lists.openwall.net   lists  /  announce  owl-users  owl-dev  john-users  john-dev  passwdqc-users  yescrypt  popa3d-users  /  oss-security  kernel-hardening  musl  sabotage  tlsify  passwords  /  crypt-dev  xvendor  /  Bugtraq  Full-Disclosure  linux-kernel  linux-netdev  linux-ext4  linux-hardening  linux-cve-announce  PHC 
Open Source and information security mailing list archives
 
Hash Suite: Windows password security audit tool. GUI, reports in PDF.
[<prev] [next>] [<thread-prev] [day] [month] [year] [list]
Message-ID: <20171121213400.GA1503@cmpxchg.org>
Date:   Tue, 21 Nov 2017 16:34:00 -0500
From:   Johannes Weiner <hannes@...xchg.org>
To:     Shakeel Butt <shakeelb@...gle.com>
Cc:     Vlastimil Babka <vbabka@...e.cz>,
        Huang Ying <ying.huang@...el.com>,
        Tim Chen <tim.c.chen@...ux.intel.com>,
        Michal Hocko <mhocko@...nel.org>,
        Greg Thelen <gthelen@...gle.com>,
        Andrew Morton <akpm@...ux-foundation.org>,
        Balbir Singh <bsingharora@...il.com>,
        Minchan Kim <minchan@...nel.org>, Shaohua Li <shli@...com>,
        Jérôme Glisse <jglisse@...hat.com>,
        Jan Kara <jack@...e.cz>, Nicholas Piggin <npiggin@...il.com>,
        Dan Williams <dan.j.williams@...el.com>,
        Mel Gorman <mgorman@...e.de>, Hugh Dickins <hughd@...gle.com>,
        Linux MM <linux-mm@...ck.org>,
        LKML <linux-kernel@...r.kernel.org>
Subject: Re: [PATCH] mm, mlock, vmscan: no more skipping pagevecs

On Tue, Nov 21, 2017 at 10:22:23AM -0800, Shakeel Butt wrote:
> On Tue, Nov 21, 2017 at 7:06 AM, Johannes Weiner <hannes@...xchg.org> wrote:
> > On Tue, Nov 21, 2017 at 01:39:57PM +0100, Vlastimil Babka wrote:
> >> On 11/04/2017 11:43 PM, Shakeel Butt wrote:
> >> > When a thread mlocks an address space backed by file, a new
> >> > page is allocated (assuming file page is not in memory), added
> >> > to the local pagevec (lru_add_pvec), I/O is triggered and the
> >> > thread then sleeps on the page. On I/O completion, the thread
> >> > can wake on a different CPU, the mlock syscall will then sets
> >> > the PageMlocked() bit of the page but will not be able to put
> >> > that page in unevictable LRU as the page is on the pagevec of
> >> > a different CPU. Even on drain, that page will go to evictable
> >> > LRU because the PageMlocked() bit is not checked on pagevec
> >> > drain.
> >> >
> >> > The page will eventually go to right LRU on reclaim but the
> >> > LRU stats will remain skewed for a long time.
> >> >
> >> > However, this issue does not happen for anon pages on swap
> >> > because unlike file pages, anon pages are not added to pagevec
> >> > until they have been fully swapped in. Also the fault handler
> >> > uses vm_flags to set the PageMlocked() bit of such anon pages
> >> > even before returning to mlock() syscall and mlocked pages will
> >> > skip pagevecs and directly be put into unevictable LRU. No such
> >> > luck for file pages.
> >> >
> >> > One way to resolve this issue, is to somehow plumb vm_flags from
> >> > filemap_fault() to add_to_page_cache_lru() which will then skip
> >> > the pagevec for pages of VM_LOCKED vma and directly put them to
> >> > unevictable LRU. However this patch took a different approach.
> >> >
> >> > All the pages, even unevictable, will be added to the pagevecs
> >> > and on the drain, the pages will be added on their LRUs correctly
> >> > by checking their evictability. This resolves the mlocked file
> >> > pages on pagevec of other CPUs issue because when those pagevecs
> >> > will be drained, the mlocked file pages will go to unevictable
> >> > LRU. Also this makes the race with munlock easier to resolve
> >> > because the pagevec drains happen in LRU lock.
> >> >
> >> > There is one (good) side effect though. Without this patch, the
> >> > pages allocated for System V shared memory segment are added to
> >> > evictable LRUs even after shmctl(SHM_LOCK) on that segment. This
> >> > patch will correctly put such pages to unevictable LRU.
> >> >
> >> > Signed-off-by: Shakeel Butt <shakeelb@...gle.com>
> >>
> >> I like the approach in general, as it seems to make the code simpler,
> >> and the diffstats support that. I found no bugs, but I can't say that
> >> with certainty that there aren't any, though. This code is rather
> >> tricky. But it should be enough for an ack, so.
> >>
> >> Acked-by: Vlastimil Babka <vbabka@...e.cz>
> >>
> >> A question below, though.
> >>
> >> ...
> >>
> >> > @@ -883,15 +855,41 @@ void lru_add_page_tail(struct page *page, struct page *page_tail,
> >> >  static void __pagevec_lru_add_fn(struct page *page, struct lruvec *lruvec,
> >> >                              void *arg)
> >> >  {
> >> > -   int file = page_is_file_cache(page);
> >> > -   int active = PageActive(page);
> >> > -   enum lru_list lru = page_lru(page);
> >> > +   enum lru_list lru;
> >> > +   int was_unevictable = TestClearPageUnevictable(page);
> >> >
> >> >     VM_BUG_ON_PAGE(PageLRU(page), page);
> >> >
> >> >     SetPageLRU(page);
> >> > +   /*
> >> > +    * Page becomes evictable in two ways:
> >> > +    * 1) Within LRU lock [munlock_vma_pages() and __munlock_pagevec()].
> >> > +    * 2) Before acquiring LRU lock to put the page to correct LRU and then
> >> > +    *   a) do PageLRU check with lock [check_move_unevictable_pages]
> >> > +    *   b) do PageLRU check before lock [isolate_lru_page]
> >> > +    *
> >> > +    * (1) & (2a) are ok as LRU lock will serialize them. For (2b), if the
> >> > +    * other thread does not observe our setting of PG_lru and fails
> >> > +    * isolation, the following page_evictable() check will make us put
> >> > +    * the page in correct LRU.
> >> > +    */
> >> > +   smp_mb();
> >>
> >> Could you elaborate on the purpose of smp_mb() here? Previously there
> >> was "The other side is TestClearPageMlocked() or shmem_lock()" in
> >> putback_lru_page(), which seems rather unclear to me (neither has an
> >> explicit barrier?).
> >
> > The TestClearPageMlocked() is an RMW operation with return value, and
> > thus an implicit full barrier (see Documentation/atomic_bitops.txt).
> >
> > The ordering is between putback and munlock:
> >
> > #0                         #1
> > list_add(&page->lru,...)   if (TestClearPageMlock())
> > SetPageLRU()                 __munlock_isolate_lru_page()
> > smp_mb()
> > if (page_evictable())
> >   rescue
> >
> > The scenario that the barrier prevents from happening is:
> >
> > list_add(&page->lru,...)
> > if (page_evictable())
> >    rescue
> >                            if (TestClearPageMlock())
> >                                __munlock_isolate_lru_page() // FAILS on !PageLRU
> > SetPageLRU()
> >
> > and now an evictable page is stranded on the unevictable LRU.
> >
> > The barrier guarantees that if #0 doesn't see the page evictable yet,
> > #1 WILL see the PageLRU and succeed in isolation and rescue.
> >
> > Shakeel, please don't drop that "the other side" comment. You mention
> > the places that make the page evictable - which is great, and please
> > keep that as well - but for barriers it's always good to know exactly
> > which operation guarantees the ordering on the other side. In fact, it
> > would be great if you could add comments to the TestClearPageMlocked()
> > sites that mention how they order against the smp_mb() in LRU putback.
> 
> Johannes, I have a question. The example you presented is valid before
> this patch as '#0' was happening outside LRU lock. This patch moves
> '#0' inside LRU lock and '#1' was already in LRU lock therefore no
> issue for this particular scenario. However there is still a
> TestClearPageMlocked() in clear_page_mlock() which happens outside LRU
> lock and same issue which you have explained can happen even with this
> patch (but without smp_mb()).
> 
> So, "the other side" for smp_mb() after this patch will only be the
> TestClearPageMlock() in clear_page_mlock() because all other
> TestClearPageMlocked() instances are serialized by LRU lock. Please
> let me know if I missed something.

You are right, I overlooked the lru lock in __munlock_pagevec(). It's
really only clear_page_mlock() that needs the ordering.

Powered by blists - more mailing lists

Powered by Openwall GNU/*/Linux Powered by OpenVZ