fetch-pack: disregard invalid pack lockfiles
PATCH 1/1 fetch-pack: disregard invalid pack lockfiles
9da69a6539 (fetch-pack: support more than one pack lockfile, 2020-06-10) started to use a string_list for pack lockfile names instead of a single string pointer. It removed a NULL check from transport_unlock_pack() as well, which is the function that eventually deletes these lockfiles and releases their name strings.
index_pack_lockfile() can return NULL if it doesn't like the contents it reads from the file descriptor passed to it. unlink(2) is declared to not accept NULL pointers (at least with glibc). Undefined Behavior Sanitizer together with Address Sanitizer detects a case where a NULL lockfile name is passed to unlink(2) by transport_unlock_pack() in t1060 (make SANITIZE=address,undefined; cd t; ./t1060-object-corruption.sh).
Reinstate the NULL check to avoid undefined behavior, but put it right at the source, so that the number of items in the string_list reflects the number of valid lockfiles.
--- fetch-pack.c | 5 +++-- 1 file changed, 3 insertions(+), 2 deletions(-)
fetch-pack.c +3 -2
@@ -915,8 +915,9 @@ static int get_pack(struct fetch_pack_args *args,if (start_command (&cmd))die (_ ("fetch-pack: unable to fork off %s" ), cmd_name);if (do_keep && pack_lockfiles) { -string_list_append_nodup (pack_lockfiles, -index_pack_lockfile (cmd.out)); +char *pack_lockfile =index_pack_lockfile (cmd.out); +if (pack_lockfile) +string_list_append_nodup (pack_lockfiles, pack_lockfile);close (cmd.out); }
--
2.29.2
Quoted 6 lines · René Scharfe
index_pack_lockfile()can return NULL if it doesn't like the contents it reads from the file descriptor passed to it. unlink(2) is declared to not accept NULL pointers (at least with glibc). Undefined Behavior Sanitizer together with Address Sanitizer detects a case where a NULL lockfile name is passed to unlink(2) bytransport_unlock_pack()in t1060 (make SANITIZE=address,undefined; cd t; ./t1060-object-corruption.sh).
Which test in t1060? I tried to reproduce this myself, but couldn't seem to coax out a failure. (Initially I thought that my ccache wasn't letting me recompile with the SANITIZE options, but running 'ccache clear' and then trying again left the test still passing).
Quoted 14 lines
Reinstate the NULL check to avoid undefined behavior, but put it right at the source, so that the number of items in the
string_listreflects the number of valid lockfiles.Signed-off-by René Scharfe <[email protected]>--- fetch-pack.c | 5 +++-- 1 file changed, 3 insertions(+), 2 deletions(-)fetch-pack.c +0
-0
fetch-pack.c static int get_pack(struct fetch_pack_args *args, · line 920
Quoted 8 lines
fetch-pack.c +3
-2@@ -915,8 +915,9 @@ static int get_pack(struct fetch_pack_args *args,if (start_command (&cmd))die (_ ("fetch-pack: unable to fork off %s" ), cmd_name);if (do_keep && pack_lockfiles) { -string_list_append_nodup (pack_lockfiles, -index_pack_lockfile (cmd.out)); +char *pack_lockfile =index_pack_lockfile (cmd.out); +if (pack_lockfile) +string_list_append_nodup (pack_lockfiles, pack_lockfile);
Makes sense.
Thanks,
Taylor
Am 30.11.20 um 20:53 schrieb Taylor Blau:
Quoted 8 lines · René Scharfe
Quoted 6 lines
index_pack_lockfile()can return NULL if it doesn't like the contents it reads from the file descriptor passed to it. unlink(2) is declared to not accept NULL pointers (at least with glibc). Undefined Behavior Sanitizer together with Address Sanitizer detects a case where a NULL lockfile name is passed to unlink(2) bytransport_unlock_pack()in t1060 (make SANITIZE=address,undefined; cd t; ./t1060-object-corruption.sh).
Quoted 4 lines
Which test in t1060? I tried to reproduce this myself, but couldn't seem to coax out a failure. (Initially I thought that my ccache wasn't letting me recompile with the SANITIZE options, but running 'ccache clear' and then trying again left the test still passing).
15 - fetch into corrupted repo with index-pack
$ cat trash\ directory.t1060-object-corruption/bit-error-cp/stderr error: inflate: data stream error (invalid distance too far back) error: unable to unpack d95f3ad14dee633a758d2e331151e950dd13e4ed header fatal: cannot read existing object info d95f3ad14dee633a758d2e331151e950dd13e4ed fatal: index-pack failed wrapper.c:568:52: runtime error: null pointer passed as argument 1, which is declared to never be null /usr/include/unistd.h:825:48: note: nonnull attribute specified here SUMMARY: UndefinedBehaviorSanitizer: undefined-behavior wrapper.c:568:52 in Aborted
Compiled with:
Debian clang version 11.0.0-5+b1
Target: x86_64-pc-linux-gnu
Thread model: posix
InstalledDir: /usr/bin
René
Taylor Blau
Reviewed-byQuoted 26 lines · René Scharfe
Am 30.11.20 um 20:53 schrieb Taylor Blau:
Quoted 8 lines · René Scharfe
Quoted 6 lines
index_pack_lockfile()can return NULL if it doesn't like the contents it reads from the file descriptor passed to it. unlink(2) is declared to not accept NULL pointers (at least with glibc). Undefined Behavior Sanitizer together with Address Sanitizer detects a case where a NULL lockfile name is passed to unlink(2) bytransport_unlock_pack()in t1060 (make SANITIZE=address,undefined; cd t; ./t1060-object-corruption.sh).Quoted 4 lines
Which test in t1060? I tried to reproduce this myself, but couldn't seem to coax out a failure. (Initially I thought that my ccache wasn't letting me recompile with the SANITIZE options, but running 'ccache clear' and then trying again left the test still passing).
15 - fetch into corrupted repo with index-pack
$ cat trash\ directory.t1060-object-corruption/bit-error-cp/stderr error: inflate: data stream error (invalid distance too far back) error: unable to unpack d95f3ad14dee633a758d2e331151e950dd13e4ed header fatal: cannot read existing object info d95f3ad14dee633a758d2e331151e950dd13e4ed fatal: index-pack failed wrapper.c:568:52: runtime error: null pointer passed as argument 1, which is declared to never be null /usr/include/unistd.h:825:48: note: nonnull attribute specified here SUMMARY: UndefinedBehaviorSanitizer: undefined-behavior wrapper.c:568:52 in Aborted
Quoted 5 lines
Compiled with:
Debian clang version 11.0.0-5+b1 Target: x86_64-pc-linux-gnu Thread model: posix InstalledDir: /usr/bin
I see. I was compiling with: gcc 10.2.0, so setting CC=clang does reproduce the error for me.
Thanks,
Taylor
Quoted 13 lines · René Scharfe
9da69a6539(fetch-pack: support more than one pack lockfile, 2020-06-10) started to use astring_listfor pack lockfile names instead of a single string pointer. It removed a NULL check fromtransport_unlock_pack()as well, which is the function that eventually deletes these lockfiles and releases their name strings.
index_pack_lockfile()can return NULL if it doesn't like the contents it reads from the file descriptor passed to it. unlink(2) is declared to not accept NULL pointers (at least with glibc). Undefined Behavior Sanitizer together with Address Sanitizer detects a case where a NULL lockfile name is passed to unlink(2) bytransport_unlock_pack()in t1060 (make SANITIZE=address,undefined; cd t; ./t1060-object-corruption.sh).
Quoted 3 lines
Reinstate the NULL check to avoid undefined behavior, but put it right at the source, so that the number of items in the
string_listreflects the number of valid lockfiles.
It took me a minute to understand how 9da69a6539 made this worse, since in the hunk you're touching here, the original "if NULL, do nothing" check was checking the pointer-to-pointer to see if the caller was interested in the lockfile name. But your "but put it right at the source" pointed me in the right direction. The hunk from 9da69a6539 that matters is this one:
- if (pack_lockfile) {
- printf("lock %s\n", pack_lockfile);
+ if (pack_lockfiles.nr) {
+ int i;
+
+ printf("lock %s\n", pack_lockfiles.items[0].string);
fflush(stdout); + for (i = 1; i < pack_lockfiles.nr; i++)
+ warning(_("Lockfile created but not reported: %s"),
+ pack_lockfiles.items
(not complaining about anything, just verbosely reviewing).
Quoted 5 lines
fetch-pack.c +0
-0
fetch-pack.c static int get_pack(struct fetch_pack_args *args, · line 922
Quoted 10 lines
fetch-pack.c +3
-2@@ -915,8 +915,9 @@ static int get_pack(struct fetch_pack_args *args,if (start_command (&cmd))die (_ ("fetch-pack: unable to fork off %s" ), cmd_name);if (do_keep && pack_lockfiles) { -string_list_append_nodup (pack_lockfiles, -index_pack_lockfile (cmd.out)); +char *pack_lockfile =index_pack_lockfile (cmd.out); +if (pack_lockfile) +string_list_append_nodup (pack_lockfiles, pack_lockfile);close (cmd.out); }
So this is an obviously correct fix, but I have to wonder whether we ought to be (even before 9da69a6539) complaining about pack-objects not correctly reporting the pack name to us. I think in practice this happens when it dies early without reporting anything to us, in which case we'd notice its non-zero exit anyway. So it's probably not a big deal, but would amount to an assertion (and if we did want to do it, it should come on top of your fix, and not hold it up).
-Peff