Re: possible deadlock in __do_page_fault

[Tetsuo Handa <penguin-kernel@i-love.sakura.ne.jp>]

On 2019/01/29 1:45, Joel Fernandes wrote:
>>  		freed += range_size(range);
>> +		mutex_unlock(&ashmem_mutex);
>> +		f->f_op->fallocate(f,
>> +				   FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE,
>> +				   start, end - start);
>> +		fput(f);
>> +		if (atomic_dec_and_test(&ashmem_shrink_inflight))
>> +			wake_up_all(&ashmem_shrink_wait);
>> +		mutex_lock(&ashmem_mutex);
> 
> Let us replace mutex_lock with mutex_trylock, as done before the loop? Here
> is there is an opportunity to not block other ashmem operations. Otherwise
> LGTM. Also, CC stable.

If shrinker succeeded to grab ashmem_mutex using mutex_trylock(), it is
guaranteed that that thread is not inside

  mutex_lock(&ashmem_mutex);
  kmalloc(GFP_KERNEL);
  mutex_unlock(&ashmem_mutex);

block. Therefore, I think that it is safe to use mutex_lock() here.

Nonetheless, although syzbot did not find other dependency, I can update this
patch to use mutex_trylock() if you worry about not-yet-discovered dependency.



From fd850fecd248951ad1ad26b37ec5bf84afe41cbb Mon Sep 17 00:00:00 2001
From: Tetsuo Handa <penguin-kernel@I-love.SAKURA.ne.jp>
Date: Tue, 29 Jan 2019 10:56:47 +0900
Subject: [PATCH v2] staging: android: ashmem: Don't call fallocate() with ashmem_mutex held.

syzbot is hitting lockdep warnings [1][2][3]. This patch tries to fix
the warning by eliminating ashmem_shrink_scan() => {shmem|vfs}_fallocate()
sequence.

[1] https://syzkaller.appspot.com/bug?id=87c399f6fa6955006080b24142e2ce7680295ad4
[2] https://syzkaller.appspot.com/bug?id=7ebea492de7521048355fc84210220e1038a7908
[3] https://syzkaller.appspot.com/bug?id=e02419c12131c24e2a957ea050c2ab6dcbbc3270

Reported-by: syzbot <syzbot+a76129f18c89f3e2ddd4@syzkaller.appspotmail.com>
Reported-by: syzbot <syzbot+148c2885d71194f18d28@syzkaller.appspotmail.com>
Reported-by: syzbot <syzbot+4b8b031b89e6b96c4b2e@syzkaller.appspotmail.com>
Signed-off-by: Tetsuo Handa <penguin-kernel@I-love.SAKURA.ne.jp>
Cc: stable@vger.kernel.org
---
 drivers/staging/android/ashmem.c | 25 ++++++++++++++++++++-----
 1 file changed, 20 insertions(+), 5 deletions(-)

diff --git a/drivers/staging/android/ashmem.c b/drivers/staging/android/ashmem.c
index 90a8a9f1ac7d..ade8438a827a 100644
--- a/drivers/staging/android/ashmem.c
+++ b/drivers/staging/android/ashmem.c
@@ -75,6 +75,9 @@ struct ashmem_range {
 /* LRU list of unpinned pages, protected by ashmem_mutex */
 static LIST_HEAD(ashmem_lru_list);
 
+static atomic_t ashmem_shrink_inflight = ATOMIC_INIT(0);
+static DECLARE_WAIT_QUEUE_HEAD(ashmem_shrink_wait);
+
 /*
  * long lru_count - The count of pages on our LRU list.
  *
@@ -438,7 +441,6 @@ static int ashmem_mmap(struct file *file, struct vm_area_struct *vma)
 static unsigned long
 ashmem_shrink_scan(struct shrinker *shrink, struct shrink_control *sc)
 {
-	struct ashmem_range *range, *next;
 	unsigned long freed = 0;
 
 	/* We might recurse into filesystem code, so bail out if necessary */
@@ -448,21 +450,33 @@ ashmem_shrink_scan(struct shrinker *shrink, struct shrink_control *sc)
 	if (!mutex_trylock(&ashmem_mutex))
 		return -1;
 
-	list_for_each_entry_safe(range, next, &ashmem_lru_list, lru) {
+	while (!list_empty(&ashmem_lru_list)) {
+		struct ashmem_range *range =
+			list_first_entry(&ashmem_lru_list, typeof(*range), lru);
 		loff_t start = range->pgstart * PAGE_SIZE;
 		loff_t end = (range->pgend + 1) * PAGE_SIZE;
+		struct file *f = range->asma->file;
 
-		range->asma->file->f_op->fallocate(range->asma->file,
-				FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE,
-				start, end - start);
+		get_file(f);
+		atomic_inc(&ashmem_shrink_inflight);
 		range->purged = ASHMEM_WAS_PURGED;
 		lru_del(range);
 
 		freed += range_size(range);
+		mutex_unlock(&ashmem_mutex);
+		f->f_op->fallocate(f,
+				   FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE,
+				   start, end - start);
+		fput(f);
+		if (atomic_dec_and_test(&ashmem_shrink_inflight))
+			wake_up_all(&ashmem_shrink_wait);
+		if (!mutex_trylock(&ashmem_mutex))
+			goto out;
 		if (--sc->nr_to_scan <= 0)
 			break;
 	}
 	mutex_unlock(&ashmem_mutex);
+out:
 	return freed;
 }
 
@@ -713,6 +727,7 @@ static int ashmem_pin_unpin(struct ashmem_area *asma, unsigned long cmd,
 		return -EFAULT;
 
 	mutex_lock(&ashmem_mutex);
+	wait_event(ashmem_shrink_wait, !atomic_read(&ashmem_shrink_inflight));
 
 	if (!asma->file)
 		goto out_unlock;
-- 
2.17.1