Skip to contentSkip to conversation

mm/swap: fix race condition in direct swapin path

PATCH In review Kairui Song 5 participants 9 messages Feb 5, 2024 – Feb 6, 2024

Subsystems (3)
lore
Share & tools
lore
#0 lore raw

PATCH 1/1 mm/swap: fix race condition in direct swapin path

From: Kairui Song <[email protected]>

In the direct swapin path, when two or more threads swapin the same entry at the same time, they get different pages (A, B) because swap cache is skipped. 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 modify and swap-out the page again, using the same entry. It break the pte_same check because PTE value is unchanged, causing ABA problem. Then thread (T0) will then install the stalled page
(A) into the PTE so new data in page (B) is lost, 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_readpage() <- read to page A   swap_readpage() <- read to page B
<slow on later locks or interrupt>  <finished swapin first>
...                                 set_pte_at()

swap_free() <- Now the entry is freed. <write to page B, now page A stalled> <swap out page B using 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!

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 busy wait since it's a rare and very short event.

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 the cached swapin path, two swapin path being used at the same time leads to a much more complex scenario.

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")

---
Huge thanks to Huang Ying and Chris Li for help finding this issue!

 mm/memory.c   | 19 +++++++++++++++++++
 mm/swap.h     |  5 +++++
 mm/swapfile.c | 16 ++++++++++++++++
 3 files changed, 40 insertions(+)
mm/memory.c +19 -0
index 7e1f4849463a..fd7c55a292f1 100644
@@ -3867,6 +3867,20 @@ vm_fault_t do_swap_page(struct vm_fault *vmf)
 	if (!folio) {
 		if (data_race(si->flags & SWP_SYNCHRONOUS_IO) &&
 		    __swap_count(entry) == 1) {
+			/*
+			 * With swap count == 1, after we read the entry,
+			 * other threads could finish swapin first, free
+			 * the entry, then swapout the modified page using
+			 * the same entry. Now the content we just read is
+			 * stalled, and it's undetectable as pte_same()
+			 * returns true due to entry reuse.
+			 *
+			 * So pin the swap entry using the cache flag even
+			 * cache is not used.
+			 */
+			if (swapcache_prepare(entry))
+				goto out;
+
 			/* skip swapcache */
 			folio = vma_alloc_folio(GFP_HIGHUSER_MOVABLE, 0,
 						vma, vmf->address, false);
@@ -4116,6 +4130,9 @@ vm_fault_t do_swap_page(struct vm_fault *vmf)
 unlock:
 	if (vmf->pte)
 		pte_unmap_unlock(vmf->pte, vmf->ptl);
+	/* Clear the swap cache pin for direct swapin after PTL unlock */
+	if (folio && !swapcache)
+		swapcache_clear(si, entry);
 out:
 	if (si)
 		put_swap_device(si);
@@ -4124,6 +4141,8 @@ vm_fault_t do_swap_page(struct vm_fault *vmf)
 	if (vmf->pte)
 		pte_unmap_unlock(vmf->pte, vmf->ptl);
 out_page:
+	if (!swapcache)
+		swapcache_clear(si, entry);
 	folio_unlock(folio);
 out_release:
 	folio_put(folio);
mm/swap.h +5 -0
index 758c46ca671e..fc2f6ade7f80 100644
@@ -41,6 +41,7 @@ void __delete_from_swap_cache(struct folio *folio,
 void delete_from_swap_cache(struct folio *folio);
 void clear_shadow_from_swap_cache(int type, unsigned long begin,
 				  unsigned long end);
+void swapcache_clear(struct swap_info_struct *si, swp_entry_t entry);
 struct folio *swap_cache_get_folio(swp_entry_t entry,
 		struct vm_area_struct *vma, unsigned long addr);
 struct folio *filemap_get_incore_folio(struct address_space *mapping,
@@ -97,6 +98,10 @@ static inline int swap_writepage(struct page *p, struct writeback_control *wbc)
 	return 0;
 }
 
+static inline void swapcache_clear(struct swap_info_struct *si, swp_entry_t entry)
+{
+}
+
 static inline struct folio *swap_cache_get_folio(swp_entry_t entry,
 		struct vm_area_struct *vma, unsigned long addr)
 {
mm/swapfile.c +16 -0
index 556ff7347d5f..f7d4ad152a7f 100644
@@ -3365,6 +3365,22 @@ int swapcache_prepare(swp_entry_t entry)
 	return __swap_duplicate(entry, SWAP_HAS_CACHE);
 }
 
+/*
+ * Clear the cache flag and release pinned entry.
+ */
+void swapcache_clear(struct swap_info_struct *si, swp_entry_t entry)
+{
+	struct swap_cluster_info *ci;
+	unsigned long offset = swp_offset(entry);
+	unsigned char usage;
+
+	ci = lock_cluster_or_swap_info(si, offset);
+	usage = __swap_entry_free_locked(si, offset, SWAP_HAS_CACHE);
+	unlock_cluster_or_swap_info(si, ci);
+	if (!usage)
+		free_swap_slot(entry);
+}
+
 struct swap_info_struct *swp_swap_info(swp_entry_t entry)
 {
 	return swap_type_to_swap_info(swp_type(entry));
Signature · 1 line
2.43.0
#1 lore raw
Quoted 8 lines · Kairui Song

From: Kairui Song <[email protected]>

In the direct swapin path, when two or more threads swapin the same entry at the same time, they get different pages (A, B) because swap cache is skipped. 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 modify and swap-out the page again, using the

Even if T0's swap_read_folio is later than T1, problems can still happen. after T1 swaps in and sets ptes, then frees the swap entry. T0 reads zRAM later. it will get zero as zRAM will fill zero for freed slot,

static int zram_read_from_zspool(struct zram *zram, struct page *page,

                                 u32 index)
{
       ...


                 value = handle ? zram_get_element(zram, index) : 0;
                mem = kmap_local_page(page);
               zram_fill_page(mem, PAGE_SIZE, value);
                 kunmap_local(mem);
                return 0;
         }

}

Even though nobody modifies the data before the page is swapped out to the same swap offset as before tT0's orig_pte, T0's pte_same check is still true and T0 will map filled zeroed page to pte.

so there is more than one risk besides modified data losses.

Quoted 38 lines

same entry. It break the pte_same check because PTE value is unchanged, causing ABA problem. Then thread (T0) will then install the stalled page
(A) into the PTE so new data in page (B) is lost, 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_readpage() <- read to page A   swap_readpage() <- read to page B
<slow on later locks or interrupt>  <finished swapin first>

.. set_pte_at()
swap_free() <- Now the entry is freed. <write to page B, now page A stalled> <swap out page B using 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!

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 busy wait since it's a rare and very short event.

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 the cached swapin path, two swapin path being used at the same time leads to a much more complex scenario.

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:

Quoted 10 lines
$ 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!

i am also reading these codes recently. It is quite unbelievable this is really happening
now. as freeing swaps is returning slot to slots_ret, but allocating swap is from slots.
so if swapfile is large, the chance that the newly allocated swap was a recently freed swap
is close to 0%. but yes, the code does have the risk.

Quoted 15 lines

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:

Quoted 7 lines
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")

I will also run your patch on my problem I reported today[1]. will update the result to you this week.

[1] https://lore.kernel.org/linux-mm/[email protected]/T/#m41701d0c0e127cdae636e97a13ab521364a810f4

Quoted 103 lines

---
Huge thanks to Huang Ying and Chris Li for help finding this issue!

 mm/memory.c   | 19 +++++++++++++++++++
 mm/swap.h     |  5 +++++
 mm/swapfile.c | 16 ++++++++++++++++
 3 files changed, 40 insertions(+)
mm/memory.c +19 -0
index 7e1f4849463a..fd7c55a292f1 100644
@@ -3867,6 +3867,20 @@ vm_fault_t do_swap_page(struct vm_fault *vmf)
        if (!folio) {
                if (data_race(si->flags & SWP_SYNCHRONOUS_IO) &&
                    __swap_count(entry) == 1) {
+                       /*
+                        * With swap count == 1, after we read the entry,
+                        * other threads could finish swapin first, free
+                        * the entry, then swapout the modified page using
+                        * the same entry. Now the content we just read is
+                        * stalled, and it's undetectable as pte_same()
+                        * returns true due to entry reuse.
+                        *
+                        * So pin the swap entry using the cache flag even
+                        * cache is not used.
+                        */
+                       if (swapcache_prepare(entry))
+                               goto out;
+
                        /* skip swapcache */
                        folio = vma_alloc_folio(GFP_HIGHUSER_MOVABLE, 0,
                                                vma, vmf->address, false);
@@ -4116,6 +4130,9 @@ vm_fault_t do_swap_page(struct vm_fault *vmf)
 unlock:
        if (vmf->pte)
                pte_unmap_unlock(vmf->pte, vmf->ptl);
+       /* Clear the swap cache pin for direct swapin after PTL unlock */
+       if (folio && !swapcache)
+               swapcache_clear(si, entry);
 out:
        if (si)
                put_swap_device(si);
@@ -4124,6 +4141,8 @@ vm_fault_t do_swap_page(struct vm_fault *vmf)
        if (vmf->pte)
                pte_unmap_unlock(vmf->pte, vmf->ptl);
 out_page:
+       if (!swapcache)
+               swapcache_clear(si, entry);
        folio_unlock(folio);
 out_release:
        folio_put(folio);
mm/swap.h +5 -0
index 758c46ca671e..fc2f6ade7f80 100644
@@ -41,6 +41,7 @@ void __delete_from_swap_cache(struct folio *folio,
 void delete_from_swap_cache(struct folio *folio);
 void clear_shadow_from_swap_cache(int type, unsigned long begin,
                                  unsigned long end);
+void swapcache_clear(struct swap_info_struct *si, swp_entry_t entry);
 struct folio *swap_cache_get_folio(swp_entry_t entry,
                struct vm_area_struct *vma, unsigned long addr);
 struct folio *filemap_get_incore_folio(struct address_space *mapping,
@@ -97,6 +98,10 @@ static inline int swap_writepage(struct page *p, struct writeback_control *wbc)
        return 0;
 }
 
+static inline void swapcache_clear(struct swap_info_struct *si, swp_entry_t entry)
+{
+}
+
 static inline struct folio *swap_cache_get_folio(swp_entry_t entry,
                struct vm_area_struct *vma, unsigned long addr)
 {
mm/swapfile.c +16 -1
index 556ff7347d5f..f7d4ad152a7f 100644
@@ -3365,6 +3365,22 @@ int swapcache_prepare(swp_entry_t entry)
        return __swap_duplicate(entry, SWAP_HAS_CACHE);
 }
 
+/*
+ * Clear the cache flag and release pinned entry.
+ */
+void swapcache_clear(struct swap_info_struct *si, swp_entry_t entry)
+{
+       struct swap_cluster_info *ci;
+       unsigned long offset = swp_offset(entry);
+       unsigned char usage;
+
+       ci = lock_cluster_or_swap_info(si, offset);
+       usage = __swap_entry_free_locked(si, offset, SWAP_HAS_CACHE);
+       unlock_cluster_or_swap_info(si, ci);
+       if (!usage)
+               free_swap_slot(entry);
+}
+
 struct swap_info_struct *swp_swap_info(swp_entry_t entry)
 {
        return swap_type_to_swap_info(swp_type(entry));
--
Quoted 3 lines

2.43.0

Thanks
Barry

#2 lore raw
Quoted 1 line · Barry Song

Hi, Barry

Thanks for the comments.

Quoted 26 lines · Kairui Song
Quoted 8 lines

From: Kairui Song <[email protected]>

In the direct swapin path, when two or more threads swapin the same entry at the same time, they get different pages (A, B) because swap cache is skipped. 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 modify and swap-out the page again, using the

Even if T0's swap_read_folio is later than T1, problems can still happen. after T1 swaps in and sets ptes, then frees the swap entry. T0 reads zRAM later. it will get zero as zRAM will fill zero for freed slot,

static int zram_read_from_zspool(struct zram *zram, struct page *page,

                                 u32 index)
{
       ...


                 value = handle ? zram_get_element(zram, index) : 0;
                mem = kmap_local_page(page);
               zram_fill_page(mem, PAGE_SIZE, value);
                 kunmap_local(mem);
                return 0;
         }
Quoted 7 lines

}

Even though nobody modifies the data before the page is swapped out to the same swap offset as before tT0's orig_pte, T0's pte_same check is still true and T0 will map filled zeroed page to pte.

so there is more than one risk besides modified data losses.

Thanks for the complement, I think this is true, and it shares the same problem of the entry reuse, so this patch also covered this potential race. I can add more words later to cover this case as well.

Quoted 50 lines
Quoted 39 lines

same entry. It break the pte_same check because PTE value is unchanged, causing ABA problem. Then thread (T0) will then install the stalled page
(A) into the PTE so new data in page (B) is lost, 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_readpage() <- read to page A   swap_readpage() <- read to page B
<slow on later locks or interrupt>  <finished swapin first>

.. set_pte_at()
swap_free() <- Now the entry is freed. <write to page B, now page A stalled> <swap out page B using 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!

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 busy wait since it's a rare and very short event.

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 the cached swapin path, two swapin path being used at the same time leads to a much more complex scenario.

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:

Quoted 10 lines
$ 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!
Quoted 7 lines

i am also reading these codes recently. It is quite unbelievable this is really happening
now. as freeing swaps is returning slot to slots_ret, but allocating swap is from slots.
so if swapfile is large, the chance that the newly allocated swap was a recently freed swap
is close to 0%. but yes, the code does have the risk.

Indeed, for reproducing I used a 32M swap device, and the data being swapped in/out is large enough to make full use of it. So the reproduce rate is increased by a lot. It's not a completely fictional test as some low end device do have smaller swaps, and real world race could happen in many strange ways.

Quoted 19 lines
Quoted 19 lines

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)
Quoted 9 lines
Quoted 3 lines
Fixes 0bcac06f27d7 ("mm, swap: skip swapcache for swapin of synchronous device")

I will also run your patch on my problem I reported today[1]. will update the result to you this week.

[1] https://lore.kernel.org/linux-mm/[email protected]/T/#m41701d0c0e127cdae636e97a13ab521364a810f4

Thanks!

Reviewed-by on v1 1/1
#3 lore raw
In reply to Kairui Song
Quoted 70 lines · Kairui Song

From: Kairui Song <[email protected]>

In the direct swapin path, when two or more threads swapin the same entry at the same time, they get different pages (A, B) because swap cache is skipped. 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 modify and swap-out the page again, using the same entry. It break the pte_same check because PTE value is unchanged, causing ABA problem. Then thread (T0) will then install the stalled page
(A) into the PTE so new data in page (B) is lost, 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_readpage() <- read to page A   swap_readpage() <- read to page B
<slow on later locks or interrupt>  <finished swapin first>
...                                 set_pte_at()

swap_free() <- Now the entry is freed. <write to page B, now page A stalled> <swap out page B using 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!

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 busy wait since it's a rare and very short event.

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 the cached swapin path, two swapin path being used at the same time leads to a much more complex scenario.

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:

Quoted 7 lines
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")
Quoted 15 lines

---
Huge thanks to Huang Ying and Chris Li for help finding this issue!

 mm/memory.c   | 19 +++++++++++++++++++
 mm/swap.h     |  5 +++++
 mm/swapfile.c | 16 ++++++++++++++++
 3 files changed, 40 insertions(+)
mm/memory.c +0 -0
index 7e1f4849463a..fd7c55a292f1 100644
@@ -3867,6 +3867,20 @@ vm_fault_t do_swap_page(struct vm_fault *vmf)
 	if (!folio) {
 		if (data_race(si->flags & SWP_SYNCHRONOUS_IO) &&

mm/memory.c vm_fault_t do_swap_page(struct vm_fault *vmf) · line 3878

Quoted 10 lines
mm/memory.c +9 -0
 		    __swap_count(entry) == 1) {
+			/*
+			 * With swap count == 1, after we read the entry,
+			 * other threads could finish swapin first, free
+			 * the entry, then swapout the modified page using
+			 * the same entry. Now the content we just read is
+			 * stalled, and it's undetectable as pte_same()
+			 * returns true due to entry reuse.
+			 *
+			 * So pin the swap entry using the cache flag even

"pin" doesn't sound intuitive here. I know that the swap entry will not be freed with this. But now, the parallel swap in will busy waiting. So, I suggest to say,

Prevent parallel swapin from proceeding with the cache flag. Otherwise, it may swapin first, free the entry, then swapout the modified page using the same entry ...

Quoted 55 lines
mm/memory.c +10 -0
+			 * cache is not used.
+			 */
+			if (swapcache_prepare(entry))
+				goto out;
+
 			/* skip swapcache */
 			folio = vma_alloc_folio(GFP_HIGHUSER_MOVABLE, 0,
 						vma, vmf->address, false);
@@ -4116,6 +4130,9 @@ vm_fault_t do_swap_page(struct vm_fault *vmf)
 unlock:
 	if (vmf->pte)
 		pte_unmap_unlock(vmf->pte, vmf->ptl);
+	/* Clear the swap cache pin for direct swapin after PTL unlock */
+	if (folio && !swapcache)
+		swapcache_clear(si, entry);
 out:
 	if (si)
 		put_swap_device(si);
@@ -4124,6 +4141,8 @@ vm_fault_t do_swap_page(struct vm_fault *vmf)
 	if (vmf->pte)
 		pte_unmap_unlock(vmf->pte, vmf->ptl);
 out_page:
+	if (!swapcache)
+		swapcache_clear(si, entry);
 	folio_unlock(folio);
 out_release:
 	folio_put(folio);
mm/swap.h +5 -0
index 758c46ca671e..fc2f6ade7f80 100644
@@ -41,6 +41,7 @@ void __delete_from_swap_cache(struct folio *folio,
 void delete_from_swap_cache(struct folio *folio);
 void clear_shadow_from_swap_cache(int type, unsigned long begin,
 				  unsigned long end);
+void swapcache_clear(struct swap_info_struct *si, swp_entry_t entry);
 struct folio *swap_cache_get_folio(swp_entry_t entry,
 		struct vm_area_struct *vma, unsigned long addr);
 struct folio *filemap_get_incore_folio(struct address_space *mapping,
@@ -97,6 +98,10 @@ static inline int swap_writepage(struct page *p, struct writeback_control *wbc)
 	return 0;
 }
 
+static inline void swapcache_clear(struct swap_info_struct *si, swp_entry_t entry)
+{
+}
+
 static inline struct folio *swap_cache_get_folio(swp_entry_t entry,
 		struct vm_area_struct *vma, unsigned long addr)
 {
mm/swapfile.c +0 -0
index 556ff7347d5f..f7d4ad152a7f 100644

mm/swapfile.c int swapcache_prepare(swp_entry_t entry) · line 3369

Quoted 5 lines
mm/swapfile.c +2 -0
@@ -3365,6 +3365,22 @@ int swapcache_prepare(swp_entry_t entry)
 	return __swap_duplicate(entry, SWAP_HAS_CACHE);
 }
 
+/*
+ * Clear the cache flag and release pinned entry.

Even if we will keep "pin" in above comments, this is hard to be understood for reader. Need a little more explanation like "release pinned entry for device with SWP_SYNCHRONOUS_IO.

Or, just remove the comments. We have comments in calling site already.

Quoted 7 lines
mm/swapfile.c +7 -0
+ */
+void swapcache_clear(struct swap_info_struct *si, swp_entry_t entry)
+{
+	struct swap_cluster_info *ci;
+	unsigned long offset = swp_offset(entry);
+	unsigned char usage;
+

mm/swapfile.c int swapcache_prepare(swp_entry_t entry) · line 3386

Quoted 10 lines
mm/swapfile.c +7 -0
+	ci = lock_cluster_or_swap_info(si, offset);
+	usage = __swap_entry_free_locked(si, offset, SWAP_HAS_CACHE);
+	unlock_cluster_or_swap_info(si, ci);
+	if (!usage)
+		free_swap_slot(entry);
+}
+
 struct swap_info_struct *swp_swap_info(swp_entry_t entry)
 {
 	return swap_type_to_swap_info(swp_type(entry));

Otherwise it looks good for me, Thanks!

#4 lore raw
Quoted 77 lines · Ying
Quoted 74 lines · Kairui Song

From: Kairui Song <[email protected]>

In the direct swapin path, when two or more threads swapin the same entry at the same time, they get different pages (A, B) because swap cache is skipped. 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 modify and swap-out the page again, using the same entry. It break the pte_same check because PTE value is unchanged, causing ABA problem. Then thread (T0) will then install the stalled page
(A) into the PTE so new data in page (B) is lost, 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_readpage() <- read to page A   swap_readpage() <- read to page B
<slow on later locks or interrupt>  <finished swapin first>
...                                 set_pte_at()

swap_free() <- Now the entry is freed. <write to page B, now page A stalled> <swap out page B using 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!

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 busy wait since it's a rare and very short event.

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 the cached swapin path, two swapin path being used at the same time leads to a much more complex scenario.

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)
Quoted 5 lines
Quoted 3 lines
Fixes 0bcac06f27d7 ("mm, swap: skip swapcache for swapin of synchronous device")

Of course :), wasn't sure about how to add your credit, will add this to V2.

Quoted 30 lines
Quoted 15 lines

---
Huge thanks to Huang Ying and Chris Li for help finding this issue!

 mm/memory.c   | 19 +++++++++++++++++++
 mm/swap.h     |  5 +++++
 mm/swapfile.c | 16 ++++++++++++++++
 3 files changed, 40 insertions(+)
mm/memory.c +0 -0
index 7e1f4849463a..fd7c55a292f1 100644
@@ -3867,6 +3867,20 @@ vm_fault_t do_swap_page(struct vm_fault *vmf)
      if (!folio) {
              if (data_race(si->flags & SWP_SYNCHRONOUS_IO) &&

mm/memory.c vm_fault_t do_swap_page(struct vm_fault *vmf) · line 3878

Quoted 10 lines
mm/memory.c +9 -0
                  __swap_count(entry) == 1) {
+                     /*
+                      * With swap count == 1, after we read the entry,
+                      * other threads could finish swapin first, free
+                      * the entry, then swapout the modified page using
+                      * the same entry. Now the content we just read is
+                      * stalled, and it's undetectable as pte_same()
+                      * returns true due to entry reuse.
+                      *
+                      * So pin the swap entry using the cache flag even

"pin" doesn't sound intuitive here. I know that the swap entry will not be freed with this. But now, the parallel swap in will busy waiting. So, I suggest to say,

Quoted 3 lines

Prevent parallel swapin from proceeding with the cache flag. Otherwise, it may swapin first, free the entry, then swapout the modified page using the same entry ...

Good suggestion.

Quoted 59 lines
Quoted 59 lines
mm/memory.c +10 -0
+                      * cache is not used.
+                      */
+                     if (swapcache_prepare(entry))
+                             goto out;
+
                      /* skip swapcache */
                      folio = vma_alloc_folio(GFP_HIGHUSER_MOVABLE, 0,
                                              vma, vmf->address, false);
@@ -4116,6 +4130,9 @@ vm_fault_t do_swap_page(struct vm_fault *vmf)
 unlock:
      if (vmf->pte)
              pte_unmap_unlock(vmf->pte, vmf->ptl);
+     /* Clear the swap cache pin for direct swapin after PTL unlock */
+     if (folio && !swapcache)
+             swapcache_clear(si, entry);
 out:
      if (si)
              put_swap_device(si);
@@ -4124,6 +4141,8 @@ vm_fault_t do_swap_page(struct vm_fault *vmf)
      if (vmf->pte)
              pte_unmap_unlock(vmf->pte, vmf->ptl);
 out_page:
+     if (!swapcache)
+             swapcache_clear(si, entry);
      folio_unlock(folio);
 out_release:
      folio_put(folio);
mm/swap.h +5 -0
index 758c46ca671e..fc2f6ade7f80 100644
@@ -41,6 +41,7 @@ void __delete_from_swap_cache(struct folio *folio,
 void delete_from_swap_cache(struct folio *folio);
 void clear_shadow_from_swap_cache(int type, unsigned long begin,
                                unsigned long end);
+void swapcache_clear(struct swap_info_struct *si, swp_entry_t entry);
 struct folio *swap_cache_get_folio(swp_entry_t entry,
              struct vm_area_struct *vma, unsigned long addr);
 struct folio *filemap_get_incore_folio(struct address_space *mapping,
@@ -97,6 +98,10 @@ static inline int swap_writepage(struct page *p, struct writeback_control *wbc)
      return 0;
 }
 
+static inline void swapcache_clear(struct swap_info_struct *si, swp_entry_t entry)
+{
+}
+
 static inline struct folio *swap_cache_get_folio(swp_entry_t entry,
              struct vm_area_struct *vma, unsigned long addr)
 {
mm/swapfile.c +0 -0
index 556ff7347d5f..f7d4ad152a7f 100644
@@ -3365,6 +3365,22 @@ int swapcache_prepare(swp_entry_t entry)
      return __swap_duplicate(entry, SWAP_HAS_CACHE);
 }
 
Quoted 8 lines

mm/swapfile.c int swapcache_prepare(swp_entry_t entry) · line 3369

Quoted 2 lines
mm/swapfile.c +2 -0
+/*
+ * Clear the cache flag and release pinned entry.

Even if we will keep "pin" in above comments, this is hard to be understood for reader. Need a little more explanation like "release pinned entry for device with SWP_SYNCHRONOUS_IO.

Or, just remove the comments. We have comments in calling site already.

Then I prefer to remove this, there is only one caller, it should be easy to understand.

Quoted 20 lines
Quoted 7 lines
mm/swapfile.c +7 -0
+ */
+void swapcache_clear(struct swap_info_struct *si, swp_entry_t entry)
+{
+     struct swap_cluster_info *ci;
+     unsigned long offset = swp_offset(entry);
+     unsigned char usage;
+

mm/swapfile.c int swapcache_prepare(swp_entry_t entry) · line 3386

Quoted 10 lines
mm/swapfile.c +7 -0
+     ci = lock_cluster_or_swap_info(si, offset);
+     usage = __swap_entry_free_locked(si, offset, SWAP_HAS_CACHE);
+     unlock_cluster_or_swap_info(si, ci);
+     if (!usage)
+             free_swap_slot(entry);
+}
+
 struct swap_info_struct *swp_swap_info(swp_entry_t entry)
 {
      return swap_type_to_swap_info(swp_type(entry));

Otherwise it looks good for me, Thanks!

Quoted 1 line

Thanks for the review.

#5 lore raw
In reply to Kairui Song

Hi Kairui,

On Mon, 5 Feb 2024 19:09:59 +0800 Kairui Song <[email protected]> wrote:

[...]

Quoted 17 lines
 mm/memory.c   | 19 +++++++++++++++++++
 mm/swap.h     |  5 +++++
 mm/swapfile.c | 16 ++++++++++++++++
 3 files changed, 40 insertions(+)
mm/memory.c +4 -0
index 7e1f4849463a..fd7c55a292f1 100644
@@ -3867,6 +3867,20 @@ vm_fault_t do_swap_page(struct vm_fault *vmf)
 	if (!folio) {
 		if (data_race(si->flags & SWP_SYNCHRONOUS_IO) &&
 		    __swap_count(entry) == 1) {
+			/*
+			 * With swap count == 1, after we read the entry,
+			 * other threads could finish swapin first, free
+			 * the entry, then swapout the modified page using

mm/memory.c vm_fault_t do_swap_page(struct vm_fault *vmf) · line 3883

Quoted 10 lines
mm/memory.c +10 -0
+			 * the same entry. Now the content we just read is
+			 * stalled, and it's undetectable as pte_same()
+			 * returns true due to entry reuse.
+			 *
+			 * So pin the swap entry using the cache flag even
+			 * cache is not used.
+			 */
+			if (swapcache_prepare(entry))
+				goto out;
+

I'm getting below build error after this patch. I guess maybe the code need to take care of CONFIG_SWAP unset case?

    .../mm/memory.c: In function 'do_swap_page':
    .../mm/memory.c:4004:8: error: implicit declaration of function 'swapcache_prepare'; did you mean 'swapcache_clear'? [-Werror=implicit-function-declaration]
     4004 |    if (swapcache_prepare(entry))
          |        ^~~~~~~~~~~~~~~~~
          |        swapcache_clear

Thanks,
SJ

[...]

#6 lore raw
Quoted 37 lines · SeongJae Park

Hi Kairui,

On Mon, 5 Feb 2024 19:09:59 +0800 Kairui Song <[email protected]> wrote:

[...]

Quoted 17 lines
 mm/memory.c   | 19 +++++++++++++++++++
 mm/swap.h     |  5 +++++
 mm/swapfile.c | 16 ++++++++++++++++
 3 files changed, 40 insertions(+)
mm/memory.c +4 -0
index 7e1f4849463a..fd7c55a292f1 100644
@@ -3867,6 +3867,20 @@ vm_fault_t do_swap_page(struct vm_fault *vmf)
      if (!folio) {
              if (data_race(si->flags & SWP_SYNCHRONOUS_IO) &&
                  __swap_count(entry) == 1) {
+                     /*
+                      * With swap count == 1, after we read the entry,
+                      * other threads could finish swapin first, free
+                      * the entry, then swapout the modified page using

mm/memory.c vm_fault_t do_swap_page(struct vm_fault *vmf) · line 3883

Quoted 10 lines
mm/memory.c +10 -0
+                      * the same entry. Now the content we just read is
+                      * stalled, and it's undetectable as pte_same()
+                      * returns true due to entry reuse.
+                      *
+                      * So pin the swap entry using the cache flag even
+                      * cache is not used.
+                      */
+                     if (swapcache_prepare(entry))
+                             goto out;
+

I'm getting below build error after this patch. I guess maybe the code need to take care of CONFIG_SWAP unset case?

Quoted 7 lines
    .../mm/memory.c: In function 'do_swap_page':
    .../mm/memory.c:4004:8: error: implicit declaration of function 'swapcache_prepare'; did you mean 'swapcache_clear'? [-Werror=implicit-function-declaration]
     4004 |    if (swapcache_prepare(entry))
          |        ^~~~~~~~~~~~~~~~~
          |        swapcache_clear

Ah, right. Thanks for the feedback.

For the CONFIG_SWAP unset case I added an empty function for swapcache_clear, but the original swapcache_prepare also needs an empty placeholder now. Will add that in V2.

Acked-by on v1 1/1
#7 lore raw
In reply to Kairui Song
Quoted 4 lines · Kairui Song

From: Kairui Song <[email protected]>

In the direct swapin path, when two or more threads swapin the same entry

There is no other places referring to that path as "direct" swapin.

I'd rephrase it as: "When skipping swapcache for SWP_SYNCHRONOUS_IO, ...", and similarly for the subject: "mm: fix race when skipping swapcache".

Quoted 67 lines

at the same time, they get different pages (A, B) because swap cache is skipped. 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 modify and swap-out the page again, using the same entry. It break the pte_same check because PTE value is unchanged, causing ABA problem. Then thread (T0) will then install the stalled page
(A) into the PTE so new data in page (B) is lost, 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_readpage() <- read to page A   swap_readpage() <- read to page B
<slow on later locks or interrupt>  <finished swapin first>
...                                 set_pte_at()

swap_free() <- Now the entry is freed. <write to page B, now page A stalled> <swap out page B using 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!

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 busy wait since it's a rare and very short event.

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 the cached swapin path, two swapin path being used at the same time leads to a much more complex scenario.

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:

Quoted 7 lines
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")
#8 lore raw
Quoted 5 lines · Yu Zhao
Quoted 3 lines · Kairui Song

From: Kairui Song <[email protected]>

Quoted 7 lines
Quoted 1 line

In the direct swapin path, when two or more threads swapin the same entry

There is no other places referring to that path as "direct" swapin.

I'd rephrase it as: "When skipping swapcache for SWP_SYNCHRONOUS_IO, ...", and similarly for the subject: "mm: fix race when skipping swapcache".

Good suggestion.

Quoted 72 lines
Quoted 72 lines

at the same time, they get different pages (A, B) because swap cache is skipped. 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 modify and swap-out the page again, using the same entry. It break the pte_same check because PTE value is unchanged, causing ABA problem. Then thread (T0) will then install the stalled page
(A) into the PTE so new data in page (B) is lost, 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_readpage() <- read to page A   swap_readpage() <- read to page B
<slow on later locks or interrupt>  <finished swapin first>
...                                 set_pte_at()

swap_free() <- Now the entry is freed. <write to page B, now page A stalled> <swap out page B using 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!

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 busy wait since it's a rare and very short event.

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 the cached swapin path, two swapin path being used at the same time leads to a much more complex scenario.

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)
Quoted 7 lines
Quoted 3 lines
Fixes 0bcac06f27d7 ("mm, swap: skip swapcache for swapin of synchronous device")

Thanks!