We currently have a race where if setup is really slow, we can be calling io_wq_destroy() before we're done setting up. This will cause the caller to get stuck waiting for the manager to set things up, but the manager already exited. Fix this by doing a sync setup of the manager. This also fixes the case where if we failed creating workers, we'd also get stuck. In practice this race window was really small, as we already wait for the manager to start. Hence someone would have to call io_wq_destroy() after the task has started, but before it started the first loop. The reported test case forked tons of these, which is why it became an issue. Reported-by: syzbot+0f1cc17f85154f400465@xxxxxxxxxxxxxxxxxxxxxxxxx Fixes: 771b53d033e8 ("io-wq: small threadpool implementation for io_uring") Signed-off-by: Jens Axboe <axboe@xxxxxxxxx> --- fs/io-wq.c | 38 ++++++++++++++++++++++++++------------ 1 file changed, 26 insertions(+), 12 deletions(-) diff --git a/fs/io-wq.c b/fs/io-wq.c index 9174007ce107..1f640c489f7c 100644 --- a/fs/io-wq.c +++ b/fs/io-wq.c @@ -33,6 +33,7 @@ enum { enum { IO_WQ_BIT_EXIT = 0, /* wq exiting */ IO_WQ_BIT_CANCEL = 1, /* cancel work on list */ + IO_WQ_BIT_ERROR = 2, /* error on setup */ }; enum { @@ -562,14 +563,14 @@ void io_wq_worker_sleeping(struct task_struct *tsk) spin_unlock_irq(&wqe->lock); } -static void create_io_worker(struct io_wq *wq, struct io_wqe *wqe, int index) +static bool create_io_worker(struct io_wq *wq, struct io_wqe *wqe, int index) { struct io_wqe_acct *acct =&wqe->acct[index]; struct io_worker *worker; worker = kcalloc_node(1, sizeof(*worker), GFP_KERNEL, wqe->node); if (!worker) - return; + return false; refcount_set(&worker->ref, 1); worker->nulls_node.pprev = NULL; @@ -581,7 +582,7 @@ static void create_io_worker(struct io_wq *wq, struct io_wqe *wqe, int index) "io_wqe_worker-%d/%d", index, wqe->node); if (IS_ERR(worker->task)) { kfree(worker); - return; + return false; } spin_lock_irq(&wqe->lock); @@ -599,6 +600,7 @@ static void create_io_worker(struct io_wq *wq, struct io_wqe *wqe, int index) atomic_inc(&wq->user->processes); wake_up_process(worker->task); + return true; } static inline bool io_wqe_need_worker(struct io_wqe *wqe, int index) @@ -606,9 +608,6 @@ static inline bool io_wqe_need_worker(struct io_wqe *wqe, int index) { struct io_wqe_acct *acct = &wqe->acct[index]; - /* always ensure we have one bounded worker */ - if (index == IO_WQ_ACCT_BOUND && !acct->nr_workers) - return true; /* if we have available workers or no work, no need */ if (!hlist_nulls_empty(&wqe->free_list) || !io_wqe_run_queue(wqe)) return false; @@ -621,10 +620,19 @@ static inline bool io_wqe_need_worker(struct io_wqe *wqe, int index) static int io_wq_manager(void *data) { struct io_wq *wq = data; + int i; - while (!kthread_should_stop()) { - int i; + /* create fixed workers */ + for (i = 0; i < wq->nr_wqes; i++) { + if (create_io_worker(wq, wq->wqes[i], IO_WQ_ACCT_BOUND)) + continue; + goto err; + } + refcount_set(&wq->refs, wq->nr_wqes); + complete(&wq->done); + + while (!kthread_should_stop()) { for (i = 0; i < wq->nr_wqes; i++) { struct io_wqe *wqe = wq->wqes[i]; bool fork_worker[2] = { false, false }; @@ -644,6 +652,10 @@ static int io_wq_manager(void *data) schedule_timeout(HZ); } + return 0; +err: + set_bit(IO_WQ_BIT_ERROR, &wq->state); + complete(&wq->done); return 0; } @@ -982,7 +994,6 @@ struct io_wq *io_wq_create(unsigned bounded, struct mm_struct *mm, wq->user = user; i = 0; - refcount_set(&wq->refs, wq->nr_wqes); for_each_online_node(node) { struct io_wqe *wqe; @@ -1020,6 +1031,10 @@ struct io_wq *io_wq_create(unsigned bounded, struct mm_struct *mm, wq->manager = kthread_create(io_wq_manager, wq, "io_wq_manager"); if (!IS_ERR(wq->manager)) { wake_up_process(wq->manager); + wait_for_completion(&wq->done); + if (test_bit(IO_WQ_BIT_ERROR, &wq->state)) + goto err; + reinit_completion(&wq->done); return wq; } @@ -1041,10 +1056,9 @@ void io_wq_destroy(struct io_wq *wq) { int i; - if (wq->manager) { - set_bit(IO_WQ_BIT_EXIT, &wq->state); + set_bit(IO_WQ_BIT_EXIT, &wq->state); + if (wq->manager) kthread_stop(wq->manager); - } rcu_read_lock(); for (i = 0; i < wq->nr_wqes; i++) { -- 2.24.0