On 2020/7/28 01:25, Sagi Grimberg wrote:
On 7/26/20 6:52 AM, Coly Li wrote:
Currently nvme_tcp_try_send_data() doesn't use kernel_sendpage() to send slab pages. But for pages allocated by __get_free_pages() without __GFP_COMP, which also have refcount as 0, they are still sent by kernel_sendpage() to remote end, this is problematic.
When bcache uses a remote NVMe SSD via nvme-over-tcp as its cache device, writing meta data e.g. cache_set->disk_buckets to remote SSD may trigger a kernel panic due to the above problem. Bcause the meta data pages for cache_set->disk_buckets are allocated by __get_free_pages() without __GFP_COMP.
This problem should be fixed both in upper layer driver (bcache) and nvme-over-tcp code. This patch fixes the nvme-over-tcp code by checking whether the page refcount is 0, if yes then don't use kernel_sendpage() and call sock_no_sendpage() to send the page into network stack.
Such check is done by macro sendpage_ok() in this patch, which is defined in include/linux/net.h as, (!PageSlab(page) && page_count(page) >= 1) If sendpage_ok() returns false, sock_no_sendpage() will handle the page other than kernel_sendpage().
The code comments in this patch is copied and modified from drbd where the similar problem already gets solved by Philipp Reisner. This is the best code comment including my own version.
Signed-off-by: Coly Li colyli@suse.de Cc: Chaitanya Kulkarni chaitanya.kulkarni@wdc.com Cc: Christoph Hellwig hch@lst.de Cc: Hannes Reinecke hare@suse.de Cc: Jan Kara jack@suse.com Cc: Jens Axboe axboe@kernel.dk Cc: Mikhail Skorzhinskii mskorzhinskiy@solarflare.com Cc: Philipp Reisner philipp.reisner@linbit.com Cc: Sagi Grimberg sagi@grimberg.me Cc: Vlastimil Babka vbabka@suse.com Cc: stable@vger.kernel.org
Changelog: v3: introduce a more common name sendpage_ok() for the open coded check v2: fix typo in patch subject. v1: the initial version.
drivers/nvme/host/tcp.c | 13 +++++++++++-- include/linux/net.h | 2 ++ 2 files changed, 13 insertions(+), 2 deletions(-)
diff --git a/drivers/nvme/host/tcp.c b/drivers/nvme/host/tcp.c index 79ef2b8e2b3c..f9952f6d94b9 100644 --- a/drivers/nvme/host/tcp.c +++ b/drivers/nvme/host/tcp.c @@ -887,8 +887,17 @@ static int nvme_tcp_try_send_data(struct nvme_tcp_request *req) else flags |= MSG_MORE | MSG_SENDPAGE_NOTLAST; - /* can't zcopy slab pages */ - if (unlikely(PageSlab(page))) { + /* + * e.g. XFS meta- & log-data is in slab pages, or bcache meta + * data pages, or other high order pages allocated by + * __get_free_pages() without __GFP_COMP, which have a page_count + * of 0 and/or have PageSlab() set. We cannot use send_page for + * those, as that does get_page(); put_page(); and would cause + * either a VM_BUG directly, or __page_cache_release a page that + * would actually still be referenced by someone, leading to some + * obscure delayed Oops somewhere else. + */
I was hoping that this comment would move to the helper as well.
Sure, I will do that.
Agree with Christoph comment as well.
I will move the inline sendpage_ok() to a separated patch.
Coly Li