io_uring: test patch for fd passing

This allows a chain link to set ->fd to IOSQE_FD_LAST_OPEN, which will
then be turned into the results from the last open (or accept) request
in the chain. If no open has been done, this isn't valid and the request
will be errored with -EBADF.

With this, we can define chains of open+read+close, where the read and
close part can work on the fd instantiated by the open request.

Totally a work in progress, POC so far.

Signed-off-by: Jens Axboe <axboe@kernel.dk>

diff --git a/fs/io_uring.c b/fs/io_uring.c
index 047bee3e8fdb0488fb3dc904070d43e473f0e1d5..176c0ba68f455d0ad750141997a3879e7b6e3136 100644 (file)
--- a/fs/io_uring.c
+++ b/fs/io_uring.c
@@ -511,6 +511,7 @@ enum {
        REQ_F_OVERFLOW_BIT,
        REQ_F_POLLED_BIT,
        REQ_F_BUFFER_SELECTED_BIT,
+       REQ_F_OPEN_FD_BIT,
 
        /* not a real bit, just to check we're not overflowing the space */
        __REQ_F_LAST_BIT,
@@ -562,6 +563,8 @@ enum {
        REQ_F_POLLED            = BIT(REQ_F_POLLED_BIT),
        /* buffer already selected */
        REQ_F_BUFFER_SELECTED   = BIT(REQ_F_BUFFER_SELECTED_BIT),
+       /* use chain previous open fd */
+       REQ_F_OPEN_FD           = BIT(REQ_F_OPEN_FD_BIT),
 };
 
 struct async_poll {
@@ -600,6 +603,9 @@ struct io_kiocb {
        bool                            needs_fixed_file;
        u8                              opcode;
 
+       int                     last_open_fd;
+       struct file             *last_open_file;
+
        struct io_ring_ctx      *ctx;
        struct list_head        list;
        unsigned int            flags;
@@ -1344,7 +1350,7 @@ static inline void io_put_file(struct io_kiocb *req, struct file *file,
 {
        if (fixed)
                percpu_ref_put(&req->ctx->file_data->refs);
-       else
+       else if (!(req->flags & REQ_F_OPEN_FD))
                fput(file);
 }
 
@@ -1487,6 +1493,14 @@ static void io_req_link_next(struct io_kiocb *req, struct io_kiocb **nxtptr)
                list_del_init(&req->link_list);
                if (!list_empty(&nxt->link_list))
                        nxt->flags |= REQ_F_LINK;
+               if (nxt->flags & REQ_F_OPEN_FD) {
+                       WARN_ON_ONCE(nxt->file);
+                       if (unlikely(!req->last_open_file))
+                               nxt->flags |= REQ_F_FAIL_LINK;
+                       nxt->last_open_file = req->last_open_file;
+                       nxt->last_open_fd = req->last_open_fd;
+                       nxt->file = req->last_open_file;
+               }
                *nxtptr = nxt;
                break;
        }
@@ -2965,8 +2979,8 @@ static int io_openat2_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
 
 static int io_openat2(struct io_kiocb *req, bool force_nonblock)
 {
+       struct file *file = NULL;
        struct open_flags op;
-       struct file *file;
        int ret;
 
        if (force_nonblock)
@@ -2991,8 +3005,12 @@ static int io_openat2(struct io_kiocb *req, bool force_nonblock)
 err:
        putname(req->open.filename);
        req->flags &= ~REQ_F_NEED_CLEANUP;
-       if (ret < 0)
+       if (ret < 0) {
                req_set_fail_links(req);
+       } else if (req->flags & REQ_F_LINK) {
+               req->last_open_file = file;
+               req->last_open_fd = ret;
+       }
        io_cqring_add_event(req, ret);
        io_put_req(req);
        return 0;
@@ -3410,6 +3428,14 @@ static int io_close(struct io_kiocb *req, bool force_nonblock)
 {
        int ret;
 
+       if (req->flags & REQ_F_OPEN_FD) {
+               if (req->close.fd != IOSQE_FD_LAST_OPEN)
+                       return -EBADF;
+               req->close.fd = req->last_open_fd;
+               req->last_open_file = NULL;
+               req->last_open_fd = -1;
+       }
+
        if (req->file->f_op == &io_uring_fops ||
            req->close.fd == req->ctx->ring_fd)
                return -EBADF;
@@ -3963,8 +3989,14 @@ static int __io_accept(struct io_kiocb *req, bool force_nonblock)
                return -EAGAIN;
        if (ret == -ERESTARTSYS)
                ret = -EINTR;
-       if (ret < 0)
+       if (ret < 0) {
                req_set_fail_links(req);
+       } else if (req->flags & REQ_F_LINK) {
+               rcu_read_lock();
+               req->last_open_file = fcheck_files(current->files, ret);
+               rcu_read_unlock();
+               req->last_open_fd = ret;
+       }
        io_cqring_add_event(req, ret);
        io_put_req(req);
        return 0;
@@ -5000,6 +5032,9 @@ static int io_issue_sqe(struct io_kiocb *req, const struct io_uring_sqe *sqe,
        struct io_ring_ctx *ctx = req->ctx;
        int ret;
 
+       if ((req->flags & REQ_F_OPEN_FD) && !req->file)
+               return -EBADF;
+
        switch (req->opcode) {
        case IORING_OP_NOP:
                ret = io_nop(req);
@@ -5336,6 +5371,14 @@ static int io_req_set_file(struct io_submit_state *state, struct io_kiocb *req,
                return 0;
 
        fixed = (flags & IOSQE_FIXED_FILE);
+       if (fd == IOSQE_FD_LAST_OPEN) {
+               if (fixed)
+                       return -EBADF;
+               req->flags |= REQ_F_OPEN_FD;
+               req->file = NULL;
+               return 0;
+       }
+
        if (unlikely(!fixed && req->needs_fixed_file))
                return -EBADF;
 
@@ -5647,6 +5690,9 @@ err_req:
                trace_io_uring_link(ctx, req, head);
                list_add_tail(&req->link_list, &head->link_list);
 
+               req->last_open_fd = -1;
+               req->last_open_file = NULL;
+
                /* last request of a link, enqueue the link */
                if (!(sqe_flags & (IOSQE_IO_LINK|IOSQE_IO_HARDLINK))) {
                        io_queue_link_head(head);
@@ -5659,6 +5705,8 @@ err_req:
                }
                if (sqe_flags & (IOSQE_IO_LINK|IOSQE_IO_HARDLINK)) {
                        req->flags |= REQ_F_LINK;
+                       req->last_open_fd = -1;
+                       req->last_open_file = NULL;
                        INIT_LIST_HEAD(&req->link_list);
                        ret = io_req_defer_prep(req, sqe);
                        if (ret)

diff --git a/include/uapi/linux/io_uring.h b/include/uapi/linux/io_uring.h
index cef4c0c0f26b553e477817dbcd0791b5c56711d5..3ee4ba230c827ddf3900c4a061241634b7ad5f16 100644 (file)
--- a/include/uapi/linux/io_uring.h
+++ b/include/uapi/linux/io_uring.h
@@ -85,6 +85,12 @@ enum {
 /* select buffer from sqe->buf_group */
 #define IOSQE_BUFFER_SELECT    (1U << IOSQE_BUFFER_SELECT_BIT)
 
+/*
+ * 'magic' ->fd values don't point to a real fd, but rather define how fds
+ * can be inherited through links in a chain
+ */
+#define IOSQE_FD_LAST_OPEN     (-4096) /* previous result is fd */
+
 /*
  * io_uring_setup() flags
  */