Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1756914Ab1F0Gz2 (ORCPT ); Mon, 27 Jun 2011 02:55:28 -0400 Received: from e7.ny.us.ibm.com ([32.97.182.137]:33523 "EHLO e7.ny.us.ibm.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1756195Ab1F0GyP (ORCPT ); Mon, 27 Jun 2011 02:54:15 -0400 Date: Mon, 27 Jun 2011 12:15:02 +0530 From: Srikar Dronamraju To: Peter Zijlstra Cc: Ingo Molnar , Steven Rostedt , Linux-mm , Arnaldo Carvalho de Melo , Linus Torvalds , Andi Kleen , Hugh Dickins , Christoph Hellwig , Jonathan Corbet , Thomas Gleixner , Masami Hiramatsu , Oleg Nesterov , LKML , Jim Keniston , Roland McGrath , Ananth N Mavinakayanahalli , Andrew Morton Subject: Re: [PATCH v4 3.0-rc2-tip 7/22] 7: uprobes: mmap and fork hooks. Message-ID: <20110627064502.GB24776@linux.vnet.ibm.com> Reply-To: Srikar Dronamraju References: <20110616130012.GL4952@linux.vnet.ibm.com> <1308248588.13240.267.camel@twins> <20110617045000.GM4952@linux.vnet.ibm.com> <1308297836.13240.380.camel@twins> <20110617090504.GN4952@linux.vnet.ibm.com> <1308303665.2355.11.camel@twins> <1308662243.26237.144.camel@twins> <20110622143906.GF16471@linux.vnet.ibm.com> <20110624020659.GA24776@linux.vnet.ibm.com> <1308901324.27849.7.camel@twins> MIME-Version: 1.0 Content-Type: text/plain; charset=iso-8859-1 Content-Disposition: inline In-Reply-To: <1308901324.27849.7.camel@twins> User-Agent: Mutt/1.5.20 (2009-06-14) Sender: linux-kernel-owner@vger.kernel.org List-ID: X-Mailing-List: linux-kernel@vger.kernel.org Content-Length: 3742 Lines: 159 > > mutex_lock(&mapping->i_mmap_mutex); > > delete_uprobe(uprobe); > > mutex_unlock(&mapping->i_mmap_mutex); > > > > inode->uprobes_count--; > > mutex_unlock(&inode->i_mutex); > > Right, so this lonesome unlock got me puzzled for a while, I always find > it best not to do asymmetric locking like this, keep the lock and unlock > in the same function. > Okay, will do. > > } > > > > int register_uprobe(...) > > { > > uprobe = alloc_uprobe(...); // find or insert in tree > > > > mutex_lock(&inode->i_mutex); // sync with register/unregister > > if (uprobe->consumers) { > > add_consumer(); > > goto put_unlock; > > } > > add_consumer(); > > inode->uprobes_count++; > > mutex_lock(&mapping->i_mmap_mutex); //sync with mmap. > > vma_prio_tree_foreach(..) { > > // get mm ref, add to list blah blah > > } > > > > mutex_unlock(&mapping->i_mmap_mutex); > > list_for_each_entry_safe() { > > if (ret) { > > // del from list etc.. > > // > > continue; > > } > > down_read(mm->mmap_sem); > > ret = install_breakpoint(); > > up_read(..); > > // del from list etc.. > > // > > if (ret && (ret == -ESRCH || ret == -EEXIST)) > > ret = 0; > > } > > > > if (ret) > > _unregister_uprobe(); > > > > put_unlock: > > mutex_unlock(&inode->i_mutex); > > You see, now this is a double unlock hmm . .will correct this. > > > put_uprobe(uprobe); > > return ret; > > } > > > > void unregister_uprobe(...) > > { > > mutex_lock(&inode->i_mutex); // sync with register/unregister > > uprobe = find_uprobe(); // ref++ > > _unregister_uprobe(); > > mutex_unlock(&inode->i_mutex); > > idem > > > put_uprobe(uprobe); > > } > > > > int mmap_uprobe(struct vm_area_struct *vma) > > { > > struct list_head tmp_list; > > struct uprobe *uprobe, *u; > > struct mm_struct *mm; > > struct inode *inode; > > int ret = 0; > > > > if (!valid_vma(vma)) > > return ret; /* Bail-out */ > > > > mm = vma->vm_mm; > > inode = vma->vm_file->f_mapping->host; > > if (inode->uprobes_count) > > return ret; > > __iget(inode); > > > > INIT_LIST_HEAD(&tmp_list); > > > > mutex_lock(&mapping->i_mmap_mutex); > > add_to_temp_list(vma, inode, &tmp_list); > > list_for_each_entry_safe(uprobe, u, &tmp_list, pending_list) { > > loff_t vaddr; > > > > list_del(&uprobe->pending_list); > > if (ret) > > continue; > > > > vaddr = vma->vm_start + uprobe->offset; > > vaddr -= vma->vm_pgoff << PAGE_SHIFT; > > ret = install_breakpoint(mm, uprobe, vaddr); > > Right, so this is the problem, you cannot do allocations under > i_mmap_mutex, however I think you can under i_mutex. I didnt know that we cannot do allocations under i_mmap_mutex. Why is this? I cant take i_mutex, because we would have already held down_write(mmap_sem) here. > > > if (ret && (ret == -ESRCH || ret == -EEXIST)) > > ret = 0; > > } > > > > mutex_unlock(&mapping->i_mmap_mutex); > > iput(inode); > > return ret; > > } > > > > int munmap_uprobe(struct vm_area_struct *vma) > > { > > struct list_head tmp_list; > > struct uprobe *uprobe, *u; > > struct mm_struct *mm; > > struct inode *inode; > > int ret = 0; > > > > if (!valid_vma(vma)) > > return ret; /* Bail-out */ > > > > mm = vma->vm_mm; > > inode = vma->vm_file->f_mapping->host; > > if (inode->uprobes_count) > > return ret; > > Should that be !->uprobes_count? Yes it should be !inode->uprobes_count. (both here and in mmap_uprobe) > > -- Thanks and Regards Srikar -- 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/