On Fri, Feb 8, 2019 at 6:34 PM Jens Axboe <axboe@xxxxxxxxx> wrote: > The submission queue (SQ) and completion queue (CQ) rings are shared > between the application and the kernel. This eliminates the need to > copy data back and forth to submit and complete IO. > > IO submissions use the io_uring_sqe data structure, and completions > are generated in the form of io_uring_cqe data structures. The SQ > ring is an index into the io_uring_sqe array, which makes it possible > to submit a batch of IOs without them being contiguous in the ring. > The CQ ring is always contiguous, as completion events are inherently > unordered, and hence any io_uring_cqe entry can point back to an > arbitrary submission. > > Two new system calls are added for this: > > io_uring_setup(entries, params) > Sets up an io_uring instance for doing async IO. On success, > returns a file descriptor that the application can mmap to > gain access to the SQ ring, CQ ring, and io_uring_sqes. > > io_uring_enter(fd, to_submit, min_complete, flags, sigset, sigsetsize) > Initiates IO against the rings mapped to this fd, or waits for > them to complete, or both. The behavior is controlled by the > parameters passed in. If 'to_submit' is non-zero, then we'll > try and submit new IO. If IORING_ENTER_GETEVENTS is set, the > kernel will wait for 'min_complete' events, if they aren't > already available. It's valid to set IORING_ENTER_GETEVENTS > and 'min_complete' == 0 at the same time, this allows the > kernel to return already completed events without waiting > for them. This is useful only for polling, as for IRQ > driven IO, the application can just check the CQ ring > without entering the kernel. > > With this setup, it's possible to do async IO with a single system > call. Future developments will enable polled IO with this interface, > and polled submission as well. The latter will enable an application > to do IO without doing ANY system calls at all. > > For IRQ driven IO, an application only needs to enter the kernel for > completions if it wants to wait for them to occur. > > Each io_uring is backed by a workqueue, to support buffered async IO > as well. We will only punt to an async context if the command would > need to wait for IO on the device side. Any data that can be accessed > directly in the page cache is done inline. This avoids the slowness > issue of usual threadpools, since cached data is accessed as quickly > as a sync interface. [...] > +static void io_commit_cqring(struct io_ring_ctx *ctx) > +{ > + struct io_cq_ring *ring = ctx->cq_ring; > + > + if (ctx->cached_cq_tail != ring->r.tail) { I know that this is very unlikely to actually matter, but both because I don't feel fuzzy about relying on compiler internals regarding when the compiler might decide to generate dangerous double-reads (if switch() can blow up, why shouldn't the compiler be able to make if() blow up if it wants to, too?), and because I would like it to be as clear as possible to the reader which memory is shared with userspace, can we please have READ_ONCE() on *every* shared memory read, not just the ones in places that look like they might plausibly blow up otherwise? Sorry, shared memory is a bit of a pet peeve of mine. > + /* order cqe stores with ring update */ > + smp_wmb(); > + WRITE_ONCE(ring->r.tail, ctx->cached_cq_tail); > + /* write side barrier of tail update, app has read side */ > + smp_wmb(); > + > + if (wq_has_sleeper(&ctx->cq_wait)) { > + wake_up_interruptible(&ctx->cq_wait); > + kill_fasync(&ctx->cq_fasync, SIGIO, POLL_IN); > + } > + } > +} [...] > +static void io_cqring_fill_event(struct io_ring_ctx *ctx, u64 ki_user_data, > + long res, unsigned ev_flags) > +{ > + struct io_uring_cqe *cqe; > + > + /* > + * If we can't get a cq entry, userspace overflowed the > + * submission (by quite a lot). Increment the overflow count in > + * the ring. > + */ > + cqe = io_get_cqring(ctx); > + if (cqe) { > + cqe->user_data = ki_user_data; > + cqe->res = res; > + cqe->flags = ev_flags; Please use WRITE_ONCE() for stores like these. > + } else > + ctx->cq_ring->overflow++; > +} [...] > +static int __io_submit_sqe(struct io_ring_ctx *ctx, struct io_kiocb *req, > + const struct sqe_submit *s, bool force_nonblock) > +{ > + ssize_t ret; > + int opcode; > + > + if (unlikely(s->index >= ctx->sq_entries)) > + return -EINVAL; > + req->user_data = READ_ONCE(s->sqe->user_data); > + > + opcode = READ_ONCE(s->sqe->opcode); There might be a sneaky bug here. Consider the following scenario: 1. request gets submitted from io_sq_wq_submit_work() with opcode IORING_OP_READV, io_read() is invoked 2. io_read() looks up the file, taking a reference to it 3. call_read_iter() returns -EAGAIN 4. io_read() returns -EAGAIN without dropping its reference on the file (because it expects that it'll be called again) 5. __io_submit_sqe() returns -EAGAIN 6. io_sq_wq_submit_work() loops back and retries __io_submit_sqe() 7. __io_submit_sqe() reads opcode again, this time it's IORING_OP_NOP 8. io_nop() gets called 9. io_nop() uses io_free_req() to delete the request without dropping its reference on the file So that's a file reference leak, I think? > + switch (opcode) { > + case IORING_OP_NOP: > + ret = io_nop(req, req->user_data); > + break; > + case IORING_OP_READV: > + ret = io_read(req, s, force_nonblock); > + break; > + case IORING_OP_WRITEV: > + ret = io_write(req, s, force_nonblock); > + break; > + default: > + ret = -EINVAL; > + break; > + } > + > + return ret; > +} [...] > +static int io_submit_sqe(struct io_ring_ctx *ctx, const struct sqe_submit *s) > +{ > + struct io_kiocb *req; > + ssize_t ret; > + > + /* enforce forwards compatibility on users */ > + if (unlikely(s->sqe->flags)) > + return -EINVAL; > + > + req = io_get_req(ctx); > + if (unlikely(!req)) > + return -EAGAIN; > + > + req->rw.ki_filp = NULL; > + > + ret = __io_submit_sqe(ctx, req, s, true); > + if (ret == -EAGAIN) { > + memcpy(&req->submit, s, sizeof(*s)); > + INIT_WORK(&req->work, io_sq_wq_submit_work); > + queue_work(ctx->sqo_wq, &req->work); > + ret = 0; > + } > + if (ret) > + io_free_req(req); > + > + return ret; > +} > + > +static void io_commit_sqring(struct io_ring_ctx *ctx) > +{ > + struct io_sq_ring *ring = ctx->sq_ring; > + > + if (ctx->cached_sq_head != ring->r.head) { > + WRITE_ONCE(ring->r.head, ctx->cached_sq_head); > + /* write side barrier of head update, app has read side */ > + smp_wmb(); Can you elaborate on what this memory barrier is doing? Don't you need some sort of memory barrier *before* the WRITE_ONCE(), to ensure that nobody sees the updated head before you're done reading the submission queue entry? Or is that barrier elsewhere? > + } > +} > + > +/* > + * Undo last io_get_sqring() > + */ > +static void io_drop_sqring(struct io_ring_ctx *ctx) > +{ > + ctx->cached_sq_head--; > +} [...] > +static void io_unaccount_mem(struct user_struct *user, unsigned long nr_pages) > +{ > + if (capable(CAP_IPC_LOCK)) > + return; Hrm... what happens if root creates a uring, drops CAP_IPC_LOCK, and then destroys the uring? Will the pages get subtracted from ->locked_vm even though they were never added to it, causing a wraparound? You might want to make sure that ctx->user is set if and only if the creator didn't have CAP_IPC_LOCK, and then just do a `user == NULL` check instead of a `capable(...)` check. Or you could do what BPF is doing (AFAICS) and not treat root specially - root can just bump the rlimit if necessary. > + atomic_long_sub(nr_pages, &user->locked_vm); > +} > + > +static int io_account_mem(struct user_struct *user, unsigned long nr_pages) > +{ > + unsigned long page_limit, cur_pages, new_pages; > + > + if (capable(CAP_IPC_LOCK)) > + return 0; > + > + /* Don't allow more pages than we can safely lock */ > + page_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT; > + > + do { > + cur_pages = atomic_long_read(&user->locked_vm); > + new_pages = cur_pages + nr_pages; > + if (new_pages > page_limit) > + return -ENOMEM; > + } while (atomic_long_cmpxchg(&user->locked_vm, cur_pages, > + new_pages) != cur_pages); > + > + return 0; > +} [...] > +config IO_URING > + bool "Enable IO uring support" if EXPERT > + select ANON_INODES > + default y > + help > + This option enables support for the io_uring interface, enabling > + applications to submit and completion IO through submission and > + completion rings that are shared between the kernel and application. Nit: I can't parse this part: "enabling applications to submit and completion IO"