hugetlbfs: initialize shared policy as part of inode allocation
authorMike Kravetz <mike.kravetz@oracle.com>
Fri, 31 Mar 2017 22:12:01 +0000 (15:12 -0700)
committerGreg Kroah-Hartman <gregkh@linuxfoundation.org>
Sun, 8 Oct 2017 08:26:09 +0000 (10:26 +0200)
[ Upstream commit 4742a35d9de745e867405b4311e1aac412f0ace1 ]

Any time after inode allocation, destroy_inode can be called.  The
hugetlbfs inode contains a shared_policy structure, and
mpol_free_shared_policy is unconditionally called as part of
hugetlbfs_destroy_inode.  Initialize the policy as part of inode
allocation so that any quick (error path) calls to destroy_inode will be
handed an initialized policy.

syzkaller fuzzer found this bug, that resulted in the following:

    BUG: KASAN: user-memory-access in atomic_inc
    include/asm-generic/atomic-instrumented.h:87 [inline] at addr
    000000131730bd7a
    BUG: KASAN: user-memory-access in __lock_acquire+0x21a/0x3a80
    kernel/locking/lockdep.c:3239 at addr 000000131730bd7a
    Write of size 4 by task syz-executor6/14086
    CPU: 3 PID: 14086 Comm: syz-executor6 Not tainted 4.11.0-rc3+ #364
    Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS Bochs 01/01/2011
    Call Trace:
     atomic_inc include/asm-generic/atomic-instrumented.h:87 [inline]
     __lock_acquire+0x21a/0x3a80 kernel/locking/lockdep.c:3239
     lock_acquire+0x1ee/0x590 kernel/locking/lockdep.c:3762
     __raw_write_lock include/linux/rwlock_api_smp.h:210 [inline]
     _raw_write_lock+0x33/0x50 kernel/locking/spinlock.c:295
     mpol_free_shared_policy+0x43/0xb0 mm/mempolicy.c:2536
     hugetlbfs_destroy_inode+0xca/0x120 fs/hugetlbfs/inode.c:952
     alloc_inode+0x10d/0x180 fs/inode.c:216
     new_inode_pseudo+0x69/0x190 fs/inode.c:889
     new_inode+0x1c/0x40 fs/inode.c:918
     hugetlbfs_get_inode+0x40/0x420 fs/hugetlbfs/inode.c:734
     hugetlb_file_setup+0x329/0x9f0 fs/hugetlbfs/inode.c:1282
     newseg+0x422/0xd30 ipc/shm.c:575
     ipcget_new ipc/util.c:285 [inline]
     ipcget+0x21e/0x580 ipc/util.c:639
     SYSC_shmget ipc/shm.c:673 [inline]
     SyS_shmget+0x158/0x230 ipc/shm.c:657
     entry_SYSCALL_64_fastpath+0x1f/0xc2

Analysis provided by Tetsuo Handa <penguin-kernel@I-love.SAKURA.ne.jp>

Link: http://lkml.kernel.org/r/1490477850-7944-1-git-send-email-mike.kravetz@oracle.com
Signed-off-by: Mike Kravetz <mike.kravetz@oracle.com>
Reported-by: Dmitry Vyukov <dvyukov@google.com>
Acked-by: Hillf Danton <hillf.zj@alibaba-inc.com>
Cc: Tetsuo Handa <penguin-kernel@I-love.SAKURA.ne.jp>
Cc: Michal Hocko <mhocko@suse.com>
Cc: Dave Hansen <dave.hansen@linux.intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Signed-off-by: Sasha Levin <alexander.levin@verizon.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
fs/hugetlbfs/inode.c

index 704fa0b..2c2f182 100644 (file)
@@ -695,14 +695,11 @@ static struct inode *hugetlbfs_get_root(struct super_block *sb,
 
        inode = new_inode(sb);
        if (inode) {
-               struct hugetlbfs_inode_info *info;
                inode->i_ino = get_next_ino();
                inode->i_mode = S_IFDIR | config->mode;
                inode->i_uid = config->uid;
                inode->i_gid = config->gid;
                inode->i_atime = inode->i_mtime = inode->i_ctime = current_time(inode);
-               info = HUGETLBFS_I(inode);
-               mpol_shared_policy_init(&info->policy, NULL);
                inode->i_op = &hugetlbfs_dir_inode_operations;
                inode->i_fop = &simple_dir_operations;
                /* directory inodes start off with i_nlink == 2 (for "." entry) */
@@ -733,7 +730,6 @@ static struct inode *hugetlbfs_get_inode(struct super_block *sb,
 
        inode = new_inode(sb);
        if (inode) {
-               struct hugetlbfs_inode_info *info;
                inode->i_ino = get_next_ino();
                inode_init_owner(inode, dir, mode);
                lockdep_set_class(&inode->i_mapping->i_mmap_rwsem,
@@ -741,15 +737,6 @@ static struct inode *hugetlbfs_get_inode(struct super_block *sb,
                inode->i_mapping->a_ops = &hugetlbfs_aops;
                inode->i_atime = inode->i_mtime = inode->i_ctime = current_time(inode);
                inode->i_mapping->private_data = resv_map;
-               info = HUGETLBFS_I(inode);
-               /*
-                * The policy is initialized here even if we are creating a
-                * private inode because initialization simply creates an
-                * an empty rb tree and calls rwlock_init(), later when we
-                * call mpol_free_shared_policy() it will just return because
-                * the rb tree will still be empty.
-                */
-               mpol_shared_policy_init(&info->policy, NULL);
                switch (mode & S_IFMT) {
                default:
                        init_special_inode(inode, mode, dev);
@@ -937,6 +924,18 @@ static struct inode *hugetlbfs_alloc_inode(struct super_block *sb)
                hugetlbfs_inc_free_inodes(sbinfo);
                return NULL;
        }
+
+       /*
+        * Any time after allocation, hugetlbfs_destroy_inode can be called
+        * for the inode.  mpol_free_shared_policy is unconditionally called
+        * as part of hugetlbfs_destroy_inode.  So, initialize policy here
+        * in case of a quick call to destroy.
+        *
+        * Note that the policy is initialized even if we are creating a
+        * private inode.  This simplifies hugetlbfs_destroy_inode.
+        */
+       mpol_shared_policy_init(&p->policy, NULL);
+
        return &p->vfs_inode;
 }