btrfs: Don't free tree_root when exiting btrfs_ioctl_get_subvol_info()
PATCH 1/1 btrfs: Don't free tree_root when exiting btrfs_ioctl_get_subvol_info()
[BUG]
When calling BTRF_IOC_GET_SUBVOL_INFO ioctl, we can easily hit the following backtrace:
BUG: kernel NULL pointer dereference, address: 0000000000000024 #PF: supervisor read access in kernel mode #PF: error_code(0x0000) - not-present page PGD 0 P4D 0 Oops: 0000 [#1] SMP PTI CPU: 0 PID: 27421 Comm: python3 Not tainted 5.6.0-rc1+ #539 Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS rel-1.12.0-59-gc9ba527-rebuilt.opensuse.org 04/01/2014 RIP: 0010:btrfs_root_node+0x7/0x30 [btrfs] Call Trace: btrfs_read_lock_root_node+0x1f/0x40 [btrfs] btrfs_search_slot+0x60f/0xa40 [btrfs] btrfs_ioctl+0x11f7/0x30b0 [btrfs] ksys_ioctl+0x82/0xc0 __x64_sys_ioctl+0x11/0x20 do_syscall_64+0x43/0x130 entry_SYSCALL_64_after_hwframe+0x44/0xa9 RIP: 0033:0x7fcb78d43387 ---[ end trace 1c21a7c6c0523b8c ]---
[CAUSE]
We're abusing local @root, it's originally a subvolume root, but in root backref search, it's re-assigned to tree_root.
Then we call "btrfs_put_root(root);" when exiting.
If that @root is reassgined to tree-root, we freed the most important tree, and cause use-after-free.
[FIX]
Don't re-assgiend @root, use fs_info->tree_root directly.
--- fs/btrfs/ioctl.c | 9 ++++----- 1 file changed, 4 insertions(+), 5 deletions(-)
fs/btrfs/ioctl.c +4 -5
@@ -2696,17 +2696,16 @@ static int btrfs_ioctl_get_subvol_info(struct file *file, void __user *argp) subvol_info->rtime.nsec =btrfs_stack_timespec_nsec (&root_item->rtime);if (key.objectid !=BTRFS_FS_TREE_OBJECTID ) { -/* Search root tree for ROOT_BACKREF of this subvolume */ - root = fs_info->tree_root; - key.type =BTRFS_ROOT_BACKREF_KEY ; key.offset =0 ; - ret =btrfs_search_slot (NULL , root, &key, path,0 ,0 ); + +/* Search root tree for ROOT_BACKREF of this subvolume */ + ret =btrfs_search_slot (NULL , fs_info->tree_root, &key, path,0 ,0 );if (ret <0 ) { goto out; }else if (path->slots[0 ] >=btrfs_header_nritems (path->nodes[0 ])) { - ret =btrfs_next_leaf (root, path); + ret =btrfs_next_leaf (fs_info->tree_root, path);if (ret <0 ) { goto out; }else if (ret >0 ) {
Signature · 1 line
2.25.0
Quoted 31 lines · Qu Wenruo
[BUG]
When callingBTRF_IOC_GET_SUBVOL_INFOioctl, we can easily hit the following backtrace:BUG: kernel NULL pointer dereference, address: 0000000000000024 #PF: supervisor read access in kernel mode #PF: error_code(0x0000) - not-present page PGD 0 P4D 0 Oops: 0000 [#1] SMP PTI CPU: 0 PID: 27421 Comm: python3 Not tainted 5.6.0-rc1+ #539 Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS rel- 1.12.0-59-gc9ba527-rebuilt.opensuse.org 04/01/2014 RIP: 0010:btrfs_root_node+0x7/0x30 [btrfs] Call Trace: btrfs_read_lock_root_node+0x1f/0x40 [btrfs] btrfs_search_slot+0x60f/0xa40 [btrfs] btrfs_ioctl+0x11f7/0x30b0 [btrfs] ksys_ioctl+0x82/0xc0 __x64_sys_ioctl+0x11/0x20 do_syscall_64+0x43/0x130 entry_SYSCALL_64_after_hwframe+0x44/0xa9 RIP: 0033:0x7fcb78d43387 ---[ end trace 1c21a7c6c0523b8c ]---[CAUSE]
We're abusing local @root, it's originally a subvolume root, but in root backref search, it's re-assigned totree_root.Then we call "
btrfs_put_root(root);" when exiting.
If that @root is reassgined to tree-root, we freed the most important tree, and cause use-after-free.
Quoted 8 lines
[FIX]
Don't re-assgiend @root, usefs_info->tree_rootdirectly.Reported-by Marcos Paulo de Souza <[email protected]>Fixes 8c319b625e0a ("btrfs: hold a ref on the root in
btrfs_ioctl_get_subvol_info")[To David: please fold the fix into that commit]Signed-off-by Qu Wenruo <[email protected]>
This patch fixed the issue, thanks Qu.
Quoted 36 lines
--- fs/btrfs/ioctl.c | 9 ++++----- 1 file changed, 4 insertions(+), 5 deletions(-)fs/btrfs/ioctl.c +3
-4@@ -2696,17 +2696,16 @@ static int btrfs_ioctl_get_subvol_info(structfile *file, void __user *argp) subvol_info->rtime.nsec = btrfs_stack_timespec_nsec(&root_item- >rtime); if (key.objectid != BTRFS_FS_TREE_OBJECTID) { -/* Search root tree for ROOT_BACKREF of this subvolume */ - root = fs_info->tree_root; - key.type =BTRFS_ROOT_BACKREF_KEY ; key.offset =0 ; - ret =btrfs_search_slot (NULL , root, &key, path,0 ,0 ); + +/* Search root tree for ROOT_BACKREF of this subvolume */ + ret =btrfs_search_slot (NULL , fs_info->tree_root, &key, path,0 ,0 );if (ret <0 ) { goto out; } else if (path->slots[0] >=fs/btrfs/ioctl.c +1
-1btrfs_header_nritems(path->nodes[0])) { - ret =btrfs_next_leaf (root, path); + ret = btrfs_next_leaf(fs_info->tree_root,path);
fs/btrfs/ioctl.c +0
-0if (ret <0 ) { goto out; }else if (ret >0 ) {
Quoted 30 lines · Qu Wenruo
[BUG]
When callingBTRF_IOC_GET_SUBVOL_INFOioctl, we can easily hit the following backtrace:BUG: kernel NULL pointer dereference, address: 0000000000000024 #PF: supervisor read access in kernel mode #PF: error_code(0x0000) - not-present page PGD 0 P4D 0 Oops: 0000 [#1] SMP PTI CPU: 0 PID: 27421 Comm: python3 Not tainted 5.6.0-rc1+ #539 Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS rel-1.12.0-59-gc9ba527-rebuilt.opensuse.org 04/01/2014 RIP: 0010:btrfs_root_node+0x7/0x30 [btrfs] Call Trace: btrfs_read_lock_root_node+0x1f/0x40 [btrfs] btrfs_search_slot+0x60f/0xa40 [btrfs] btrfs_ioctl+0x11f7/0x30b0 [btrfs] ksys_ioctl+0x82/0xc0 __x64_sys_ioctl+0x11/0x20 do_syscall_64+0x43/0x130 entry_SYSCALL_64_after_hwframe+0x44/0xa9 RIP: 0033:0x7fcb78d43387 ---[ end trace 1c21a7c6c0523b8c ]---[CAUSE]
We're abusing local @root, it's originally a subvolume root, but in root backref search, it's re-assigned totree_root.Then we call "
btrfs_put_root(root);" when exiting.
If that @root is reassgined to tree-root, we freed the most important tree, and cause use-after-free.
Quoted 7 lines
[FIX]
Don't re-assgiend @root, usefs_info->tree_rootdirectly.Reported-by Marcos Paulo de Souza <[email protected]>Fixes 8c319b625e0a ("btrfs: hold a ref on the root in btrfs_ioctl_get_subvol_info")[To David: please fold the fix into that commit]Signed-off-by Qu Wenruo <[email protected]>
Ooops
Thanks,
Josef
Quoted 30 lines · Qu Wenruo
[BUG]
When callingBTRF_IOC_GET_SUBVOL_INFOioctl, we can easily hit the following backtrace:BUG: kernel NULL pointer dereference, address: 0000000000000024 #PF: supervisor read access in kernel mode #PF: error_code(0x0000) - not-present page PGD 0 P4D 0 Oops: 0000 [#1] SMP PTI CPU: 0 PID: 27421 Comm: python3 Not tainted 5.6.0-rc1+ #539 Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS rel-1.12.0-59-gc9ba527-rebuilt.opensuse.org 04/01/2014 RIP: 0010:btrfs_root_node+0x7/0x30 [btrfs] Call Trace: btrfs_read_lock_root_node+0x1f/0x40 [btrfs] btrfs_search_slot+0x60f/0xa40 [btrfs] btrfs_ioctl+0x11f7/0x30b0 [btrfs] ksys_ioctl+0x82/0xc0 __x64_sys_ioctl+0x11/0x20 do_syscall_64+0x43/0x130 entry_SYSCALL_64_after_hwframe+0x44/0xa9 RIP: 0033:0x7fcb78d43387 ---[ end trace 1c21a7c6c0523b8c ]---[CAUSE]
We're abusing local @root, it's originally a subvolume root, but in root backref search, it's re-assigned totree_root.Then we call "
btrfs_put_root(root);" when exiting.
If that @root is reassgined to tree-root, we freed the most important tree, and cause use-after-free.
Quoted 6 lines
[FIX]
Don't re-assgiend @root, usefs_info->tree_rootdirectly.Reported-by Marcos Paulo de Souza <[email protected]>Fixes 8c319b625e0a ("btrfs: hold a ref on the root in btrfs_ioctl_get_subvol_info")[To David: please fold the fix into that commit]
Folded, thanks for the report and fix.