Re: [PATCH 1/3] io_uring: don't use req->work.creds for inline requests

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

 



hi,

On 26/05/2020 09:43, Xiaoguang Wang wrote:
In io_init_req(), if uers requires a new credentials, currently we'll
save it in req->work.creds, but indeed io_wq_work is designed to describe
needed running environment for requests that will go to io-wq, if one
request is going to be submitted inline, we'd better not touch io_wq_work.
Here add a new 'const struct cred *creds' in io_kiocb, if uers requires a
new credentials, inline requests can use it.

This patch is also a preparation for later patch.

What's the difference from keeping only one creds field in io_kiocb (i.e.
req->work.creds), but handling it specially (i.e. always initialising)? It will
be a lot easier than tossing it around.

Also, the patch doubles {get,put}_creds() for sqe->personality case, and that's
extra atomics without a good reason.
You're right, thanks.
The original motivation for this patch is that it's just a preparation later patch
"io_uring: avoid whole io_wq_work copy for inline requests", I can use io_wq_work.func
to determine whether to drop io_wq_work in io_req_work_drop_env(), so if io_wq_work.func
is NULL, I don't want io_wq_work has a valid creds.
I'll look into whether we can just assign req->creds's pointer to io_wq_work.creds to
reduce the atomic operations.

Regards,
Xiaoguang Wang



Signed-off-by: Xiaoguang Wang <xiaoguang.wang@xxxxxxxxxxxxxxxxx>
---
  fs/io_uring.c | 23 +++++++++++++++--------
  1 file changed, 15 insertions(+), 8 deletions(-)

diff --git a/fs/io_uring.c b/fs/io_uring.c
index 2af87f73848e..788d960abc69 100644
--- a/fs/io_uring.c
+++ b/fs/io_uring.c
@@ -635,6 +635,7 @@ struct io_kiocb {
  	unsigned int		flags;
  	refcount_t		refs;
  	struct task_struct	*task;
+	const struct cred	*creds;
  	unsigned long		fsize;
  	u64			user_data;
  	u32			result;
@@ -1035,8 +1036,10 @@ static inline void io_req_work_grab_env(struct io_kiocb *req,
  		mmgrab(current->mm);
  		req->work.mm = current->mm;
  	}
-	if (!req->work.creds)
+	if (!req->creds)
  		req->work.creds = get_current_cred();
+	else
+		req->work.creds = get_cred(req->creds);
  	if (!req->work.fs && def->needs_fs) {
  		spin_lock(&current->fs->lock);
  		if (!current->fs->in_exec) {
@@ -1368,6 +1371,9 @@ static void __io_req_aux_free(struct io_kiocb *req)
  	if (req->flags & REQ_F_NEED_CLEANUP)
  		io_cleanup_req(req);
+ if (req->creds)
+		put_cred(req->creds);
+
  	kfree(req->io);
  	if (req->file)
  		io_put_file(req, req->file, (req->flags & REQ_F_FIXED_FILE));
@@ -5673,13 +5679,13 @@ static void __io_queue_sqe(struct io_kiocb *req, const struct io_uring_sqe *sqe)
  again:
  	linked_timeout = io_prep_linked_timeout(req);
- if (req->work.creds && req->work.creds != current_cred()) {
+	if (req->creds && req->creds != current_cred()) {
  		if (old_creds)
  			revert_creds(old_creds);
-		if (old_creds == req->work.creds)
+		if (old_creds == req->creds)
  			old_creds = NULL; /* restored original creds */
  		else
-			old_creds = override_creds(req->work.creds);
+			old_creds = override_creds(req->creds);
  	}
ret = io_issue_sqe(req, sqe, true);
@@ -5970,11 +5976,12 @@ static int io_init_req(struct io_ring_ctx *ctx, struct io_kiocb *req,
id = READ_ONCE(sqe->personality);
  	if (id) {
-		req->work.creds = idr_find(&ctx->personality_idr, id);
-		if (unlikely(!req->work.creds))
+		req->creds = idr_find(&ctx->personality_idr, id);
+		if (unlikely(!req->creds))
  			return -EINVAL;
-		get_cred(req->work.creds);
-	}
+		get_cred(req->creds);
+	} else
+		req->creds = NULL;
/* same numerical values with corresponding REQ_F_*, safe to copy */
  	req->flags |= sqe_flags;





[Index of Archives]     [Linux Samsung SoC]     [Linux Rockchip SoC]     [Linux Actions SoC]     [Linux for Synopsys ARC Processors]     [Linux NFS]     [Linux NILFS]     [Linux USB Devel]     [Video for Linux]     [Linux Audio Users]     [Yosemite News]     [Linux Kernel]     [Linux SCSI]


  Powered by Linux