On Mon, Feb 19, 2024 at 11:09 AM Kairui Song <ryncsn@xxxxxxxxx> wrote:
>
> 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@xxxxxxxxx/
> > >>> >
> > >>> > 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.
I just forgot 1 jiffy >= 1 ms here, and uninterruptible should make it
unable to wakeup until timeout...
> > 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.
But this should be still the same though, the minor/major fault ratio
in previous test result [1] shows the race on ZRAM even with threads
set to race on purpose, the chance is low, and thanks for the info on
mentioning another commit!
[1] https://lore.kernel.org/all/CAMgjq7BvTJmxrWQOJvkLt4g_jnvmx07NdU63sGeRMGde4Ov=gA@xxxxxxxxxxxxxx/
