The patch titled
Subject: mm/gup_benchmark: take the mmap lock around GUP
has been added to the -mm tree. Its filename is
mm-gup_benchmark-take-the-mmap-lock-around-gup.patch
This patch should soon appear at
https://ozlabs.org/~akpm/mmots/broken-out/mm-gup_benchmark-take-the-mmap-lock-around-gup.patch
and later at
https://ozlabs.org/~akpm/mmotm/broken-out/mm-gup_benchmark-take-the-mmap-lock-around-gup.patch
Before you just go and hit "reply", please:
a) Consider who else should be cc'ed
b) Prefer to cc a suitable mailing list as well
c) Ideally: find the original patch on the mailing list and do a
reply-to-all to that, adding suitable additional cc's
*** Remember to use Documentation/process/submit-checklist.rst when testing your code ***
The -mm tree is included into linux-next and is updated
there every 3-4 working days
------------------------------------------------------
From: Jann Horn <jannh@xxxxxxxxxx>
Subject: mm/gup_benchmark: take the mmap lock around GUP
To be safe against concurrent changes to the VMA tree, we must take the
mmap lock around GUP operations (excluding the GUP-fast family of
operations, which will take the mmap lock by themselves if necessary).
This code is only for testing, and it's only reachable by root through
debugfs, so this doesn't really have any impact; however, if we want to
add lockdep asserts into the GUP path, we need to have clean locking here.
Link: https://lkml.kernel.org/r/CAG48ez3SG6ngZLtasxJ6LABpOnqCz5-QHqb0B4k44TQ8F9n6+w@xxxxxxxxxxxxxx
Signed-off-by: Jann Horn <jannh@xxxxxxxxxx>
Reviewed-by: Jason Gunthorpe <jgg@xxxxxxxxxx>
Reviewed-by: John Hubbard <jhubbard@xxxxxxxxxx>
Cc: "Eric W . Biederman" <ebiederm@xxxxxxxxxxxx>
Cc: Michel Lespinasse <walken@xxxxxxxxxx>
Cc: Mauro Carvalho Chehab <mchehab@xxxxxxxxxx>
Cc: Sakari Ailus <sakari.ailus@xxxxxxxxxxxxxxx>
Signed-off-by: Andrew Morton <akpm@xxxxxxxxxxxxxxxxxxxx>
---
mm/gup_benchmark.c | 15 ++++++++++++---
1 file changed, 12 insertions(+), 3 deletions(-)
--- a/mm/gup_benchmark.c~mm-gup_benchmark-take-the-mmap-lock-around-gup
+++ a/mm/gup_benchmark.c
@@ -72,6 +72,8 @@ static int __gup_benchmark_ioctl(unsigne
int nr;
struct page **pages;
int ret = 0;
+ bool needs_mmap_lock =
+ cmd != GUP_FAST_BENCHMARK && cmd != PIN_FAST_BENCHMARK;
if (gup->size > ULONG_MAX)
return -EINVAL;
@@ -81,6 +83,11 @@ static int __gup_benchmark_ioctl(unsigne
if (!pages)
return -ENOMEM;
+ if (needs_mmap_lock && mmap_read_lock_killable(current->mm)) {
+ ret = -EINTR;
+ goto free_pages;
+ }
+
i = 0;
nr = gup->nr_pages_per_call;
start_time = ktime_get();
@@ -120,9 +127,8 @@ static int __gup_benchmark_ioctl(unsigne
pages + i, NULL);
break;
default:
- kvfree(pages);
ret = -EINVAL;
- goto out;
+ goto unlock;
}
if (nr <= 0)
@@ -150,8 +156,11 @@ static int __gup_benchmark_ioctl(unsigne
end_time = ktime_get();
gup->put_delta_usec = ktime_us_delta(end_time, start_time);
+unlock:
+ if (needs_mmap_lock)
+ mmap_read_unlock(current->mm);
+free_pages:
kvfree(pages);
-out:
return ret;
}
_
Patches currently in -mm which might be from jannh@xxxxxxxxxx are
mm-mmu_notifier-fix-mmget-assert-in-__mmu_interval_notifier_insert.patch
binfmt_elf_fdpic-stop-using-dump_emit-on-user-pointers-on-mmu.patch
coredump-let-dump_emit-bail-out-on-short-writes.patch
coredump-refactor-page-range-dumping-into-common-helper.patch
coredump-rework-elf-elf_fdpic-vma_dump_size-into-common-helper.patch
binfmt_elf-binfmt_elf_fdpic-use-a-vma-list-snapshot.patch
mm-gup-take-mmap_lock-in-get_dump_page.patch
mm-remove-the-now-unnecessary-mmget_still_valid-hack.patch
mm-gup_benchmark-take-the-mmap-lock-around-gup.patch
binfmt_elf-take-the-mmap-lock-around-find_extend_vma.patch
mmap-locking-api-dont-check-locking-if-the-mm-isnt-live-yet.patch
mm-gup-assert-that-the-mmap-lock-is-held-in-__get_user_pages.patch
