]> git.baikalelectronics.ru Git - kernel.git/commitdiff
io_uring: add lookup table for various opcode needs
authorJens Axboe <axboe@kernel.dk>
Wed, 18 Dec 2019 16:50:26 +0000 (09:50 -0700)
committerJens Axboe <axboe@kernel.dk>
Tue, 21 Jan 2020 00:03:59 +0000 (17:03 -0700)
We currently have various switch statements that check if an opcode needs
a file, mm, etc. These are hard to keep in sync as opcodes are added. Add
a struct io_op_def that holds all of this information, so we have just
one spot to update when opcodes are added.

This also enables us to NOT allocate req->io if a deferred command
doesn't need it, and corrects some mistakes we had in terms of what
commands need mm context.

Signed-off-by: Jens Axboe <axboe@kernel.dk>
fs/io_uring.c

index b086a2dc47e21f4cf5942d94ecffe8b5dd0b716c..244aaccbc82bdb011002e8676dbf96fad95443c5 100644 (file)
@@ -516,6 +516,135 @@ struct io_submit_state {
        unsigned int            ios_left;
 };
 
+struct io_op_def {
+       /* needs req->io allocated for deferral/async */
+       unsigned                async_ctx : 1;
+       /* needs current->mm setup, does mm access */
+       unsigned                needs_mm : 1;
+       /* needs req->file assigned */
+       unsigned                needs_file : 1;
+       /* needs req->file assigned IFF fd is >= 0 */
+       unsigned                fd_non_neg : 1;
+       /* hash wq insertion if file is a regular file */
+       unsigned                hash_reg_file : 1;
+       /* unbound wq insertion if file is a non-regular file */
+       unsigned                unbound_nonreg_file : 1;
+};
+
+static const struct io_op_def io_op_defs[] = {
+       {
+               /* IORING_OP_NOP */
+       },
+       {
+               /* IORING_OP_READV */
+               .async_ctx              = 1,
+               .needs_mm               = 1,
+               .needs_file             = 1,
+               .unbound_nonreg_file    = 1,
+       },
+       {
+               /* IORING_OP_WRITEV */
+               .async_ctx              = 1,
+               .needs_mm               = 1,
+               .needs_file             = 1,
+               .hash_reg_file          = 1,
+               .unbound_nonreg_file    = 1,
+       },
+       {
+               /* IORING_OP_FSYNC */
+               .needs_file             = 1,
+       },
+       {
+               /* IORING_OP_READ_FIXED */
+               .needs_file             = 1,
+               .unbound_nonreg_file    = 1,
+       },
+       {
+               /* IORING_OP_WRITE_FIXED */
+               .needs_file             = 1,
+               .hash_reg_file          = 1,
+               .unbound_nonreg_file    = 1,
+       },
+       {
+               /* IORING_OP_POLL_ADD */
+               .needs_file             = 1,
+               .unbound_nonreg_file    = 1,
+       },
+       {
+               /* IORING_OP_POLL_REMOVE */
+       },
+       {
+               /* IORING_OP_SYNC_FILE_RANGE */
+               .needs_file             = 1,
+       },
+       {
+               /* IORING_OP_SENDMSG */
+               .async_ctx              = 1,
+               .needs_mm               = 1,
+               .needs_file             = 1,
+               .unbound_nonreg_file    = 1,
+       },
+       {
+               /* IORING_OP_RECVMSG */
+               .async_ctx              = 1,
+               .needs_mm               = 1,
+               .needs_file             = 1,
+               .unbound_nonreg_file    = 1,
+       },
+       {
+               /* IORING_OP_TIMEOUT */
+               .async_ctx              = 1,
+               .needs_mm               = 1,
+       },
+       {
+               /* IORING_OP_TIMEOUT_REMOVE */
+       },
+       {
+               /* IORING_OP_ACCEPT */
+               .needs_mm               = 1,
+               .needs_file             = 1,
+               .unbound_nonreg_file    = 1,
+       },
+       {
+               /* IORING_OP_ASYNC_CANCEL */
+       },
+       {
+               /* IORING_OP_LINK_TIMEOUT */
+               .async_ctx              = 1,
+               .needs_mm               = 1,
+       },
+       {
+               /* IORING_OP_CONNECT */
+               .async_ctx              = 1,
+               .needs_mm               = 1,
+               .needs_file             = 1,
+               .unbound_nonreg_file    = 1,
+       },
+       {
+               /* IORING_OP_FALLOCATE */
+               .needs_file             = 1,
+       },
+       {
+               /* IORING_OP_OPENAT */
+               .needs_file             = 1,
+               .fd_non_neg             = 1,
+       },
+       {
+               /* IORING_OP_CLOSE */
+               .needs_file             = 1,
+       },
+       {
+               /* IORING_OP_FILES_UPDATE */
+               .needs_mm               = 1,
+       },
+       {
+               /* IORING_OP_STATX */
+               .needs_mm               = 1,
+               .needs_file             = 1,
+               .fd_non_neg             = 1,
+       },
+};
+
 static void io_wq_submit_work(struct io_wq_work **workptr);
 static void io_cqring_fill_event(struct io_kiocb *req, long res);
 static void io_put_req(struct io_kiocb *req);
@@ -671,41 +800,20 @@ static void __io_commit_cqring(struct io_ring_ctx *ctx)
        }
 }
 
-static inline bool io_req_needs_user(struct io_kiocb *req)
-{
-       return !(req->opcode == IORING_OP_READ_FIXED ||
-                req->opcode == IORING_OP_WRITE_FIXED);
-}
-
 static inline bool io_prep_async_work(struct io_kiocb *req,
                                      struct io_kiocb **link)
 {
+       const struct io_op_def *def = &io_op_defs[req->opcode];
        bool do_hashed = false;
 
-       switch (req->opcode) {
-       case IORING_OP_WRITEV:
-       case IORING_OP_WRITE_FIXED:
-               /* only regular files should be hashed for writes */
-               if (req->flags & REQ_F_ISREG)
+       if (req->flags & REQ_F_ISREG) {
+               if (def->hash_reg_file)
                        do_hashed = true;
-               /* fall-through */
-       case IORING_OP_READV:
-       case IORING_OP_READ_FIXED:
-       case IORING_OP_SENDMSG:
-       case IORING_OP_RECVMSG:
-       case IORING_OP_ACCEPT:
-       case IORING_OP_POLL_ADD:
-       case IORING_OP_CONNECT:
-               /*
-                * We know REQ_F_ISREG is not set on some of these
-                * opcodes, but this enables us to keep the check in
-                * just one place.
-                */
-               if (!(req->flags & REQ_F_ISREG))
+       } else {
+               if (def->unbound_nonreg_file)
                        req->work.flags |= IO_WQ_WORK_UNBOUND;
-               break;
        }
-       if (io_req_needs_user(req))
+       if (def->needs_mm)
                req->work.flags |= IO_WQ_WORK_NEEDS_USER;
 
        *link = io_prep_linked_timeout(req);
@@ -1826,6 +1934,8 @@ static void io_req_map_rw(struct io_kiocb *req, ssize_t io_size,
 
 static int io_alloc_async_ctx(struct io_kiocb *req)
 {
+       if (!io_op_defs[req->opcode].async_ctx)
+               return 0;
        req->io = kmalloc(sizeof(*req->io), GFP_KERNEL);
        return req->io == NULL;
 }
@@ -3765,29 +3875,13 @@ static void io_wq_submit_work(struct io_wq_work **workptr)
                io_wq_assign_next(workptr, nxt);
 }
 
-static bool io_req_op_valid(int op)
-{
-       return op >= IORING_OP_NOP && op < IORING_OP_LAST;
-}
-
 static int io_req_needs_file(struct io_kiocb *req, int fd)
 {
-       switch (req->opcode) {
-       case IORING_OP_NOP:
-       case IORING_OP_POLL_REMOVE:
-       case IORING_OP_TIMEOUT:
-       case IORING_OP_TIMEOUT_REMOVE:
-       case IORING_OP_ASYNC_CANCEL:
-       case IORING_OP_LINK_TIMEOUT:
+       if (!io_op_defs[req->opcode].needs_file)
                return 0;
-       case IORING_OP_OPENAT:
-       case IORING_OP_STATX:
-               return fd != -1;
-       default:
-               if (io_req_op_valid(req->opcode))
-                       return 1;
-               return -EINVAL;
-       }
+       if (fd == -1 && io_op_defs[req->opcode].fd_non_neg)
+               return 0;
+       return 1;
 }
 
 static inline struct file *io_file_from_index(struct io_ring_ctx *ctx,
@@ -3804,7 +3898,7 @@ static int io_req_set_file(struct io_submit_state *state, struct io_kiocb *req,
 {
        struct io_ring_ctx *ctx = req->ctx;
        unsigned flags;
-       int fd, ret;
+       int fd;
 
        flags = READ_ONCE(sqe->flags);
        fd = READ_ONCE(sqe->fd);
@@ -3812,9 +3906,8 @@ static int io_req_set_file(struct io_submit_state *state, struct io_kiocb *req,
        if (flags & IOSQE_IO_DRAIN)
                req->flags |= REQ_F_IO_DRAIN;
 
-       ret = io_req_needs_file(req, fd);
-       if (ret <= 0)
-               return ret;
+       if (!io_req_needs_file(req, fd))
+               return 0;
 
        if (flags & IOSQE_FIXED_FILE) {
                if (unlikely(!ctx->file_data ||
@@ -4240,7 +4333,16 @@ static int io_submit_sqes(struct io_ring_ctx *ctx, unsigned int nr,
                        break;
                }
 
-               if (io_req_needs_user(req) && !*mm) {
+               /* will complete beyond this point, count as submitted */
+               submitted++;
+
+               if (unlikely(req->opcode >= IORING_OP_LAST)) {
+                       io_cqring_add_event(req, -EINVAL);
+                       io_double_put_req(req);
+                       break;
+               }
+
+               if (io_op_defs[req->opcode].needs_mm && !*mm) {
                        mm_fault = mm_fault || !mmget_not_zero(ctx->sqo_mm);
                        if (!mm_fault) {
                                use_mm(ctx->sqo_mm);
@@ -4248,7 +4350,6 @@ static int io_submit_sqes(struct io_ring_ctx *ctx, unsigned int nr,
                        }
                }
 
-               submitted++;
                req->ring_file = ring_file;
                req->ring_fd = ring_fd;
                req->has_user = *mm != NULL;
@@ -6104,6 +6205,7 @@ out_fput:
 
 static int __init io_uring_init(void)
 {
+       BUILD_BUG_ON(ARRAY_SIZE(io_op_defs) != IORING_OP_LAST);
        req_cachep = KMEM_CACHE(io_kiocb, SLAB_HWCACHE_ALIGN | SLAB_PANIC);
        return 0;
 };