drbd: a few more GFP_KERNEL -> GFP_NOIO

This has not yet been observed, but conceivably, when using GFP_KERNEL
allocations from drbd_md_sync(), drbd_flush_after_epoch() or
receive_SyncParam(), we could trigger additional IO to our own device,
or an other device in a criss-cross setup, and end up in a local
deadlock, or potentially a distributed deadlock in a criss-cross setup
involving the peer blocked in a similar way waiting for us to make
progress.

Signed-off-by: Philipp Reisner <philipp.reisner@linbit.com>
Signed-off-by: Lars Ellenberg <lars.ellenberg@linbit.com>
This commit is contained in:
Lars Ellenberg 2012-09-26 14:22:40 +02:00 committed by Philipp Reisner
parent bc891c9ae3
commit 8747d30af9
2 changed files with 2 additions and 2 deletions

View File

@ -2294,7 +2294,7 @@ static inline void drbd_md_flush(struct drbd_conf *mdev)
if (test_bit(MD_NO_FUA, &mdev->flags))
return;
r = blkdev_issue_flush(mdev->ldev->md_bdev, GFP_KERNEL, NULL);
r = blkdev_issue_flush(mdev->ldev->md_bdev, GFP_NOIO, NULL);
if (r) {
set_bit(MD_NO_FUA, &mdev->flags);
dev_err(DEV, "meta data flush failed with status %d, disabling md-flushes\n", r);

View File

@ -473,7 +473,7 @@ struct fifo_buffer *fifo_alloc(int fifo_size)
{
struct fifo_buffer *fb;
fb = kzalloc(sizeof(struct fifo_buffer) + sizeof(int) * fifo_size, GFP_KERNEL);
fb = kzalloc(sizeof(struct fifo_buffer) + sizeof(int) * fifo_size, GFP_NOIO);
if (!fb)
return NULL;