Re: [RFC PATCH] fs/lock: increase the filp's reference for Posix-style locks

[Jeff Layton <jlayton@kernel.org>]

On Mon, 2022-11-07 at 20:03 +0800, Xiubo Li wrote:
> On 07/11/2022 18:33, Jeff Layton wrote:
> > On Mon, 2022-11-07 at 17:52 +0800, xiubli@redhat.com wrote:
> > > From: Xiubo Li <xiubli@redhat.com>
> > > 
> > > When closing the file descripters in parallel in multiple threads,
> > > who are sharing the same file descripters, the filp_close() will
> > > remove all the Posix-style locks. But if two threads both calling
> > > the filp_close() it may race and cause use-after-free crash:
> > > 
> > >   PID: 327771   TASK: ffff952aa1db3180  CPU: 8    COMMAND: "db2fmp"
> > >    #0 [ffff95202f33b960] machine_kexec at ffffffff890662f4
> > >    #1 [ffff95202f33b9c0] __crash_kexec at ffffffff89122b82
> > >    #2 [ffff95202f33ba90] crash_kexec at ffffffff89122c70
> > >    #3 [ffff95202f33baa8] oops_end at ffffffff89791798
> > >    #4 [ffff95202f33bad0] no_context at ffffffff89075d14
> > >    #5 [ffff95202f33bb20] __bad_area_nosemaphore at ffffffff89075fe2
> > >    #6 [ffff95202f33bb70] bad_area_nosemaphore at ffffffff89076104
> > >    #7 [ffff95202f33bb80] __do_page_fault at ffffffff89794750
> > >    #8 [ffff95202f33bbf0] do_page_fault at ffffffff89794975
> > >    #9 [ffff95202f33bc20] page_fault at ffffffff89790778
> > >       [exception RIP: ceph_fl_release_lock+20]
> > >       RIP: ffffffffc08247a4  RSP: ffff95202f33bcd0  RFLAGS: 00010286
> > >       RAX: ffff952d4ebd8a00  RBX: 0000000000000000  RCX: dead000000000200
> > >       RDX: ffff95202f33bd60  RSI: ffff95202f33bd60  RDI: ffff9526b6ac5b00
> > >       RBP: ffff95202f33bce0   R8: ffff9526b6ac5b18   R9: ffffffffc083c368
> > >       R10: 0000000000001109  R11: 0000000000000000  R12: ffff95202f33bd60
> > >       R13: ffff9526b6ac5b00  R14: 0000000000000000  R15: 0000000000000000
> > >       ORIG_RAX: ffffffffffffffff  CS: 0010  SS: 0018
> > >   #10 [ffff95202f33bce8] locks_release_private at ffffffff892ab3d7
> > >   #11 [ffff95202f33bd00] locks_free_lock at ffffffff892ac34d
> > >   #12 [ffff95202f33bd18] locks_dispose_list at ffffffff892ac44b
> > >   #13 [ffff95202f33bd40] __posix_lock_file at ffffffff892acdfa
> > >   #14 [ffff95202f33bda8] posix_lock_file at ffffffff892ad146
> > >   #15 [ffff95202f33bdb8] ceph_lock at ffffffffc0824e8a [ceph]
> > >   #16 [ffff95202f33bdf8] vfs_lock_file at ffffffff892ad185
> > >   #17 [ffff95202f33be08] locks_remove_posix at ffffffff892ad239
> > >   #18 [ffff95202f33bee0] locks_remove_posix at ffffffff892ad2a0
> > >   #19 [ffff95202f33bef0] filp_close at ffffffff8924baa6
> > >   #20 [ffff95202f33bf18] __close_fd at ffffffff8926f89c
> > >   #21 [ffff95202f33bf40] sys_close at ffffffff8924d503
> > >   #22 [ffff95202f33bf50] system_call_fastpath at ffffffff89799f92
> > >       RIP: 00007f806ec446ab  RSP: 00007f80517f0d90  RFLAGS: 00010206
> > >       RAX: 0000000000000003  RBX: 00007f8030001a20  RCX: 00007f80300386b0
> > >       RDX: 00007f806ef0d880  RSI: 0000000000000001  RDI: 0000000000000006
> > >       RBP: 00007f806ef0e3c0   R8: 00007f80517fa700   R9: 0000000000000000
> > >       R10: 0000000000000000  R11: 0000000000000206  R12: 0000000000000000
> > >       R13: 00007f80300035b0  R14: 00007f80517f1104  R15: 000000000000006c
> > >       ORIG_RAX: 0000000000000003  CS: 0033  SS: 002b
> > > 
> > > We need to make sure that the filp in the file_lock shouldn't be
> > > release when any file_lock is still referring to it.
> > > 
> > > For the Posix-style locks, whose owner will be the thread ids, we
> > > will increase the filp's reference.
> > > 
> > > URL: https://tracker.ceph.com/issues/57986
> > > Signed-off-by: Xiubo Li <xiubli@redhat.com>
> > > ---
> > >   drivers/android/binder.c |  2 +-
> > >   fs/file.c                | 15 ++++++++++-----
> > >   fs/locks.c               | 18 +++++++++++++++---
> > >   include/linux/fs.h       | 14 ++++++++++++++
> > >   io_uring/openclose.c     |  3 ++-
> > >   5 files changed, 42 insertions(+), 10 deletions(-)
> > > 
> > > diff --git a/drivers/android/binder.c b/drivers/android/binder.c
> > > index 880224ec6abb..03692564d940 100644
> > > --- a/drivers/android/binder.c
> > > +++ b/drivers/android/binder.c
> > > @@ -1924,7 +1924,7 @@ static void binder_deferred_fd_close(int fd)
> > >   	if (twcb->file) {
> > >   		// pin it until binder_do_fd_close(); see comments there
> > >   		get_file(twcb->file);
> > > -		filp_close(twcb->file, current->files);
> > > +		filp_close(twcb->file, file_lock_make_thread_owner(current->files));
> > >   		task_work_add(current, &twcb->twork, TWA_RESUME);
> > >   	} else {
> > >   		kfree(twcb);
> > > diff --git a/fs/file.c b/fs/file.c
> > > index 5f9c802a5d8d..39ad8e74a8d9 100644
> > > --- a/fs/file.c
> > > +++ b/fs/file.c
> > > @@ -417,6 +417,7 @@ static struct fdtable *close_files(struct files_struct * files)
> > >   	 * files structure.
> > >   	 */
> > >   	struct fdtable *fdt = rcu_dereference_raw(files->fdt);
> > > +	fl_owner_t owner = file_lock_make_thread_owner(files);
> > >   	unsigned int i, j = 0;
> > >   
> > >   	for (;;) {
> > > @@ -429,7 +430,7 @@ static struct fdtable *close_files(struct files_struct * files)
> > >   			if (set & 1) {
> > >   				struct file * file = xchg(&fdt->fd[i], NULL);
> > >   				if (file) {
> > > -					filp_close(file, files);
> > > +					filp_close(file, owner);
> > >   					cond_resched();
> > >   				}
> > >   			}
> > > @@ -653,6 +654,7 @@ static struct file *pick_file(struct files_struct *files, unsigned fd)
> > >   int close_fd(unsigned fd)
> > >   {
> > >   	struct files_struct *files = current->files;
> > > +	fl_owner_t owner = file_lock_make_thread_owner(files);
> > >   	struct file *file;
> > >   
> > >   	spin_lock(&files->file_lock);
> > > @@ -661,7 +663,7 @@ int close_fd(unsigned fd)
> > >   	if (!file)
> > >   		return -EBADF;
> > >   
> > > -	return filp_close(file, files);
> > > +	return filp_close(file, owner);
> > >   }
> > >   EXPORT_SYMBOL(close_fd); /* for ksys_close() */
> > >   
> > > @@ -695,6 +697,7 @@ static inline void __range_cloexec(struct files_struct *cur_fds,
> > >   static inline void __range_close(struct files_struct *cur_fds, unsigned int fd,
> > >   				 unsigned int max_fd)
> > >   {
> > > +	fl_owner_t owner = file_lock_make_thread_owner(cur_fds);
> > >   	unsigned n;
> > >   
> > >   	rcu_read_lock();
> > > @@ -711,7 +714,7 @@ static inline void __range_close(struct files_struct *cur_fds, unsigned int fd,
> > >   
> > >   		if (file) {
> > >   			/* found a valid file to close */
> > > -			filp_close(file, cur_fds);
> > > +			filp_close(file, owner);
> > >   			cond_resched();
> > >   		}
> > >   	}
> > > @@ -816,6 +819,7 @@ struct file *close_fd_get_file(unsigned int fd)
> > >   
> > >   void do_close_on_exec(struct files_struct *files)
> > >   {
> > > +	fl_owner_t owner = file_lock_make_thread_owner(files);
> > >   	unsigned i;
> > >   	struct fdtable *fdt;
> > >   
> > > @@ -841,7 +845,7 @@ void do_close_on_exec(struct files_struct *files)
> > >   			rcu_assign_pointer(fdt->fd[fd], NULL);
> > >   			__put_unused_fd(files, fd);
> > >   			spin_unlock(&files->file_lock);
> > > -			filp_close(file, files);
> > > +			filp_close(file, owner);
> > >   			cond_resched();
> > >   			spin_lock(&files->file_lock);
> > >   		}
> > > @@ -1080,6 +1084,7 @@ static int do_dup2(struct files_struct *files,
> > >   	struct file *file, unsigned fd, unsigned flags)
> > >   __releases(&files->file_lock)
> > >   {
> > > +	fl_owner_t owner = file_lock_make_thread_owner(files);
> > >   	struct file *tofree;
> > >   	struct fdtable *fdt;
> > >   
> > > @@ -1111,7 +1116,7 @@ __releases(&files->file_lock)
> > >   	spin_unlock(&files->file_lock);
> > >   
> > >   	if (tofree)
> > > -		filp_close(tofree, files);
> > > +		filp_close(tofree, owner);
> > >   
> > >   	return fd;
> > >   
> > > diff --git a/fs/locks.c b/fs/locks.c
> > > index 607f94a0e789..e8b67f87e0ee 100644
> > > --- a/fs/locks.c
> > > +++ b/fs/locks.c
> > > @@ -331,6 +331,8 @@ EXPORT_SYMBOL_GPL(locks_owner_has_blockers);
> > >   /* Free a lock which is not in use. */
> > >   void locks_free_lock(struct file_lock *fl)
> > >   {
> > > +	if (fl->fl_file && file_lock_is_thread_owner(fl->fl_owner))
> > > +		fput(fl->fl_file);
> > >   	locks_release_private(fl);
> > >   	kmem_cache_free(filelock_cache, fl);
> > >   }
> > > @@ -384,7 +386,10 @@ void locks_copy_lock(struct file_lock *new, struct file_lock *fl)
> > >   
> > >   	locks_copy_conflock(new, fl);
> > >   
> > > -	new->fl_file = fl->fl_file;
> > > +	if (file_lock_is_thread_owner(new->fl_owner))
> > > +		new->fl_file = get_file(fl->fl_file);
> > > +	else
> > > +		new->fl_file = fl->fl_file;
> > >   	new->fl_ops = fl->fl_ops;
> > >   
> > >   	if (fl->fl_ops) {
> > > @@ -488,13 +493,14 @@ static int flock64_to_posix_lock(struct file *filp, struct file_lock *fl,
> > >   	} else
> > >   		fl->fl_end = OFFSET_MAX;
> > >   
> > > -	fl->fl_owner = current->files;
> > > +	fl->fl_owner = file_lock_make_thread_owner(current->files);
> > >   	fl->fl_pid = current->tgid;
> > > -	fl->fl_file = filp;
> > > +	fl->fl_file = get_file(filp);
> > >   	fl->fl_flags = FL_POSIX;
> > >   	fl->fl_ops = NULL;
> > >   	fl->fl_lmops = NULL;
> > >   
> > > +
> > >   	return assign_type(fl, l->l_type);
> > >   }
> > >   
> > > @@ -2243,6 +2249,7 @@ int fcntl_getlk(struct file *filp, unsigned int cmd, struct flock *flock)
> > >   
> > >   		fl->fl_flags |= FL_OFDLCK;
> > >   		fl->fl_owner = filp;
> > > +		fput(filp);
> > >   	}
> > >   
> > >   	error = vfs_test_lock(filp, fl);
> > > @@ -2376,6 +2383,7 @@ int fcntl_setlk(unsigned int fd, struct file *filp, unsigned int cmd,
> > >   		cmd = F_SETLK;
> > >   		file_lock->fl_flags |= FL_OFDLCK;
> > >   		file_lock->fl_owner = filp;
> > > +		fput(filp);
> > >   		break;
> > >   	case F_OFD_SETLKW:
> > >   		error = -EINVAL;
> > > @@ -2385,6 +2393,7 @@ int fcntl_setlk(unsigned int fd, struct file *filp, unsigned int cmd,
> > >   		cmd = F_SETLKW;
> > >   		file_lock->fl_flags |= FL_OFDLCK;
> > >   		file_lock->fl_owner = filp;
> > > +		fput(filp);
> > >   		fallthrough;
> > >   	case F_SETLKW:
> > >   		file_lock->fl_flags |= FL_SLEEP;
> > > @@ -2450,6 +2459,7 @@ int fcntl_getlk64(struct file *filp, unsigned int cmd, struct flock64 *flock)
> > >   		cmd = F_GETLK64;
> > >   		fl->fl_flags |= FL_OFDLCK;
> > >   		fl->fl_owner = filp;
> > > +		fput(filp);
> > >   	}
> > >   
> > >   	error = vfs_test_lock(filp, fl);
> > > @@ -2499,6 +2509,7 @@ int fcntl_setlk64(unsigned int fd, struct file *filp, unsigned int cmd,
> > >   		cmd = F_SETLK64;
> > >   		file_lock->fl_flags |= FL_OFDLCK;
> > >   		file_lock->fl_owner = filp;
> > > +		fput(filp);
> > >   		break;
> > >   	case F_OFD_SETLKW:
> > >   		error = -EINVAL;
> > > @@ -2508,6 +2519,7 @@ int fcntl_setlk64(unsigned int fd, struct file *filp, unsigned int cmd,
> > >   		cmd = F_SETLKW64;
> > >   		file_lock->fl_flags |= FL_OFDLCK;
> > >   		file_lock->fl_owner = filp;
> > > +		fput(filp);
> > >   		fallthrough;
> > >   	case F_SETLKW64:
> > >   		file_lock->fl_flags |= FL_SLEEP;
> > > diff --git a/include/linux/fs.h b/include/linux/fs.h
> > > index e654435f1651..d7d81962a863 100644
> > > --- a/include/linux/fs.h
> > > +++ b/include/linux/fs.h
> > > @@ -1028,6 +1028,20 @@ static inline struct file *get_file(struct file *f)
> > >   /* legacy typedef, should eventually be removed */
> > >   typedef void *fl_owner_t;
> > >   
> > > +/*
> > > + * Set the last significant bit to 1 to mark that
> > > + * we have get a reference of the fl->fl_file.
> > > + */
> > > +static inline fl_owner_t file_lock_make_thread_owner(fl_owner_t owner)
> > > +{
> > > +	return (fl_owner_t)((unsigned long)owner | 1UL);
> > > +}
> > > +
> > > +static inline bool file_lock_is_thread_owner(fl_owner_t owner)
> > > +{
> > > +	return ((unsigned long)owner & 1UL);
> > > +}
> > > +
> > >   struct file_lock;
> > >   
> > >   struct file_lock_operations {
> > > diff --git a/io_uring/openclose.c b/io_uring/openclose.c
> > > index 67178e4bb282..5a12cdf7f8d0 100644
> > > --- a/io_uring/openclose.c
> > > +++ b/io_uring/openclose.c
> > > @@ -212,6 +212,7 @@ int io_close_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
> > >   int io_close(struct io_kiocb *req, unsigned int issue_flags)
> > >   {
> > >   	struct files_struct *files = current->files;
> > > +	fl_owner_t owner = file_lock_make_thread_owner(files);
> > >   	struct io_close *close = io_kiocb_to_cmd(req, struct io_close);
> > >   	struct fdtable *fdt;
> > >   	struct file *file;
> > > @@ -247,7 +248,7 @@ int io_close(struct io_kiocb *req, unsigned int issue_flags)
> > >   		goto err;
> > >   
> > >   	/* No ->flush() or already async, safely close from here */
> > > -	ret = filp_close(file, current->files);
> > > +	ret = filp_close(file, owner);
> > >   err:
> > >   	if (ret < 0)
> > >   		req_set_fail(req);
> > I think this is the wrong approach to fixing this. It also looks like
> > you could hit a similar problem with OFD locks and this patch wouldn't
> > address that issue.
> 
> For the OFD locks they will set the 'file' struct as the owner just as 
> the flock does, it should be okay and I don't think it has this issue if 
> my understanding is correct here.
> 

They set the the owner to "file", but they don't hold a reference to it.
With OFD locks, the file is what holds references to the lock, not the
reverse.

> > The real bug seems to be that ceph_fl_release_lock dereferences fl_file,
> > at a point when it shouldn't rely on that being valid. Most filesystems
> > stash some info in fl->fl_u if they need to do bookkeeping after
> > releasing a lock. Perhaps ceph should be doing something similar?
> 
> This is the 'filp' memory in filp_close(filp, ...):
> 
> crash> file.f_path.dentry,f_inode 0xffff952d7ab46200
>    f_path.dentry = 0xffff9521b121cb40
>    f_inode = 0xffff951f3ea33550,
> 
> We can see the 'f_inode' is pointing to the correct inode memory.
> 
> 
> 
> While later in 'ceph_fl_release_lock()':
> 
> 41 static void ceph_fl_release_lock(struct file_lock *fl)
> 42 {
> 43     struct ceph_file_info *fi = fl->fl_file->private_data;
> 44     struct inode *inode = file_inode(fl->fl_file);
> 45     struct ceph_inode_info *ci = ceph_inode(inode);
> 46     atomic_dec(&fi->num_locks);
> 47     if (atomic_dec_and_test(&ci->i_filelock_ref)) {
> 48         /* clear error when all locks are released */
> 49         spin_lock(&ci->i_ceph_lock);
> 50         ci->i_ceph_flags &= ~CEPH_I_ERROR_FILELOCK;
> 51         spin_unlock(&ci->i_ceph_lock);
> 52     }
> 53 }
> 

You only need the inode for most of this. The exception is
fi->num_locks, so you may need to test for that in a different way.

> It crashed in Line#47 and the 'fl->fl_file' memory is:
> 
> crash> file.f_path.dentry,f_inode 0xffff952d4ebd8a00
>    f_path.dentry = 0x0
>    f_inode = 0x0,
> 
> Please NOTE: the 'filp' and 'fl->fl_file' are two different 'file struct'.
> 

Yep, I understand the bug. I just don't like the proposed fix. :)

> Can we fix this by using 'fl->fl_u' here ?
> 

Probably. You could take and hold an inode reference in there, and maybe
add a function that looks at whether there are any locks held against a
particular file, rather than trying to count locks in ceph_file_info.

> I was also thinking I could just call the 'get_file(file)' in 
> ceph_lock() and then in ceph_fl_release_lock() release the reference 
> counter. How about this ?
> 

That may work too, though again, I'd be worried about cyclical
dependencies, particularly with OFD locks. If the lock holds a reference
to the file, then can the file's refcount ever go to zero if the lock is
never explicitly released? I think not.

You may also need to consider flock locks too, since they have similar
ownership semantics to OFD locks.
-- 
Jeff Layton <jlayton@kernel.org>