On 1/29/19 9:36 AM, Jann Horn wrote: > On Mon, Jan 28, 2019 at 10:36 PM Jens Axboe <axboe@xxxxxxxxx> wrote: >> We normally have to fget/fput for each IO we do on a file. Even with >> the batching we do, the cost of the atomic inc/dec of the file usage >> count adds up. >> >> This adds IORING_REGISTER_FILES, and IORING_UNREGISTER_FILES opcodes >> for the io_uring_register(2) system call. The arguments passed in must >> be an array of __s32 holding file descriptors, and nr_args should hold >> the number of file descriptors the application wishes to pin for the >> duration of the io_uring context (or until IORING_UNREGISTER_FILES is >> called). >> >> When used, the application must set IOSQE_FIXED_FILE in the sqe->flags >> member. Then, instead of setting sqe->fd to the real fd, it sets sqe->fd >> to the index in the array passed in to IORING_REGISTER_FILES. >> >> Files are automatically unregistered when the io_uring context is >> torn down. An application need only unregister if it wishes to >> register a few set of fds. > > s/few/new/ ? Indeed, thanks. >> diff --git a/fs/io_uring.c b/fs/io_uring.c >> index 682714d6f217..77993972879b 100644 >> --- a/fs/io_uring.c >> +++ b/fs/io_uring.c >> @@ -98,6 +98,10 @@ struct io_ring_ctx { >> struct fasync_struct *cq_fasync; >> } ____cacheline_aligned_in_smp; >> >> + /* if used, fixed file set */ >> + struct file **user_files; >> + unsigned nr_user_files; > > It'd be nice if you could add a comment about locking rules here - > something like "writers must ensure that ->refs is dead, readers must > ensure that ->refs is alive as long as the file* is used". Will add. > [...] >> @@ -612,7 +625,14 @@ static int io_prep_rw(struct io_kiocb *req, const struct io_uring_sqe *sqe, >> struct kiocb *kiocb = &req->rw; >> int ret; >> >> - kiocb->ki_filp = io_file_get(state, sqe->fd); >> + if (sqe->flags & IOSQE_FIXED_FILE) { >> + if (unlikely(!ctx->user_files || sqe->fd >= ctx->nr_user_files)) >> + return -EBADF; >> + kiocb->ki_filp = ctx->user_files[sqe->fd]; > > It doesn't really matter as long as ctx->nr_user_files<=INT_MAX, but > it'd be nice if you could explicitly cast sqe->fd to unsigned here. OK, will do. >> + req->flags |= REQ_F_FIXED_FILE; >> + } else { >> + kiocb->ki_filp = io_file_get(state, sqe->fd); >> + } > [...] >> +static int io_sqe_files_register(struct io_ring_ctx *ctx, void __user *arg, >> + unsigned nr_args) >> +{ >> + __s32 __user *fds = (__s32 __user *) arg; >> + int fd, i, ret = 0; >> + >> + if (ctx->user_files) >> + return -EBUSY; >> + if (!nr_args) >> + return -EINVAL; >> + >> + ctx->user_files = kcalloc(nr_args, sizeof(struct file *), GFP_KERNEL); >> + if (!ctx->user_files) >> + return -ENOMEM; >> + >> + for (i = 0; i < nr_args; i++) { >> + ret = -EFAULT; >> + if (copy_from_user(&fd, &fds[i], sizeof(fd))) >> + break; > > "i" is signed, but "nr_args" is unsigned. You can't get through that > kcalloc() call with nr_args>=0x80000000 on a normal kernel, someone > would have to set CONFIG_FORCE_MAX_ZONEORDER really high for that, but > still, in theory, if you reach this copy_to_user(..., &fds[i], ...) > with a negative "i", that'd be bad. You might want to make "i" > unsigned here and check that it's at least smaller than UINT_MAX... Done. Thanks for your review! -- Jens Axboe