From mboxrd@z Thu Jan 1 00:00:00 1970 From: Jens Axboe Subject: Re: IORING_REGISTER_CREDS[_UPDATE]() and credfd_create()? Date: Tue, 28 Jan 2020 17:18:47 -0700 Message-ID: References: <688e187a-75dd-89d9-921c-67de228605ce@samba.org> <1ac31828-e915-6180-cdb4-36685442ea75@kernel.dk> <0d4f43d8-a0c4-920b-5b8f-127c1c5a3fad@kernel.dk> <2d7e7fa2-e725-8beb-90b9-6476d48bdb33@gmail.com> <6c401e23-de7c-1fc1-4122-33d53fcf9700@kernel.dk> <35eebae7-76dd-52ee-58b2-4f9e85caee40@kernel.dk> <6415ae98-e205-5374-296d-0442e1ed2034@kernel.dk> Mime-Version: 1.0 Content-Type: text/plain; charset=utf-8 Content-Transfer-Encoding: 7bit Return-path: In-Reply-To: <6415ae98-e205-5374-296d-0442e1ed2034-tSWWG44O7X1aa/9Udqfwiw@public.gmane.org> Content-Language: en-US Sender: linux-api-owner-u79uwXL29TY76Z2rM5mHXA@public.gmane.org To: Pavel Begunkov , Stefan Metzmacher Cc: io-uring , Linux API Mailing List List-Id: linux-api@vger.kernel.org On 1/28/20 5:15 PM, Jens Axboe wrote: > On 1/28/20 5:10 PM, Pavel Begunkov wrote: >> On 29/01/2020 02:51, Jens Axboe wrote: >>> On 1/28/20 4:40 PM, Jens Axboe wrote: >>>> On 1/28/20 4:36 PM, Pavel Begunkov wrote: >>>>> On 28/01/2020 22:42, Jens Axboe wrote: >>>>>> I didn't like it becoming a bit too complicated, both in terms of >>>>>> implementation and use. And the fact that we'd have to jump through >>>>>> hoops to make this work for a full chain. >>>>>> >>>>>> So I punted and just added sqe->personality and IOSQE_PERSONALITY. >>>>>> This makes it way easier to use. Same branch: >>>>>> >>>>>> https://git.kernel.dk/cgit/linux-block/log/?h=for-5.6/io_uring-vfs-creds >>>>>> >>>>>> I'd feel much better with this variant for 5.6. >>>>>> >>>>> >>>>> Checked out ("don't use static creds/mm assignments") >>>>> >>>>> 1. do we miscount cred refs? We grab one in get_current_cred() for each async >>>>> request, but if (worker->creds != work->creds) it will never be put. >>>> >>>> Yeah I think you're right, that needs a bit of fixing up. >>> >> >> Hmm, it seems it leaks it unconditionally, as it grabs in a ref in >> override_creds(). >> >>> I think this may have gotten fixed with the later addition posted today? >>> I'll double check. But for the newer stuff, we put it for both cases >>> when the request is freed. >> >> Yeah, maybe. I got tangled trying to verify both at once and decided to start >> with the old one. >> >> >>>>> 2. shouldn't worker->creds be named {old,saved,etc}_creds? It's set as >>>>> >>>>> worker->creds = override_creds(work->creds); >>>>> >>>>> Where override_creds() returns previous creds. And if so, then the following >>>>> fast check looks strange: >>>>> >>>>> worker->creds != work->creds >>>> >>>> Don't care too much about the naming, but the logic does appear off. >>>> I'll take a look at both of these tonight, unless you beat me to it. >> >> Apparently, you're faster :) >> >>> >>> Testing this now, what a braino. >>> >>> diff --git a/fs/io-wq.c b/fs/io-wq.c >>> index ee49e8852d39..8fbbadf04cc3 100644 >>> --- a/fs/io-wq.c >>> +++ b/fs/io-wq.c >>> @@ -56,7 +56,8 @@ struct io_worker { >>> >>> struct rcu_head rcu; >>> struct mm_struct *mm; >>> - const struct cred *creds; >>> + const struct cred *cur_creds; >>> + const struct cred *saved_creds; >>> struct files_struct *restore_files; >>> }; >>> >>> @@ -135,9 +136,9 @@ static bool __io_worker_unuse(struct io_wqe *wqe, struct io_worker *worker) >>> { >>> bool dropped_lock = false; >>> >>> - if (worker->creds) { >>> - revert_creds(worker->creds); >>> - worker->creds = NULL; >>> + if (worker->saved_creds) { >>> + revert_creds(worker->saved_creds); >>> + worker->cur_creds = worker->saved_creds = NULL; >>> } >>> >>> if (current->files != worker->restore_files) { >>> @@ -424,10 +425,11 @@ static void io_wq_switch_mm(struct io_worker *worker, struct io_wq_work *work) >>> static void io_wq_switch_creds(struct io_worker *worker, >>> struct io_wq_work *work) >>> { >>> - if (worker->creds) >>> - revert_creds(worker->creds); >>> + if (worker->saved_creds) >>> + revert_creds(worker->saved_creds); >>> >>> - worker->creds = override_creds(work->creds); >>> + worker->saved_creds = override_creds(work->creds); >>> + worker->cur_creds = work->creds; >>> } >> >> How about as follows? rever_creds() is a bit heavier than put_creds(). >> >> static void io_wq_switch_creds(struct io_worker *worker, >> struct io_wq_work *work) >> { >> const struct cred *old_creds = override_creds(work->creds); >> >> if (worker->saved_creds) >> put_cred(old_creds); >> else >> worker->saved_creds = old; >> worker->cur_creds = work->creds; >> } > > Looks good to me, I'll fold. Actually, it doesn't clear current->cred then, which seems a bit.. unfortunate. Could be a source of issues. -- Jens Axboe