From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: X-Spam-Checker-Version: SpamAssassin 3.4.0 (2014-02-07) on aws-us-west-2-korg-lkml-1.web.codeaurora.org X-Spam-Level: X-Spam-Status: No, score=-10.3 required=3.0 tests=BAYES_00,DKIM_SIGNED, DKIM_VALID,HEADER_FROM_DIFFERENT_DOMAINS,INCLUDES_CR_TRAILER, MAILING_LIST_MULTI,NICE_REPLY_A,SPF_HELO_NONE,SPF_PASS,URIBL_BLOCKED, USER_AGENT_SANE_1 autolearn=unavailable autolearn_force=no version=3.4.0 Received: from mail.kernel.org (mail.kernel.org [198.145.29.99]) by smtp.lore.kernel.org (Postfix) with ESMTP id 60694C433E0 for ; Sat, 23 Jan 2021 17:38:34 +0000 (UTC) Received: from vger.kernel.org (vger.kernel.org [23.128.96.18]) by mail.kernel.org (Postfix) with ESMTP id 3566D224DE for ; Sat, 23 Jan 2021 17:38:34 +0000 (UTC) Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1725765AbhAWRiL (ORCPT ); Sat, 23 Jan 2021 12:38:11 -0500 Received: from lindbergh.monkeyblade.net ([23.128.96.19]:34586 "EHLO lindbergh.monkeyblade.net" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1726111AbhAWRiI (ORCPT ); Sat, 23 Jan 2021 12:38:08 -0500 Received: from mail-pg1-x52e.google.com (mail-pg1-x52e.google.com [IPv6:2607:f8b0:4864:20::52e]) by lindbergh.monkeyblade.net (Postfix) with ESMTPS id 5C433C0613D6 for ; Sat, 23 Jan 2021 09:37:28 -0800 (PST) Received: by mail-pg1-x52e.google.com with SMTP id i5so6084591pgo.1 for ; Sat, 23 Jan 2021 09:37:28 -0800 (PST) DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=kernel-dk.20150623.gappssmtp.com; s=20150623; h=subject:to:cc:references:from:message-id:date:user-agent :mime-version:in-reply-to:content-language:content-transfer-encoding; bh=zyXonzKaM8k6OOnkDrxM5pAHWQEmapPZdcJnRLpZOUY=; b=1LTdiGYKSeYUIoLVn2hyU/Aq/Ef1TzWLSzSD+FI3sw9U60YDuJtu7JswiavSVAD2FR /wx1dUpnFU0P9teymLn8RbM+6wPArUxVrpJAPyj/jwbC2ndC3SpT9hNPRiYKROFvr3eh jo39gcx0X///m5tIIj2w+vq9yqm47ecbGqOjENWReEygvWp0ejCcUTiBXs6s8NrnVjTj wW0ngs+NBKLdPx/bgkZuDVOp8VdXKJ/YqylCYbsv37ZaJns9BPK33g90be2br47t+mjt iDORfZdudab53iEvgWcrL9UwiAl7jKSW5+SHajADgP23KGzNjnnTvN9VZI6CElBFs/L6 8b7A== X-Google-DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=1e100.net; s=20161025; h=x-gm-message-state:subject:to:cc:references:from:message-id:date :user-agent:mime-version:in-reply-to:content-language :content-transfer-encoding; bh=zyXonzKaM8k6OOnkDrxM5pAHWQEmapPZdcJnRLpZOUY=; b=ECHZpgMPZCY0kviTf7xINIH/8AU8CKiETlhmjmi1/dH1sJuKZibz02zCeGNqJRsJXF UL3oooNeC40y+Oleam7uvwMdhCv04mcMctQCt0XGBsd0jY+eiqPPKDEb7fb+SYj848oV sHDXYyt4dzMbRgCJoD7xIFwROR+NS/zoIsTueL3A10MraxPDhvexX14PvFY+S/hdJvbA pigyHg0YUQkIqC+BF+/ZQHnOhWdV9n6Ao+6ZtUyHun76jo4n3wtQ3fTrcwqQrCJwzvlo HnifaRoVxcaP2uPbDVs5S5I80afVH4Ta6fDHfH+2Qecrvmq/g56EiGMcpX+jnB11F+yD OgOA== X-Gm-Message-State: AOAM531c6rO9W4LpzSMJMJCrs9P4SygWsGcVcIuW3wcbVJIWWU2QxoFL ujIDIsJkcfYBkNtely7VqSQ0gQ== X-Google-Smtp-Source: ABdhPJxRf5qjlvC4skWicX7l2XytqhLPOkLd7uFj0eop1zrDNaLQPmktHwtTCqnRJB5DzSvImpcCcw== X-Received: by 2002:a63:5b1a:: with SMTP id p26mr10584754pgb.76.1611423447701; Sat, 23 Jan 2021 09:37:27 -0800 (PST) Received: from [192.168.4.41] (cpe-72-132-29-68.dc.res.rr.com. [72.132.29.68]) by smtp.gmail.com with ESMTPSA id r14sm12893895pgi.27.2021.01.23.09.37.26 (version=TLS1_3 cipher=TLS_AES_128_GCM_SHA256 bits=128/128); Sat, 23 Jan 2021 09:37:26 -0800 (PST) Subject: Re: [RFC PATCH] io_uring: add support for IORING_OP_GETDENTS64 To: Lennert Buytenhek Cc: linux-kernel@vger.kernel.org, io-uring@vger.kernel.org, linux-fsdevel@vger.kernel.org, linux-btrfs@vger.kernel.org References: <20210123114152.GA120281@wantstofly.org> From: Jens Axboe Message-ID: Date: Sat, 23 Jan 2021 10:37:25 -0700 User-Agent: Mozilla/5.0 (X11; Linux x86_64; rv:68.0) Gecko/20100101 Thunderbird/68.10.0 MIME-Version: 1.0 In-Reply-To: <20210123114152.GA120281@wantstofly.org> Content-Type: text/plain; charset=utf-8 Content-Language: en-US Content-Transfer-Encoding: 7bit Precedence: bulk List-ID: X-Mailing-List: io-uring@vger.kernel.org On 1/23/21 4:41 AM, Lennert Buytenhek wrote: > IORING_OP_GETDENTS64 behaves like getdents64(2) and takes the same > arguments. > > Signed-off-by: Lennert Buytenhek > --- > This seems to work OK, but I'd appreciate a review from someone more > familiar with io_uring internals than I am, as I'm not entirely sure > I did everything quite right. > > A dumb test program for IORING_OP_GETDENTS64 is available here: > > https://krautbox.wantstofly.org/~buytenh/uringfind.c > > This does more or less what find(1) does: it scans recursively through > a directory tree and prints the names of all directories and files it > encounters along the way -- but then using io_uring. (The uring version > prints the names of encountered files and directories in an order that's > determined by SQE completion order, which is somewhat nondeterministic > and likely to differ between runs.) > > On a directory tree with 14-odd million files in it that's on a > six-drive (spinning disk) btrfs raid, find(1) takes: > > # echo 3 > /proc/sys/vm/drop_caches > # time find /mnt/repo > /dev/null > > real 24m7.815s > user 0m15.015s > sys 0m48.340s > # > > And the io_uring version takes: > > # echo 3 > /proc/sys/vm/drop_caches > # time ./uringfind /mnt/repo > /dev/null > > real 10m29.064s > user 0m4.347s > sys 0m1.677s > # > > These timings are repeatable and consistent to within a few seconds. > > (btrfs seems to be sending most metadata reads to the same drive in the > array during this test, even though this filesystem is using the raid1c4 > profile for metadata, so I suspect that more drive-level parallelism can > be extracted with some btrfs tweaks.) > > The fully cached case also shows some speedup for the io_uring version: > > # time find /mnt/repo > /dev/null > > real 0m5.223s > user 0m1.926s > sys 0m3.268s > # > > vs: > > # time ./uringfind /mnt/repo > /dev/null > > real 0m3.604s > user 0m2.417s > sys 0m0.793s > # > > That said, the point of this patch isn't primarily to enable > lightning-fast find(1) or du(1), but more to complete the set of > filesystem I/O primitives available via io_uring, so that applications > can do all of their filesystem I/O using the same mechanism, without > having to manually punt some of their work out to worker threads -- and > indeed, an object storage backend server that I wrote a while ago can > run with a pure io_uring based event loop with this patch. The results look nice for sure. Once concern is that io_uring generally guarantees that any state passed in is stable once submit is done. For the below implementation, that doesn't hold as the linux_dirent64 isn't used until later in the process. That means if you do: submit_getdents64(ring) { struct linux_dirent64 dent; struct io_uring_sqe *sqe; sqe = io_uring_get_sqe(ring); io_uring_prep_getdents64(sqe, ..., &dent); io_uring_submit(ring); } other_func(ring) { struct io_uring_cqe *cqe; submit_getdents64(ring); io_uring_wait_cqe(ring, &cqe); } then the kernel side might get garbage by the time the sqe is actually submitted. This is true because you don't use it inline, only from the out-of-line async context. Usually this is solved by having the prep side copy in the necessary state, eg see io_openat2_prep() for how we make filename and open_how stable by copying them into kernel memory. That ensures that if/when these operations need to go async and finish out-of-line, the contents are stable and there's no requirement for the application to keep them valid once submission is done. Not sure how best to solve that, since the vfs side relies heavily on linux_dirent64 being a user pointer... Outside of that, implementation looks straight forward. -- Jens Axboe