smb: client: fix use-after-free in smb2_query_info_compound()
authorPaulo Alcantara <pc@manguebit.com>
Mon, 30 Oct 2023 20:19:56 +0000 (17:19 -0300)
committerGreg Kroah-Hartman <gregkh@linuxfoundation.org>
Tue, 28 Nov 2023 17:20:09 +0000 (17:20 +0000)
commit 5c86919455c1edec99ebd3338ad213b59271a71b upstream.

The following UAF was triggered when running fstests generic/072 with
KASAN enabled against Windows Server 2022 and mount options
'multichannel,max_channels=2,vers=3.1.1,mfsymlinks,noperm'

  BUG: KASAN: slab-use-after-free in smb2_query_info_compound+0x423/0x6d0 [cifs]
  Read of size 8 at addr ffff888014941048 by task xfs_io/27534

  CPU: 0 PID: 27534 Comm: xfs_io Not tainted 6.6.0-rc7 #1
  Hardware name: QEMU Standard PC (Q35 + ICH9, 2009), BIOS
  rel-1.16.2-3-gd478f380-rebuilt.opensuse.org 04/01/2014
  Call Trace:
   dump_stack_lvl+0x4a/0x80
   print_report+0xcf/0x650
   ? srso_alias_return_thunk+0x5/0x7f
   ? srso_alias_return_thunk+0x5/0x7f
   ? __phys_addr+0x46/0x90
   kasan_report+0xda/0x110
   ? smb2_query_info_compound+0x423/0x6d0 [cifs]
   ? smb2_query_info_compound+0x423/0x6d0 [cifs]
   smb2_query_info_compound+0x423/0x6d0 [cifs]
   ? __pfx_smb2_query_info_compound+0x10/0x10 [cifs]
   ? srso_alias_return_thunk+0x5/0x7f
   ? __stack_depot_save+0x39/0x480
   ? kasan_save_stack+0x33/0x60
   ? kasan_set_track+0x25/0x30
   ? ____kasan_slab_free+0x126/0x170
   smb2_queryfs+0xc2/0x2c0 [cifs]
   ? __pfx_smb2_queryfs+0x10/0x10 [cifs]
   ? __pfx___lock_acquire+0x10/0x10
   smb311_queryfs+0x210/0x220 [cifs]
   ? __pfx_smb311_queryfs+0x10/0x10 [cifs]
   ? srso_alias_return_thunk+0x5/0x7f
   ? __lock_acquire+0x480/0x26c0
   ? lock_release+0x1ed/0x640
   ? srso_alias_return_thunk+0x5/0x7f
   ? do_raw_spin_unlock+0x9b/0x100
   cifs_statfs+0x18c/0x4b0 [cifs]
   statfs_by_dentry+0x9b/0xf0
   fd_statfs+0x4e/0xb0
   __do_sys_fstatfs+0x7f/0xe0
   ? __pfx___do_sys_fstatfs+0x10/0x10
   ? srso_alias_return_thunk+0x5/0x7f
   ? lockdep_hardirqs_on_prepare+0x136/0x200
   ? srso_alias_return_thunk+0x5/0x7f
   do_syscall_64+0x3f/0x90
   entry_SYSCALL_64_after_hwframe+0x6e/0xd8

  Allocated by task 27534:
   kasan_save_stack+0x33/0x60
   kasan_set_track+0x25/0x30
   __kasan_kmalloc+0x8f/0xa0
   open_cached_dir+0x71b/0x1240 [cifs]
   smb2_query_info_compound+0x5c3/0x6d0 [cifs]
   smb2_queryfs+0xc2/0x2c0 [cifs]
   smb311_queryfs+0x210/0x220 [cifs]
   cifs_statfs+0x18c/0x4b0 [cifs]
   statfs_by_dentry+0x9b/0xf0
   fd_statfs+0x4e/0xb0
   __do_sys_fstatfs+0x7f/0xe0
   do_syscall_64+0x3f/0x90
   entry_SYSCALL_64_after_hwframe+0x6e/0xd8

  Freed by task 27534:
   kasan_save_stack+0x33/0x60
   kasan_set_track+0x25/0x30
   kasan_save_free_info+0x2b/0x50
   ____kasan_slab_free+0x126/0x170
   slab_free_freelist_hook+0xd0/0x1e0
   __kmem_cache_free+0x9d/0x1b0
   open_cached_dir+0xff5/0x1240 [cifs]
   smb2_query_info_compound+0x5c3/0x6d0 [cifs]
   smb2_queryfs+0xc2/0x2c0 [cifs]

This is a race between open_cached_dir() and cached_dir_lease_break()
where the cache entry for the open directory handle receives a lease
break while creating it.  And before returning from open_cached_dir(),
we put the last reference of the new @cfid because of
!@cfid->has_lease.

Besides the UAF, while running xfstests a lot of missed lease breaks
have been noticed in tests that run several concurrent statfs(2) calls
on those cached fids

  CIFS: VFS: \\w22-root1.gandalf.test No task to wake, unknown frame...
  CIFS: VFS: \\w22-root1.gandalf.test Cmd: 18 Err: 0x0 Flags: 0x1...
  CIFS: VFS: \\w22-root1.gandalf.test smb buf 00000000715bfe83 len 108
  CIFS: VFS: Dump pending requests:
  CIFS: VFS: \\w22-root1.gandalf.test No task to wake, unknown frame...
  CIFS: VFS: \\w22-root1.gandalf.test Cmd: 18 Err: 0x0 Flags: 0x1...
  CIFS: VFS: \\w22-root1.gandalf.test smb buf 000000005aa7316e len 108
  ...

To fix both, in open_cached_dir() ensure that @cfid->has_lease is set
right before sending out compounded request so that any potential
lease break will be get processed by demultiplex thread while we're
still caching @cfid.  And, if open failed for some reason, re-check
@cfid->has_lease to decide whether or not put lease reference.

Cc: stable@vger.kernel.org
Signed-off-by: Paulo Alcantara (SUSE) <pc@manguebit.com>
Signed-off-by: Steve French <stfrench@microsoft.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
fs/smb/client/cached_dir.c

index fe1bf5b..59f6b8e 100644 (file)
@@ -32,7 +32,7 @@ static struct cached_fid *find_or_create_cached_dir(struct cached_fids *cfids,
                         * fully cached or it may be in the process of
                         * being deleted due to a lease break.
                         */
-                       if (!cfid->has_lease) {
+                       if (!cfid->time || !cfid->has_lease) {
                                spin_unlock(&cfids->cfid_list_lock);
                                return NULL;
                        }
@@ -193,10 +193,20 @@ int open_cached_dir(unsigned int xid, struct cifs_tcon *tcon,
        npath = path_no_prefix(cifs_sb, path);
        if (IS_ERR(npath)) {
                rc = PTR_ERR(npath);
-               kfree(utf16_path);
-               return rc;
+               goto out;
        }
 
+       if (!npath[0]) {
+               dentry = dget(cifs_sb->root);
+       } else {
+               dentry = path_to_dentry(cifs_sb, npath);
+               if (IS_ERR(dentry)) {
+                       rc = -ENOENT;
+                       goto out;
+               }
+       }
+       cfid->dentry = dentry;
+
        /*
         * We do not hold the lock for the open because in case
         * SMB2_open needs to reconnect.
@@ -249,6 +259,15 @@ int open_cached_dir(unsigned int xid, struct cifs_tcon *tcon,
 
        smb2_set_related(&rqst[1]);
 
+       /*
+        * Set @cfid->has_lease to true before sending out compounded request so
+        * its lease reference can be put in cached_dir_lease_break() due to a
+        * potential lease break right after the request is sent or while @cfid
+        * is still being cached.  Concurrent processes won't be to use it yet
+        * due to @cfid->time being zero.
+        */
+       cfid->has_lease = true;
+
        rc = compound_send_recv(xid, ses, server,
                                flags, 2, rqst,
                                resp_buftype, rsp_iov);
@@ -263,6 +282,8 @@ int open_cached_dir(unsigned int xid, struct cifs_tcon *tcon,
        cfid->tcon = tcon;
        cfid->is_open = true;
 
+       spin_lock(&cfids->cfid_list_lock);
+
        o_rsp = (struct smb2_create_rsp *)rsp_iov[0].iov_base;
        oparms.fid->persistent_fid = o_rsp->PersistentFileId;
        oparms.fid->volatile_fid = o_rsp->VolatileFileId;
@@ -270,18 +291,25 @@ int open_cached_dir(unsigned int xid, struct cifs_tcon *tcon,
        oparms.fid->mid = le64_to_cpu(o_rsp->hdr.MessageId);
 #endif /* CIFS_DEBUG2 */
 
-       if (o_rsp->OplockLevel != SMB2_OPLOCK_LEVEL_LEASE)
+       rc = -EINVAL;
+       if (o_rsp->OplockLevel != SMB2_OPLOCK_LEVEL_LEASE) {
+               spin_unlock(&cfids->cfid_list_lock);
                goto oshr_free;
+       }
 
        smb2_parse_contexts(server, o_rsp,
                            &oparms.fid->epoch,
                            oparms.fid->lease_key, &oplock,
                            NULL, NULL);
-       if (!(oplock & SMB2_LEASE_READ_CACHING_HE))
+       if (!(oplock & SMB2_LEASE_READ_CACHING_HE)) {
+               spin_unlock(&cfids->cfid_list_lock);
                goto oshr_free;
+       }
        qi_rsp = (struct smb2_query_info_rsp *)rsp_iov[1].iov_base;
-       if (le32_to_cpu(qi_rsp->OutputBufferLength) < sizeof(struct smb2_file_all_info))
+       if (le32_to_cpu(qi_rsp->OutputBufferLength) < sizeof(struct smb2_file_all_info)) {
+               spin_unlock(&cfids->cfid_list_lock);
                goto oshr_free;
+       }
        if (!smb2_validate_and_copy_iov(
                                le16_to_cpu(qi_rsp->OutputBufferOffset),
                                sizeof(struct smb2_file_all_info),
@@ -289,37 +317,24 @@ int open_cached_dir(unsigned int xid, struct cifs_tcon *tcon,
                                (char *)&cfid->file_all_info))
                cfid->file_all_info_is_valid = true;
 
-       if (!npath[0])
-               dentry = dget(cifs_sb->root);
-       else {
-               dentry = path_to_dentry(cifs_sb, npath);
-               if (IS_ERR(dentry)) {
-                       rc = -ENOENT;
-                       goto oshr_free;
-               }
-       }
-       spin_lock(&cfids->cfid_list_lock);
-       cfid->dentry = dentry;
        cfid->time = jiffies;
-       cfid->has_lease = true;
        spin_unlock(&cfids->cfid_list_lock);
+       /* At this point the directory handle is fully cached */
+       rc = 0;
 
 oshr_free:
-       kfree(utf16_path);
        SMB2_open_free(&rqst[0]);
        SMB2_query_info_free(&rqst[1]);
        free_rsp_buf(resp_buftype[0], rsp_iov[0].iov_base);
        free_rsp_buf(resp_buftype[1], rsp_iov[1].iov_base);
-       spin_lock(&cfids->cfid_list_lock);
-       if (!cfid->has_lease) {
-               if (rc) {
-                       if (cfid->on_list) {
-                               list_del(&cfid->entry);
-                               cfid->on_list = false;
-                               cfids->num_entries--;
-                       }
-                       rc = -ENOENT;
-               } else {
+       if (rc) {
+               spin_lock(&cfids->cfid_list_lock);
+               if (cfid->on_list) {
+                       list_del(&cfid->entry);
+                       cfid->on_list = false;
+                       cfids->num_entries--;
+               }
+               if (cfid->has_lease) {
                        /*
                         * We are guaranteed to have two references at this
                         * point. One for the caller and one for a potential
@@ -327,25 +342,24 @@ oshr_free:
                         * will be closed when the caller closes the cached
                         * handle.
                         */
+                       cfid->has_lease = false;
                        spin_unlock(&cfids->cfid_list_lock);
                        kref_put(&cfid->refcount, smb2_close_cached_fid);
                        goto out;
                }
+               spin_unlock(&cfids->cfid_list_lock);
        }
-       spin_unlock(&cfids->cfid_list_lock);
+out:
        if (rc) {
                if (cfid->is_open)
                        SMB2_close(0, cfid->tcon, cfid->fid.persistent_fid,
                                   cfid->fid.volatile_fid);
                free_cached_dir(cfid);
-               cfid = NULL;
-       }
-out:
-       if (rc == 0) {
+       } else {
                *ret_cfid = cfid;
                atomic_inc(&tcon->num_remote_opens);
        }
-
+       kfree(utf16_path);
        return rc;
 }