On 2023/9/29 0:42, Marco Pagani wrote:
On 2023-09-28 11:14, Jinjie Ruan wrote:
As Marco pointed out, commit 2810c1e99867 ("kunit: Fix wild-memory-access bug in kunit_free_suite_set()") causes test suites to run while the test module is still in MODULE_STATE_COMING. In that state, the module is not fully initialized, lacking sysfs, module_memory, args, init function which causes null-ptr-deref of using fake devices below.
Since load_module() notify MODULE_STATE_COMING in prepare_coming_module(), and then init sysfs and args etc. in parse_args() and mod_sysfs_setup(), after that it notify MODULE_STATE_LIVE in do_init_module(), and fake driver in the test suits depend on them. So the test suits should be executed when notify MODULE_STATE_LIVE.
But the kunit_free_suite_set() in kunit_module_exit() depends on the success of kunit_filter_suites() in kunit_module_init(). The best practice is to alloc and init resource when notify MODULE_STATE_COMING and free them when notify MODULE_STATE_GOING. So split the kunit_module_exec() from kunit_module_init() to run test suits when MODULE_STATE_LIVE, call kunit_filter_suites() and allocate memory in kunit_module_init() and call kunit_free_suite_set() in kunit_module_exit() to free the memory.
So if load_module() succeeds and notify module state as below, it calls kunit_module_init(), kunit_module_exec() and kunit_module_exit(), which will work ok. The mod->state state machine when load_module() succeeds:
kunit_filter_suites() kunit_module_exec() MODULE_STATE_UNFORMED ---> MODULE_STATE_COMING ---> MODULE_STATE_LIVE ^ | | | +---------------- MODULE_STATE_GOING <---------+ kunit_free_suite_set()If load_module() fails and notify module state as below, it call kunit_module_init() and kunit_module_exit(), which will also work ok. The mod->state state machine when load_module() fails at mod_sysfs_setup():
kunit_filter_suites() kunit_free_suite_set() MODULE_STATE_UNFORMED ---> MODULE_STATE_COMING ---> MODULE_STATE_GOING ^ | | | +-----------------------------------------------+general protection fault, probably for non-canonical address 0xdffffc0000000003: 0000 [#1] PREEMPT SMP KASAN KASAN: null-ptr-deref in range [0x0000000000000018-0x000000000000001f] CPU: 1 PID: 1868 Comm: modprobe Tainted: G W N 6.6.0-rc3+ #61 Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS 1.15.0-1 04/01/2014 RIP: 0010:kobject_namespace+0x71/0x150 Code: 89 fa 48 c1 ea 03 80 3c 02 00 0f 85 cd 00 00 00 48 b8 00 00 00 00 00 fc ff df 49 8b 5c 24 28 48 8d 7b 18 48 89 fa 48 c1 ea 03 <80> 3c 02 00 0f 85 c1 00 00 00 48 8b 43 18 48 85 c0 74 79 4c 89 e7 RSP: 0018:ffff88810f797288 EFLAGS: 00010206 RAX: dffffc0000000000 RBX: 0000000000000000 RCX: 0000000000000000 RDX: 0000000000000003 RSI: ffffffff847b4900 RDI: 0000000000000018 RBP: ffff88810ba08940 R08: 0000000000000001 R09: ffffed1021ef2e0f R10: ffff88810f79707f R11: 746e756f63666572 R12: ffffffffa0241990 R13: ffff88810ba08958 R14: ffff88810ba08968 R15: ffffffff84ac6c20 FS: 00007ff9f2186540(0000) GS:ffff888119c80000(0000) knlGS:0000000000000000 CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033 CR2: 00007fff73a2cff8 CR3: 000000010b77b002 CR4: 0000000000770ee0 DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000 DR3: 0000000000000000 DR6: 00000000fffe0ff0 DR7: 0000000000000400 PKRU: 55555554 Call Trace:
<TASK> ? die_addr+0x3d/0xa0 ? exc_general_protection+0x144/0x220 ? asm_exc_general_protection+0x22/0x30 ? kobject_namespace+0x71/0x150 kobject_add_internal+0x267/0x870 kobject_add+0x120/0x1f0 ? kset_create_and_add+0x160/0x160 ? __kmem_cache_alloc_node+0x1d2/0x350 ? _raw_spin_lock+0x87/0xe0 ? kobject_create_and_add+0x3c/0xb0 kobject_create_and_add+0x68/0xb0 module_add_driver+0x260/0x350 bus_add_driver+0x2c9/0x580 driver_register+0x133/0x460 kunit_run_tests+0xdb/0xef0 ? _prb_read_valid+0x3e3/0x550 ? _raw_spin_lock+0x87/0xe0 ? _raw_spin_lock_bh+0xe0/0xe0 ? __send_ipi_mask+0x1ba/0x450 ? __pte_offset_map+0x19/0x1f0 ? __pte_offset_map_lock+0xd6/0x1b0 ? __kunit_test_suites_exit+0x30/0x30 ? kvm_smp_send_call_func_ipi+0x68/0xc0 ? do_sync_core+0x22/0x30 ? smp_call_function_many_cond+0x1be/0xcf0 ? __text_poke+0x890/0x890 ? __text_poke+0x890/0x890 ? on_each_cpu_cond_mask+0x46/0x70 ? text_poke_bp_batch+0x413/0x570 ? do_sync_core+0x30/0x30 ? __jump_label_patch+0x34c/0x350 ? mutex_unlock+0x80/0xd0 ? __mutex_unlock_slowpath.constprop.0+0x2a0/0x2a0 __kunit_test_suites_init+0xc4/0x120 kunit_module_notify+0x36c/0x3b0 ? __kunit_test_suites_init+0x120/0x120 ? preempt_count_add+0x79/0x150 notifier_call_chain+0xbf/0x280 ? kasan_quarantine_put+0x21/0x1a0 blocking_notifier_call_chain_robust+0xbb/0x140 ? notifier_call_chain+0x280/0x280 ? 0xffffffffa0238000 load_module+0x4af0/0x67d0 ? module_frob_arch_sections+0x20/0x20 ? rwsem_down_write_slowpath+0x11a0/0x11a0 ? kernel_read_file+0x3ca/0x510 ? __x64_sys_fspick+0x2a0/0x2a0 ? init_module_from_file+0xd2/0x130 init_module_from_file+0xd2/0x130 ? __ia32_sys_init_module+0xa0/0xa0 ? userfaultfd_unmap_prep+0x3d0/0x3d0 ? _raw_spin_lock_bh+0xe0/0xe0 idempotent_init_module+0x339/0x610 ? init_module_from_file+0x130/0x130 ? __fget_light+0x57/0x500 __x64_sys_finit_module+0xba/0x130 do_syscall_64+0x35/0x80 entry_SYSCALL_64_after_hwframe+0x46/0xb0
Fixes: 2810c1e99867 ("kunit: Fix wild-memory-access bug in kunit_free_suite_set()") Reported-by: Marco Pagani marpagan@redhat.com Signed-off-by: Jinjie Ruan ruanjinjie@huawei.com
lib/kunit/test.c | 25 ++++++++++++++++--------- 1 file changed, 16 insertions(+), 9 deletions(-)
diff --git a/lib/kunit/test.c b/lib/kunit/test.c index 145f70219f46..8fac4783c676 100644 --- a/lib/kunit/test.c +++ b/lib/kunit/test.c @@ -739,7 +739,6 @@ static int kunit_module_init(struct module *mod) struct kunit_suite_set suite_set = { mod->kunit_suites, mod->kunit_suites + mod->num_kunit_suites, };
- const char *action = kunit_action(); int err = 0;
suite_set = kunit_filter_suites(&suite_set, @@ -752,16 +751,28 @@ static int kunit_module_init(struct module *mod) mod->kunit_suites = (struct kunit_suite **)suite_set.start; mod->num_kunit_suites = suite_set.end - suite_set.start;
- if (!action)
- return err;
+}
+static void kunit_module_exec(struct module *mod) +{
- struct kunit_suite_set suite_set = {
mod->kunit_suites, mod->kunit_suites + mod->num_kunit_suites,- };
- const char *action = kunit_action();
- if (!action) { kunit_exec_run_tests(&suite_set, false);
__kunit_test_suites_exit(mod->kunit_suites,mod->num_kunit_suites);- }
I don't think destroying debugfs right after running the tests is advisable.
The reason why I sent an RFC is to leave room for a discussion on which is the best way to solve the issue. I think it would be better to have a discussion before rushing patches.
Maybe there is a more appropriate fix method.
Thanks, Marco
else if (!strcmp(action, "list")) kunit_exec_list_tests(&suite_set, false); else if (!strcmp(action, "list_attr")) kunit_exec_list_tests(&suite_set, true); else pr_err("kunit: unknown action '%s'\n", action);
- return err;
} static void kunit_module_exit(struct module *mod) @@ -769,11 +780,6 @@ static void kunit_module_exit(struct module *mod) struct kunit_suite_set suite_set = { mod->kunit_suites, mod->kunit_suites + mod->num_kunit_suites, };
- const char *action = kunit_action();
- if (!action)
__kunit_test_suites_exit(mod->kunit_suites,mod->num_kunit_suites);kunit_free_suite_set(suite_set); } @@ -789,6 +795,7 @@ static int kunit_module_notify(struct notifier_block *nb, unsigned long val, ret = kunit_module_init(mod); break; case MODULE_STATE_LIVE:
kunit_module_exec(mod);