Message ID | 330f0214f1d8150e25dc609477e89534e3da961a.1728527388.git.wqu@suse.com (mailing list archive) |
---|---|
State | New, archived |
Headers | show |
Series | btrfs: handle NULL as device path for btrfs_scan_one_device() | expand |
Looks good to me,
Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
On Thu, Oct 10, 2024 at 01:00:57PM +1030, Qu Wenruo wrote: > [BUG] > Syzbot reports a crash when mounting a btrfs with NULL as @path for > btrfs_scan_one_device(): > > Oops: general protection fault, probably for non-canonical address 0xdffffc0000000000: 0000 [#1] PREEMPT SMP KASAN PTI > KASAN: null-ptr-deref in range [0x0000000000000000-0x0000000000000007] > CPU: 0 UID: 0 PID: 5235 Comm: syz-executor338 Not tainted 6.12.0-rc2-next-20241008-syzkaller #0 > RIP: 0010:strlen+0x2c/0x70 lib/string.c:402 > Call Trace: > <TASK> > getname_kernel+0x1d/0x2f0 fs/namei.c:232 > kern_path+0x1d/0x50 fs/namei.c:2716 > is_good_dev_path fs/btrfs/volumes.c:760 [inline] > btrfs_scan_one_device+0x19e/0xd90 fs/btrfs/volumes.c:1484 > btrfs_get_tree_super fs/btrfs/super.c:1841 [inline] > btrfs_get_tree+0x30e/0x1920 fs/btrfs/super.c:2114 > vfs_get_tree+0x90/0x2b0 fs/super.c:1800 > fc_mount+0x1b/0xb0 fs/namespace.c:1231 > btrfs_get_tree_subvol fs/btrfs/super.c:2077 [inline] > btrfs_get_tree+0x652/0x1920 fs/btrfs/super.c:2115 > vfs_get_tree+0x90/0x2b0 fs/super.c:1800 > vfs_cmd_create+0xa0/0x1f0 fs/fsopen.c:225 > __do_sys_fsconfig fs/fsopen.c:472 [inline] > __se_sys_fsconfig+0xa1f/0xf70 fs/fsopen.c:344 > do_syscall_x64 arch/x86/entry/common.c:52 [inline] > do_syscall_64+0xf3/0x230 arch/x86/entry/common.c:83 > entry_SYSCALL_64_after_hwframe+0x77/0x7f > RIP: 0033:0x7fe8c78542a9 > ---[ end trace 0000000000000000 ]--- > > [CAUSE] > The reproducer passes NULL as the device path for > btrfs_scan_one_device(). > > Normally such case will be rejected by bdev_file_open_by_path(), which > will return -EINVAL and we reject the mount. > > But since commit ("btrfs: avoid unnecessary device path update for the > same device") and commit ("btrfs: canonicalize the device path before > adding") will do extra kern_path() lookup before > bdev_file_open_by_path(). > > Unlike bdev_file_open_by_path(), kern_path() doesn't do the NULL pointer > check and triggers the above strlen() on NULL pointer. > > [FIX] > For function is_good_dev_path() and get_canonical_dev_path(), do the > extra NULL pointer checks. > > This will be folded into the two offending patches, and I hope this is > really the last fix I need to fold... After you fold it please manually close the syzbot reports as this won't be done by the automatic detection of the reported-by links.
diff --git a/fs/btrfs/volumes.c b/fs/btrfs/volumes.c index 5f5748ab6f9d..dc9f54849f39 100644 --- a/fs/btrfs/volumes.c +++ b/fs/btrfs/volumes.c @@ -749,6 +749,9 @@ static bool is_good_dev_path(const char *dev_path) bool is_good = false; int ret; + if (!dev_path) + goto out; + path_buf = kmalloc(PATH_MAX, GFP_KERNEL); if (!path_buf) goto out; @@ -779,6 +782,11 @@ static int get_canonical_dev_path(const char *dev_path, char *canonical) char *resolved_path; int ret; + if (!dev_path) { + ret = -EINVAL; + goto out; + } + path_buf = kmalloc(PATH_MAX, GFP_KERNEL); if (!path_buf) { ret = -ENOMEM;
[BUG] Syzbot reports a crash when mounting a btrfs with NULL as @path for btrfs_scan_one_device(): Oops: general protection fault, probably for non-canonical address 0xdffffc0000000000: 0000 [#1] PREEMPT SMP KASAN PTI KASAN: null-ptr-deref in range [0x0000000000000000-0x0000000000000007] CPU: 0 UID: 0 PID: 5235 Comm: syz-executor338 Not tainted 6.12.0-rc2-next-20241008-syzkaller #0 RIP: 0010:strlen+0x2c/0x70 lib/string.c:402 Call Trace: <TASK> getname_kernel+0x1d/0x2f0 fs/namei.c:232 kern_path+0x1d/0x50 fs/namei.c:2716 is_good_dev_path fs/btrfs/volumes.c:760 [inline] btrfs_scan_one_device+0x19e/0xd90 fs/btrfs/volumes.c:1484 btrfs_get_tree_super fs/btrfs/super.c:1841 [inline] btrfs_get_tree+0x30e/0x1920 fs/btrfs/super.c:2114 vfs_get_tree+0x90/0x2b0 fs/super.c:1800 fc_mount+0x1b/0xb0 fs/namespace.c:1231 btrfs_get_tree_subvol fs/btrfs/super.c:2077 [inline] btrfs_get_tree+0x652/0x1920 fs/btrfs/super.c:2115 vfs_get_tree+0x90/0x2b0 fs/super.c:1800 vfs_cmd_create+0xa0/0x1f0 fs/fsopen.c:225 __do_sys_fsconfig fs/fsopen.c:472 [inline] __se_sys_fsconfig+0xa1f/0xf70 fs/fsopen.c:344 do_syscall_x64 arch/x86/entry/common.c:52 [inline] do_syscall_64+0xf3/0x230 arch/x86/entry/common.c:83 entry_SYSCALL_64_after_hwframe+0x77/0x7f RIP: 0033:0x7fe8c78542a9 ---[ end trace 0000000000000000 ]--- [CAUSE] The reproducer passes NULL as the device path for btrfs_scan_one_device(). Normally such case will be rejected by bdev_file_open_by_path(), which will return -EINVAL and we reject the mount. But since commit ("btrfs: avoid unnecessary device path update for the same device") and commit ("btrfs: canonicalize the device path before adding") will do extra kern_path() lookup before bdev_file_open_by_path(). Unlike bdev_file_open_by_path(), kern_path() doesn't do the NULL pointer check and triggers the above strlen() on NULL pointer. [FIX] For function is_good_dev_path() and get_canonical_dev_path(), do the extra NULL pointer checks. This will be folded into the two offending patches, and I hope this is really the last fix I need to fold... Reported-by: syzbot+cee29f5a48caf10cd475@syzkaller.appspotmail.com Link: https://lore.kernel.org/linux-btrfs/670689a8.050a0220.67064.0046.GAE@google.com/ Signed-off-by: Qu Wenruo <wqu@suse.com> --- fs/btrfs/volumes.c | 8 ++++++++ 1 file changed, 8 insertions(+)