msg_ring fixes
cover msg_ring fixes
First two patches are msg_ring fixes. 3/3 can go 6.3
Pavel Begunkov (3):
io_uring/msg_ring: fix flagging remote execution io_uring/msg_ring: fix remote queue to disabled ring io_uring/msg_ring: optimise with correct tw notify method
io_uring/io_uring.c | 4 ++-- io_uring/msg_ring.c | 51 ++++++++++++++++++++++++++++++--------------- 2 files changed, 36 insertions(+), 19 deletions(-)
Signature · 1 line
2.38.1
PATCH 1/3 io_uring/msg_ring: fix flagging remote execution
There is a couple of problems with queueing a tw in io_msg_ring_data() for remote execution. First, once we queue it the target ring can go away and so setting IORING_SQ_TASKRUN there is not safe. Secondly, the userspace might not expect IORING_SQ_TASKRUN.
Extract a helper and uniformly use TWA_SIGNAL without TWA_SIGNAL_NO_IPI tricks for now, just as it was done in the original patch.
--- io_uring/msg_ring.c | 40 +++++++++++++++++++++++----------------- 1 file changed, 23 insertions(+), 17 deletions(-)
io_uring/msg_ring.c +23 -17
@@ -58,6 +58,25 @@ void io_msg_ring_cleanup(struct io_kiocb *req) msg->src_file =NULL ; } +static inline bool io_msg_need_remote (struct io_ring_ctx *target_ctx) +{ +if (!target_ctx->task_complete) +return false; +return current != target_ctx->submitter_task; +} + +static int io_msg_exec_remote (struct io_kiocb *req,task_work_func_t func) +{ +struct io_ring_ctx *ctx = req->file->private_data; + struct io_msg *msg = io_kiocb_to_cmd(req, struct io_msg); + +init_task_work (&msg->tw, func); +if (task_work_add (ctx->submitter_task, &msg->tw,TWA_SIGNAL )) +return -EOWNERDEAD ; + +return IOU_ISSUE_SKIP_COMPLETE ; +} +static void io_msg_tw_complete (struct callback_head *head) { struct io_msg *msg = container_of(head, struct io_msg, tw);
@@ -96,15 +115,8 @@ static int io_msg_ring_data(struct io_kiocb *req, unsigned int issue_flags)if (msg->src_fd || msg->dst_fd || msg->flags)return -EINVAL ; -if (target_ctx->task_complete && current != target_ctx->submitter_task) { -init_task_work (&msg->tw, io_msg_tw_complete); -if (task_work_add (target_ctx->submitter_task, &msg->tw, -TWA_SIGNAL_NO_IPI )) -return -EOWNERDEAD ; - -atomic_or (IORING_SQ_TASKRUN , &target_ctx->rings->sq_flags); -return IOU_ISSUE_SKIP_COMPLETE ; - } +if (io_msg_need_remote (target_ctx)) +return io_msg_exec_remote (req, io_msg_tw_complete); ret = -EOVERFLOW ;if (target_ctx->flags &IORING_SETUP_IOPOLL ) {
@@ -202,14 +214,8 @@ static int io_msg_send_fd(struct io_kiocb *req, unsigned int issue_flags) req->flags |=REQ_F_NEED_CLEANUP ; } -if (target_ctx->task_complete && current != target_ctx->submitter_task) { -init_task_work (&msg->tw, io_msg_tw_fd_complete); -if (task_work_add (target_ctx->submitter_task, &msg->tw, -TWA_SIGNAL )) -return -EOWNERDEAD ; - -return IOU_ISSUE_SKIP_COMPLETE ; - } +if (io_msg_need_remote (target_ctx)) +return io_msg_exec_remote (req, io_msg_tw_fd_complete);return io_msg_install_complete (req, issue_flags); }
Signature · 1 line
2.38.1
PATCH 2/3 io_uring/msg_ring: fix remote queue to disabled ring
In reply to Pavel BegunkovIORING_SETUP_R_DISABLED rings don't have the submitter task set, so it's not always safe to use ->submitter_task and we have to check if it has already been set.
--- io_uring/io_uring.c | 4 ++-- io_uring/msg_ring.c | 4 ++++ 2 files changed, 6 insertions(+), 2 deletions(-)
io_uring/io_uring.c +2 -2
@@ -3674,7 +3674,7 @@ static __cold int io_uring_create(unsigned entries, struct io_uring_params *p,if (ctx->flags &IORING_SETUP_SINGLE_ISSUER && !(ctx->flags &IORING_SETUP_R_DISABLED )) - ctx->submitter_task =get_task_struct (current); +WRITE_ONCE (ctx->submitter_task,get_task_struct (current)); file =io_uring_get_file (ctx);if (IS_ERR (file)) {
@@ -3868,7 +3868,7 @@ static int io_register_enable_rings(struct io_ring_ctx *ctx)return -EBADFD ;if (ctx->flags &IORING_SETUP_SINGLE_ISSUER && !ctx->submitter_task) - ctx->submitter_task =get_task_struct (current); +WRITE_ONCE (ctx->submitter_task,get_task_struct (current));if (ctx->restrictions.registered) ctx->restricted =1 ;
io_uring/msg_ring.c +4 -0
@@ -69,6 +69,10 @@ static int io_msg_exec_remote(struct io_kiocb *req, task_work_func_t func) {struct io_ring_ctx *ctx = req->file->private_data; struct io_msg *msg = io_kiocb_to_cmd(req, struct io_msg); +struct task_struct *task =READ_ONCE (ctx->submitter_task); + +if (unlikely (!task)) +return -EOWNERDEAD ;init_task_work (&msg->tw, func);if (task_work_add (ctx->submitter_task, &msg->tw,TWA_SIGNAL ))
Signature · 1 line
2.38.1
PATCH 3/3 io_uring/msg_ring: optimise with correct tw notify method
In reply to Pavel BegunkovWe may want to use TWA_SIGNAL_NO_IPI instead of TWA_SIGNAL if the target ring is configured with IORING_SETUP_COOP_TASKRUN, change io_msg_exec_remote() to use the target ring's ->notify_method.
The caveat is that we have to set IORING_SQ_TASKRUN if the rings asks for it. However, once task_work_add() succeeds the target ring might go away and so we grab a ctx reference to pin the ring until we set IORING_SQ_TASKRUN.
--- io_uring/msg_ring.c | 15 +++++++++++---- 1 file changed, 11 insertions(+), 4 deletions(-)
io_uring/msg_ring.c +11 -4
@@ -70,15 +70,22 @@ static int io_msg_exec_remote(struct io_kiocb *req, task_work_func_t func)struct io_ring_ctx *ctx = req->file->private_data; struct io_msg *msg = io_kiocb_to_cmd(req, struct io_msg);struct task_struct *task =READ_ONCE (ctx->submitter_task); +int ret =IOU_ISSUE_SKIP_COMPLETE ;if (unlikely (!task))return -EOWNERDEAD ; +percpu_ref_get (&ctx->refs);init_task_work (&msg->tw, func); -if (task_work_add (ctx->submitter_task, &msg->tw,TWA_SIGNAL )) -return -EOWNERDEAD ; - -return IOU_ISSUE_SKIP_COMPLETE ; +if (task_work_add (ctx->submitter_task, &msg->tw, ctx->notify_method)) { + ret = -EOWNERDEAD ; + goto out; + } +if (ctx->flags &IORING_SETUP_TASKRUN_FLAG ) +atomic_or (IORING_SQ_TASKRUN , &ctx->rings->sq_flags); +out: +percpu_ref_put (&ctx->refs); +return ret; }static void io_msg_tw_complete (struct callback_head *head)
Signature · 1 line
2.38.1
Jens Axboe
maintainerRe: [PATCH for-6.2 2/3] io_uring/msg_ring: fix remote queue to disabled ring
In reply to Pavel BegunkovQuoted 3 lines · Pavel Begunkov
IORING_SETUP_R_DISABLEDrings don't have the submitter task set, so it's not always safe to use ->submitter_task and we have to check if it has already been set.
As per private discussion, can we just forbid it in general?
Signature · 1 line
Jens Axboe