Re: git pull --set-upstream segfaults on branchless repo.

["Jan Pokorný" <poki@fnusa.cz>]

Hello,

Clemens Fruhwirth píše v Po 05. 07. 2021 v 17:46 +0200:
> What did you do before the bug happened? (Steps to reproduce your
> issue)
> 
> Run "git pull origin nixos-unstable --rebase --set-upstream"
> on a repo that had no branch set, e.g. when running "git branch" gave
> "* (no branch)"
> 
> What did you expect to happen? (Expected behavior)
> Pull from upstream and set upstream.
> 
> What happened instead? (Actual behavior)
> Segfault with the following trace:
> (gdb) bt
> #0  0x000000000044a8c9 in do_fetch (transport=0x2557920, rs=rs@entry=0x7ffd42335c00) at builtin/fetch.c:1568
> #1  0x000000000044ac61 in fetch_one (remote=<optimized out>, argc=1110662144, argv=0x7ffd42335fc8, prune_tags_ok=<optimized out>, use_stdin_refspecs=0) at builtin/fetch.c:1892
> #2  0x000000000044af15 in cmd_fetch (argc=0, argv=0x7ffd42335fc8, prefix=0x0) at builtin/fetch.c:1992
> #3  0x0000000000406354 in run_builtin (p=0x70d3e0 <commands+960>, argc=argc@entry=5, argv=argv@entry=0x7ffd42335fc0) at git.c:453
> #4  0x00000000004065c3 in handle_builtin (argc=5, argv=0x7ffd42335fc0) at git.c:704
> #5  0x0000000000407c4a in run_argv (argcp=argcp@entry=0x7ffd42335e8c, argv=argv@entry=0x7ffd42335e80) at git.c:771
> #6  0x00000000004080a4 in cmd_main (argc=<optimized out>, argc@entry=6, argv=<optimized out>, argv@entry=0x7ffd42335fb8) at git.c:902
> #7  0x00000000004c614c in main (argc=6, argv=0x7ffd42335fb8) at common-main.c:52
> (gdb) p branch
> $1 = (struct branch *) 0x0
> 
> What's different between what you expected and what actually happened?
> Not segfault
> 
> Anything else you want to add:
> 
> Dropping the "--set-upstream" makes the segfault go away. Looking at
> builtin/fetch.c around L1568 I see,
> 
>         if (set_upstream) {
>                 struct branch *branch = branch_get("HEAD");
>                 [..]
>                 if (source_ref) {
>                         if (!strcmp(source_ref->name, "HEAD") ||
>                             starts_with(source_ref->name, "refs/heads/"))
>                                 install_branch_config(0,
>                                                       branch->name, // <- SEGFAULT HERE
>                                                       transport->remote->name,
>                                                       source_ref->name);
>                         [..]
>               }
>         }
> 
> It's rather clear that branch is just NULL from the gdb session above,
> and the branch->name dereference fails.
> It might be useful to catch branch == NULL.
> 
> Please review the rest of the bug report below.
> You can delete any lines you don't wish to share.
> 
> [System Info]
> git version:
> git version 2.31.1
> [...]

observed something pretty similar on Fedora, with both
git-2.32.0-0.3.rc2.fc35.x86_64 and git-2.32.0-1.fc35.x86_64, e.g.:

$ git clone https://github.com/git/git.git
[...]
$ cd git
$ git checkout --detach v2.1.0
HEAD is now at 6c4ab27f23 Git 2.1
$ git pull --set-upstream origin master
From https://github.com/git/git
 * branch                  master     -> FETCH_HEAD
error: fetch died of signal 11

  1625			if (source_ref) {
  1626				if (!strcmp(source_ref->name, "HEAD") ||
  1627				   starts_with(source_ref->name, "refs/heads/"))
=>1628					install_branch_config(0,
  1629							     branch->name,
  1630							     transport->remote->name,
  1631							     source_ref->name);

(gdb) p branch
$1 = <optimized out>
(gdb) p branch->name
value has been optimized out

   0x000055b5bceba921 <+8705>:	nopl   0x0(%rax)
   0x000055b5bceba928 <+8712>:	mov    (%rsp),%rax
=> 0x000055b5bceba92c <+8716>:	mov    (%r8),%rsi
   0x000055b5bceba92f <+8719>:	xor    %edi,%edi
   0x000055b5bceba931 <+8721>:	mov    0x8(%rax),%rax
   0x000055b5bceba935 <+8725>:	mov    0x10(%rax),%rdx
   0x000055b5bceba939 <+8729>:	call   0x55b5bcf75d40 <install_branch_config>

(gdb) info registers r8 rsi
r8             0x0                 0
rsi            0x55b5bdbe430b      94239060804363

This is just a generic reproducer; original use case revolved around
the use of submodules (also happened to be in a detached state within,
"--set-upstream" attempt was then my rushed response to "You are
not currently on a branch." error message on a plain pull there).

> (Please cc me on replies, not subscribe to the mailing list).

(ditto, please)

-- poki