Skip to contentSkip to conversation

userfaultfd: fix a race between writeprotect and exit_mmap()

PATCH Applied Nadav Amit 3 participants 4 messages Sep 22, 2021 – Sep 24, 2021

Subsystems (3)
Stable branch Memory management VFS
lore
Share & tools
lore

Nadav Amit

author
#0 lore raw

PATCH 1/1 userfaultfd: fix a race between writeprotect and exit_mmap()

From: Nadav Amit <[email protected]>

A race is possible when a process exits, its VMAs are removed by exit_mmap() and at the same time userfaultfd_writeprotect() is called.

The race was detected by KASAN on a development kernel, but it appears to be possible on vanilla kernels as well.

Use mmget_not_zero() to prevent the race as done in other userfaultfd operations.

Cc Peter Xu <[email protected]>
Cc Andrea Arcangeli <[email protected]>
Cc [email protected]
Fixes 63b2d4174c4ad ("userfaultfd: wp: add the writeprotect API to userfaultfd ioctl")
Signed-off-by Nadav Amit <[email protected]>
---
 fs/userfaultfd.c | 12 +++++++++---
 1 file changed, 9 insertions(+), 3 deletions(-)
fs/userfaultfd.c +9 -3
index 003f0d31743e..22bf14ab2d16 100644
@@ -1827,9 +1827,15 @@ static int userfaultfd_writeprotect(struct userfaultfd_ctx *ctx,
 	if (mode_wp && mode_dontwake)
 		return -EINVAL;
 
-	ret = mwriteprotect_range(ctx->mm, uffdio_wp.range.start,
-				  uffdio_wp.range.len, mode_wp,
-				  &ctx->mmap_changing);
+	if (mmget_not_zero(ctx->mm)) {
+		ret = mwriteprotect_range(ctx->mm, uffdio_wp.range.start,
+					  uffdio_wp.range.len, mode_wp,
+					  &ctx->mmap_changing);
+		mmput(ctx->mm);
+	} else {
+		return -ESRCH;
+	}
+
 	if (ret)
 		return ret;
Signature · 1 line
2.25.1
Tested-by on v1 1/1
#1 lore raw

Hi,

I confirmed this patch (applied on 5.14) gets rid of the below userfaultfd test failure.

# ./userfaultfd anon 16 2
nr_pages: 4096, nr_pages_per_cpu: 256
bounces: 1, mode: rnd read, userfaults: 313 missing
(51+34+37+26+41+28+15+20+16+12+13+7+10+2+0+1) 995 wp
(121+79+96+53+90+104+48+61+56+82+56+41+49+26+11+22)
bounces: 0, mode: read, userfaults: 64 missing
(15+8+10+6+5+2+4+3+3+1+4+0+0+2+0+1) 2157 wp
(223+274+189+141+116+132+203+153+143+126+110+114+101+66+42+24)
testing uffd-wp with pagemap (pgsize=4096): done
testing uffd-wp with pagemap (pgsize=2097152): done
testing UFFDIO_ZEROPAGE: done.
testing signal delivery: done.
testing events (fork, remap, remove): ERROR: nr 3933 memory corruption 0 1
 (errno=0, line=963)
ERROR: faulting process failed (errno=0, line=1117)
Quoted 12 lines · Nadav Amit

From: Nadav Amit <[email protected]>

A race is possible when a process exits, its VMAs are removed by exit_mmap() and at the same time userfaultfd_writeprotect() is called.

The race was detected by KASAN on a development kernel, but it appears to be possible on vanilla kernels as well.

Use mmget_not_zero() to prevent the race as done in other userfaultfd operations.

Quoted 7 lines
Cc Peter Xu <[email protected]>
Cc Andrea Arcangeli <[email protected]>
Cc [email protected]
Fixes 63b2d4174c4ad ("userfaultfd: wp: add the writeprotect API to

userfaultfd ioctl")

Signed-off-by Nadav Amit <[email protected]>
Tested-by Li Wang <[email protected]>
Quoted 33 lines
---
 fs/userfaultfd.c | 12 +++++++++---
 1 file changed, 9 insertions(+), 3 deletions(-)
fs/userfaultfd.c +9 -4
index 003f0d31743e..22bf14ab2d16 100644
@@ -1827,9 +1827,15 @@ static int userfaultfd_writeprotect(struct
 userfaultfd_ctx *ctx,
        if (mode_wp && mode_dontwake)
                return -EINVAL;
 
-       ret = mwriteprotect_range(ctx->mm, uffdio_wp.range.start,
-                                 uffdio_wp.range.len, mode_wp,
-                                 &ctx->mmap_changing);
+       if (mmget_not_zero(ctx->mm)) {
+               ret = mwriteprotect_range(ctx->mm, uffdio_wp.range.start,
+                                         uffdio_wp.range.len, mode_wp,
+                                         &ctx->mmap_changing);
+               mmput(ctx->mm);
+       } else {
+               return -ESRCH;
+       }
+
        if (ret)
                return ret;
 
--

2.25.1

Signature · 2 lines
Regards,
Li Wang
#2 lore raw
Quoted 1 line · Li Wang

Hi,

Li,

Quoted 4 lines

I confirmed this patch (applied on 5.14) gets rid of the below userfaultfd test failure.

Quoted 15 lines
# ./userfaultfd anon 16 2
nr_pages: 4096, nr_pages_per_cpu: 256
bounces: 1, mode: rnd read, userfaults: 313 missing
(51+34+37+26+41+28+15+20+16+12+13+7+10+2+0+1) 995 wp
(121+79+96+53+90+104+48+61+56+82+56+41+49+26+11+22)
bounces: 0, mode: read, userfaults: 64 missing
(15+8+10+6+5+2+4+3+3+1+4+0+0+2+0+1) 2157 wp
(223+274+189+141+116+132+203+153+143+126+110+114+101+66+42+24)
testing uffd-wp with pagemap (pgsize=4096): done
testing uffd-wp with pagemap (pgsize=2097152): done
testing UFFDIO_ZEROPAGE: done.
testing signal delivery: done.
testing events (fork, remap, remove): ERROR: nr 3933 memory corruption 0 1
 (errno=0, line=963)
ERROR: faulting process failed (errno=0, line=1117)

Just to keep a record within this thread - my understanding is above issue is a separate issue from what Nadav has fixed. The other fix could be:

https://lore.kernel.org/lkml/[email protected]/

When verify with Nadav's patch, please check whether you have thp enabled globally:

  # echo always > /sys/kernel/mm/transparent_hugepage/enabled

Thanks,

Signature · 1 line
Peter Xu
Reviewed-by on v1 1/1
#3 lore raw
In reply to Nadav Amit
Quoted 9 lines · Nadav Amit

From: Nadav Amit <[email protected]>

A race is possible when a process exits, its VMAs are removed by exit_mmap() and at the same time userfaultfd_writeprotect() is called.

The race was detected by KASAN on a development kernel, but it appears to be possible on vanilla kernels as well.

Quoted 8 lines

Use mmget_not_zero() to prevent the race as done in other userfaultfd operations.

Cc Peter Xu <[email protected]>
Cc Andrea Arcangeli <[email protected]>
Cc [email protected]
Fixes 63b2d4174c4ad ("userfaultfd: wp: add the writeprotect API to userfaultfd ioctl")
Signed-off-by Nadav Amit <[email protected]>
Reviewed-by Peter Xu <[email protected]>

Thanks!

Signature · 1 line
Peter Xu