Am Freitag, 15. November 2019, 14:27:40 CET schrieb Al Viro: > On Fri, Nov 15, 2019 at 02:01:55PM +0100, Dietmar Hahn wrote: > > > Later a user tool dumped with SIGSEGV and the linux system crashed. > > I investigated the crash dump and found the cause. > > > > Via format_corename() in fs/coredump.c the helper_argv[] with 3 entries is > > created and helper_argv[0] == "" (because of the ' ' after the '|') > > ispipe is set to 1. > > Later in call_usermodehelper_setup(): > > sub_info->path = path; == helper_argv[0] == "" > > This leads in call_usermodehelper_exec() to: > > if (strlen(sub_info->path) == 0) > > goto out; > > with a return value of 0. > > But no pipe is created and thus cprm.file == NULL. > > This leads in file_start_write() to the panic because of dereferencing > > file_inode(file)->i_mode) > > > > I'am not sure what's the best way to fix this so I've no patch. > > Thanks. > > Check in the caller of format_corename() for **argv being '\0' and fail > if it is? I mean, turn that > if (ispipe < 0) { > printk(KERN_WARNING "format_corename failed\n"); > printk(KERN_WARNING "Aborting core\n"); > goto fail_unlock; > } > in there into > if (ispipe < 0 || !**argv) { > printk(KERN_WARNING "format_corename failed\n"); > printk(KERN_WARNING "Aborting core\n"); > goto fail_unlock; > } Unfortunately this doesn't work because argv[0] is always 0 in case of ispipe in format_corename(): if (ispipe) { int argvs = sizeof(core_pattern) / 2; (*argv) = kmalloc_array(argvs, sizeof(**argv), GFP_KERNEL); if (!(*argv)) return -ENOMEM; (*argv)[(*argc)++] = 0; ++pat_ptr; } The manpage says: The program must be ..., and must immediately follow the '|' character. Why not check this in format_corename(), maybe: @@ -211,6 +211,8 @@ static int format_corename(struct core_name *cn, struct coredump_params *cprm, return -ENOMEM; (*argv)[(*argc)++] = 0; ++pat_ptr; + if (isspace(*pat_ptr)) + return -EINVAL; } Dietmar.