lists.openwall.net   lists  /  announce  owl-users  owl-dev  john-users  john-dev  passwdqc-users  yescrypt  popa3d-users  /  oss-security  kernel-hardening  musl  sabotage  tlsify  passwords  /  crypt-dev  xvendor  /  Bugtraq  Full-Disclosure  linux-kernel  linux-netdev  linux-ext4  linux-hardening  linux-cve-announce  PHC 
Open Source and information security mailing list archives
 
Hash Suite: Windows password security audit tool. GUI, reports in PDF.
[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Date:	Mon, 18 Jul 2011 14:50:55 +0530
From:	Srikar Dronamraju <srikar@...ux.vnet.ibm.com>
To:	Peter Zijlstra <peterz@...radead.org>
Cc:	Ingo Molnar <mingo@...e.hu>, Steven Rostedt <rostedt@...dmis.org>,
	Linux-mm <linux-mm@...ck.org>,
	Arnaldo Carvalho de Melo <acme@...radead.org>,
	Linus Torvalds <torvalds@...ux-foundation.org>,
	Andi Kleen <andi@...stfloor.org>,
	Hugh Dickins <hughd@...gle.com>,
	Christoph Hellwig <hch@...radead.org>,
	Jonathan Corbet <corbet@....net>,
	Thomas Gleixner <tglx@...utronix.de>,
	Masami Hiramatsu <masami.hiramatsu.pt@...achi.com>,
	Oleg Nesterov <oleg@...hat.com>,
	LKML <linux-kernel@...r.kernel.org>,
	Jim Keniston <jkenisto@...ux.vnet.ibm.com>,
	Roland McGrath <roland@...k.frob.com>,
	Ananth N Mavinakayanahalli <ananth@...ibm.com>,
	Andrew Morton <akpm@...ux-foundation.org>
Subject: Re: [PATCH v4 3.0-rc2-tip 7/22]  7: uprobes: mmap and fork hooks.

Here is another take at locking for uprobes.

/*
 *  Register/Unregister:
 *  	- walks thro vma-rmap creates a temp list of vmas.
 *	- Cannot sleep while walking thro vma-rmap; hence temp list.
 * 	- Iterate thro the temp list and insert/delete breakpoints.
 *	- Insertion/deletion of breakpoint needs mmap_sem to be held and
 *	  dropped in the loop.
 *
 * mmap/munmap:
 *	- called with down_write(mmap_sem)
 * 	- walk thro the uprobe rbtree and create a temp list of uprobes.
 *	- Cannot sleep while walking thro rbtree; hence temp list.
 * 	- Iterate thro the temp list and insert/delete breakpoints.
 *
 * Issues:
 *	- Lock ordering issues since mmap/munmap are called with mmap_sem
 *	  held for write unlike register/unregister where they need to
 *	  held just before insertion of breakpoints.
 * 	- Cannot allocate when i_mmap_mutex is held.
 *	- Vma objects can vanish after munmap.
 *	- Multiple vmas mapping to same text area of an inode in the same mm
 *	  struct.
 *
 * Changes:
 *
 *  - Introduce uprobes_list and uprobes_vaddr in vm_area_struct.
 *    uprobes_list is a node in the temp list of vmas while
 *    registering/unregistering uprobes. uprobes_vaddr caches the vaddr to
 *    insert/remove the breakpoint.
 *
 *  - Introduce srcu to synchronize vma deletion with walking the list of
 *    vma in register/unregister_uprobe.
 *
 *  - Introduce uprobes_mmap_mutex to synchronize uprobe deletion and
 *    mmap_uprobe().
 *
 * Locking:
 *   hierarcy:
 *    --> inode->i_mutex
 *    		--> mm->mmap_sem
 *    			--> mapping->i_mmap_mutex
 *    			--> uprobes_mmap_mutex
 *    				--> treelock
 *
 *   i_mutex: serializes register/unregister on the same inode.
 *
 *   i_mmap_mutex: allows to walk the vma-rmap but cannot be held across
 *  		kzalloc.
 *
 *   mmap_sem : insert an instruction into the address-space.
 *
 *   uprobes_mmap_mutex : serializes multiple mmap_uprobes since the
 *			  same uprobe cannot be in more than one list.
 *   			: serializes walking the uprobe list with deleting
 *   			  the uprobe. (this can be achieved by taking a
 *   			  reference to uprobe while we walk).
 *
 *   treelock (spinlock) : modification/walking thro the rb tree.
 *
 *   srcu : synchronize walking the vma list with munmap.
 *
 *  Assumptions:
 *  Can sleep including sleeping for page allocations/copying original
 *  instructions while holding the srcu.
 *
 * Advantages:
 * 1. No need to drop mmap_sem.
 * 2. Now register/unregister can run in parallel except unless when uprobe
 *    gets deleted.
 *
 * Disadvantages:
 * 1. All calls to mmap_uprobe are serialized.
 * 2. Locking gets complicated with 3 mutexs, mmap_sem, treelock and srcu.
 *
 * Note:
 *      Can avoid uprobes_mmap_mutex/srcu/serializing the mmap_uprobes by
 * 	try_mutex_lock(i_mutex) followed by dropping mmap_sem; acquiring
 * 	i_mutex and mmap_sem in order and verifying the vma in both
 * 	mmap_uprobe and munmap_uprobe but Peter Zijlstra not happy with that
 * 	approach.
 */

void _unregister_uprobe(...)
{
	if (!del_consumer(...)) {	// includes tree removal on last consumer
		return;
	}
	if (uprobe->consumers)
		return;		// Not the last consumer.

	INIT_LIST_HEAD(&tmp_list);

	mutex_lock(&mapping->i_mmap_mutex);	//sync with mmap.
	vma_prio_tree_foreach() {
		list_add(&vma->uprobes_list, &tmplist);
		// set vma->uprobes_vaddr;
	}

	mutex_unlock(&mapping->i_mmap_mutex);

	srcu_read_lock(&uprobes_srcu);	// synch with munmap_uprobe.
	list_for_each_entry_safe() {
		mm = vma->vm_mm;
		vaddr = vma->uprobes_vaddr;
		// remove from list
		down_read(&mm->mmap_sem);
		remove_breakpoint();	// unconditional, if it wasn't there
		up_read(&mm->mmap_sem);
	}
	srcu_read_unlock(&uprobes_srcu, ..);

	inode->uprobes_count--;

	mutex_lock(&uprobe_mmap_mutex);	// sync with mmap_uprobe.
	delete_uprobe(uprobe);
	mutex_unlock(&uprobe_mmap_mutex);
}

int register_uprobe(...)
{
	uprobe = alloc_uprobe(...);	// find or insert in tree

	INIT_LIST_HEAD(&tmp_list);

	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_uprobe
	vma_prio_tree_foreach(..) {
		list_add(&vma->uprobes_list, &tmplist);
		// get mm ref, add to list blah blah
		// set vma->uprobes_vaddr;
	}

	mutex_unlock(&mapping->i_mmap_mutex);

	srcu_read_lock(&uprobes_srcu);	// synch with munmap_uprobe.
	list_for_each_entry_safe() {
		mm = vma->vm_mm;
		vaddr = vma->uprobes_vaddr;
		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;
	}
	srcu_read_unlock(&uprobes_srcu, ..);

	if (ret)
		_unregister_uprobe();

      put_unlock:
	mutex_unlock(&inode->i_mutex);
	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);
	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(&uprobes_mmap_mutex);
	add_to_temp_list(vma, inode, &tmp_list);
	list_for_each_entry_safe(uprobe, u, &tmp_list, pending_list) {
		loff_t vaddr;
		if (!uprobe->consumer)
			continue;

		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);

		if (ret && (ret == -ESRCH || ret == -EEXIST))
			ret = 0;
	}

	mutex_unlock(&uprobes_mmap_mutex);
	iput(inode);
	return ret;
}

void munmap_uprobe(struct vm_area_struct *vma)
{
	struct list_head tmp_list;
	struct uprobe *uprobe, *u;
	struct mm_struct *mm;
	struct inode *inode;

	if (!valid_vma(vma))
		return;	/* Bail-out */

	mm = vma->vm_mm;
	inode = vma->vm_file->f_mapping->host;
	if (!inode->uprobes_count)
		return;

	list_del(&vma->uprobes_list);
	synchronize_srcu(..);	// synchronize with (un)register_uprobe

	if (!mm->uprobes_count)
		return;

//      walk thro RB tree and decrement mm->uprobes_count
	walk_rbtree_and_dec_uprobes_count();	//hold treelock.

	return;
}
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@...r.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/

Powered by blists - more mailing lists

Powered by Openwall GNU/*/Linux Powered by OpenVZ