From: Nicholas Piggin <npiggin@gmail.com>
To: Al Viro <viro@ZenIV.linux.org.uk>
Cc: Linus Torvalds <torvalds@linux-foundation.org>,
Dave Chinner <david@fromorbit.com>, CAI Qian <caiqian@redhat.com>,
linux-xfs <linux-xfs@vger.kernel.org>,
xfs@oss.sgi.com, Jens Axboe <axboe@kernel.dk>
Subject: Re: xfs_file_splice_read: possible circular locking dependency detected
Date: Wed, 14 Sep 2016 13:39:25 +1000 [thread overview]
Message-ID: <20160914133925.2fba4629@roar.ozlabs.ibm.com> (raw)
In-Reply-To: <20160914031648.GB2356@ZenIV.linux.org.uk>
On Wed, 14 Sep 2016 04:16:48 +0100
Al Viro <viro@ZenIV.linux.org.uk> wrote:
> [Jens and Nick Cc'd]
>
> On Fri, Sep 09, 2016 at 07:06:29PM -0700, Linus Torvalds wrote:
>
> > I think you'd be better off with just a really small on-stack case
> > (like maybe 2-3 entries), and just allocate anything bigger
> > dynamically. Or you could even see how bad it is if you just
> > force-limit it to max 4 entries or something like that and just do
> > partial writes.
>
> Umm... Right now it tries to allocate as much as the output pipe could
> possibly hold. With default being 16 buffers, you'll end up with doing
> dynamic allocation in all cases (it doesn't even look at the amount of
> data we want to transfer).
>
> The situation with splice_pipe_desc looks very odd:
>
> * all but one instance are on stack frames of some ->splice_read()
> or something called by it (exception is in vmsplice)
>
> * all but one instance (a different one - see below) go through
> splice_grow_spd / splice_to_pipe / splice_shrink_spd sequence and
> nothing else sees them. The exception is skb_splice_bits() and there we
> have MAX_SKB_FRAGS for size, don't bother with grow/shrink and the only
> thing done to that spd is splice_to_pipe() (from the callback passed to
> skb_splice_bits()).
>
> * only one ->splice_read() instance does _not_ create
> splice_pipe_descriptor. It's fuse_dev_splice_read(), and it pays for that
> by open-coding splice_to_pipe(). The only reason for open-coding is that
> we don't have a "stronger SPLICE_F_NONBLOCK" that would fail if the data
> wouldn't fit. SPLICE_F_NONBLOCK stuffs as much as possible and buggers off
> without waiting, fuse_dev_splice_read() wants all or nothing (and no waiting).
>
> * incidentally, we *can't* add new flags - splice(2)/tee(2)/vmsplice(2)
> quietly ignore all bits they do not recognize. In fact, splice(2) ends up
> passing them (unsanitized) to ->splice_read and ->splice_write instances.
>
> * for splice(2) the IO size is limited by nominal capacity of output
> pipe. Looks fairly arbitrary (the limit is the same whether the pipe is
> full or empty), but I wouldn't be surprised if userland programmers would
> get unhappy if they have to take more iterations through their loops.
>
> * the other caller of ->splice_read() is splice_direct_to_actor() and
> that can be called on a fairly deep stack. However, there we loop ourselves
> and smaller chunk size is not a problem.
>
> * in case of skb_splice_bits(), we probably want a grow/shrink pair
> as well, with well below MAX_SKB_FRAGS for a default - what's the typical
> number of fragments per skb?
>
> > So feel free to try maxing out using only a small handful of
> > pipe_buffer entries. Returning partial IO from splice() is fine.
>
> Are you sure that nobody's growing the output pipe buffer before
> doing splice() into it as a way to reduce the amount of iterations?
>
> FWIW, I would love to replace these array of page * + array of
> <offset,len,private> triples with array of pipe_buffer; for one thing,
> this ridiculous ->sbd_release() goes away (we simply call ->ops->release()
> on all unwanted buffers), which gets rid of wonders like
> static void buffer_spd_release(struct splice_pipe_desc *spd, unsigned int i)
> {
> struct buffer_ref *ref =
> (struct buffer_ref *)spd->partial[i].private;
>
> if (--ref->ref)
> return;
>
> ring_buffer_free_read_page(ref->buffer, ref->page);
> kfree(ref);
> spd->partial[i].private = 0;
> }
> static void buffer_pipe_buf_release(struct pipe_inode_info *pipe,
> struct pipe_buffer *buf)
> {
> struct buffer_ref *ref = (struct buffer_ref *)buf->private;
>
> if (--ref->ref)
> return;
>
> ring_buffer_free_read_page(ref->buffer, ref->page);
> kfree(ref);
> buf->private = 0;
> }
>
> pairs that need to be kept in sync, etc.
>
> One inconvenience created by that is stuff like
> spd.nr_pages = find_get_pages_contig(mapping, index, nr_pages, spd.pages);
> in there; granted, this one will go away with __generic_file_splice_read(),
> but e.g. get_iovec_page_array() is using get_user_pages_fast(), which wants
> to put pages next to each other. That one is from vmsplice_to_pipe() guts,
> and I've no idea what the normal use patterns are. OTOH, how much overhead
> would we get from repeated calls of get_user_pages_fast() for e.g. 16 pages
> or so, compared to larger chunks? It is on a shallow stack, so it's not
> as if we couldn't afford a 16-element array of struct page * in there...
Should not be so bad, but I don't have hard numbers for you. PAGEVEC_SIZE
is 14, and that's conceptually rather similar operation (walk radix tree;
grab pages). OTOH many archs are heavier and do locking and vmas walking etc.
Documentation/features/vm/pte_special/arch-support.txt
But even for those, at 16 entries, the bulk of the cost *should* be hitting
struct page cachelines and refcounting. The rest should mostly stay in cache.
next prev parent reply other threads:[~2016-09-14 3:49 UTC|newest]
Thread overview: 135+ messages / expand[flat|nested] mbox.gz Atom feed top
[not found] <723420070.1340881.1472835555274.JavaMail.zimbra@redhat.com>
[not found] ` <1832555471.1341372.1472835736236.JavaMail.zimbra@redhat.com>
2016-09-03 0:39 ` xfs_file_splice_read: possible circular locking dependency detected Dave Chinner
2016-09-03 0:57 ` Linus Torvalds
2016-09-03 1:45 ` Al Viro
2016-09-06 23:59 ` Dave Chinner
2016-09-08 20:35 ` Al Viro
2016-09-06 21:53 ` CAI Qian
2016-09-06 23:34 ` Dave Chinner
2016-09-08 15:29 ` CAI Qian
2016-09-08 17:56 ` Al Viro
2016-09-08 18:12 ` Linus Torvalds
2016-09-08 18:18 ` Linus Torvalds
2016-09-08 20:44 ` Al Viro
2016-09-08 20:57 ` Al Viro
2016-09-08 21:23 ` Al Viro
2016-09-08 21:38 ` Dave Chinner
2016-09-08 23:55 ` Al Viro
2016-09-09 1:53 ` Dave Chinner
2016-09-09 2:22 ` Linus Torvalds
2016-09-09 2:26 ` Linus Torvalds
2016-09-09 2:34 ` Al Viro
2016-09-09 2:50 ` Linus Torvalds
2016-09-09 22:19 ` Al Viro
2016-09-10 2:06 ` Linus Torvalds
2016-09-14 3:16 ` Al Viro
2016-09-14 3:39 ` Nicholas Piggin [this message]
2016-09-14 4:01 ` Linus Torvalds
2016-09-18 5:33 ` Al Viro
2016-09-19 3:08 ` Nicholas Piggin
2016-09-19 6:11 ` Al Viro
2016-09-19 7:26 ` Nicholas Piggin
2016-09-14 3:49 ` Linus Torvalds
2016-09-14 4:26 ` Al Viro
2016-09-17 8:20 ` Al Viro
2016-09-17 19:00 ` Al Viro
2016-09-17 20:15 ` Linus Torvalds
2016-09-23 19:00 ` [RFC][CFT] splice_read reworked Al Viro
2016-09-23 19:01 ` [PATCH 01/11] fix memory leaks in tracing_buffers_splice_read() Al Viro
2016-09-23 19:02 ` [PATCH 02/11] splice_to_pipe(): don't open-code wakeup_pipe_readers() Al Viro
2016-09-23 19:02 ` [PATCH 03/11] splice: switch get_iovec_page_array() to iov_iter Al Viro
2016-09-23 19:03 ` [PATCH 04/11] splice: lift pipe_lock out of splice_to_pipe() Al Viro
2016-09-23 19:45 ` Linus Torvalds
2016-09-23 20:10 ` Al Viro
2016-09-23 20:36 ` Linus Torvalds
2016-09-24 3:59 ` Al Viro
2016-09-24 17:29 ` Al Viro
2016-09-27 15:38 ` Nicholas Piggin
2016-09-27 15:53 ` Chuck Lever
2016-09-24 3:59 ` [PATCH 04/12] " Al Viro
2016-09-26 13:35 ` Miklos Szeredi
2016-09-27 4:14 ` Al Viro
2016-12-17 19:54 ` Andreas Schwab
2016-12-18 19:28 ` Linus Torvalds
2016-12-18 19:57 ` Andreas Schwab
2016-12-18 20:12 ` Al Viro
2016-12-18 20:30 ` Al Viro
2016-12-18 22:10 ` Linus Torvalds
2016-12-18 22:18 ` Al Viro
2016-12-18 22:22 ` Linus Torvalds
2016-12-18 22:49 ` Andreas Schwab
2016-12-21 18:56 ` Andreas Schwab
2016-12-21 19:12 ` Linus Torvalds
2016-09-24 4:00 ` [PATCH 06/12] new helper: add_to_pipe() Al Viro
2016-09-26 13:49 ` Miklos Szeredi
2016-09-24 4:01 ` [PATCH 10/12] new iov_iter flavour: pipe-backed Al Viro
2016-09-29 20:53 ` Miklos Szeredi
2016-09-29 22:50 ` Al Viro
2016-09-30 7:30 ` Miklos Szeredi
2016-10-03 3:34 ` [RFC] O_DIRECT vs EFAULT (was Re: [PATCH 10/12] new iov_iter flavour: pipe-backed) Al Viro
2016-10-03 17:07 ` Linus Torvalds
2016-10-03 18:54 ` Al Viro
2016-09-24 4:01 ` [PATCH 11/12] switch generic_file_splice_read() to use of ->read_iter() Al Viro
2016-09-24 4:02 ` [PATCH 12/12] switch default_file_splice_read() to use of pipe-backed iov_iter Al Viro
2016-09-23 19:03 ` [PATCH 05/11] skb_splice_bits(): get rid of callback Al Viro
2016-09-23 19:04 ` [PATCH 06/11] new helper: add_to_pipe() Al Viro
2016-09-23 19:04 ` [PATCH 07/11] fuse_dev_splice_read(): switch to add_to_pipe() Al Viro
2016-09-23 19:06 ` [PATCH 08/11] cifs: don't use memcpy() to copy struct iov_iter Al Viro
2016-09-23 19:08 ` [PATCH 09/11] fuse_ioctl_copy_user(): don't open-code copy_page_{to,from}_iter() Al Viro
2016-09-26 9:31 ` Miklos Szeredi
2016-09-23 19:09 ` [PATCH 10/11] new iov_iter flavour: pipe-backed Al Viro
2016-09-23 19:10 ` [PATCH 11/11] switch generic_file_splice_read() to use of ->read_iter() Al Viro
2016-09-30 13:32 ` [RFC][CFT] splice_read reworked CAI Qian
2016-09-30 17:42 ` CAI Qian
2016-09-30 18:33 ` CAI Qian
2016-10-03 1:37 ` Al Viro
2016-10-03 17:49 ` CAI Qian
2016-10-04 17:39 ` local DoS - systemd hang or timeout (WAS: Re: [RFC][CFT] splice_read reworked) CAI Qian
2016-10-04 21:42 ` tj
2016-10-05 14:09 ` CAI Qian
2016-10-05 15:30 ` tj
2016-10-05 15:54 ` CAI Qian
2016-10-05 18:57 ` CAI Qian
2016-10-05 20:05 ` Al Viro
2016-10-06 12:20 ` CAI Qian
2016-10-06 12:25 ` CAI Qian
2016-10-06 16:11 ` CAI Qian
2016-10-06 17:00 ` Linus Torvalds
2016-10-06 18:12 ` CAI Qian
2016-10-07 9:57 ` Dave Chinner
2016-10-07 15:25 ` Linus Torvalds
2016-10-07 7:08 ` Jan Kara
2016-10-07 14:43 ` CAI Qian
2016-10-07 15:27 ` CAI Qian
2016-10-07 18:56 ` CAI Qian
2016-10-09 21:54 ` Dave Chinner
2016-10-10 14:10 ` CAI Qian
2016-10-10 20:14 ` CAI Qian
2016-10-10 21:57 ` Dave Chinner
2016-10-12 19:50 ` [bisected] " CAI Qian
2016-10-12 20:59 ` Dave Chinner
2016-10-13 16:25 ` CAI Qian
2016-10-13 20:49 ` Dave Chinner
2016-10-13 20:56 ` CAI Qian
2016-10-09 21:51 ` Dave Chinner
2016-10-07 9:27 ` Dave Chinner
2016-10-03 1:42 ` [RFC][CFT] splice_read reworked Al Viro
2016-10-03 14:06 ` CAI Qian
2016-10-03 15:20 ` CAI Qian
2016-10-03 21:12 ` Dave Chinner
2016-10-04 13:57 ` CAI Qian
2016-10-03 20:32 ` CAI Qian
2016-10-03 20:35 ` Al Viro
2016-10-04 13:29 ` CAI Qian
2016-10-04 14:28 ` Al Viro
2016-10-04 16:21 ` CAI Qian
2016-10-04 20:12 ` Al Viro
2016-10-05 14:30 ` CAI Qian
2016-10-05 16:07 ` Al Viro
2016-09-09 2:31 ` xfs_file_splice_read: possible circular locking dependency detected Al Viro
2016-09-09 2:39 ` Linus Torvalds
2016-09-09 2:26 ` Al Viro
2016-09-09 2:19 ` Al Viro
2016-09-08 18:01 ` Linus Torvalds
2016-09-08 20:39 ` CAI Qian
2016-09-08 21:19 ` Dave Chinner
2016-09-08 21:30 ` Al Viro
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=20160914133925.2fba4629@roar.ozlabs.ibm.com \
--to=npiggin@gmail.com \
--cc=axboe@kernel.dk \
--cc=caiqian@redhat.com \
--cc=david@fromorbit.com \
--cc=linux-xfs@vger.kernel.org \
--cc=torvalds@linux-foundation.org \
--cc=viro@ZenIV.linux.org.uk \
--cc=xfs@oss.sgi.com \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).