Some hardware crypto drivers use GFP_ATOMIC allocations in the request
routine. These allocations can randomly fail - for example, they fail if
too many network packets are received.
If we propagated the failure up to the I/O stack, it would cause I/O
errors and data corruption. So, we sleep and retry.
Signed-off-by: Mikulas Patocka <mpato...@redhat.com>
Cc: sta...@vger.kernel.org
Index: linux-2.6/drivers/md/dm-crypt.c
===================================================================
--- linux-2.6.orig/drivers/md/dm-crypt.c
+++ linux-2.6/drivers/md/dm-crypt.c
@@ -1534,6 +1534,7 @@ static blk_status_t crypt_convert(struct
crypt_alloc_req(cc, ctx);
atomic_inc(&ctx->cc_pending);
+again:
if (crypt_integrity_aead(cc))
r = crypt_convert_block_aead(cc, ctx, ctx->r.req_aead,
tag_offset);
else
@@ -1541,6 +1542,17 @@ static blk_status_t crypt_convert(struct
switch (r) {
/*
+ * Some hardware crypto drivers use GFP_ATOMIC allocations in
+ * the request routine. These allocations can randomly fail. If
+ * we propagated the failure up to the I/O stack, it would cause
+ * I/O errors and data corruption.
+ *
+ * So, we sleep and retry.
+ */
+ case -ENOMEM:
+ msleep(1);
+ goto again;
+ /*
* The request was queued by a crypto driver
* but the driver request queue is full, let's wait.
*/
