[PATCH v2] ocfs2: validate dx entry list counts on read
ZhengYuan Huang <[email protected]>
| Newsgroups | dev.linux.lists.ocfs2-devel,org.kernel.vger.linux-kernel |
|---|---|
| Message-ID | <[email protected]> |
[BUG] A corrupt indexed directory dx leaf can carry de_num_used larger than de_count. When unlinkat() removes an indexed entry, that corrupt count reaches ocfs2_dx_list_remove_entry() and drives an oversized memmove(): BUG: KASAN: use-after-free in ocfs2_dx_list_remove_entry fs/ocfs2/dir.c:1246 [inline] BUG: KASAN: use-after-free in ocfs2_delete_entry_dx fs/ocfs2/dir.c:1363 [inline] BUG: KASAN: use-after-free in ocfs2_delete_entry+0xb47/0x1070 fs/ocfs2/dir.c:1417 Read of size 401648 at addr ffff888020b8bd20 by task syz.0.1/347 Call Trace: __dump_stack lib/dump_stack.c:94 [inline] dump_stack_lvl+0xbe/0x130 lib/dump_stack.c:120 print_address_description mm/kasan/report.c:378 [inline] print_report+0xd1/0x650 mm/kasan/report.c:482 kasan_report+0xfb/0x140 mm/kasan/report.c:595 check_region_inline mm/kasan/generic.c:186 [inline] kasan_check_range+0x11c/0x200 mm/kasan/generic.c:200 __asan_memmove+0x23/0x80 mm/kasan/shadow.c:94 ocfs2_dx_list_remove_entry fs/ocfs2/dir.c:1246 [inline] ocfs2_delete_entry_dx fs/ocfs2/dir.c:1363 [inline] ocfs2_delete_entry+0xb47/0x1070 fs/ocfs2/dir.c:1417 ocfs2_unlink+0xb79/0x1900 fs/ocfs2/namei.c:991 vfs_unlink+0x2c2/0x900 fs/namei.c:4673 do_unlinkat+0x474/0x680 fs/namei.c:4737 __do_sys_unlinkat fs/namei.c:4778 [inline] __se_sys_unlinkat fs/namei.c:4771 [inline] __x64_sys_unlinkat+0xae/0x130 fs/namei.c:4771 ... [CAUSE] DX leaf and inline dx root blocks validate signatures and ECC, but they do not validate the geometry of struct ocfs2_dx_entry_list. Corrupt de_count/de_num_used values therefore survive block read and are later used as traversal bounds and memmove/memset lengths in lookup, delete and rebalance paths. In the unlink case, lookup can still find the target entry before the bad bound matters, and the same corrupt de_num_used then reaches ocfs2_dx_list_remove_entry(). [FIX] Validate dx entry-list geometry in the metadata read callbacks. de_count must match the block layout for the current superblock, and de_num_used must not exceed de_count. Do this for both dx leaves and inline dx roots so corrupted metadata is rejected before any caller walks or rewrites de_entries[]. Signed-off-by: ZhengYuan Huang <[email protected]> --- No Fixes tag is included because this is a missing metadata validation check, not a regression attributable to a single introducing commit. v2: - Reframe the bug report around the unlinkat-triggered UAF - Keep the code change unchanged because it fixes the same root cause --- fs/ocfs2/dir.c | 40 ++++++++++++++++++++++++++++++++++++++-- 1 file changed, 38 insertions(+), 2 deletions(-) diff --git a/fs/ocfs2/dir.c b/fs/ocfs2/dir.c index 8e6b03238327..a89634e20b32 100644 --- a/fs/ocfs2/dir.c +++ b/fs/ocfs2/dir.c @@ -578,6 +578,29 @@ static int ocfs2_read_dir_block_direct(struct inode *dir, u64 phys, return ret; } +static int ocfs2_validate_dx_entry_list(struct super_block *sb, u64 blkno, + struct ocfs2_dx_entry_list *entry_list, + unsigned int expected, + const char *owner) +{ + unsigned int count = le16_to_cpu(entry_list->de_count); + unsigned int num_used = le16_to_cpu(entry_list->de_num_used); + + if (count != expected) + return ocfs2_error(sb, + "%s # %llu has invalid de_count %u (expected %u)\n", + owner, (unsigned long long)blkno, count, + expected); + + if (num_used > count) + return ocfs2_error(sb, + "%s # %llu has invalid de_num_used %u (de_count %u)\n", + owner, (unsigned long long)blkno, num_used, + count); + + return 0; +} + static int ocfs2_validate_dx_root(struct super_block *sb, struct buffer_head *bh) { @@ -604,7 +625,14 @@ static int ocfs2_validate_dx_root(struct super_block *sb, goto bail; } - if (!(dx_root->dr_flags & OCFS2_DX_FLAG_INLINE)) { + if (dx_root->dr_flags & OCFS2_DX_FLAG_INLINE) { + ret = ocfs2_validate_dx_entry_list(sb, bh->b_blocknr, + &dx_root->dr_entries, + ocfs2_dx_entries_per_root(sb), + "Dir Index Root"); + if (ret) + goto bail; + } else { struct ocfs2_extent_list *el = &dx_root->dr_list; if (le16_to_cpu(el->l_count) != ocfs2_extent_recs_per_dx_root(sb)) { @@ -660,13 +688,21 @@ static int ocfs2_validate_dx_leaf(struct super_block *sb, mlog(ML_ERROR, "Checksum failed for dir index leaf block %llu\n", (unsigned long long)bh->b_blocknr); - return ret; + goto bail; } if (!OCFS2_IS_VALID_DX_LEAF(dx_leaf)) { ret = ocfs2_error(sb, "Dir Index Leaf has bad signature %.*s\n", 7, dx_leaf->dl_signature); + goto bail; } + ret = ocfs2_validate_dx_entry_list(sb, bh->b_blocknr, + &dx_leaf->dl_list, + ocfs2_dx_entries_per_leaf(sb), + "Dir Index Leaf"); + +bail: + return ret; } -- 2.43.0