Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1754049AbYGUQS0 (ORCPT ); Mon, 21 Jul 2008 12:18:26 -0400 Received: (majordomo@vger.kernel.org) by vger.kernel.org id S1753818AbYGUQSM (ORCPT ); Mon, 21 Jul 2008 12:18:12 -0400 Received: from x346.tv-sign.ru ([89.108.83.215]:49664 "EHLO mail.screens.ru" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1751787AbYGUQSK (ORCPT ); Mon, 21 Jul 2008 12:18:10 -0400 Date: Mon, 21 Jul 2008 20:21:39 +0400 From: Oleg Nesterov To: Roland McGrath Cc: Andrew Morton , David Howells , linux-kernel@vger.kernel.org Subject: killable/interruptible coredumps Message-ID: <20080721162139.GA340@tv-sign.ru> References: <20080716135359.GA1219@tv-sign.ru> <20080720023258.26D2015421D@magilla.localdomain> <20080720122354.GD143@tv-sign.ru> <20080721000842.94DAE15421D@magilla.localdomain> Mime-Version: 1.0 Content-Type: text/plain; charset=us-ascii Content-Disposition: inline In-Reply-To: <20080721000842.94DAE15421D@magilla.localdomain> User-Agent: Mutt/1.5.11 Sender: linux-kernel-owner@vger.kernel.org List-ID: X-Mailing-List: linux-kernel@vger.kernel.org Content-Length: 5422 Lines: 168 On 07/20, Roland McGrath wrote: > > Then have the dumping and the waiting > for it be killable. I think it is easy to make the coredumping killable right now, --- fs/exec.c~ 2008-07-21 19:47:22.000000000 +0400 +++ fs/exec.c 2008-07-21 19:56:44.000000000 +0400 @@ -1523,6 +1523,7 @@ static inline int zap_threads(struct tas spin_lock_irq(&tsk->sighand->siglock); if (!signal_group_exit(tsk->signal)) { mm->core_state = core_state; + clear_thread_flag(TIF_SIGPENDING); tsk->signal->group_exit_code = exit_code; nr = zap_process(tsk); } @@ -1735,12 +1736,6 @@ int do_coredump(long signr, int exit_cod goto fail; /* - * Clear any false indication of pending signals that might - * be seen by the filesystem code called to write the core file. - */ - clear_thread_flag(TIF_SIGPENDING); - - /* * lock_kernel() because format_corename() is controlled by sysctl, which * uses lock_kernel() */ --- fs/binfmt_elf.c~ 2008-07-13 20:52:25.000000000 +0400 +++ fs/binfmt_elf.c 2008-07-20 19:53:08.000000000 +0400 @@ -1105,11 +1105,17 @@ out: */ static int dump_write(struct file *file, const void *addr, int nr) { + if (fatal_signal_pending(current)) + return 0; + return file->f_op->write(file, addr, nr, &file->f_pos) == nr; } static int dump_seek(struct file *file, loff_t off) { + if (fatal_signal_pending(current)) + return 0; + if (file->f_op->llseek && file->f_op->llseek != no_llseek) { if (file->f_op->llseek(file, off, SEEK_CUR) < 0) return 0; This relies on the fact SIGKILL will find the coredumping thread because it is not PF_EXITING, even if SIGNAL_GROUP_EXIT is set. This is a bit racy though, SIGKILL may come after zap_process(current) but before other threads have "died". I wonder if it makes sense to go futher. Currently the coredumping thread sets SIGNAL_GROUP_EXIT. We can set signal->group_exit_task instead, this way it can be killed by any fatal signal, not only by SIGKILL. For example the user can just use ^C (unless it is ignored/etc). What is your opinion about the patch below? I can't decide wether this change is good or bad. It complicates the code, but otoh it makes things more consistent, imho. Note also we can change oom_kill to check SIGNAL_GROUP_EXIT instead of PF_EXITING to detect the process which should release its ->mm and probably free the memory "soon". It is not clear what should be ->group_exit_code if ->core_dump() is interrupted. This patch sets it = exit_code, but perhaps we should keep the original signr in that case. Oleg. --- 26-rc2/fs/exec.c~1_MAKE_KILLABLE 2008-07-20 17:31:40.000000000 +0400 +++ 26-rc2/fs/exec.c 2008-07-20 19:46:21.000000000 +0400 @@ -1498,7 +1498,6 @@ static int zap_process(struct task_struc struct task_struct *t; int nr = 0; - start->signal->flags = SIGNAL_GROUP_EXIT; start->signal->group_stop_count = 0; t = start; @@ -1523,7 +1522,8 @@ static inline int zap_threads(struct tas spin_lock_irq(&tsk->sighand->siglock); if (!signal_group_exit(tsk->signal)) { mm->core_state = core_state; - tsk->signal->group_exit_code = exit_code; + clear_thread_flag(TIF_SIGPENDING); + tsk->signal->group_exit_task = tsk; nr = zap_process(tsk); } spin_unlock_irq(&tsk->sighand->siglock); @@ -1574,6 +1574,7 @@ static inline int zap_threads(struct tas if (p->mm) { if (unlikely(p->mm == mm)) { lock_task_sighand(p, &flags); + p->signal->flags = SIGNAL_GROUP_EXIT; nr += zap_process(p); unlock_task_sighand(p, &flags); } @@ -1619,7 +1620,7 @@ fail: return core_waiters; } -static void coredump_finish(struct mm_struct *mm) +static inline void core_state_finish(struct mm_struct *mm) { struct core_thread *curr, *next; struct task_struct *task; @@ -1640,6 +1641,17 @@ static void coredump_finish(struct mm_st mm->core_state = NULL; } +static void coredump_finish(struct mm_struct *mm, int exit_code) +{ + spin_lock_irq(¤t->sighand->siglock); + current->signal->flags = SIGNAL_GROUP_EXIT; + current->signal->group_exit_code = exit_code; + current->signal->group_exit_task = NULL; + spin_unlock_irq(¤t->sighand->siglock); + + core_state_finish(mm); +} + /* * set_dumpable converts traditional three-value dumpable to two flags and * stores them into mm->flags. It modifies lower two bits of mm->flags, but @@ -1735,12 +1747,6 @@ int do_coredump(long signr, int exit_cod goto fail; /* - * Clear any false indication of pending signals that might - * be seen by the filesystem code called to write the core file. - */ - clear_thread_flag(TIF_SIGPENDING); - - /* * lock_kernel() because format_corename() is controlled by sysctl, which * uses lock_kernel() */ @@ -1814,9 +1820,8 @@ int do_coredump(long signr, int exit_cod goto close_fail; retval = binfmt->core_dump(signr, regs, file, core_limit); - if (retval) - current->signal->group_exit_code |= 0x80; + exit_code |= 0x80; close_fail: filp_close(file, NULL); fail_unlock: @@ -1824,7 +1829,7 @@ fail_unlock: argv_free(helper_argv); current->fsuid = fsuid; - coredump_finish(mm); + coredump_finish(mm, exit_code); fail: return retval; } -- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/