On 8/4/20 11:15 AM, Jens Axboe wrote: > On 8/4/20 11:02 AM, xiao lin wrote: >> 在 2020年8月4日星期二，Jens Axboe <axboe@xxxxxxxxx <mailto:axboe@xxxxxxxxx>> 写道： >> >> On 8/4/20 7:18 AM, Pavel Begunkov wrote: >> > On 04/08/2020 15:56, Liu Yong wrote: >> >> In io_send_recvmsg(), there is no check for the req->file. >> >> User can change the opcode from IORING_OP_NOP to IORING_OP_SENDMSG >> >> through competition after the io_req_set_file(). >> > >> > After sqe->opcode is read and copied in io_init_req(), it only uses >> > in-kernel req->opcode. Also, io_init_req() should check for req->file >> > NULL, so shouldn't happen after. >> > >> > Do you have a reproducer? What kernel version did you use? >> >> Was looking at this too, and I'm guessing this is some 5.4 based kernel. >> Unfortunately the oops doesn't include that information. > >> Sorry, I forgot to mention that the kernel version I am using is 5.4.55. > > I think there are two options here: > > 1) Backport the series that ensured we only read those important bits once > 2) Make s->sqe a full sqe, and memcpy it in Something like this should close the ->opcode re-read for 5.4-stable. diff --git a/fs/io_uring.c b/fs/io_uring.c index e0200406765c..8bb5e19b7c3c 100644 --- a/fs/io_uring.c +++ b/fs/io_uring.c @@ -279,6 +279,7 @@ struct sqe_submit { bool has_user; bool needs_lock; bool needs_fixed_file; + u8 opcode; }; /* @@ -505,7 +506,7 @@ static inline void io_queue_async_work(struct io_ring_ctx *ctx, int rw = 0; if (req->submit.sqe) { - switch (req->submit.sqe->opcode) { + switch (req->submit.opcode) { case IORING_OP_WRITEV: case IORING_OP_WRITE_FIXED: rw = !(req->rw.ki_flags & IOCB_DIRECT); @@ -1254,23 +1255,15 @@ static int io_import_fixed(struct io_ring_ctx *ctx, int rw, } static ssize_t io_import_iovec(struct io_ring_ctx *ctx, int rw, - const struct sqe_submit *s, struct iovec **iovec, + struct io_kiocb *req, struct iovec **iovec, struct iov_iter *iter) { - const struct io_uring_sqe *sqe = s->sqe; + const struct io_uring_sqe *sqe = req->submit.sqe; void __user *buf = u64_to_user_ptr(READ_ONCE(sqe->addr)); size_t sqe_len = READ_ONCE(sqe->len); u8 opcode; - /* - * We're reading ->opcode for the second time, but the first read - * doesn't care whether it's _FIXED or not, so it doesn't matter - * whether ->opcode changes concurrently. The first read does care - * about whether it is a READ or a WRITE, so we don't trust this read - * for that purpose and instead let the caller pass in the read/write - * flag. - */ - opcode = READ_ONCE(sqe->opcode); + opcode = req->submit.opcode; if (opcode == IORING_OP_READ_FIXED || opcode == IORING_OP_WRITE_FIXED) { ssize_t ret = io_import_fixed(ctx, rw, sqe, iter); @@ -1278,7 +1271,7 @@ static ssize_t io_import_iovec(struct io_ring_ctx *ctx, int rw, return ret; } - if (!s->has_user) + if (!req->submit.has_user) return -EFAULT; #ifdef CONFIG_COMPAT @@ -1425,7 +1418,7 @@ static int io_read(struct io_kiocb *req, const struct sqe_submit *s, if (unlikely(!(file->f_mode & FMODE_READ))) return -EBADF; - ret = io_import_iovec(req->ctx, READ, s, &iovec, &iter); + ret = io_import_iovec(req->ctx, READ, req, &iovec, &iter); if (ret < 0) return ret; @@ -1490,7 +1483,7 @@ static int io_write(struct io_kiocb *req, const struct sqe_submit *s, if (unlikely(!(file->f_mode & FMODE_WRITE))) return -EBADF; - ret = io_import_iovec(req->ctx, WRITE, s, &iovec, &iter); + ret = io_import_iovec(req->ctx, WRITE, req, &iovec, &iter); if (ret < 0) return ret; @@ -2109,15 +2102,14 @@ static int io_req_defer(struct io_ring_ctx *ctx, struct io_kiocb *req, static int __io_submit_sqe(struct io_ring_ctx *ctx, struct io_kiocb *req, const struct sqe_submit *s, bool force_nonblock) { - int ret, opcode; + int ret; req->user_data = READ_ONCE(s->sqe->user_data); if (unlikely(s->index >= ctx->sq_entries)) return -EINVAL; - opcode = READ_ONCE(s->sqe->opcode); - switch (opcode) { + switch (req->submit.opcode) { case IORING_OP_NOP: ret = io_nop(req, req->user_data); break; @@ -2181,10 +2173,10 @@ static int __io_submit_sqe(struct io_ring_ctx *ctx, struct io_kiocb *req, return 0; } -static struct async_list *io_async_list_from_sqe(struct io_ring_ctx *ctx, - const struct io_uring_sqe *sqe) +static struct async_list *io_async_list_from_req(struct io_ring_ctx *ctx, + struct io_kiocb *req) { - switch (sqe->opcode) { + switch (req->submit.opcode) { case IORING_OP_READV: case IORING_OP_READ_FIXED: return &ctx->pending_async[READ]; @@ -2196,12 +2188,10 @@ static struct async_list *io_async_list_from_sqe(struct io_ring_ctx *ctx, } } -static inline bool io_sqe_needs_user(const struct io_uring_sqe *sqe) +static inline bool io_req_needs_user(struct io_kiocb *req) { - u8 opcode = READ_ONCE(sqe->opcode); - - return !(opcode == IORING_OP_READ_FIXED || - opcode == IORING_OP_WRITE_FIXED); + return !(req->submit.opcode == IORING_OP_READ_FIXED || + req->submit.opcode == IORING_OP_WRITE_FIXED); } static void io_sq_wq_submit_work(struct work_struct *work) @@ -2217,7 +2207,7 @@ static void io_sq_wq_submit_work(struct work_struct *work) int ret; old_cred = override_creds(ctx->creds); - async_list = io_async_list_from_sqe(ctx, req->submit.sqe); + async_list = io_async_list_from_req(ctx, req); allow_kernel_signal(SIGINT); restart: @@ -2239,7 +2229,7 @@ static void io_sq_wq_submit_work(struct work_struct *work) } ret = 0; - if (io_sqe_needs_user(sqe) && !cur_mm) { + if (io_req_needs_user(req) && !cur_mm) { if (!mmget_not_zero(ctx->sqo_mm)) { ret = -EFAULT; } else { @@ -2387,11 +2377,9 @@ static bool io_add_to_prev_work(struct async_list *list, struct io_kiocb *req) return ret; } -static bool io_op_needs_file(const struct io_uring_sqe *sqe) +static bool io_op_needs_file(struct io_kiocb *req) { - int op = READ_ONCE(sqe->opcode); - - switch (op) { + switch (req->submit.opcode) { case IORING_OP_NOP: case IORING_OP_POLL_REMOVE: case IORING_OP_TIMEOUT: @@ -2419,7 +2407,7 @@ static int io_req_set_file(struct io_ring_ctx *ctx, const struct sqe_submit *s, */ req->sequence = s->sequence; - if (!io_op_needs_file(s->sqe)) + if (!io_op_needs_file(req)) return 0; if (flags & IOSQE_FIXED_FILE) { @@ -2460,7 +2448,7 @@ static int __io_queue_sqe(struct io_ring_ctx *ctx, struct io_kiocb *req, s->sqe = sqe_copy; memcpy(&req->submit, s, sizeof(*s)); - list = io_async_list_from_sqe(ctx, s->sqe); + list = io_async_list_from_req(ctx, req); if (!io_add_to_prev_work(list, req)) { if (list) atomic_inc(&list->cnt); @@ -2582,7 +2570,7 @@ static void io_submit_sqe(struct io_ring_ctx *ctx, struct sqe_submit *s, req->user_data = s->sqe->user_data; #if defined(CONFIG_NET) - switch (READ_ONCE(s->sqe->opcode)) { + switch (req->submit.opcode) { case IORING_OP_SENDMSG: case IORING_OP_RECVMSG: spin_lock(¤t->fs->lock); @@ -2697,6 +2685,7 @@ static bool io_get_sqring(struct io_ring_ctx *ctx, struct sqe_submit *s) if (head < ctx->sq_entries) { s->index = head; s->sqe = &ctx->sq_sqes[head]; + s->opcode = READ_ONCE(s->sqe->opcode); s->sequence = ctx->cached_sq_head; ctx->cached_sq_head++; return true; -- Jens Axboe
