).
    +    To safely call our new replay_opts_release() we'll need to stop
    +    calling it in sequencer_remove_state(), and instead call it where we
    +    allocate the "struct replay_opts" itself.
     
         This is because in e.g. do_interactive_rebase() we construct a "struct
         replay_opts" with "get_replay_opts()", and then call
    @@ Commit message
     
         But if we encounter errors anywhere along the way we'd punt out early,
         and not free() the memory we allocated. Remembering whether we
    -    previously called sequencer_remove_state() would be a hassle, so let's
    -    make it safe to re-invoke replay_opts_release() instead.
    +    previously called sequencer_remove_state() would be a hassle.
     
    -    I experimented with a change to be more paranoid instead, i.e. to
    -    exhaustively check our state via an enum. We could make sure that we:
    -
    -    - Only allow calling "replay_opts_release()" after
    -      "sequencer_remove_state()", but not the other way around.
    -
    -    - Forbid invoking either function twice in a row.
    -
    -    But such paranoia isn't warranted here, let's instead take the easy
    -    way out and FREE_AND_NULL() this.
    +    Using a FREE_AND_NULL() pattern would also work, as it would be safe
    +    replay_opts_release() repeatedly, but let's fix this properly instead,
    +    by having the owner of the data free() it.
     
         See [2] for the initial implementation of "sequencer_remove_state()",
         which assumed that it should be removing the full (including on-disk)
    @@ sequencer.c: static const char *gpg_sign_opt_quoted(struct replay_opts *opts)
      {
      	free(opts->gpg_sign);
      	free(opts->reflog_action);
    -@@ sequencer.c: static void replay_opts_release(struct replay_opts *opts)
    - 	free(opts->strategy);
    - 	for (size_t i = 0; i < opts->xopts_nr; i++)
    - 		free(opts->xopts[i]);
    -+	opts->xopts_nr = 0;
    - 	free(opts->xopts);
    - 	strbuf_release(&opts->current_fixups);
    - }
     @@ sequencer.c: int sequencer_remove_state(struct replay_opts *opts)
      		}
      	}
 4:  e83bdfab046 !  4:  9f72cc6e46b builtin/revert.c: move free-ing of "revs" to replay_opts_release()
    @@ Commit message
         rather than having these users reach into the struct to free its
         individual members.
     
    -    As explained in earlier change we should be using FREE_AND_NULL() in
    -    replay_opts_release() rather than free().
    -
         1. d1ec656d68f (cherry-pick: free "struct replay_opts" members,
            2022-11-08)
         2. fd74ac95ac3 (revert: free "struct replay_opts" members, 2022-07-01)
    @@ builtin/revert.c: int cmd_cherry_pick(int argc, const char **argv, const char *p
     
      ## sequencer.c ##
     @@ sequencer.c: void replay_opts_release(struct replay_opts *opts)
    - 	opts->xopts_nr = 0;
    + 		free(opts->xopts[i]);
      	free(opts->xopts);
      	strbuf_release(&opts->current_fixups);
     +	if (opts->revs)
 5:  4fea2b77c6d <  -:  ----------- builtin/rebase.c: rename "squash_onto_name" to "to_free"
 6:  898bb7698fc !  5:  3d5c3152f69 builtin/rebase.c: fix "options.onto_name" leak
    @@ Metadata
      ## Commit message ##
         builtin/rebase.c: fix "options.onto_name" leak
     
    -    In [1] we started saving away the earlier xstrdup()'d
    -    "options.onto_name" assignment to free() it, but when [2] added this
    -    "keep_base" branch it didn't free() the already assigned value before
    -    re-assigning to "options.onto_name". Let's do that, and fix the memory
    -    leak.
    +    Similar to the existing "squash_onto_name" added in [1] we need to
    +    free() the xstrdup()'d "options.onto.name" added for "--keep-base" in
    +    [2]..
     
         1. 9dba809a69a (builtin rebase: support --root, 2018-09-04)
         2. 414d924beb4 (rebase: teach rebase --keep-base, 2019-08-27)
    @@ Commit message
         Signed-off-by: Ævar Arnfjörð Bjarmason <avarab@xxxxxxxxx>
     
      ## builtin/rebase.c ##
    +@@ builtin/rebase.c: int cmd_rebase(int argc, const char **argv, const char *prefix)
    + 	struct string_list strategy_options = STRING_LIST_INIT_NODUP;
    + 	struct object_id squash_onto;
    + 	char *squash_onto_name = NULL;
    ++	char *keep_base_onto_name = NULL;
    + 	int reschedule_failed_exec = -1;
    + 	int allow_preemptive_ff = 1;
    + 	int preserve_merges_selected = 0;
     @@ builtin/rebase.c: int cmd_rebase(int argc, const char **argv, const char *prefix)
      		strbuf_addstr(&buf, options.upstream_name);
      		strbuf_addstr(&buf, "...");
      		strbuf_addstr(&buf, branch_name);
     -		options.onto_name = xstrdup(buf.buf);
    -+		free(to_free);
    -+		options.onto_name = to_free = xstrdup(buf.buf);
    ++		options.onto_name = keep_base_onto_name = xstrdup(buf.buf);
      	} else if (!options.onto_name)
      		options.onto_name = options.upstream_name;
      	if (strstr(options.onto_name, "...")) {
    +@@ builtin/rebase.c: int cmd_rebase(int argc, const char **argv, const char *prefix)
    + 	free(options.strategy);
    + 	strbuf_release(&options.git_format_patch_opt);
    + 	free(squash_onto_name);
    ++	free(keep_base_onto_name);
    + 	string_list_clear(&exec, 0);
    + 	string_list_clear(&strategy_options, 0);
    + 	return !!ret;
     
      ## t/t3416-rebase-onto-threedots.sh ##
     @@ t/t3416-rebase-onto-threedots.sh: test_description='git rebase --onto A...B'
 7:  fb38dc873f9 =  6:  c07dc006c6d sequencer.c: always free() the "msgbuf" in do_pick_commit()
 8:  d4b0e2a5c83 !  7:  ee8262ab22a builtin/rebase.c: free() "options.strategy_opts"
    @@ builtin/rebase.c: int cmd_rebase(int argc, const char **argv, const char *prefix
      	free(options.strategy);
     +	free(options.strategy_opts);
      	strbuf_release(&options.git_format_patch_opt);
    - 	free(to_free);
    - 	string_list_clear(&exec, 0);
    + 	free(squash_onto_name);
    + 	free(keep_base_onto_name);
 9:  fd9c7a5547c =  8:  84343ea6bf6 commit.c: free() revs.commit in get_fork_point()
-- 
2.39.0.1225.g30a3d88132d




[Index of Archives]     [Linux Kernel Development]     [Gcc Help]     [IETF Annouce]     [DCCP]     [Netdev]     [Networking]     [Security]     [V4L]     [Bugtraq]     [Yosemite]     [MIPS Linux]     [ARM Linux]     [Linux Security]     [Linux RAID]     [Linux SCSI]     [Fedora Users]

  Powered by Linux