Re: [PATCH v3] mm/swap: fix race when skipping swapcache

From: Kairui Song
Date: Sun Feb 18 2024 - 22:10:27 EST

On Mon, Feb 19, 2024 at 10:35 AM Huang, Ying <ying.huang@xxxxxxxxx> wrote:
> "Huang, Ying" <ying.huang@xxxxxxxxx> writes:
> > Kairui Song <ryncsn@xxxxxxxxx> writes:
> >
> >> On Sun, Feb 18, 2024 at 4:34 PM Huang, Ying <ying.huang@xxxxxxxxx> wrote:
> >>>
> >>> Kairui Song <ryncsn@xxxxxxxxx> writes:
> >>>
> >>> > From: Kairui Song <kasong@xxxxxxxxxxx>
> >>> >
> >>> > When skipping swapcache for SWP_SYNCHRONOUS_IO, if two or more threads
> >>> > swapin the same entry at the same time, they get different pages (A, B).
> >>> > Before one thread (T0) finishes the swapin and installs page (A)
> >>> > to the PTE, another thread (T1) could finish swapin of page (B),
> >>> > swap_free the entry, then swap out the possibly modified page
> >>> > reusing the same entry. It breaks the pte_same check in (T0) because
> >>> > PTE value is unchanged, causing ABA problem. Thread (T0) will
> >>> > install a stalled page (A) into the PTE and cause data corruption.
> >>> >
> >>> > One possible callstack is like this:
> >>> >
> >>> > CPU0 CPU1
> >>> > ---- ----
> >>> > do_swap_page() do_swap_page() with same entry
> >>> > <direct swapin path> <direct swapin path>
> >>> > <alloc page A> <alloc page B>
> >>> > swap_read_folio() <- read to page A swap_read_folio() <- read to page B
> >>> > <slow on later locks or interrupt> <finished swapin first>
> >>> > ... set_pte_at()
> >>> > swap_free() <- entry is free
> >>> > <write to page B, now page A stalled>
> >>> > <swap out page B to same swap entry>
> >>> > pte_same() <- Check pass, PTE seems
> >>> > unchanged, but page A
> >>> > is stalled!
> >>> > swap_free() <- page B content lost!
> >>> > set_pte_at() <- staled page A installed!
> >>> >
> >>> > And besides, for ZRAM, swap_free() allows the swap device to discard
> >>> > the entry content, so even if page (B) is not modified, if
> >>> > swap_read_folio() on CPU0 happens later than swap_free() on CPU1,
> >>> > it may also cause data loss.
> >>> >
> >>> > To fix this, reuse swapcache_prepare which will pin the swap entry using
> >>> > the cache flag, and allow only one thread to pin it. Release the pin
> >>> > after PT unlocked. Racers will simply wait since it's a rare and very
> >>> > short event. A schedule() call is added to avoid wasting too much CPU
> >>> > or adding too much noise to perf statistics
> >>> >
> >>> > Other methods like increasing the swap count don't seem to be a good
> >>> > idea after some tests, that will cause racers to fall back to use the
> >>> > swap cache again. Parallel swapin using different methods leads to
> >>> > a much more complex scenario.
> >>>
> >>> The swap entry may be put in swap cache by some parallel code path
> >>> anyway. So, we always need to consider that when reasoning the code.
> >>>
> >>> > Reproducer:
> >>> >
> >>> > This race issue can be triggered easily using a well constructed
> >>> > reproducer and patched brd (with a delay in read path) [1]:
> >>> >
> >>> > With latest 6.8 mainline, race caused data loss can be observed easily:
> >>> > $ gcc -g -lpthread test-thread-swap-race.c && ./a.out
> >>> > Polulating 32MB of memory region...
> >>> > Keep swapping out...
> >>> > Starting round 0...
> >>> > Spawning 65536 workers...
> >>> > 32746 workers spawned, wait for done...
> >>> > Round 0: Error on 0x5aa00, expected 32746, got 32743, 3 data loss!
> >>> > Round 0: Error on 0x395200, expected 32746, got 32743, 3 data loss!
> >>> > Round 0: Error on 0x3fd000, expected 32746, got 32737, 9 data loss!
> >>> > Round 0 Failed, 15 data loss!
> >>> >
> >>> > This reproducer spawns multiple threads sharing the same memory region
> >>> > using a small swap device. Every two threads updates mapped pages one by
> >>> > one in opposite direction trying to create a race, with one dedicated
> >>> > thread keep swapping out the data out using madvise.
> >>> >
> >>> > The reproducer created a reproduce rate of about once every 5 minutes,
> >>> > so the race should be totally possible in production.
> >>> >
> >>> > After this patch, I ran the reproducer for over a few hundred rounds
> >>> > and no data loss observed.
> >>> >
> >>> > Performance overhead is minimal, microbenchmark swapin 10G from 32G
> >>> > zram:
> >>> >
> >>> > Before: 10934698 us
> >>> > After: 11157121 us
> >>> > Non-direct: 13155355 us (Dropping SWP_SYNCHRONOUS_IO flag)
> >>> >
> >>> > Fixes: 0bcac06f27d7 ("mm, swap: skip swapcache for swapin of synchronous device")
> >>> > Link: https://github.com/ryncsn/emm-test-project/tree/master/swap-stress-race [1]
> >>> > Reported-by: "Huang, Ying" <ying.huang@xxxxxxxxx>
> >>> > Closes: https://lore.kernel.org/lkml/87bk92gqpx.fsf_-_@xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx/
> >>> > Signed-off-by: Kairui Song <kasong@xxxxxxxxxxx>
> >>> > Cc: stable@xxxxxxxxxxxxxxx
> >>> >
> >>> > ---
> >>> > Update from V2:
> >>> > - Add a schedule() if raced to prevent repeated page faults wasting CPU
> >>> > and add noise to perf statistics.
> >>> > - Use a bool to state the special case instead of reusing existing
> >>> > variables fixing error handling [Minchan Kim].
> >>> >
> >>> > V2: https://lore.kernel.org/all/20240206182559.32264-1-ryncsn@gmailcom/
> >>> >
> >>> > Update from V1:
> >>> > - Add some words on ZRAM case, it will discard swap content on swap_free so the race window is a bit different but cure is the same. [Barry Song]
> >>> > - Update comments make it cleaner [Huang, Ying]
> >>> > - Add a function place holder to fix CONFIG_SWAP=n built [SeongJae Park]
> >>> > - Update the commit message and summary, refer to SWP_SYNCHRONOUS_IO instead of "direct swapin path" [Yu Zhao]
> >>> > - Update commit message.
> >>> > - Collect Review and Acks.
> >>> >
> >>> > V1: https://lore.kernel.org/all/20240205110959.4021-1-ryncsn@xxxxxxxxx/
> >>> >
> >>> > include/linux/swap.h | 5 +++++
> >>> > mm/memory.c | 20 ++++++++++++++++++++
> >>> > mm/swap.h | 5 +++++
> >>> > mm/swapfile.c | 13 +++++++++++++
> >>> > 4 files changed, 43 insertions(+)
> >>> >
> >>> > diff --git a/include/linux/swap.h b/include/linux/swap.h
> >>> > index 4db00ddad261..8d28f6091a32 100644
> >>> > --- a/include/linux/swap.h
> >>> > +++ b/include/linux/swap.h
> >>> > @@ -549,6 +549,11 @@ static inline int swap_duplicate(swp_entry_t swp)
> >>> > return 0;
> >>> > }
> >>> >
> >>> > +static inline int swapcache_prepare(swp_entry_t swp)
> >>> > +{
> >>> > + return 0;
> >>> > +}
> >>> > +
> >>> > static inline void swap_free(swp_entry_t swp)
> >>> > {
> >>> > }
> >>> > diff --git a/mm/memory.c b/mm/memory.c
> >>> > index 7e1f4849463a..7059230d0a54 100644
> >>> > --- a/mm/memory.c
> >>> > +++ b/mm/memory.c
> >>> > @@ -3799,6 +3799,7 @@ vm_fault_t do_swap_page(struct vm_fault *vmf)
> >>> > struct page *page;
> >>> > struct swap_info_struct *si = NULL;
> >>> > rmap_t rmap_flags = RMAP_NONE;
> >>> > + bool need_clear_cache = false;
> >>> > bool exclusive = false;
> >>> > swp_entry_t entry;
> >>> > pte_t pte;
> >>> > @@ -3867,6 +3868,20 @@ vm_fault_t do_swap_page(struct vm_fault *vmf)
> >>> > if (!folio) {
> >>> > if (data_race(si->flags & SWP_SYNCHRONOUS_IO) &&
> >>> > __swap_count(entry) == 1) {
> >>> > + /*
> >>> > + * Prevent parallel swapin from proceeding with
> >>> > + * the cache flag. Otherwise, another thread may
> >>> > + * finish swapin first, free the entry, and swapout
> >>> > + * reusing the same entry. It's undetectable as
> >>> > + * pte_same() returns true due to entry reuse.
> >>> > + */
> >>> > + if (swapcache_prepare(entry)) {
> >>> > + /* Relax a bit to prevent rapid repeated page faults */
> >>> > + schedule();
> >>>
> >>> The current task may be chosen in schedule(). So, I think that we
> >>> should use cond_resched() here.
> >>>
> >>
> >> I think if we are worried about current task got chosen again we can
> >> use schedule_timeout_uninterruptible(1) here. Isn't cond_resched still
> >> __schedule() and and it can even get omitted, so it should be "weaker"
> >> IIUC.
> >
> > schedule_timeout_uninterruptible(1) will introduce 1ms latency for the
> > second task. That may kill performance of some workloads.

It actually calls schedule_timeout so it should be a 1 jiffy latency,
not 1ms, right?

/**
* schedule_timeout - sleep until timeout
* @timeout: timeout value in jiffies
..

But I think what we really want here is actually the set_current_state
to force yield CPU for a short period. The latency should be mild.

>
> Just found that the cond_sched() in __read_swap_cache_async() has been
> changed to schedule_timeout_uninterruptible(1) to fix some live lock.
> Details are in the description of commit 029c4628b2eb ("mm: swap: get
> rid of livelock in swapin readahead"). I think the similar issue may
> happen here too. So, we must use schedule_timeout_uninterruptible(1)
> here until some other better idea becomes available.

Indeed, I'll switch to schedule_timeout_uninterruptible(1). I've
tested and posted the result with schedule_timeout_uninterruptible(1)
before, it looked fine, or even better.