On Tue, Jun 22, 2021 at 9:53 AM Hillf Danton <hdanton@xxxxxxxx> wrote: > > On Mon, 21 Jun 2021 22:32:28 -0400 Mimi Zohar wrote: > >On Fri, 2021-06-18 at 12:01 +0800, Hillf Danton wrote: > >> On Sun, 4 Apr 2021 11:10:48 +0300 Amir Goldstein wrote: > >> >On Sat, Apr 3, 2021 at 10:18 PM syzbot wrote: > >> >> > >> >> syzbot has found a reproducer for the following issue on: > >> >> > >> >> HEAD commit: 454c576c Add linux-next specific files for 20210401 > >> >> git tree: linux-next > >> >> console output: https://syzkaller.appspot.com/x/log.txt?x=1616e07ed00000 > >> >> kernel config: https://syzkaller.appspot.com/x/.config?x=920cc274cae812a5 > >> >> dashboard link: https://syzkaller.appspot.com/bug?extid=c18f2f6a7b08c51e3025 > >> >> syz repro: https://syzkaller.appspot.com/x/repro.syz?x=13da365ed00000 > >> >> C reproducer: https://syzkaller.appspot.com/x/repro.c?x=13ca9d16d00000 > >> >> > >> >> IMPORTANT: if you fix the issue, please add the following tag to the commit: > >> >> Reported-by: syzbot+c18f2f6a7b08c51e3025@xxxxxxxxxxxxxxxxxxxxxxxxx > >> >> > >> >> ====================================================== > >> >> WARNING: possible circular locking dependency detected > >> >> 5.12.0-rc5-next-20210401-syzkaller #0 Not tainted > >> >> ------------------------------------------------------ > >> >> syz-executor144/9166 is trying to acquire lock: > >> >> ffff888144cf0460 (sb_writers#5){.+.+}-{0:0}, at: ovl_maybe_copy_up+0x11f/0x190 fs/overlayfs/copy_up.c:995 > >> >> > >> >> but task is already holding lock: > >> >> ffff8880256d42c0 (&iint->mutex){+.+.}-{3:3}, at: process_measurement+0x3a8/0x17e0 security/integrity/ima/ima_main.c:253 > >> >> > >> >> which lock already depends on the new lock. > >> >> > >> >> > >> >> the existing dependency chain (in reverse order) is: > >> >> > >> >> -> #1 (&iint->mutex){+.+.}-{3:3}: > >> >> __mutex_lock_common kernel/locking/mutex.c:949 [inline] > >> >> __mutex_lock+0x139/0x1120 kernel/locking/mutex.c:1096 > >> >> process_measurement+0x3a8/0x17e0 security/integrity/ima/ima_main.c:253 > >> >> ima_file_check+0xb9/0x100 security/integrity/ima/ima_main.c:499 > >> >> do_open fs/namei.c:3361 [inline] > >> >> path_openat+0x15b5/0x27e0 fs/namei.c:3492 > >> >> do_filp_open+0x17e/0x3c0 fs/namei.c:3519 > >> >> do_sys_openat2+0x16d/0x420 fs/open.c:1187 > >> >> do_sys_open fs/open.c:1203 [inline] > >> >> __do_sys_open fs/open.c:1211 [inline] > >> >> __se_sys_open fs/open.c:1207 [inline] > >> >> __x64_sys_open+0x119/0x1c0 fs/open.c:1207 > >> >> do_syscall_64+0x2d/0x70 arch/x86/entry/common.c:46 > >> >> entry_SYSCALL_64_after_hwframe+0x44/0xae > >> >> > >> >> -> #0 (sb_writers#5){.+.+}-{0:0}: > >> >> check_prev_add kernel/locking/lockdep.c:2938 [inline] > >> >> check_prevs_add kernel/locking/lockdep.c:3061 [inline] > >> >> validate_chain kernel/locking/lockdep.c:3676 [inline] > >> >> __lock_acquire+0x2a17/0x5230 kernel/locking/lockdep.c:4902 > >> >> lock_acquire kernel/locking/lockdep.c:5512 [inline] > >> >> lock_acquire+0x1ab/0x740 kernel/locking/lockdep.c:5477 > >> >> percpu_down_read include/linux/percpu-rwsem.h:51 [inline] > >> >> __sb_start_write include/linux/fs.h:1758 [inline] > >> >> sb_start_write include/linux/fs.h:1828 [inline] > >> >> mnt_want_write+0x6e/0x3e0 fs/namespace.c:375 > >> >> ovl_maybe_copy_up+0x11f/0x190 fs/overlayfs/copy_up.c:995 > >> >> ovl_open+0xba/0x270 fs/overlayfs/file.c:149 > >> >> do_dentry_open+0x4b9/0x11b0 fs/open.c:826 > >> >> vfs_open fs/open.c:940 [inline] > >> >> dentry_open+0x132/0x1d0 fs/open.c:956 > >> >> ima_calc_file_hash+0x2d2/0x4b0 security/integrity/ima/ima_crypto.c:557 > >> >> ima_collect_measurement+0x4ca/0x570 security/integrity/ima/ima_api.c:252 > >> >> process_measurement+0xd1c/0x17e0 security/integrity/ima/ima_main.c:330 > >> >> ima_file_check+0xb9/0x100 security/integrity/ima/ima_main.c:499 > >> >> do_open fs/namei.c:3361 [inline] > >> >> path_openat+0x15b5/0x27e0 fs/namei.c:3492 > >> >> do_filp_open+0x17e/0x3c0 fs/namei.c:3519 > >> >> do_sys_openat2+0x16d/0x420 fs/open.c:1187 > >> >> do_sys_open fs/open.c:1203 [inline] > >> >> __do_sys_open fs/open.c:1211 [inline] > >> >> __se_sys_open fs/open.c:1207 [inline] > >> >> __x64_sys_open+0x119/0x1c0 fs/open.c:1207 > >> >> do_syscall_64+0x2d/0x70 arch/x86/entry/common.c:46 > >> >> entry_SYSCALL_64_after_hwframe+0x44/0xae > >> >> > >> >> other info that might help us debug this: > >> >> > >> >> Possible unsafe locking scenario: > >> >> > >> >> CPU0 CPU1 > >> >> ---- ---- > >> >> lock(&iint->mutex); > >> >> lock(sb_writers#5); > >> >> lock(&iint->mutex); > >> >> lock(sb_writers#5); > >> >> > >> >> *** DEADLOCK *** > >> >> > >> >> 1 lock held by syz-executor144/9166: > >> >> #0: ffff8880256d42c0 (&iint->mutex){+.+.}-{3:3}, at: process_measurement+0x3a8/0x17e0 security/integrity/ima/ima_main.c:253 > >> >> > >> > >> It is reported again. > >> https://lore.kernel.org/lkml/00000000000067d24205c4d0e599@xxxxxxxxxx/ > >> > > >> >It's a false positive lockdep warning due to missing annotation of > >> >stacking layer on iint->mutex in IMA code. > >> > >> Add it by copying what's created for ovl, see below. > >> > > >> >To fix it properly, iint->mutex, which can be taken in any of the > >> >stacking fs layers, should be annotated with stacking depth like > >> >ovl_lockdep_annotate_inode_mutex_key() > >> > > >> >I think it's the same root cause as: > >> >https://syzkaller.appspot.com/bug?extid=18a1619cceea30ed45af > >> >https://syzkaller.appspot.com/bug?extid=ae82084b07d0297e566b > >> > > >> >I think both of the above were marked "fixed" by a paper over. > >> >The latter was marked "fixed" by "ovl: detect overlapping layers" > >> >but if you look at the repro, the fact that 'workdir' overlaps with > >> >'lowerdir' has nothing to do with the lockdep warning, so said > >> >"fix" just papered over the IMA lockdep warning. > >> > > >> >Thanks, > >> >Amir. > >> > >> +++ x/security/integrity/iint.c > >> @@ -85,6 +85,45 @@ static void iint_free(struct integrity_i > >> kmem_cache_free(iint_cache, iint); > >> } > >> > >> +/* > >> + * a copy from ovl_lockdep_annotate_inode_mutex_key() in a bit to fix > >> + > >> + Possible unsafe locking scenario: > >> + > >> + CPU0 CPU1 > >> + ---- ---- > >> + lock(&iint->mutex); > >> + lock(sb_writers#5); > >> + lock(&iint->mutex); > >> + lock(sb_writers#5); > >> + > >> + *** DEADLOCK *** > >> + > >> +It's a false positive lockdep warning due to missing annotation of > >> +stacking layer on iint->mutex in IMA code. [1] > >> + > >> +[1] https://lore.kernel.org/linux-unionfs/CAOQ4uxjk4XYuwz5HCmN-Ge=Ld=tM1f7ZxVrd5U1AC2Wisc9MTA@xxxxxxxxxxxxxx/ > >> +*/ > >> +static void iint_annotate_mutex_key(struct integrity_iint_cache *iint, > >> + struct inode *inode) > >> +{ > >> +#ifdef CONFIG_LOCKDEP > >> + static struct lock_class_key > >> + iint_mutex_key[FILESYSTEM_MAX_STACK_DEPTH], > >> + iint_mutex_dir_key[FILESYSTEM_MAX_STACK_DEPTH]; > >> + > >> + int depth = inode->i_sb->s_stack_depth - 1; > >> + > >> + if (WARN_ON_ONCE(depth < 0 || depth >= FILESYSTEM_MAX_STACK_DEPTH)) > >> + depth = 0; > >> + > >> + if (S_ISDIR(inode->i_mode)) > >> + lockdep_set_class(&iint->mutex, &iint_mutex_dir_key[depth]); > >> + else > >> + lockdep_set_class(&iint->mutex, &iint_mutex_key[depth]); > >> +#endif > >> +} > > > >The iint cache is only for regular files. > > Yes you are right. > > > >> + > >> /** > >> * integrity_inode_get - find or allocate an iint associated with an inode > >> * @inode: pointer to the inode > >> @@ -113,6 +152,7 @@ struct integrity_iint_cache *integrity_i > >> iint = kmem_cache_alloc(iint_cache, GFP_NOFS); > >> if (!iint) > >> return NULL; > >> + iint_annotate_mutex_key(iint, inode); > >> > >> write_lock(&integrity_iint_lock); > > > >Should annotating the iint be limited to files on overlay filesystems? > > Yes but it is more difficult to address than thought without adding to > the vfs is_ovl_inode(inode). > > Aside from adding is_xfs_inode(inode), another option is move the > dentry_open() in ima_calc_file_hash() out of and before the iint->mutex in > process_measurement(), and that will keep the AB locking order intact. > > Thoughts are welcome. There is no need to detect overlayfs check for (inode->i_sb->s_stack_depth > 0) Thanks, Amir.
