CVE-2021-47508

In the Linux kernel, the following vulnerability has been resolved:<br /> <br /> btrfs: free exchange changeset on failures<br /> <br /> Fstests runs on my VMs have show several kmemleak reports like the following.<br /> <br /> unreferenced object 0xffff88811ae59080 (size 64):<br /> comm "xfs_io", pid 12124, jiffies 4294987392 (age 6.368s)<br /> hex dump (first 32 bytes):<br /> 00 c0 1c 00 00 00 00 00 ff cf 1c 00 00 00 00 00 ................<br /> 90 97 e5 1a 81 88 ff ff 90 97 e5 1a 81 88 ff ff ................<br /> backtrace:<br /> [] ulist_add_merge+0x60/0x150 [btrfs]<br /> [] set_state_bits+0x86/0xc0 [btrfs]<br /> [] set_extent_bit+0x270/0x690 [btrfs]<br /> [] set_record_extent_bits+0x19/0x20 [btrfs]<br /> [] qgroup_reserve_data+0x274/0x310 [btrfs]<br /> [] btrfs_check_data_free_space+0x5c/0xa0 [btrfs]<br /> [] btrfs_delalloc_reserve_space+0x1b/0xa0 [btrfs]<br /> [] btrfs_dio_iomap_begin+0x415/0x970 [btrfs]<br /> [] iomap_iter+0x161/0x1e0<br /> [] __iomap_dio_rw+0x1df/0x700<br /> [] iomap_dio_rw+0x5/0x20<br /> [] btrfs_file_write_iter+0x290/0x530 [btrfs]<br /> [] new_sync_write+0x106/0x180<br /> [] vfs_write+0x24d/0x2f0<br /> [] __x64_sys_pwrite64+0x69/0xa0<br /> [] do_syscall_64+0x43/0x90<br /> <br /> In case brtfs_qgroup_reserve_data() or btrfs_delalloc_reserve_metadata()<br /> fail the allocated extent_changeset will not be freed.<br /> <br /> So in btrfs_check_data_free_space() and btrfs_delalloc_reserve_space()<br /> free the allocated extent_changeset to get rid of the allocated memory.<br /> <br /> The issue currently only happens in the direct IO write path, but only<br /> after 65b3c08606e5 ("btrfs: fix ENOSPC failure when attempting direct IO<br /> write into NOCOW range"), and also at defrag_one_locked_target(). Every<br /> other place is always calling extent_changeset_free() even if its call<br /> to btrfs_delalloc_reserve_space() or btrfs_check_data_free_space() has<br /> failed.