]> git.baikalelectronics.ru Git - kernel.git/commitdiff
ext4: allow ext4_get_group_info() to fail
authorTheodore Ts'o <tytso@mit.edu>
Sat, 29 Apr 2023 04:06:28 +0000 (00:06 -0400)
committerGreg Kroah-Hartman <gregkh@linuxfoundation.org>
Wed, 24 May 2023 16:32:33 +0000 (17:32 +0100)
[ Upstream commit 5354b2af34064a4579be8bc0e2f15a7b70f14b5f ]

Previously, ext4_get_group_info() would treat an invalid group number
as BUG(), since in theory it should never happen.  However, if a
malicious attaker (or fuzzer) modifies the superblock via the block
device while it is the file system is mounted, it is possible for
s_first_data_block to get set to a very large number.  In that case,
when calculating the block group of some block number (such as the
starting block of a preallocation region), could result in an
underflow and very large block group number.  Then the BUG_ON check in
ext4_get_group_info() would fire, resutling in a denial of service
attack that can be triggered by root or someone with write access to
the block device.

For a quality of implementation perspective, it's best that even if
the system administrator does something that they shouldn't, that it
will not trigger a BUG.  So instead of BUG'ing, ext4_get_group_info()
will call ext4_error and return NULL.  We also add fallback code in
all of the callers of ext4_get_group_info() that it might NULL.

Also, since ext4_get_group_info() was already borderline to be an
inline function, un-inline it.  The results in a next reduction of the
compiled text size of ext4 by roughly 2k.

Cc: stable@kernel.org
Link: https://lore.kernel.org/r/20230430154311.579720-2-tytso@mit.edu
Reported-by: syzbot+e2efa3efc15a1c9e95c3@syzkaller.appspotmail.com
Link: https://syzkaller.appspot.com/bug?id=69b28112e098b070f639efb356393af3ffec4220
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
Reviewed-by: Jan Kara <jack@suse.cz>
Signed-off-by: Sasha Levin <sashal@kernel.org>
fs/ext4/balloc.c
fs/ext4/ext4.h
fs/ext4/ialloc.c
fs/ext4/mballoc.c
fs/ext4/super.c

index f2c415f31b7557f200355ba012661334eb49141b..a38aa33af08efe00b4e23bdcdcc0f8637e41ff2b 100644 (file)
@@ -319,6 +319,22 @@ static ext4_fsblk_t ext4_valid_block_bitmap_padding(struct super_block *sb,
        return (next_zero_bit < bitmap_size ? next_zero_bit : 0);
 }
 
+struct ext4_group_info *ext4_get_group_info(struct super_block *sb,
+                                           ext4_group_t group)
+{
+        struct ext4_group_info **grp_info;
+        long indexv, indexh;
+
+        if (unlikely(group >= EXT4_SB(sb)->s_groups_count)) {
+                ext4_error(sb, "invalid group %u", group);
+                return NULL;
+        }
+        indexv = group >> (EXT4_DESC_PER_BLOCK_BITS(sb));
+        indexh = group & ((EXT4_DESC_PER_BLOCK(sb)) - 1);
+        grp_info = sbi_array_rcu_deref(EXT4_SB(sb), s_group_info, indexv);
+        return grp_info[indexh];
+}
+
 /*
  * Return the block number which was discovered to be invalid, or 0 if
  * the block bitmap is valid.
@@ -393,7 +409,7 @@ static int ext4_validate_block_bitmap(struct super_block *sb,
 
        if (buffer_verified(bh))
                return 0;
-       if (EXT4_MB_GRP_BBITMAP_CORRUPT(grp))
+       if (!grp || EXT4_MB_GRP_BBITMAP_CORRUPT(grp))
                return -EFSCORRUPTED;
 
        ext4_lock_group(sb, block_group);
index a2bc440743ae48fde20dfce1701cc46c6fcdd8e7..2156d2a1a7700a58ed855fc3a4bd71ef7864dbdf 100644 (file)
@@ -2716,6 +2716,8 @@ extern void ext4_check_blocks_bitmap(struct super_block *);
 extern struct ext4_group_desc * ext4_get_group_desc(struct super_block * sb,
                                                    ext4_group_t block_group,
                                                    struct buffer_head ** bh);
+extern struct ext4_group_info *ext4_get_group_info(struct super_block *sb,
+                                                  ext4_group_t group);
 extern int ext4_should_retry_alloc(struct super_block *sb, int *retries);
 
 extern struct buffer_head *ext4_read_block_bitmap_nowait(struct super_block *sb,
@@ -3322,19 +3324,6 @@ static inline void ext4_isize_set(struct ext4_inode *raw_inode, loff_t i_size)
        raw_inode->i_size_high = cpu_to_le32(i_size >> 32);
 }
 
-static inline
-struct ext4_group_info *ext4_get_group_info(struct super_block *sb,
-                                           ext4_group_t group)
-{
-        struct ext4_group_info **grp_info;
-        long indexv, indexh;
-        BUG_ON(group >= EXT4_SB(sb)->s_groups_count);
-        indexv = group >> (EXT4_DESC_PER_BLOCK_BITS(sb));
-        indexh = group & ((EXT4_DESC_PER_BLOCK(sb)) - 1);
-        grp_info = sbi_array_rcu_deref(EXT4_SB(sb), s_group_info, indexv);
-        return grp_info[indexh];
-}
-
 /*
  * Reading s_groups_count requires using smp_rmb() afterwards.  See
  * the locking protocol documented in the comments of ext4_group_add()
index e9bc46684106b5cb2c3922711f6008584535f882..e09c74927a43042df8d2d437e5a074be441b0a1b 100644 (file)
@@ -91,7 +91,7 @@ static int ext4_validate_inode_bitmap(struct super_block *sb,
 
        if (buffer_verified(bh))
                return 0;
-       if (EXT4_MB_GRP_IBITMAP_CORRUPT(grp))
+       if (!grp || EXT4_MB_GRP_IBITMAP_CORRUPT(grp))
                return -EFSCORRUPTED;
 
        ext4_lock_group(sb, block_group);
@@ -293,7 +293,7 @@ void ext4_free_inode(handle_t *handle, struct inode *inode)
        }
        if (!(sbi->s_mount_state & EXT4_FC_REPLAY)) {
                grp = ext4_get_group_info(sb, block_group);
-               if (unlikely(EXT4_MB_GRP_IBITMAP_CORRUPT(grp))) {
+               if (!grp || unlikely(EXT4_MB_GRP_IBITMAP_CORRUPT(grp))) {
                        fatal = -EFSCORRUPTED;
                        goto error_return;
                }
@@ -1047,7 +1047,7 @@ got_group:
                         * Skip groups with already-known suspicious inode
                         * tables
                         */
-                       if (EXT4_MB_GRP_IBITMAP_CORRUPT(grp))
+                       if (!grp || EXT4_MB_GRP_IBITMAP_CORRUPT(grp))
                                goto next_group;
                }
 
@@ -1185,6 +1185,10 @@ got:
 
                if (!(sbi->s_mount_state & EXT4_FC_REPLAY)) {
                        grp = ext4_get_group_info(sb, group);
+                       if (!grp) {
+                               err = -EFSCORRUPTED;
+                               goto out;
+                       }
                        down_read(&grp->alloc_sem); /*
                                                     * protect vs itable
                                                     * lazyinit
@@ -1528,7 +1532,7 @@ int ext4_init_inode_table(struct super_block *sb, ext4_group_t group,
        }
 
        gdp = ext4_get_group_desc(sb, group, &group_desc_bh);
-       if (!gdp)
+       if (!gdp || !grp)
                goto out;
 
        /*
index b1b63fc84ddbaeb6b8ea6c4d192d336fb9e3da4b..09bb1dae40cf0ebbd2d3f8fc11daa6cb8ad06241 100644 (file)
@@ -745,6 +745,8 @@ static int __mb_check_buddy(struct ext4_buddy *e4b, char *file,
        MB_CHECK_ASSERT(e4b->bd_info->bb_fragments == fragments);
 
        grp = ext4_get_group_info(sb, e4b->bd_group);
+       if (!grp)
+               return NULL;
        list_for_each(cur, &grp->bb_prealloc_list) {
                ext4_group_t groupnr;
                struct ext4_prealloc_space *pa;
@@ -1060,9 +1062,9 @@ mb_set_largest_free_order(struct super_block *sb, struct ext4_group_info *grp)
 
 static noinline_for_stack
 void ext4_mb_generate_buddy(struct super_block *sb,
-                               void *buddy, void *bitmap, ext4_group_t group)
+                           void *buddy, void *bitmap, ext4_group_t group,
+                           struct ext4_group_info *grp)
 {
-       struct ext4_group_info *grp = ext4_get_group_info(sb, group);
        struct ext4_sb_info *sbi = EXT4_SB(sb);
        ext4_grpblk_t max = EXT4_CLUSTERS_PER_GROUP(sb);
        ext4_grpblk_t i = 0;
@@ -1183,6 +1185,8 @@ static int ext4_mb_init_cache(struct page *page, char *incore, gfp_t gfp)
                        break;
 
                grinfo = ext4_get_group_info(sb, group);
+               if (!grinfo)
+                       continue;
                /*
                 * If page is uptodate then we came here after online resize
                 * which added some new uninitialized group info structs, so
@@ -1248,6 +1252,10 @@ static int ext4_mb_init_cache(struct page *page, char *incore, gfp_t gfp)
                                group, page->index, i * blocksize);
                        trace_ext4_mb_buddy_bitmap_load(sb, group);
                        grinfo = ext4_get_group_info(sb, group);
+                       if (!grinfo) {
+                               err = -EFSCORRUPTED;
+                               goto out;
+                       }
                        grinfo->bb_fragments = 0;
                        memset(grinfo->bb_counters, 0,
                               sizeof(*grinfo->bb_counters) *
@@ -1258,7 +1266,7 @@ static int ext4_mb_init_cache(struct page *page, char *incore, gfp_t gfp)
                        ext4_lock_group(sb, group);
                        /* init the buddy */
                        memset(data, 0xff, blocksize);
-                       ext4_mb_generate_buddy(sb, data, incore, group);
+                       ext4_mb_generate_buddy(sb, data, incore, group, grinfo);
                        ext4_unlock_group(sb, group);
                        incore = NULL;
                } else {
@@ -1372,6 +1380,9 @@ int ext4_mb_init_group(struct super_block *sb, ext4_group_t group, gfp_t gfp)
        might_sleep();
        mb_debug(sb, "init group %u\n", group);
        this_grp = ext4_get_group_info(sb, group);
+       if (!this_grp)
+               return -EFSCORRUPTED;
+
        /*
         * This ensures that we don't reinit the buddy cache
         * page which map to the group from which we are already
@@ -1446,6 +1457,8 @@ ext4_mb_load_buddy_gfp(struct super_block *sb, ext4_group_t group,
 
        blocks_per_page = PAGE_SIZE / sb->s_blocksize;
        grp = ext4_get_group_info(sb, group);
+       if (!grp)
+               return -EFSCORRUPTED;
 
        e4b->bd_blkbits = sb->s_blocksize_bits;
        e4b->bd_info = grp;
@@ -2162,6 +2175,8 @@ int ext4_mb_find_by_goal(struct ext4_allocation_context *ac,
        struct ext4_group_info *grp = ext4_get_group_info(ac->ac_sb, group);
        struct ext4_free_extent ex;
 
+       if (!grp)
+               return -EFSCORRUPTED;
        if (!(ac->ac_flags & (EXT4_MB_HINT_TRY_GOAL | EXT4_MB_HINT_GOAL_ONLY)))
                return 0;
        if (grp->bb_free == 0)
@@ -2386,7 +2401,7 @@ static bool ext4_mb_good_group(struct ext4_allocation_context *ac,
 
        BUG_ON(cr < 0 || cr >= 4);
 
-       if (unlikely(EXT4_MB_GRP_BBITMAP_CORRUPT(grp)))
+       if (unlikely(EXT4_MB_GRP_BBITMAP_CORRUPT(grp) || !grp))
                return false;
 
        free = grp->bb_free;
@@ -2455,6 +2470,8 @@ static int ext4_mb_good_group_nolock(struct ext4_allocation_context *ac,
        ext4_grpblk_t free;
        int ret = 0;
 
+       if (!grp)
+               return -EFSCORRUPTED;
        if (sbi->s_mb_stats)
                atomic64_inc(&sbi->s_bal_cX_groups_considered[ac->ac_criteria]);
        if (should_lock) {
@@ -2535,7 +2552,7 @@ ext4_group_t ext4_mb_prefetch(struct super_block *sb, ext4_group_t group,
                 * prefetch once, so we avoid getblk() call, which can
                 * be expensive.
                 */
-               if (!EXT4_MB_GRP_TEST_AND_SET_READ(grp) &&
+               if (gdp && grp && !EXT4_MB_GRP_TEST_AND_SET_READ(grp) &&
                    EXT4_MB_GRP_NEED_INIT(grp) &&
                    ext4_free_group_clusters(sb, gdp) > 0 &&
                    !(ext4_has_group_desc_csum(sb) &&
@@ -2579,7 +2596,7 @@ void ext4_mb_prefetch_fini(struct super_block *sb, ext4_group_t group,
                group--;
                grp = ext4_get_group_info(sb, group);
 
-               if (EXT4_MB_GRP_NEED_INIT(grp) &&
+               if (grp && gdp && EXT4_MB_GRP_NEED_INIT(grp) &&
                    ext4_free_group_clusters(sb, gdp) > 0 &&
                    !(ext4_has_group_desc_csum(sb) &&
                      (gdp->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT)))) {
@@ -2838,6 +2855,8 @@ static int ext4_mb_seq_groups_show(struct seq_file *seq, void *v)
                sizeof(struct ext4_group_info);
 
        grinfo = ext4_get_group_info(sb, group);
+       if (!grinfo)
+               return 0;
        /* Load the group info in memory only if not already loaded. */
        if (unlikely(EXT4_MB_GRP_NEED_INIT(grinfo))) {
                err = ext4_mb_load_buddy(sb, group, &e4b);
@@ -2848,7 +2867,7 @@ static int ext4_mb_seq_groups_show(struct seq_file *seq, void *v)
                buddy_loaded = 1;
        }
 
-       memcpy(&sg, ext4_get_group_info(sb, group), i);
+       memcpy(&sg, grinfo, i);
 
        if (buddy_loaded)
                ext4_mb_unload_buddy(&e4b);
@@ -3210,8 +3229,12 @@ static int ext4_mb_init_backend(struct super_block *sb)
 
 err_freebuddy:
        cachep = get_groupinfo_cache(sb->s_blocksize_bits);
-       while (i-- > 0)
-               kmem_cache_free(cachep, ext4_get_group_info(sb, i));
+       while (i-- > 0) {
+               struct ext4_group_info *grp = ext4_get_group_info(sb, i);
+
+               if (grp)
+                       kmem_cache_free(cachep, grp);
+       }
        i = sbi->s_group_info_size;
        rcu_read_lock();
        group_info = rcu_dereference(sbi->s_group_info);
@@ -3525,6 +3548,8 @@ int ext4_mb_release(struct super_block *sb)
                for (i = 0; i < ngroups; i++) {
                        cond_resched();
                        grinfo = ext4_get_group_info(sb, i);
+                       if (!grinfo)
+                               continue;
                        mb_group_bb_bitmap_free(grinfo);
                        ext4_lock_group(sb, i);
                        count = ext4_mb_cleanup_pa(grinfo);
@@ -4454,6 +4479,8 @@ static void ext4_mb_generate_from_freelist(struct super_block *sb, void *bitmap,
        struct ext4_free_data *entry;
 
        grp = ext4_get_group_info(sb, group);
+       if (!grp)
+               return;
        n = rb_first(&(grp->bb_free_root));
 
        while (n) {
@@ -4481,6 +4508,9 @@ void ext4_mb_generate_from_pa(struct super_block *sb, void *bitmap,
        int preallocated = 0;
        int len;
 
+       if (!grp)
+               return;
+
        /* all form of preallocation discards first load group,
         * so the only competing code is preallocation use.
         * we don't need any locking here
@@ -4672,6 +4702,8 @@ ext4_mb_new_inode_pa(struct ext4_allocation_context *ac)
 
        ei = EXT4_I(ac->ac_inode);
        grp = ext4_get_group_info(sb, ac->ac_b_ex.fe_group);
+       if (!grp)
+               return;
 
        pa->pa_obj_lock = &ei->i_prealloc_lock;
        pa->pa_inode = ac->ac_inode;
@@ -4725,6 +4757,8 @@ ext4_mb_new_group_pa(struct ext4_allocation_context *ac)
        atomic_add(pa->pa_free, &EXT4_SB(sb)->s_mb_preallocated);
 
        grp = ext4_get_group_info(sb, ac->ac_b_ex.fe_group);
+       if (!grp)
+               return;
        lg = ac->ac_lg;
        BUG_ON(lg == NULL);
 
@@ -4853,6 +4887,8 @@ ext4_mb_discard_group_preallocations(struct super_block *sb,
        int err;
        int free = 0;
 
+       if (!grp)
+               return 0;
        mb_debug(sb, "discard preallocation for group %u\n", group);
        if (list_empty(&grp->bb_prealloc_list))
                goto out_dbg;
@@ -5090,6 +5126,9 @@ static inline void ext4_mb_show_pa(struct super_block *sb)
                struct ext4_prealloc_space *pa;
                ext4_grpblk_t start;
                struct list_head *cur;
+
+               if (!grp)
+                       continue;
                ext4_lock_group(sb, i);
                list_for_each(cur, &grp->bb_prealloc_list) {
                        pa = list_entry(cur, struct ext4_prealloc_space,
@@ -5895,6 +5934,7 @@ static void ext4_mb_clear_bb(handle_t *handle, struct inode *inode,
        struct buffer_head *bitmap_bh = NULL;
        struct super_block *sb = inode->i_sb;
        struct ext4_group_desc *gdp;
+       struct ext4_group_info *grp;
        unsigned int overflow;
        ext4_grpblk_t bit;
        struct buffer_head *gd_bh;
@@ -5920,8 +5960,8 @@ do_more:
        overflow = 0;
        ext4_get_group_no_and_offset(sb, block, &block_group, &bit);
 
-       if (unlikely(EXT4_MB_GRP_BBITMAP_CORRUPT(
-                       ext4_get_group_info(sb, block_group))))
+       grp = ext4_get_group_info(sb, block_group);
+       if (unlikely(!grp || EXT4_MB_GRP_BBITMAP_CORRUPT(grp)))
                return;
 
        /*
@@ -6523,6 +6563,8 @@ int ext4_trim_fs(struct super_block *sb, struct fstrim_range *range)
 
        for (group = first_group; group <= last_group; group++) {
                grp = ext4_get_group_info(sb, group);
+               if (!grp)
+                       continue;
                /* We only do this if the grp has never been initialized */
                if (unlikely(EXT4_MB_GRP_NEED_INIT(grp))) {
                        ret = ext4_mb_init_group(sb, group, GFP_NOFS);
index c56192a84bf41af6beda726e85c0bbf9f8e3a6b4..bd2e803d653f7bed6bdb77d0f2c4089c222c4223 100644 (file)
@@ -1049,6 +1049,8 @@ void ext4_mark_group_bitmap_corrupted(struct super_block *sb,
        struct ext4_group_desc *gdp = ext4_get_group_desc(sb, group, NULL);
        int ret;
 
+       if (!grp || !gdp)
+               return;
        if (flags & EXT4_GROUP_INFO_BBITMAP_CORRUPT) {
                ret = ext4_test_and_set_bit(EXT4_GROUP_INFO_BBITMAP_CORRUPT_BIT,
                                            &grp->bb_state);