btrfs: add btrfs_inode_disk_i_size() helper to prevent torn reads of disk_i_size
PATCH 1/1 btrfs: add btrfs_inode_disk_i_size() helper to prevent torn reads of disk_i_size
From: zzzccc <[email protected]>
btrfs_inode::disk_i_size is a u64 field updated under inode->lock by btrfs_inode_safe_disk_i_size_write(), but several read sites access it without holding that lock. On 64-bit platforms this is fine because aligned u64 loads are architecturally atomic, but on 32-bit platforms a u64 load is performed as two 32-bit loads which can tear if a concurrent write updates both halves.
A torn read of disk_i_size is dangerous in the metadata-serialization paths (fill_inode_item, fill_stack_inode_item) because the torn value gets persisted to the B-tree on disk. After a crash, fsck / mount would see a file size that never existed:
- If the torn value is too large, stale data beyond the real EOF is exposed (information leak).
- If the torn value is too small (e.g. zero), file data is silently lost.
--- fs/btrfs/btrfs_inode.h | 24 ++++++++++++++++++++++++ fs/btrfs/delayed-inode.c | 2 +- fs/btrfs/file.c | 2 +- fs/btrfs/inode.c | 6 +++--- 4 files changed, 29 insertions(+), 5 deletions(-)
fs/btrfs/btrfs_inode.h +24 -0
@@ -418,6 +418,30 @@ static inline void btrfs_i_size_write(struct btrfs_inode *inode, u64 size) inode->disk_i_size = size; } +/* +* Get the on-disk file size safely without holding inode->lock. +* +* disk_i_size is protected by inode->lock when being written (see +* btrfs_inode_safe_disk_i_size_write()), but several read sites access +* it without that lock. On 64-bit platforms a plain READ_ONCE() is +* sufficient because aligned u64 loads are atomic. On 32-bit platforms +* a u64 load can tear, so we take the spinlock to guarantee a consistent +* snapshot. +*/ +static inline u64 btrfs_inode_disk_i_size (struct btrfs_inode *inode) +{ +#if BITS_PER_LONG ==32 +u64 size; + +spin_lock (&inode->lock); + size = inode->disk_i_size; +spin_unlock (&inode->lock); +return size; +#else +return READ_ONCE (inode->disk_i_size); +#endif +} +static inline bool btrfs_is_free_space_inode (const struct btrfs_inode *inode) {return test_bit (BTRFS_INODE_FREE_SPACE_INODE , &inode->runtime_flags);
fs/btrfs/delayed-inode.c +1 -1
@@ -1841,7 +1841,7 @@ static void fill_stack_inode_item(struct btrfs_trans_handle *trans,btrfs_set_stack_inode_uid (inode_item,i_uid_read (vfs_inode));btrfs_set_stack_inode_gid (inode_item,i_gid_read (vfs_inode)); -btrfs_set_stack_inode_size (inode_item, inode->disk_i_size); +btrfs_set_stack_inode_size (inode_item,btrfs_inode_disk_i_size (inode));btrfs_set_stack_inode_mode (inode_item, vfs_inode->i_mode);btrfs_set_stack_inode_nlink (inode_item, vfs_inode->i_nlink);btrfs_set_stack_inode_nbytes (inode_item,inode_get_bytes (vfs_inode));
fs/btrfs/file.c +1 -1
@@ -178,7 +178,7 @@ int btrfs_drop_extents(struct btrfs_trans_handle *trans,if (args->drop_cache)btrfs_drop_extent_map_range (inode, args->start, args->end -1 , false); -if (data_race (args->start >= inode->disk_i_size) && !args->replace_extent) +if (args->start >=btrfs_inode_disk_i_size (inode) && !args->replace_extent) modify_tree =0 ; update_refs = (btrfs_root_id (root) !=BTRFS_TREE_LOG_OBJECTID );
fs/btrfs/inode.c +3 -3
@@ -837,7 +837,7 @@ static inline void inode_should_defrag(struct btrfs_inode *inode, {/* If this is a small write inside eof, kick off a defrag */ if (num_bytes < small_write && - (start >0 || end +1 < inode->disk_i_size)) + (start >0 || end +1 <btrfs_inode_disk_i_size (inode)))btrfs_add_inode_defrag (inode, small_write); }
@@ -4264,7 +4264,7 @@ static void fill_inode_item(struct btrfs_trans_handle *trans,btrfs_set_inode_uid (leaf, item,i_uid_read (inode));btrfs_set_inode_gid (leaf, item,i_gid_read (inode)); -btrfs_set_inode_size (leaf, item,BTRFS_I (inode)->disk_i_size); +btrfs_set_inode_size (leaf, item,btrfs_inode_disk_i_size (BTRFS_I (inode)));btrfs_set_inode_mode (leaf, item, inode->i_mode);btrfs_set_inode_nlink (leaf, item, inode->i_nlink);
@@ -5455,7 +5455,7 @@ static int btrfs_setsize(struct inode *inode, struct iattr *attr) ret2 =btrfs_wait_ordered_range (BTRFS_I (inode),0 , (u64)-1 );if (ret2)return ret2; -i_size_write (inode,BTRFS_I (inode)->disk_i_size); +i_size_write (inode,btrfs_inode_disk_i_size (BTRFS_I (inode))); } }
Signature · 1 line
2.34.1