Decouple Linux files from the belonging character device right after open

in the LinuxKPI. This is done by calling finit() just before returning a magic
value of ENXIO in the "linux_dev_fdopen" function.

The Linux file structure should mimic the BSD file structure as much as
possible. This patch decouples the Linux file structure from the belonging
character device right after the "linux_dev_fdopen" function has returned.
This fixes an issue which allows a Linux file handle to exist after a
character device has been destroyed and removed from the directory index
of /dev. Only when the reference count of the BSD file handle reaches zero,
the Linux file handle is destroyed. This fixes use-after-free issues related
to accessing the Linux file structure after the character device has been
destroyed.

While at it add a missing NULL check for non-present file operation.
Calling a NULL pointer will result in a segmentation fault.

Reviewed by:	kib @
MFC after:	1 week
Sponsored by:	Mellanox Technologies
This commit is contained in:
Hans Petter Selasky 2018-01-26 10:49:02 +00:00
parent 919736d252
commit e23ae408c0

View File

@ -45,6 +45,7 @@ __FBSDID("$FreeBSD$");
#include <sys/file.h>
#include <sys/filio.h>
#include <sys/rwlock.h>
#include <sys/mman.h>
#include <vm/vm.h>
#include <vm/pmap.h>
@ -390,18 +391,6 @@ kobject_init_and_add(struct kobject *kobj, const struct kobj_type *ktype,
return kobject_add_complete(kobj, parent);
}
static void
linux_file_dtor(void *cdp)
{
struct linux_file *filp;
linux_set_current(curthread);
filp = cdp;
filp->f_op->release(filp->f_vnode, filp);
vdrop(filp->f_vnode);
kfree(filp);
}
static void
linux_kq_lock(void *arg)
{
@ -438,7 +427,7 @@ linux_kq_lock_unowned(void *arg)
}
static void
linux_dev_kqfilter_poll(struct linux_file *, int);
linux_file_kqfilter_poll(struct linux_file *, int);
struct linux_file *
linux_file_alloc(void)
@ -678,24 +667,31 @@ static struct cdev_pager_ops linux_cdev_pager_ops[2] = {
},
};
#define OPW(fp,td,code) ({ \
struct file *__fpop; \
int __retval; \
\
__fpop = (td)->td_fpop; \
(td)->td_fpop = (fp); \
__retval = (code); \
(td)->td_fpop = __fpop; \
__retval; \
})
static int
linux_dev_open(struct cdev *dev, int oflags, int devtype, struct thread *td)
linux_dev_fdopen(struct cdev *dev, int fflags, struct thread *td, struct file *file)
{
struct linux_cdev *ldev;
struct linux_file *filp;
struct file *file;
int error;
file = td->td_fpop;
ldev = dev->si_drv1;
if (ldev == NULL)
return (ENODEV);
filp = linux_file_alloc();
filp->f_dentry = &filp->f_dentry_store;
filp->f_op = ldev->ops;
filp->f_mode = file->f_flag;
filp->f_flags = file->f_flag;
vhold(file->f_vnode);
filp->f_vnode = file->f_vnode;
filp->_file = file;
@ -704,35 +700,17 @@ linux_dev_open(struct cdev *dev, int oflags, int devtype, struct thread *td)
if (filp->f_op->open) {
error = -filp->f_op->open(file->f_vnode, filp);
if (error) {
vdrop(filp->f_vnode);
kfree(filp);
goto done;
}
}
error = devfs_set_cdevpriv(filp, linux_file_dtor);
if (error) {
filp->f_op->release(file->f_vnode, filp);
vdrop(filp->f_vnode);
kfree(filp);
}
done:
return (error);
}
}
static int
linux_dev_close(struct cdev *dev, int fflag, int devtype, struct thread *td)
{
struct linux_file *filp;
struct file *file;
int error;
/* hold on to the vnode - used for fstat() */
vhold(filp->f_vnode);
file = td->td_fpop;
if ((error = devfs_get_cdevpriv((void **)&filp)) != 0)
return (error);
filp->f_flags = file->f_flag;
devfs_clear_cdevpriv();
return (0);
/* release the file from devfs */
finit(file, filp->f_mode, DTYPE_DEV, filp, &linuxfileops);
return (ENXIO);
}
#define LINUX_IOCTL_MIN_PTR 0x10000UL
@ -850,24 +828,12 @@ linux_access_ok(int rw, const void *uaddr, size_t len)
}
static int
linux_dev_ioctl(struct cdev *dev, u_long cmd, caddr_t data, int fflag,
struct thread *td)
linux_file_ioctl_sub(struct file *fp, struct linux_file *filp,
u_long cmd, caddr_t data, struct thread *td)
{
struct linux_file *filp;
struct file *file;
unsigned size;
int error;
file = td->td_fpop;
if ((error = devfs_get_cdevpriv((void **)&filp)) != 0)
return (error);
filp->f_flags = file->f_flag;
/* the LinuxKPI supports blocking and non-blocking I/O */
if (cmd == FIONBIO || cmd == FIOASYNC)
return (0);
linux_set_current(td);
size = IOCPARM_LEN(cmd);
/* refer to logic in sys_ioctl() */
if (size > 0) {
@ -888,17 +854,17 @@ linux_dev_ioctl(struct cdev *dev, u_long cmd, caddr_t data, int fflag,
if (td->td_proc->p_elf_machine == EM_386) {
/* try the compat IOCTL handler first */
if (filp->f_op->compat_ioctl != NULL)
error = -filp->f_op->compat_ioctl(filp, cmd, (u_long)data);
error = -OPW(fp, td, filp->f_op->compat_ioctl(filp, cmd, (u_long)data));
else
error = ENOTTY;
/* fallback to the regular IOCTL handler, if any */
if (error == ENOTTY && filp->f_op->unlocked_ioctl != NULL)
error = -filp->f_op->unlocked_ioctl(filp, cmd, (u_long)data);
error = -OPW(fp, td, filp->f_op->unlocked_ioctl(filp, cmd, (u_long)data));
} else
#endif
if (filp->f_op->unlocked_ioctl != NULL)
error = -filp->f_op->unlocked_ioctl(filp, cmd, (u_long)data);
error = -OPW(fp, td, filp->f_op->unlocked_ioctl(filp, cmd, (u_long)data));
else
error = ENOTTY;
if (size > 0) {
@ -908,118 +874,15 @@ linux_dev_ioctl(struct cdev *dev, u_long cmd, caddr_t data, int fflag,
if (error == EWOULDBLOCK) {
/* update kqfilter status, if any */
linux_dev_kqfilter_poll(filp,
linux_file_kqfilter_poll(filp,
LINUX_KQ_FLAG_HAS_READ | LINUX_KQ_FLAG_HAS_WRITE);
} else if (error == ERESTARTSYS)
error = ERESTART;
return (error);
}
static int
linux_dev_read(struct cdev *dev, struct uio *uio, int ioflag)
{
struct linux_file *filp;
struct thread *td;
struct file *file;
ssize_t bytes;
int error;
td = curthread;
file = td->td_fpop;
if ((error = devfs_get_cdevpriv((void **)&filp)) != 0)
return (error);
filp->f_flags = file->f_flag;
/* XXX no support for I/O vectors currently */
if (uio->uio_iovcnt != 1)
return (EOPNOTSUPP);
linux_set_current(td);
if (filp->f_op->read) {
bytes = filp->f_op->read(filp, uio->uio_iov->iov_base,
uio->uio_iov->iov_len, &uio->uio_offset);
if (bytes >= 0) {
uio->uio_iov->iov_base =
((uint8_t *)uio->uio_iov->iov_base) + bytes;
uio->uio_iov->iov_len -= bytes;
uio->uio_resid -= bytes;
} else {
error = -bytes;
if (error == ERESTARTSYS)
error = ERESTART;
}
} else
error = ENXIO;
/* update kqfilter status, if any */
linux_dev_kqfilter_poll(filp, LINUX_KQ_FLAG_HAS_READ);
return (error);
}
static int
linux_dev_write(struct cdev *dev, struct uio *uio, int ioflag)
{
struct linux_file *filp;
struct thread *td;
struct file *file;
ssize_t bytes;
int error;
td = curthread;
file = td->td_fpop;
if ((error = devfs_get_cdevpriv((void **)&filp)) != 0)
return (error);
filp->f_flags = file->f_flag;
/* XXX no support for I/O vectors currently */
if (uio->uio_iovcnt != 1)
return (EOPNOTSUPP);
linux_set_current(td);
if (filp->f_op->write) {
bytes = filp->f_op->write(filp, uio->uio_iov->iov_base,
uio->uio_iov->iov_len, &uio->uio_offset);
if (bytes >= 0) {
uio->uio_iov->iov_base =
((uint8_t *)uio->uio_iov->iov_base) + bytes;
uio->uio_iov->iov_len -= bytes;
uio->uio_resid -= bytes;
} else {
error = -bytes;
if (error == ERESTARTSYS)
error = ERESTART;
}
} else
error = ENXIO;
/* update kqfilter status, if any */
linux_dev_kqfilter_poll(filp, LINUX_KQ_FLAG_HAS_WRITE);
return (error);
}
#define LINUX_POLL_TABLE_NORMAL ((poll_table *)1)
static int
linux_dev_poll(struct cdev *dev, int events, struct thread *td)
{
struct linux_file *filp;
struct file *file;
int revents;
if (devfs_get_cdevpriv((void **)&filp) != 0)
goto error;
file = td->td_fpop;
filp->f_flags = file->f_flag;
linux_set_current(td);
if (filp->f_op->poll != NULL)
revents = filp->f_op->poll(filp, LINUX_POLL_TABLE_NORMAL) & events;
else
revents = 0;
return (revents);
error:
return (events & (POLLHUP|POLLIN|POLLRDNORM|POLLOUT|POLLWRNORM));
}
/*
* This function atomically updates the poll wakeup state and returns
* the previous state at the time of update.
@ -1127,7 +990,7 @@ linux_poll_wakeup(struct linux_file *filp)
}
static void
linux_dev_kqfilter_detach(struct knote *kn)
linux_file_kqfilter_detach(struct knote *kn)
{
struct linux_file *filp = kn->kn_hook;
@ -1137,7 +1000,7 @@ linux_dev_kqfilter_detach(struct knote *kn)
}
static int
linux_dev_kqfilter_read_event(struct knote *kn, long hint)
linux_file_kqfilter_read_event(struct knote *kn, long hint)
{
struct linux_file *filp = kn->kn_hook;
@ -1147,7 +1010,7 @@ linux_dev_kqfilter_read_event(struct knote *kn, long hint)
}
static int
linux_dev_kqfilter_write_event(struct knote *kn, long hint)
linux_file_kqfilter_write_event(struct knote *kn, long hint)
{
struct linux_file *filp = kn->kn_hook;
@ -1158,24 +1021,26 @@ linux_dev_kqfilter_write_event(struct knote *kn, long hint)
static struct filterops linux_dev_kqfiltops_read = {
.f_isfd = 1,
.f_detach = linux_dev_kqfilter_detach,
.f_event = linux_dev_kqfilter_read_event,
.f_detach = linux_file_kqfilter_detach,
.f_event = linux_file_kqfilter_read_event,
};
static struct filterops linux_dev_kqfiltops_write = {
.f_isfd = 1,
.f_detach = linux_dev_kqfilter_detach,
.f_event = linux_dev_kqfilter_write_event,
.f_detach = linux_file_kqfilter_detach,
.f_event = linux_file_kqfilter_write_event,
};
static void
linux_dev_kqfilter_poll(struct linux_file *filp, int kqflags)
linux_file_kqfilter_poll(struct linux_file *filp, int kqflags)
{
int temp;
if (filp->f_kqflags & kqflags) {
struct thread *td = curthread;
/* get the latest polling state */
temp = filp->f_op->poll(filp, NULL);
temp = OPW(filp->_file, td, filp->f_op->poll(filp, NULL));
spin_lock(&filp->f_kqlock);
/* clear kqflags */
@ -1196,17 +1061,14 @@ linux_dev_kqfilter_poll(struct linux_file *filp, int kqflags)
}
static int
linux_dev_kqfilter(struct cdev *dev, struct knote *kn)
linux_file_kqfilter(struct file *file, struct knote *kn)
{
struct linux_file *filp;
struct file *file;
struct thread *td;
int error;
td = curthread;
file = td->td_fpop;
if ((error = devfs_get_cdevpriv((void **)&filp)) != 0)
return (error);
filp = (struct linux_file *)file->f_data;
filp->f_flags = file->f_flag;
if (filp->f_op->poll == NULL)
return (EINVAL);
@ -1218,12 +1080,14 @@ linux_dev_kqfilter(struct cdev *dev, struct knote *kn)
kn->kn_fop = &linux_dev_kqfiltops_read;
kn->kn_hook = filp;
knlist_add(&filp->f_selinfo.si_note, kn, 1);
error = 0;
break;
case EVFILT_WRITE:
filp->f_kqflags |= LINUX_KQ_FLAG_HAS_WRITE;
kn->kn_fop = &linux_dev_kqfiltops_write;
kn->kn_hook = filp;
knlist_add(&filp->f_selinfo.si_note, kn, 1);
error = 0;
break;
default:
error = EINVAL;
@ -1235,32 +1099,28 @@ linux_dev_kqfilter(struct cdev *dev, struct knote *kn)
linux_set_current(td);
/* update kqfilter status, if any */
linux_dev_kqfilter_poll(filp,
linux_file_kqfilter_poll(filp,
LINUX_KQ_FLAG_HAS_READ | LINUX_KQ_FLAG_HAS_WRITE);
}
return (error);
}
static int
linux_dev_mmap_single(struct cdev *dev, vm_ooffset_t *offset,
vm_size_t size, struct vm_object **object, int nprot)
linux_file_mmap_single(struct file *fp, vm_ooffset_t *offset,
vm_size_t size, struct vm_object **object, int nprot,
struct thread *td)
{
struct vm_area_struct *vmap;
struct mm_struct *mm;
struct linux_file *filp;
struct thread *td;
struct file *file;
vm_memattr_t attr;
int error;
td = curthread;
file = td->td_fpop;
if ((error = devfs_get_cdevpriv((void **)&filp)) != 0)
return (error);
filp->f_flags = file->f_flag;
filp = (struct linux_file *)fp->f_data;
filp->f_flags = fp->f_flag;
if (filp->f_op->mmap == NULL)
return (ENODEV);
return (EOPNOTSUPP);
linux_set_current(td);
@ -1288,7 +1148,7 @@ linux_dev_mmap_single(struct cdev *dev, vm_ooffset_t *offset,
if (unlikely(down_write_killable(&vmap->vm_mm->mmap_sem))) {
error = EINTR;
} else {
error = -filp->f_op->mmap(filp, vmap);
error = -OPW(fp, td, filp->f_op->mmap(filp, vmap));
up_write(&vmap->vm_mm->mmap_sem);
}
@ -1351,11 +1211,11 @@ linux_dev_mmap_single(struct cdev *dev, vm_ooffset_t *offset,
if (vm_no_fault) {
*object = cdev_pager_allocate(vm_private_data, OBJT_DEVICE,
&linux_cdev_pager_ops[1], size, nprot, *offset,
curthread->td_ucred);
td->td_ucred);
} else {
*object = cdev_pager_allocate(vm_private_data, OBJT_MGTDEVICE,
&linux_cdev_pager_ops[0], size, nprot, *offset,
curthread->td_ucred);
td->td_ucred);
}
/* check if allocating the VM object failed */
@ -1376,7 +1236,7 @@ linux_dev_mmap_single(struct cdev *dev, vm_ooffset_t *offset,
(vm_paddr_t)vmap->vm_pfn << PAGE_SHIFT, vmap->vm_len);
*object = vm_pager_allocate(OBJT_SG, sg, vmap->vm_len,
nprot, 0, curthread->td_ucred);
nprot, 0, td->td_ucred);
linux_cdev_handle_free(vmap);
@ -1397,15 +1257,7 @@ linux_dev_mmap_single(struct cdev *dev, vm_ooffset_t *offset,
struct cdevsw linuxcdevsw = {
.d_version = D_VERSION,
.d_flags = D_TRACKCLOSE,
.d_open = linux_dev_open,
.d_close = linux_dev_close,
.d_read = linux_dev_read,
.d_write = linux_dev_write,
.d_ioctl = linux_dev_ioctl,
.d_mmap_single = linux_dev_mmap_single,
.d_poll = linux_dev_poll,
.d_kqfilter = linux_dev_kqfilter,
.d_fdopen = linux_dev_fdopen,
.d_name = "lkpidev",
};
@ -1425,18 +1277,61 @@ linux_file_read(struct file *file, struct uio *uio, struct ucred *active_cred,
return (EOPNOTSUPP);
linux_set_current(td);
if (filp->f_op->read) {
bytes = filp->f_op->read(filp, uio->uio_iov->iov_base,
uio->uio_iov->iov_len, &uio->uio_offset);
bytes = OPW(file, td, filp->f_op->read(filp, uio->uio_iov->iov_base,
uio->uio_iov->iov_len, &uio->uio_offset));
if (bytes >= 0) {
uio->uio_iov->iov_base =
((uint8_t *)uio->uio_iov->iov_base) + bytes;
uio->uio_iov->iov_len -= bytes;
uio->uio_resid -= bytes;
} else
} else {
error = -bytes;
if (error == ERESTARTSYS)
error = ERESTART;
}
} else
error = ENXIO;
/* update kqfilter status, if any */
linux_file_kqfilter_poll(filp, LINUX_KQ_FLAG_HAS_READ);
return (error);
}
static int
linux_file_write(struct file *file, struct uio *uio, struct ucred *active_cred,
int flags, struct thread *td)
{
struct linux_file *filp;
ssize_t bytes;
int error;
error = 0;
filp = (struct linux_file *)file->f_data;
filp->f_flags = file->f_flag;
/* XXX no support for I/O vectors currently */
if (uio->uio_iovcnt != 1)
return (EOPNOTSUPP);
linux_set_current(td);
if (filp->f_op->write) {
bytes = OPW(file, td, filp->f_op->write(filp, uio->uio_iov->iov_base,
uio->uio_iov->iov_len, &uio->uio_offset));
if (bytes >= 0) {
uio->uio_iov->iov_base =
((uint8_t *)uio->uio_iov->iov_base) + bytes;
uio->uio_iov->iov_len -= bytes;
uio->uio_resid -= bytes;
} else {
error = -bytes;
if (error == ERESTARTSYS)
error = ERESTART;
}
} else
error = ENXIO;
/* update kqfilter status, if any */
linux_file_kqfilter_poll(filp, LINUX_KQ_FLAG_HAS_WRITE);
return (error);
}
@ -1451,7 +1346,7 @@ linux_file_poll(struct file *file, int events, struct ucred *active_cred,
filp->f_flags = file->f_flag;
linux_set_current(td);
if (filp->f_op->poll != NULL)
revents = filp->f_op->poll(filp, LINUX_POLL_TABLE_NORMAL) & events;
revents = OPW(file, td, filp->f_op->poll(filp, LINUX_POLL_TABLE_NORMAL)) & events;
else
revents = 0;
@ -1465,11 +1360,16 @@ linux_file_close(struct file *file, struct thread *td)
int error;
filp = (struct linux_file *)file->f_data;
KASSERT(file_count(filp) == 0, ("File refcount(%d) is not zero", file_count(filp)));
filp->f_flags = file->f_flag;
linux_set_current(td);
linux_poll_wait_dequeue(filp);
error = -filp->f_op->release(NULL, filp);
error = -OPW(file, td, filp->f_op->release(filp->f_vnode, filp));
funsetown(&filp->f_sigio);
if (filp->f_vnode != NULL)
vdrop(filp->f_vnode);
kfree(filp);
return (error);
@ -1493,30 +1393,128 @@ linux_file_ioctl(struct file *fp, u_long cmd, void *data, struct ucred *cred,
case FIOASYNC:
if (filp->f_op->fasync == NULL)
break;
error = filp->f_op->fasync(0, filp, fp->f_flag & FASYNC);
error = -OPW(fp, td, filp->f_op->fasync(0, filp, fp->f_flag & FASYNC));
break;
case FIOSETOWN:
error = fsetown(*(int *)data, &filp->f_sigio);
if (error == 0)
error = filp->f_op->fasync(0, filp,
fp->f_flag & FASYNC);
if (error == 0) {
if (filp->f_op->fasync == NULL)
break;
error = -OPW(fp, td, filp->f_op->fasync(0, filp,
fp->f_flag & FASYNC));
}
break;
case FIOGETOWN:
*(int *)data = fgetown(&filp->f_sigio);
break;
default:
error = ENOTTY;
error = linux_file_ioctl_sub(fp, filp, cmd, data, td);
break;
}
return (error);
}
static int
linux_file_mmap_sub(struct thread *td, vm_size_t objsize, vm_prot_t prot,
vm_prot_t *maxprotp, int *flagsp, struct file *fp,
vm_ooffset_t *foff, vm_object_t *objp)
{
/*
* Character devices do not provide private mappings
* of any kind:
*/
if ((*maxprotp & VM_PROT_WRITE) == 0 &&
(prot & VM_PROT_WRITE) != 0)
return (EACCES);
if ((*flagsp & (MAP_PRIVATE | MAP_COPY)) != 0)
return (EINVAL);
return (linux_file_mmap_single(fp, foff, objsize, objp, (int)prot, td));
}
static int
linux_file_mmap(struct file *fp, vm_map_t map, vm_offset_t *addr, vm_size_t size,
vm_prot_t prot, vm_prot_t cap_maxprot, int flags, vm_ooffset_t foff,
struct thread *td)
{
struct linux_file *filp;
struct mount *mp;
struct vnode *vp;
vm_object_t object;
vm_prot_t maxprot;
int error;
filp = (struct linux_file *)fp->f_data;
vp = filp->f_vnode;
if (vp == NULL)
return (EOPNOTSUPP);
/*
* Ensure that file and memory protections are
* compatible.
*/
mp = vp->v_mount;
if (mp != NULL && (mp->mnt_flag & MNT_NOEXEC) != 0) {
maxprot = VM_PROT_NONE;
if ((prot & VM_PROT_EXECUTE) != 0)
return (EACCES);
} else
maxprot = VM_PROT_EXECUTE;
if ((fp->f_flag & FREAD) != 0)
maxprot |= VM_PROT_READ;
else if ((prot & VM_PROT_READ) != 0)
return (EACCES);
/*
* If we are sharing potential changes via MAP_SHARED and we
* are trying to get write permission although we opened it
* without asking for it, bail out.
*
* Note that most character devices always share mappings.
*
* Rely on linux_file_mmap_sub() to fail invalid MAP_PRIVATE
* requests rather than doing it here.
*/
if ((flags & MAP_SHARED) != 0) {
if ((fp->f_flag & FWRITE) != 0)
maxprot |= VM_PROT_WRITE;
else if ((prot & VM_PROT_WRITE) != 0)
return (EACCES);
}
maxprot &= cap_maxprot;
error = linux_file_mmap_sub(td, size, prot, &maxprot, &flags, fp, &foff,
&object);
if (error != 0)
return (error);
error = vm_mmap_object(map, addr, size, prot, maxprot, flags, object,
foff, FALSE, td);
if (error != 0)
vm_object_deallocate(object);
return (error);
}
static int
linux_file_stat(struct file *fp, struct stat *sb, struct ucred *active_cred,
struct thread *td)
{
struct linux_file *filp;
struct vnode *vp;
int error;
filp = (struct linux_file *)fp->f_data;
if (filp->f_vnode == NULL)
return (EOPNOTSUPP);
vp = filp->f_vnode;
vn_lock(vp, LK_SHARED | LK_RETRY);
error = vn_stat(vp, sb, td->td_ucred, NOCRED, td);
VOP_UNLOCK(vp, 0);
return (error);
}
static int
@ -1544,14 +1542,15 @@ linux_iminor(struct inode *inode)
struct fileops linuxfileops = {
.fo_read = linux_file_read,
.fo_write = invfo_rdwr,
.fo_write = linux_file_write,
.fo_truncate = invfo_truncate,
.fo_kqfilter = invfo_kqfilter,
.fo_kqfilter = linux_file_kqfilter,
.fo_stat = linux_file_stat,
.fo_fill_kinfo = linux_file_fill_kinfo,
.fo_poll = linux_file_poll,
.fo_close = linux_file_close,
.fo_ioctl = linux_file_ioctl,
.fo_mmap = linux_file_mmap,
.fo_chmod = invfo_chmod,
.fo_chown = invfo_chown,
.fo_sendfile = invfo_sendfile,