From 9392a27d88b9707145d713654eb26f0c29789e50 Mon Sep 17 00:00:00 2001 From: Jens Axboe Date: Thu, 6 Feb 2020 21:42:51 -0700 Subject: io-wq: add support for inheriting ->fs Some work items need this for relative path lookup, make it available like the other inherited credentials/mm/etc. Cc: stable@vger.kernel.org # 5.3+ Signed-off-by: Jens Axboe --- fs/io-wq.c | 8 ++++++++ 1 file changed, 8 insertions(+) (limited to 'fs/io-wq.c') diff --git a/fs/io-wq.c b/fs/io-wq.c index cb60a42b9fdfa5..7ac4a8876a50af 100644 --- a/fs/io-wq.c +++ b/fs/io-wq.c @@ -16,6 +16,7 @@ #include #include #include +#include #include "io-wq.h" @@ -59,6 +60,7 @@ struct io_worker { const struct cred *cur_creds; const struct cred *saved_creds; struct files_struct *restore_files; + struct fs_struct *restore_fs; }; #if BITS_PER_LONG == 64 @@ -151,6 +153,9 @@ static bool __io_worker_unuse(struct io_wqe *wqe, struct io_worker *worker) task_unlock(current); } + if (current->fs != worker->restore_fs) + current->fs = worker->restore_fs; + /* * If we have an active mm, we need to drop the wq lock before unusing * it. If we do, return true and let the caller retry the idle loop. @@ -311,6 +316,7 @@ static void io_worker_start(struct io_wqe *wqe, struct io_worker *worker) worker->flags |= (IO_WORKER_F_UP | IO_WORKER_F_RUNNING); worker->restore_files = current->files; + worker->restore_fs = current->fs; io_wqe_inc_running(wqe, worker); } @@ -481,6 +487,8 @@ static void io_worker_handle_work(struct io_worker *worker) current->files = work->files; task_unlock(current); } + if (work->fs && current->fs != work->fs) + current->fs = work->fs; if (work->mm != worker->mm) io_wq_switch_mm(worker, work); if (worker->cur_creds != work->creds) -- cgit 1.2.3-korg From 00bcda13dcbf6bf7fa6f2a5886dd555362de8cfa Mon Sep 17 00:00:00 2001 From: Jens Axboe Date: Sat, 8 Feb 2020 19:13:32 -0700 Subject: io-wq: make io_wqe_cancel_work() take a match handler We want to use the cancel functionality for canceling based on not just the work itself. Instead of matching on the work address manually, allow a match handler to tell us if we found the right work item or not. No functional changes in this patch. Signed-off-by: Jens Axboe --- fs/io-wq.c | 33 ++++++++++++++++++++++----------- 1 file changed, 22 insertions(+), 11 deletions(-) (limited to 'fs/io-wq.c') diff --git a/fs/io-wq.c b/fs/io-wq.c index 7ac4a8876a50af..df78de33ff841e 100644 --- a/fs/io-wq.c +++ b/fs/io-wq.c @@ -937,17 +937,19 @@ enum io_wq_cancel io_wq_cancel_cb(struct io_wq *wq, work_cancel_fn *cancel, return ret; } +struct work_match { + bool (*fn)(struct io_wq_work *, void *data); + void *data; +}; + static bool io_wq_worker_cancel(struct io_worker *worker, void *data) { - struct io_wq_work *work = data; + struct work_match *match = data; unsigned long flags; bool ret = false; - if (worker->cur_work != work) - return false; - spin_lock_irqsave(&worker->lock, flags); - if (worker->cur_work == work && + if (match->fn(worker->cur_work, match->data) && !(worker->cur_work->flags & IO_WQ_WORK_NO_CANCEL)) { send_sig(SIGINT, worker->task, 1); ret = true; @@ -958,15 +960,13 @@ static bool io_wq_worker_cancel(struct io_worker *worker, void *data) } static enum io_wq_cancel io_wqe_cancel_work(struct io_wqe *wqe, - struct io_wq_work *cwork) + struct work_match *match) { struct io_wq_work_node *node, *prev; struct io_wq_work *work; unsigned long flags; bool found = false; - cwork->flags |= IO_WQ_WORK_CANCEL; - /* * First check pending list, if we're lucky we can just remove it * from there. CANCEL_OK means that the work is returned as-new, @@ -976,7 +976,7 @@ static enum io_wq_cancel io_wqe_cancel_work(struct io_wqe *wqe, wq_list_for_each(node, prev, &wqe->work_list) { work = container_of(node, struct io_wq_work, list); - if (work == cwork) { + if (match->fn(work, match->data)) { wq_node_del(&wqe->work_list, node, prev); found = true; break; @@ -997,20 +997,31 @@ static enum io_wq_cancel io_wqe_cancel_work(struct io_wqe *wqe, * completion will run normally in this case. */ rcu_read_lock(); - found = io_wq_for_each_worker(wqe, io_wq_worker_cancel, cwork); + found = io_wq_for_each_worker(wqe, io_wq_worker_cancel, match); rcu_read_unlock(); return found ? IO_WQ_CANCEL_RUNNING : IO_WQ_CANCEL_NOTFOUND; } +static bool io_wq_work_match(struct io_wq_work *work, void *data) +{ + return work == data; +} + enum io_wq_cancel io_wq_cancel_work(struct io_wq *wq, struct io_wq_work *cwork) { + struct work_match match = { + .fn = io_wq_work_match, + .data = cwork + }; enum io_wq_cancel ret = IO_WQ_CANCEL_NOTFOUND; int node; + cwork->flags |= IO_WQ_WORK_CANCEL; + for_each_node(node) { struct io_wqe *wqe = wq->wqes[node]; - ret = io_wqe_cancel_work(wqe, cwork); + ret = io_wqe_cancel_work(wqe, &match); if (ret != IO_WQ_CANCEL_NOTFOUND) break; } -- cgit 1.2.3-korg From 36282881a795cbf717aca79392ae9cdf0fef59c9 Mon Sep 17 00:00:00 2001 From: Jens Axboe Date: Sat, 8 Feb 2020 19:16:39 -0700 Subject: io-wq: add io_wq_cancel_pid() to cancel based on a specific pid Add a helper that allows the caller to cancel work based on what mm it belongs to. This allows io_uring to cancel work from a given task or thread when it exits. Signed-off-by: Jens Axboe --- fs/io-wq.c | 29 +++++++++++++++++++++++++++++ fs/io-wq.h | 2 ++ 2 files changed, 31 insertions(+) (limited to 'fs/io-wq.c') diff --git a/fs/io-wq.c b/fs/io-wq.c index df78de33ff841e..182aa17dc2cabe 100644 --- a/fs/io-wq.c +++ b/fs/io-wq.c @@ -1029,6 +1029,35 @@ enum io_wq_cancel io_wq_cancel_work(struct io_wq *wq, struct io_wq_work *cwork) return ret; } +static bool io_wq_pid_match(struct io_wq_work *work, void *data) +{ + pid_t pid = (pid_t) (unsigned long) data; + + if (work) + return work->task_pid == pid; + return false; +} + +enum io_wq_cancel io_wq_cancel_pid(struct io_wq *wq, pid_t pid) +{ + struct work_match match = { + .fn = io_wq_pid_match, + .data = (void *) (unsigned long) pid + }; + enum io_wq_cancel ret = IO_WQ_CANCEL_NOTFOUND; + int node; + + for_each_node(node) { + struct io_wqe *wqe = wq->wqes[node]; + + ret = io_wqe_cancel_work(wqe, &match); + if (ret != IO_WQ_CANCEL_NOTFOUND) + break; + } + + return ret; +} + struct io_wq_flush_data { struct io_wq_work work; struct completion done; diff --git a/fs/io-wq.h b/fs/io-wq.h index f152ba677d8f57..ccc7d84af57de1 100644 --- a/fs/io-wq.h +++ b/fs/io-wq.h @@ -76,6 +76,7 @@ struct io_wq_work { const struct cred *creds; struct fs_struct *fs; unsigned flags; + pid_t task_pid; }; #define INIT_IO_WORK(work, _func) \ @@ -109,6 +110,7 @@ void io_wq_flush(struct io_wq *wq); void io_wq_cancel_all(struct io_wq *wq); enum io_wq_cancel io_wq_cancel_work(struct io_wq *wq, struct io_wq_work *cwork); +enum io_wq_cancel io_wq_cancel_pid(struct io_wq *wq, pid_t pid); typedef bool (work_cancel_fn)(struct io_wq_work *, void *); -- cgit 1.2.3-korg From 7563439adfae153b20331f1567c8b5d0e5cbd8a7 Mon Sep 17 00:00:00 2001 From: Jens Axboe Date: Tue, 11 Feb 2020 06:30:06 -0700 Subject: io-wq: don't call kXalloc_node() with non-online node Glauber reports a crash on init on a box he has: RIP: 0010:__alloc_pages_nodemask+0x132/0x340 Code: 18 01 75 04 41 80 ce 80 89 e8 48 8b 54 24 08 8b 74 24 1c c1 e8 0c 48 8b 3c 24 83 e0 01 88 44 24 20 48 85 d2 0f 85 74 01 00 00 <3b> 77 08 0f 82 6b 01 00 00 48 89 7c 24 10 89 ea 48 8b 07 b9 00 02 RSP: 0018:ffffb8be4d0b7c28 EFLAGS: 00010246 RAX: 0000000000000000 RBX: 0000000000000000 RCX: 000000000000e8e8 RDX: 0000000000000000 RSI: 0000000000000002 RDI: 0000000000002080 RBP: 0000000000012cc0 R08: 0000000000000000 R09: 0000000000000002 R10: 0000000000000dc0 R11: ffff995c60400100 R12: 0000000000000000 R13: 0000000000012cc0 R14: 0000000000000001 R15: ffff995c60db00f0 FS: 00007f4d115ca900(0000) GS:ffff995c60d80000(0000) knlGS:0000000000000000 CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033 CR2: 0000000000002088 CR3: 00000017cca66002 CR4: 00000000007606e0 DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000 DR3: 0000000000000000 DR6: 00000000fffe0ff0 DR7: 0000000000000400 PKRU: 55555554 Call Trace: alloc_slab_page+0x46/0x320 new_slab+0x9d/0x4e0 ___slab_alloc+0x507/0x6a0 ? io_wq_create+0xb4/0x2a0 __slab_alloc+0x1c/0x30 kmem_cache_alloc_node_trace+0xa6/0x260 io_wq_create+0xb4/0x2a0 io_uring_setup+0x97f/0xaa0 ? io_remove_personalities+0x30/0x30 ? io_poll_trigger_evfd+0x30/0x30 do_syscall_64+0x5b/0x1c0 entry_SYSCALL_64_after_hwframe+0x44/0xa9 RIP: 0033:0x7f4d116cb1ed which is due to the 'wqe' and 'worker' allocation being node affine. But it isn't valid to call the node affine allocation if the node isn't online. Setup structures for even offline nodes, as usual, but skip them in terms of thread setup to not waste resources. If the node isn't online, just alloc memory with NUMA_NO_NODE. Reported-by: Glauber Costa Tested-by: Glauber Costa Signed-off-by: Jens Axboe --- fs/io-wq.c | 22 ++++++++++++++++++---- 1 file changed, 18 insertions(+), 4 deletions(-) (limited to 'fs/io-wq.c') diff --git a/fs/io-wq.c b/fs/io-wq.c index 182aa17dc2cabe..0a5ab1a8f69a48 100644 --- a/fs/io-wq.c +++ b/fs/io-wq.c @@ -699,11 +699,16 @@ static int io_wq_manager(void *data) /* create fixed workers */ refcount_set(&wq->refs, workers_to_create); for_each_node(node) { + if (!node_online(node)) + continue; if (!create_io_worker(wq, wq->wqes[node], IO_WQ_ACCT_BOUND)) goto err; workers_to_create--; } + while (workers_to_create--) + refcount_dec(&wq->refs); + complete(&wq->done); while (!kthread_should_stop()) { @@ -711,6 +716,9 @@ static int io_wq_manager(void *data) struct io_wqe *wqe = wq->wqes[node]; bool fork_worker[2] = { false, false }; + if (!node_online(node)) + continue; + spin_lock_irq(&wqe->lock); if (io_wqe_need_worker(wqe, IO_WQ_ACCT_BOUND)) fork_worker[IO_WQ_ACCT_BOUND] = true; @@ -829,7 +837,9 @@ static bool io_wq_for_each_worker(struct io_wqe *wqe, list_for_each_entry_rcu(worker, &wqe->all_list, all_list) { if (io_worker_get(worker)) { - ret = func(worker, data); + /* no task if node is/was offline */ + if (worker->task) + ret = func(worker, data); io_worker_release(worker); if (ret) break; @@ -1084,6 +1094,8 @@ void io_wq_flush(struct io_wq *wq) for_each_node(node) { struct io_wqe *wqe = wq->wqes[node]; + if (!node_online(node)) + continue; init_completion(&data.done); INIT_IO_WORK(&data.work, io_wq_flush_func); data.work.flags |= IO_WQ_WORK_INTERNAL; @@ -1115,12 +1127,15 @@ struct io_wq *io_wq_create(unsigned bounded, struct io_wq_data *data) for_each_node(node) { struct io_wqe *wqe; + int alloc_node = node; - wqe = kzalloc_node(sizeof(struct io_wqe), GFP_KERNEL, node); + if (!node_online(alloc_node)) + alloc_node = NUMA_NO_NODE; + wqe = kzalloc_node(sizeof(struct io_wqe), GFP_KERNEL, alloc_node); if (!wqe) goto err; wq->wqes[node] = wqe; - wqe->node = node; + wqe->node = alloc_node; wqe->acct[IO_WQ_ACCT_BOUND].max_workers = bounded; atomic_set(&wqe->acct[IO_WQ_ACCT_BOUND].nr_running, 0); if (wq->user) { @@ -1128,7 +1143,6 @@ struct io_wq *io_wq_create(unsigned bounded, struct io_wq_data *data) task_rlimit(current, RLIMIT_NPROC); } atomic_set(&wqe->acct[IO_WQ_ACCT_UNBOUND].nr_running, 0); - wqe->node = node; wqe->wq = wq; spin_lock_init(&wqe->lock); INIT_WQ_LIST(&wqe->work_list); -- cgit 1.2.3-korg From 3030fd4cb783377eca0e2a3eee63724a5c66ee15 Mon Sep 17 00:00:00 2001 From: Jens Axboe Date: Tue, 25 Feb 2020 08:47:30 -0700 Subject: io-wq: remove spin-for-work optimization Andres reports that buffered IO seems to suck up more cycles than we would like, and he narrowed it down to the fact that the io-wq workers will briefly spin for more work on completion of a work item. This was a win on the networking side, but apparently some other cases take a hit because of it. Remove the optimization to avoid burning more CPU than we have to for disk IO. Reported-by: Andres Freund Signed-off-by: Jens Axboe --- fs/io-wq.c | 19 ------------------- 1 file changed, 19 deletions(-) (limited to 'fs/io-wq.c') diff --git a/fs/io-wq.c b/fs/io-wq.c index 0a5ab1a8f69a48..bf8ed1b0b90a01 100644 --- a/fs/io-wq.c +++ b/fs/io-wq.c @@ -535,42 +535,23 @@ static void io_worker_handle_work(struct io_worker *worker) } while (1); } -static inline void io_worker_spin_for_work(struct io_wqe *wqe) -{ - int i = 0; - - while (++i < 1000) { - if (io_wqe_run_queue(wqe)) - break; - if (need_resched()) - break; - cpu_relax(); - } -} - static int io_wqe_worker(void *data) { struct io_worker *worker = data; struct io_wqe *wqe = worker->wqe; struct io_wq *wq = wqe->wq; - bool did_work; io_worker_start(wqe, worker); - did_work = false; while (!test_bit(IO_WQ_BIT_EXIT, &wq->state)) { set_current_state(TASK_INTERRUPTIBLE); loop: - if (did_work) - io_worker_spin_for_work(wqe); spin_lock_irq(&wqe->lock); if (io_wqe_run_queue(wqe)) { __set_current_state(TASK_RUNNING); io_worker_handle_work(worker); - did_work = true; goto loop; } - did_work = false; /* drops the lock on success, retry */ if (__io_worker_idle(wqe, worker)) { __release(&wqe->lock); -- cgit 1.2.3-korg From fc04c39bae01a607454f7619665309870c60937a Mon Sep 17 00:00:00 2001 From: Pavel Begunkov Date: Sun, 1 Mar 2020 19:18:19 +0300 Subject: io-wq: fix IO_WQ_WORK_NO_CANCEL cancellation To cancel a work, io-wq sets IO_WQ_WORK_CANCEL and executes the callback. However, IO_WQ_WORK_NO_CANCEL works will just execute and may return next work, which will be ignored and lost. Cancel the whole link. Signed-off-by: Pavel Begunkov Signed-off-by: Jens Axboe --- fs/io-wq.c | 20 ++++++++++++++------ 1 file changed, 14 insertions(+), 6 deletions(-) (limited to 'fs/io-wq.c') diff --git a/fs/io-wq.c b/fs/io-wq.c index bf8ed1b0b90a01..9a7aacc96d8414 100644 --- a/fs/io-wq.c +++ b/fs/io-wq.c @@ -747,6 +747,17 @@ static bool io_wq_can_queue(struct io_wqe *wqe, struct io_wqe_acct *acct, return true; } +static void io_run_cancel(struct io_wq_work *work) +{ + do { + struct io_wq_work *old_work = work; + + work->flags |= IO_WQ_WORK_CANCEL; + work->func(&work); + work = (work == old_work) ? NULL : work; + } while (work); +} + static void io_wqe_enqueue(struct io_wqe *wqe, struct io_wq_work *work) { struct io_wqe_acct *acct = io_work_get_acct(wqe, work); @@ -760,8 +771,7 @@ static void io_wqe_enqueue(struct io_wqe *wqe, struct io_wq_work *work) * It's close enough to not be an issue, fork() has the same delay. */ if (unlikely(!io_wq_can_queue(wqe, acct, work))) { - work->flags |= IO_WQ_WORK_CANCEL; - work->func(&work); + io_run_cancel(work); return; } @@ -900,8 +910,7 @@ static enum io_wq_cancel io_wqe_cancel_cb_work(struct io_wqe *wqe, spin_unlock_irqrestore(&wqe->lock, flags); if (found) { - work->flags |= IO_WQ_WORK_CANCEL; - work->func(&work); + io_run_cancel(work); return IO_WQ_CANCEL_OK; } @@ -976,8 +985,7 @@ static enum io_wq_cancel io_wqe_cancel_work(struct io_wqe *wqe, spin_unlock_irqrestore(&wqe->lock, flags); if (found) { - work->flags |= IO_WQ_WORK_CANCEL; - work->func(&work); + io_run_cancel(work); return IO_WQ_CANCEL_OK; } -- cgit 1.2.3-korg From 80ad894382bf1d73eb688c29714fa10c0afcf2e7 Mon Sep 17 00:00:00 2001 From: Pavel Begunkov Date: Mon, 2 Mar 2020 23:46:10 +0300 Subject: io-wq: remove io_wq_flush and IO_WQ_WORK_INTERNAL io_wq_flush() is buggy, during cancelation of a flush, the associated work may be passed to the caller's (i.e. io_uring) @match callback. That callback is expecting it to be embedded in struct io_kiocb. Cancelation of internal work probably doesn't make a lot of sense to begin with. As the flush helper is no longer used, just delete it and the associated work flag. Signed-off-by: Pavel Begunkov Signed-off-by: Jens Axboe --- fs/io-wq.c | 38 +------------------------------------- fs/io-wq.h | 2 -- 2 files changed, 1 insertion(+), 39 deletions(-) (limited to 'fs/io-wq.c') diff --git a/fs/io-wq.c b/fs/io-wq.c index 9a7aacc96d8414..5cef075c0b379c 100644 --- a/fs/io-wq.c +++ b/fs/io-wq.c @@ -502,7 +502,7 @@ static void io_worker_handle_work(struct io_worker *worker) if (worker->mm) work->flags |= IO_WQ_WORK_HAS_MM; - if (wq->get_work && !(work->flags & IO_WQ_WORK_INTERNAL)) { + if (wq->get_work) { put_work = work; wq->get_work(work); } @@ -1057,42 +1057,6 @@ enum io_wq_cancel io_wq_cancel_pid(struct io_wq *wq, pid_t pid) return ret; } -struct io_wq_flush_data { - struct io_wq_work work; - struct completion done; -}; - -static void io_wq_flush_func(struct io_wq_work **workptr) -{ - struct io_wq_work *work = *workptr; - struct io_wq_flush_data *data; - - data = container_of(work, struct io_wq_flush_data, work); - complete(&data->done); -} - -/* - * Doesn't wait for previously queued work to finish. When this completes, - * it just means that previously queued work was started. - */ -void io_wq_flush(struct io_wq *wq) -{ - struct io_wq_flush_data data; - int node; - - for_each_node(node) { - struct io_wqe *wqe = wq->wqes[node]; - - if (!node_online(node)) - continue; - init_completion(&data.done); - INIT_IO_WORK(&data.work, io_wq_flush_func); - data.work.flags |= IO_WQ_WORK_INTERNAL; - io_wqe_enqueue(wqe, &data.work); - wait_for_completion(&data.done); - } -} - struct io_wq *io_wq_create(unsigned bounded, struct io_wq_data *data) { int ret = -ENOMEM, node; diff --git a/fs/io-wq.h b/fs/io-wq.h index 33baba4370c5f2..e5e15f2c93eca4 100644 --- a/fs/io-wq.h +++ b/fs/io-wq.h @@ -8,7 +8,6 @@ enum { IO_WQ_WORK_HAS_MM = 2, IO_WQ_WORK_HASHED = 4, IO_WQ_WORK_UNBOUND = 32, - IO_WQ_WORK_INTERNAL = 64, IO_WQ_WORK_CB = 128, IO_WQ_WORK_NO_CANCEL = 256, IO_WQ_WORK_CONCURRENT = 512, @@ -100,7 +99,6 @@ void io_wq_destroy(struct io_wq *wq); void io_wq_enqueue(struct io_wq *wq, struct io_wq_work *work); void io_wq_enqueue_hashed(struct io_wq *wq, struct io_wq_work *work, void *val); -void io_wq_flush(struct io_wq *wq); void io_wq_cancel_all(struct io_wq *wq); enum io_wq_cancel io_wq_cancel_work(struct io_wq *wq, struct io_wq_work *cwork); -- cgit 1.2.3-korg