"value":"In the Linux kernel, the following vulnerability has been resolved:\n\nio_uring/rw: fix missing NOWAIT check for O_DIRECT start write\n\nWhen io_uring starts a write, it'll call kiocb_start_write() to bump the\nsuper block rwsem, preventing any freezes from happening while that\nwrite is in-flight. The freeze side will grab that rwsem for writing,\nexcluding any new writers from happening and waiting for existing writes\nto finish. But io_uring unconditionally uses kiocb_start_write(), which\nwill block if someone is currently attempting to freeze the mount point.\nThis causes a deadlock where freeze is waiting for previous writes to\ncomplete, but the previous writes cannot complete, as the task that is\nsupposed to complete them is blocked waiting on starting a new write.\nThis results in the following stuck trace showing that dependency with\nthe write blocked starting a new write:\n\ntask:fio state:D stack:0 pid:886 tgid:886 ppid:876\nCall trace:\n __switch_to+0x1d8/0x348\n __schedule+0x8e8/0x2248\n schedule+0x110/0x3f0\n percpu_rwsem_wait+0x1e8/0x3f8\n __percpu_down_read+0xe8/0x500\n io_write+0xbb8/0xff8\n io_issue_sqe+0x10c/0x1020\n io_submit_sqes+0x614/0x2110\n __arm64_sys_io_uring_enter+0x524/0x1038\n invoke_syscall+0x74/0x268\n el0_svc_common.constprop.0+0x160/0x238\n do_el0_svc+0x44/0x60\n el0_svc+0x44/0xb0\n el0t_64_sync_handler+0x118/0x128\n el0t_64_sync+0x168/0x170\nINFO: task fsfreeze:7364 blocked for more than 15 seconds.\n Not tainted 6.12.0-rc5-00063-g76aaf945701c #7963\n\nwith the attempting freezer stuck trying to grab the rwsem:\n\ntask:fsfreeze state:D stack:0 pid:7364 tgid:7364 ppid:995\nCall trace:\n __switch_to+0x1d8/0x348\n __schedule+0x8e8/0x2248\n schedule+0x110/0x3f0\n percpu_down_write+0x2b0/0x680\n freeze_super+0x248/0x8a8\n do_vfs_ioctl+0x149c/0x1b18\n __arm64_sys_ioctl+0xd0/0x1a0\n invoke_syscall+0x74/0x268\n el0_svc_common.constprop.0+0x160/0x238\n do_el0_svc+0x44/0x60\n el0_svc+0x44/0xb0\n el0t_64_sync_handler+0x118/0x128\n el0t_64_sync+0x168/0x170\n\nFix this by having the io_uring side honor IOCB_NOWAIT, and only attempt a\nblocking grab of the super block rwsem if it isn't set. For normal issue\nwhere IOCB_NOWAIT would always be set, this returns -EAGAIN which will\nhave io_uring core issue a blocking attempt of the write. That will in\nturn also get completions run, ensuring forward progress.\n\nSince freezing requires CAP_SYS_ADMIN in the first place, this isn't\nsomething that can be triggered by a regular user."
