Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1755454AbZCVSf1 (ORCPT ); Sun, 22 Mar 2009 14:35:27 -0400 Received: (majordomo@vger.kernel.org) by vger.kernel.org id S1751673AbZCVSfO (ORCPT ); Sun, 22 Mar 2009 14:35:14 -0400 Received: from extu-mxob-2.symantec.com ([216.10.194.135]:55326 "EHLO extu-mxob-2.symantec.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1751201AbZCVSfM (ORCPT ); Sun, 22 Mar 2009 14:35:12 -0400 Date: Sun, 22 Mar 2009 18:33:04 +0000 (GMT) From: Hugh Dickins X-X-Sender: hugh@blonde.anvils To: "Eric W. Biederman" cc: Benjamin Herrenschmidt , Andrew Morton , Jesse Barnes , Tejun Heo , Kay Sievers , Greg Kroah-Hartman , Nick Piggin , linux-kernel@vger.kernel.org Subject: [PATCH next] sysfs: fix some bin_vm_ops errors Message-ID: MIME-Version: 1.0 Content-Type: TEXT/PLAIN; charset=US-ASCII Sender: linux-kernel-owner@vger.kernel.org List-ID: X-Mailing-List: linux-kernel@vger.kernel.org Content-Length: 3177 Lines: 91 Commit 86c9508eb1c0ce5aa07b5cf1d36b60c54efc3d7a "sysfs: don't block indefinitely for unmapped files" in linux-next crashes the PowerMac G5 when X starts up. It's caught out by the way powerpc's pci_mmap of legacy_mem uses shmem_zero_setup(), substituting a new vma->vm_file whose private_data no longer points to the bin_buffer (substitution done because some versions of X crash if that mmap fails). The fix to this is straightforward: the original vm_file is fput() in that case, so this mmap won't block sysfs at all, so just don't switch over to bin_vm_ops if vm_file has changed. But more fixes made before realizing that was the problem:- It should not be an error if bin_page_mkwrite() finds no underlying page_mkwrite(). Check that a file already mmap'ed has the same underlying vm_ops _before_ pointing vma->vm_ops at bin_vm_ops. If the file being mmap'ed is a shmem/tmpfs file, don't fail the mmap on CONFIG_NUMA=y, just because that has a set_policy and get_policy. They probably won't be called, and it's not a disaster if they don't work on sysfs. vm_ops->migrate? I couldn't find any example. Just delete CONFIG_NUMA tests, easy enough to add stubs later if required. Signed-off-by: Hugh Dickins --- I have only tested that X now starts up fine on the G5: I don't think I've messed up the intent of your patch, but not tested it still works. fs/sysfs/bin.c | 18 ++++++++---------- 1 file changed, 8 insertions(+), 10 deletions(-) --- 2.6.29-rc8-next/fs/sysfs/bin.c 2009-03-20 12:41:45.000000000 +0000 +++ linux/fs/sysfs/bin.c 2009-03-21 18:01:26.000000000 +0000 @@ -241,9 +241,12 @@ static int bin_page_mkwrite(struct vm_ar struct sysfs_dirent *attr_sd = file->f_path.dentry->d_fsdata; int ret; - if (!bb->vm_ops || !bb->vm_ops->page_mkwrite) + if (!bb->vm_ops) return -EINVAL; + if (!bb->vm_ops->page_mkwrite) + return 0; + if (!sysfs_get_active_two(attr_sd)) return -EINVAL; @@ -287,7 +290,6 @@ static int mmap(struct file *file, struc struct sysfs_dirent *attr_sd = file->f_path.dentry->d_fsdata; struct bin_attribute *attr = attr_sd->s_bin_attr.bin_attr; struct kobject *kobj = attr_sd->s_parent->s_dir.kobj; - struct vm_operations_struct *vm_ops; int rc; mutex_lock(&bb->mutex); @@ -302,24 +304,20 @@ static int mmap(struct file *file, struc goto out_put; rc = attr->mmap(kobj, attr, vma); - vm_ops = vma->vm_ops; - vma->vm_ops = &bin_vm_ops; if (rc) goto out_put; - rc = -EINVAL; - if (bb->mmapped && bb->vm_ops != vma->vm_ops) + if (vma->vm_file != file) goto out_put; -#ifdef CONFIG_NUMA rc = -EINVAL; - if (vm_ops && ((vm_ops->set_policy || vm_ops->get_policy || vm_ops->migrate))) + if (bb->mmapped && bb->vm_ops != vma->vm_ops) goto out_put; -#endif rc = 0; bb->mmapped = 1; - bb->vm_ops = vm_ops; + bb->vm_ops = vma->vm_ops; + vma->vm_ops = &bin_vm_ops; out_put: sysfs_put_active_two(attr_sd); out_unlock: -- 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/