| Message ID | 555ec6717ecab0fe6ef5660bcf0d61d59f84ef8b.1641552500.git.ps@pks.im (mailing list archive) |
|---|---|
| State | Accepted |
| Commit | ff0ae63f07f018b813326f6baf0bdcb766f24a21 |
| Headers | show |
| Series | [1/1] fetch: fix deadlock when cleaning up lockfiles in async signals | expand |
On 2022-01-07 at 10:55:47, Patrick Steinhardt wrote: > [Resend with the correct In-Reply-To header set to fix threading] > > When fetching packfiles, we write a bunch of lockfiles for the packfiles > we're writing into the repository. In order to not leave behind any > cruft in case we exit or receive a signal, we register both an exit > handler as well as signal handlers for common signals like SIGINT. These > handlers will then unlink the locks and free the data structure tracking > them. We have observed a deadlock in this logic though: > > (gdb) bt > #0 __lll_lock_wait_private () at ../sysdeps/unix/sysv/linux/x86_64/lowlevellock.S:95 > #1 0x00007f4932bea2cd in _int_free (av=0x7f4932f2eb20 <main_arena>, p=0x3e3e4200, have_lock=0) at malloc.c:3969 > #2 0x00007f4932bee58c in __GI___libc_free (mem=<optimized out>) at malloc.c:2975 > #3 0x0000000000662ab1 in string_list_clear () > #4 0x000000000044f5bc in unlock_pack_on_signal () > #5 <signal handler called> > #6 _int_free (av=0x7f4932f2eb20 <main_arena>, p=<optimized out>, have_lock=0) at malloc.c:4024 > #7 0x00007f4932bee58c in __GI___libc_free (mem=<optimized out>) at malloc.c:2975 > #8 0x000000000065afd5 in strbuf_release () > #9 0x000000000066ddb9 in delete_tempfile () > #10 0x0000000000610d0b in files_transaction_cleanup.isra () > #11 0x0000000000611718 in files_transaction_abort () > #12 0x000000000060d2ef in ref_transaction_abort () > #13 0x000000000060d441 in ref_transaction_prepare () > #14 0x000000000060e0b5 in ref_transaction_commit () > #15 0x00000000004511c2 in fetch_and_consume_refs () > #16 0x000000000045279a in cmd_fetch () > #17 0x0000000000407c48 in handle_builtin () > #18 0x0000000000408df2 in cmd_main () > #19 0x00000000004078b5 in main () > > The process was killed with a signal, which caused the signal handler to > kick in and try free the data structures after we have unlinked the > locks. It then deadlocks while calling free(3P). > > The root cause of this is that it is not allowed to call certain > functions in async-signal handlers, as specified by signal-safety(7). > Next to most I/O functions, this list of disallowed functions also > includes memory-handling functions like malloc(3P) and free(3P) because > they may not be reentrant. As a result, if we execute such functions in > the signal handler, then they may operate on inconistent state and fail > in unexpected ways. > > Fix this bug by not calling non-async-signal-safe functions when running > in the signal handler. We're about to re-raise the signal anyway and > will thus exit, so it's not much of a problem to keep the string list of > lockfiles untouched. Note that it's fine though to call unlink(2), so > we'll still clean up the lockfiles correctly. I took a look, and this seems reasonable to me. I know in the non-signal case, we'd want to clean up because it means we can check for leaks, but I don't see the utility of running Git under Valgrind and then sending it a signal, and I think it's just safe to ignore that case.
On Fri, Jan 07, 2022 at 11:55:47AM +0100, Patrick Steinhardt wrote: > diff --git a/transport.c b/transport.c > index 92ab9a3fa6..2a3e324154 100644 > --- a/transport.c > +++ b/transport.c > @@ -1456,13 +1456,18 @@ int transport_fetch_refs(struct transport *transport, struct ref *refs) > return rc; > } > > -void transport_unlock_pack(struct transport *transport) > +void transport_unlock_pack(struct transport *transport, unsigned int flags) > { > + int in_signal_handler = !!(flags & TRANSPORT_UNLOCK_PACK_IN_SIGNAL_HANDLER); > int i; > > for (i = 0; i < transport->pack_lockfiles.nr; i++) > - unlink_or_warn(transport->pack_lockfiles.items[i].string); > - string_list_clear(&transport->pack_lockfiles, 0); > + if (in_signal_handler) > + unlink(transport->pack_lockfiles.items[i].string); > + else > + unlink_or_warn(transport->pack_lockfiles.items[i].string); This puzzled me when I first read it. But unlink_or_warn() isn't reentrant because it performs buffered IO on stderr, so if we reached this signal handler while executing another function call modifying those same buffers, the call within the signal handler would have undefined behavior. So that makes sense: freeing (with string_list_clear() below) and performing buffered IO (with unlink_or_warn() here as just described) are certainly off the table. But is unlink() safe as-is? I'm not so sure. Reading signal-safety(7), it's clearly on the list of functions that are OK to call. But reading the errno section: errno Fetching and setting the value of errno is async-signal-safe provided that the signal handler saves errno on entry and restores its value before returning. We certainly not doing that, though that's nothing new, and so I wonder why it doesn't seem to be an issue in practice. Thanks, Taylor
Hi Taylor On 07/01/2022 22:41, Taylor Blau wrote: > On Fri, Jan 07, 2022 at 11:55:47AM +0100, Patrick Steinhardt wrote: >> diff --git a/transport.c b/transport.c >> index 92ab9a3fa6..2a3e324154 100644 >> --- a/transport.c >> +++ b/transport.c >> @@ -1456,13 +1456,18 @@ int transport_fetch_refs(struct transport *transport, struct ref *refs) >> return rc; >> } >> >> -void transport_unlock_pack(struct transport *transport) >> +void transport_unlock_pack(struct transport *transport, unsigned int flags) >> { >> + int in_signal_handler = !!(flags & TRANSPORT_UNLOCK_PACK_IN_SIGNAL_HANDLER); >> int i; >> >> for (i = 0; i < transport->pack_lockfiles.nr; i++) >> - unlink_or_warn(transport->pack_lockfiles.items[i].string); >> - string_list_clear(&transport->pack_lockfiles, 0); >> + if (in_signal_handler) >> + unlink(transport->pack_lockfiles.items[i].string); >> + else >> + unlink_or_warn(transport->pack_lockfiles.items[i].string); > > This puzzled me when I first read it. But unlink_or_warn() isn't > reentrant because it performs buffered IO on stderr, so if we reached > this signal handler while executing another function call modifying > those same buffers, the call within the signal handler would have > undefined behavior. > > So that makes sense: freeing (with string_list_clear() below) and > performing buffered IO (with unlink_or_warn() here as just described) > are certainly off the table. > > But is unlink() safe as-is? I'm not so sure. Reading signal-safety(7), > it's clearly on the list of functions that are OK to call. But reading > the errno section: > > errno > Fetching and setting the value of errno is async-signal-safe > provided that the signal handler saves errno on entry and restores > its value before returning. > > We certainly not doing that, though that's nothing new, and so I wonder > why it doesn't seem to be an issue in practice. Because in this case we re-raise the signal and exit it does not matter if unlink() clobbers errno. If instead the program were to continue after handling the signal then we would have to save and restore errno to avoid interfering with the code that was running when the signal handler was called. Best Wishes Phillip > Thanks, > Taylor
On Sat, Jan 08, 2022 at 10:54:35AM +0000, Phillip Wood wrote: > > But is unlink() safe as-is? I'm not so sure. Reading signal-safety(7), > > it's clearly on the list of functions that are OK to call. But reading > > the errno section: > > > > (snip) > > > > We certainly not doing that, though that's nothing new, and so I wonder > > why it doesn't seem to be an issue in practice. > > Because in this case we re-raise the signal and exit it does not matter if > unlink() clobbers errno. If instead the program were to continue after > handling the signal then we would have to save and restore errno to avoid > interfering with the code that was running when the signal handler was > called. That makes perfect sense to me. Thanks for a clear explanation. Taylor
diff --git a/builtin/clone.c b/builtin/clone.c index 5bed37f8b5..585eef9b9a 100644 --- a/builtin/clone.c +++ b/builtin/clone.c @@ -1290,7 +1290,7 @@ int cmd_clone(int argc, const char **argv, const char *prefix) */ submodule_progress = transport->progress; - transport_unlock_pack(transport); + transport_unlock_pack(transport, 0); transport_disconnect(transport); if (option_dissociate) { diff --git a/builtin/fetch.c b/builtin/fetch.c index f1fe73a3e0..4bc04522ed 100644 --- a/builtin/fetch.c +++ b/builtin/fetch.c @@ -223,17 +223,22 @@ static struct option builtin_fetch_options[] = { OPT_END() }; -static void unlock_pack(void) +static void unlock_pack(unsigned int flags) { if (gtransport) - transport_unlock_pack(gtransport); + transport_unlock_pack(gtransport, flags); if (gsecondary) - transport_unlock_pack(gsecondary); + transport_unlock_pack(gsecondary, flags); +} + +static void unlock_pack_atexit(void) +{ + unlock_pack(0); } static void unlock_pack_on_signal(int signo) { - unlock_pack(); + unlock_pack(TRANSPORT_UNLOCK_PACK_IN_SIGNAL_HANDLER); sigchain_pop(signo); raise(signo); } @@ -1329,7 +1334,7 @@ static int fetch_and_consume_refs(struct transport *transport, trace2_region_leave("fetch", "consume_refs", the_repository); out: - transport_unlock_pack(transport); + transport_unlock_pack(transport, 0); return ret; } @@ -1978,7 +1983,7 @@ static int fetch_one(struct remote *remote, int argc, const char **argv, gtransport->server_options = &server_options; sigchain_push_common(unlock_pack_on_signal); - atexit(unlock_pack); + atexit(unlock_pack_atexit); sigchain_push(SIGPIPE, SIG_IGN); exit_code = do_fetch(gtransport, &rs); sigchain_pop(SIGPIPE); diff --git a/transport.c b/transport.c index 92ab9a3fa6..2a3e324154 100644 --- a/transport.c +++ b/transport.c @@ -1456,13 +1456,18 @@ int transport_fetch_refs(struct transport *transport, struct ref *refs) return rc; } -void transport_unlock_pack(struct transport *transport) +void transport_unlock_pack(struct transport *transport, unsigned int flags) { + int in_signal_handler = !!(flags & TRANSPORT_UNLOCK_PACK_IN_SIGNAL_HANDLER); int i; for (i = 0; i < transport->pack_lockfiles.nr; i++) - unlink_or_warn(transport->pack_lockfiles.items[i].string); - string_list_clear(&transport->pack_lockfiles, 0); + if (in_signal_handler) + unlink(transport->pack_lockfiles.items[i].string); + else + unlink_or_warn(transport->pack_lockfiles.items[i].string); + if (!in_signal_handler) + string_list_clear(&transport->pack_lockfiles, 0); } int transport_connect(struct transport *transport, const char *name, diff --git a/transport.h b/transport.h index 8bb4c8bbc8..3f16e50c19 100644 --- a/transport.h +++ b/transport.h @@ -279,7 +279,19 @@ const struct ref *transport_get_remote_refs(struct transport *transport, */ const struct git_hash_algo *transport_get_hash_algo(struct transport *transport); int transport_fetch_refs(struct transport *transport, struct ref *refs); -void transport_unlock_pack(struct transport *transport); + +/* + * If this flag is set, unlocking will avoid to call non-async-signal-safe + * functions. This will necessarily leave behind some data structures which + * cannot be cleaned up. + */ +#define TRANSPORT_UNLOCK_PACK_IN_SIGNAL_HANDLER (1 << 0) + +/* + * Unlock all packfiles locked by the transport. + */ +void transport_unlock_pack(struct transport *transport, unsigned int flags); + int transport_disconnect(struct transport *transport); char *transport_anonymize_url(const char *url); void transport_take_over(struct transport *transport,
[Resend with the correct In-Reply-To header set to fix threading] When fetching packfiles, we write a bunch of lockfiles for the packfiles we're writing into the repository. In order to not leave behind any cruft in case we exit or receive a signal, we register both an exit handler as well as signal handlers for common signals like SIGINT. These handlers will then unlink the locks and free the data structure tracking them. We have observed a deadlock in this logic though: (gdb) bt #0 __lll_lock_wait_private () at ../sysdeps/unix/sysv/linux/x86_64/lowlevellock.S:95 #1 0x00007f4932bea2cd in _int_free (av=0x7f4932f2eb20 <main_arena>, p=0x3e3e4200, have_lock=0) at malloc.c:3969 #2 0x00007f4932bee58c in __GI___libc_free (mem=<optimized out>) at malloc.c:2975 #3 0x0000000000662ab1 in string_list_clear () #4 0x000000000044f5bc in unlock_pack_on_signal () #5 <signal handler called> #6 _int_free (av=0x7f4932f2eb20 <main_arena>, p=<optimized out>, have_lock=0) at malloc.c:4024 #7 0x00007f4932bee58c in __GI___libc_free (mem=<optimized out>) at malloc.c:2975 #8 0x000000000065afd5 in strbuf_release () #9 0x000000000066ddb9 in delete_tempfile () #10 0x0000000000610d0b in files_transaction_cleanup.isra () #11 0x0000000000611718 in files_transaction_abort () #12 0x000000000060d2ef in ref_transaction_abort () #13 0x000000000060d441 in ref_transaction_prepare () #14 0x000000000060e0b5 in ref_transaction_commit () #15 0x00000000004511c2 in fetch_and_consume_refs () #16 0x000000000045279a in cmd_fetch () #17 0x0000000000407c48 in handle_builtin () #18 0x0000000000408df2 in cmd_main () #19 0x00000000004078b5 in main () The process was killed with a signal, which caused the signal handler to kick in and try free the data structures after we have unlinked the locks. It then deadlocks while calling free(3P). The root cause of this is that it is not allowed to call certain functions in async-signal handlers, as specified by signal-safety(7). Next to most I/O functions, this list of disallowed functions also includes memory-handling functions like malloc(3P) and free(3P) because they may not be reentrant. As a result, if we execute such functions in the signal handler, then they may operate on inconistent state and fail in unexpected ways. Fix this bug by not calling non-async-signal-safe functions when running in the signal handler. We're about to re-raise the signal anyway and will thus exit, so it's not much of a problem to keep the string list of lockfiles untouched. Note that it's fine though to call unlink(2), so we'll still clean up the lockfiles correctly. Signed-off-by: Patrick Steinhardt <ps@pks.im> --- builtin/clone.c | 2 +- builtin/fetch.c | 17 +++++++++++------ transport.c | 11 ++++++++--- transport.h | 14 +++++++++++++- 4 files changed, 33 insertions(+), 11 deletions(-)