On Tue, Jul 10, 2018 at 04:04:55PM +0100, Al Viro wrote:
> First of all, I'm still not at all convinced that this "noaccount" thing is
> sane, especially since path_open() is exported. But that aside, __get_empty_filp()
> needs to be shot, just for the name and calling conventions alone.
>
> It gets a bullshit argument (bool account) *AND* does not get the argument it
> does need. Note that the first thing get_empty_filp() (now __get...) does
> is
> const struct cred *cred = current_cred();
> followed by
> f->f_cred = get_cred(cred);
>
> Now look at path_open(). What happens to the cred argument it gets? It goes
> to do_dentry_open(), where it gets passed to security_file_open() and not
> used by anything else. In security_file_open() we have it passed to
> ret = call_int_hook(file_open, 0, file, cred);
> and there are three instances of ->file_open() - apparmor, selinux and tomoyo.
> The last one ignores cred entirely; the other two do checks based on it,
> but *all* of them leave file->f_cred as it was.
>
> This is not a new crap. It had been inherited from dentry_open(), which got it
> from "CRED: Pass credentials through dentry_open()" back in 2008. Note that
> * among the callers of dentry_open() (20) and vfs_open() (2 more)
> only these
> fs/cachefiles/rdwr.c:913: file = dentry_open(&path, O_RDWR | O_LARGEFILE, cache->cache_cred);
> security/apparmor/file.c:695: devnull = dentry_open(&aa_null, O_RDWR, cred);
> security/selinux/hooks.c:2628: devnull = dentry_open(&selinux_null, O_RDWR, cred);
> get cred != current_cred(). Which helps masking the issue, but makes the
> decision to add that argument (instead of a separate helper) rather dubious.
> * overlayfs itself appears to *have* run into the problem, judging
> by
> old_cred = ovl_override_creds(inode->i_sb);
> realfile = path_open(&file->f_path, file->f_flags | O_NOATIME,
> realinode, current_cred(), false);
> revert_creds(old_cred);
> in there.
>
> Folks, if you have to go to that kind of contortions, why not do it right?
> * add static __alloc_file(cred), which would get cred pointer (and not
> use current_cred() internally), allocated a file (without bothering with
> nr_files) and returned it
> * have alloc_empty_file(cred) that would do the song and dance
> with nr_files (and used __alloc_file() internally).
> * use that as a replacement for get_empty_filp() - path_openat() would
> *probably* use current_cred() for argument, alloc_file() definitely would and
> dentry_open() would pass its cred argument.
> * in internal.h, static inline alloc_empty_file_noaccount(cred) would
> use __alloc_file() and set FMODE_NOACCOUNT in case of success.
> * do_dentry_open() loses the fucking cred argument - it should be in
> file->f_cred.
> * vfs_open() goes away - in your branch it's absolutely pointless.
> * path_open() loses its 'account' argument - it's always false.
> Uses alloc_empty_file_noaccount() to allocate the sucker. And for fsck
> sake, pass it the creds you want to use rather than playing that kind of
> games with override/revert.
FWIW, see vfs.git#work.open2 for part of that program. I have not pulled
the overlayfs stuff in, but doing the rest based at circa e9cf4c40af4c
("fold put_filp() into fput()") is trivial. Remains to be done out of the
above:
* add static __alloc_file(cred), which would be basically
alloc_empty_file() sans nr_files-related parts. Make alloc_empty_file()
call it for actual allocation.
* in internal.h, static inline alloc_empty_file_noaccount(cred) would
use __alloc_file() and set FMODE_NOACCOUNT in case of success.
* path_open() loses its 'account' argument - it's always false.
Uses alloc_empty_file_noaccount() to allocate the sucker. And for fsck
sake, pass it the creds you want to use rather than playing that kind of
games with override/revert.
* (after the overlayfs changes to vfs_open()) vfs_open() goes away,
expanded into callers.
Stuff currently in work.open2:
Preparatory fixes (this cycle fodder):
drm_mode_create_lease_ioctl(): fix open-coded filp_clone_open()
cxl_getfile(): fix double-iput() on alloc_file() failures
ocxlflash_getfile(): fix double-iput() on alloc_file() failures
More preparatory massage:
make get_empty_filp() to call file_free_rcu() directly
fold security_file_free() into file_free()
turn filp_clone_open() into inline wrapper for dentry_open()
create_pipe_files(): use fput() if allocation of the second file fails
make sure do_dentry_open() won't return positive as an error
Providing the right ->f_cred:
pass creds to get_empty_filp(), make sure dentry_open() passes the right creds
get rid of cred argument of vfs_open() and do_dentry_open()
security_file_open(): lose cred argument
->file_open(): lose cred argument
Mirror the "do we need fput() or do we need put_filp()" in ->f_mode:
introduce FMODE_OPENED
fold put_filp() into fput()
At that point we can start simplifying open-related paths, now that
cleanups are a lot more uniform:
lift fput() on late failures into path_openat()
now we can fold open_check_o_direct() into do_dentry_open()
switch all remaining checks for FILE_OPENED to FMODE_OPENED
Half of the reasons for ->atomic_open() 'opened' argument is gone, let's deal with
the rest:
introduce FMODE_CREATED and switch to it
IMA: don't propagate opened through the entire thing
getting rid of 'opened' argument of ->atomic_open() - step 1
getting rid of 'opened' argument of ->atomic_open() - part 2
get rid of 'opened' argument of ->atomic_open() - part 3
get rid of 'opened' in path_openat() and the helpers downstream
->atomic_open(): return 0 in all success cases
document ->atomic_open() changes
switch atomic_open() and lookup_open() to returning 0 in all success cases
kill FILE_{CREATED,OPENED}
At that point we've got _much_ simpler ->atomic_open() calling conventions as well
as the code using it. Next part - alloc_file() calling conventions:
new wrapper: alloc_file_pseudo()
__shmem_file_setup(): reorder allocations
... and switch shmem_file_setup() to alloc_file_pseudo()
cxl_getfile(): switch to alloc_file_pseudo()
ocxlflash_getfile(): switch to alloc_file_pseudo()
hugetlb_file_setup(): switch to alloc_file_pseudo()
anon_inode_getfile(): switch to alloc_file_pseudo()
create_pipe_files(): switch the first allocation to alloc_file_pseudo()
new helper: alloc_file_clone()
do_shmat(): grab shp->shm_file earlier, switch to alloc_file_clone()
make alloc_file() static
document alloc_file() changes
And cleanups in pathname-resolving loops from better calling conventions
for path_init()/link_path_walk():
make path_init() unconditionally paired with terminate_walk()
allow link_path_walk() to take ERR_PTR()
few more cleanups of link_path_walk() callers
IMO the diffstat is not too bad -
Documentation/filesystems/Locking | 2 +-
Documentation/filesystems/porting | 20 +++
Documentation/filesystems/vfs.txt | 18 +--
drivers/gpu/drm/drm_lease.c | 16 +--
drivers/misc/cxl/api.c | 21 +---
drivers/scsi/cxlflash/ocxl_hw.c | 24 +---
fs/9p/vfs_inode.c | 7 +-
fs/9p/vfs_inode_dotl.c | 7 +-
fs/aio.c | 26 +---
fs/anon_inodes.c | 29 +----
fs/bad_inode.c | 2 +-
fs/binfmt_misc.c | 2 +-
fs/ceph/file.c | 7 +-
fs/ceph/super.h | 3 +-
fs/cifs/cifsfs.h | 3 +-
fs/cifs/dir.c | 7 +-
fs/file_table.c | 71 +++++++----
fs/fuse/dir.c | 10 +-
fs/gfs2/inode.c | 32 +++--
fs/hugetlbfs/inode.c | 55 +++------
fs/internal.h | 6 +-
fs/namei.c | 223 +++++++++++++---------------------
fs/nfs/dir.c | 14 ++-
fs/nfs/nfs4_fs.h | 2 +-
fs/nfs/nfs4proc.c | 2 +-
fs/nfsd/vfs.c | 2 +-
fs/open.c | 86 ++++---------
fs/pipe.c | 38 ++----
include/linux/file.h | 8 +-
include/linux/fs.h | 16 +--
include/linux/ima.h | 4 +-
include/linux/lsm_hooks.h | 2 +-
include/linux/security.h | 5 +-
ipc/shm.c | 39 +++---
mm/shmem.c | 50 ++------
net/socket.c | 27 +---
security/apparmor/lsm.c | 4 +-
security/integrity/ima/ima.h | 4 +-
security/integrity/ima/ima_appraise.c | 4 +-
security/integrity/ima/ima_main.c | 16 +--
security/security.c | 4 +-
security/selinux/hooks.c | 4 +-
security/smack/smack_lsm.c | 6 +-
security/tomoyo/tomoyo.c | 2 +-
44 files changed, 360 insertions(+), 570 deletions(-)
