On Fri, Dec 6, 2024 at 10:13 PM David Howells <dhowells@xxxxxxxxxx> wrote: > Can you get some tracing for me? Here's a Brotli-compressed trace (of 6.12.3 + dhowells/netfs-writeback). I can reproduce it this way: 1. log in via SSH (cm4all-lukko is our SSH server) 2. "cp" a file on the Ceph home directory 3. log out - bash hangs inside write(): [<0>] folio_wait_bit_common+0x23a/0x4f0 [<0>] folio_wait_private_2_killable+0x3b/0x80 [<0>] netfs_write_begin+0x5a5/0xb90 [<0>] ceph_write_begin+0x6f/0x150 [<0>] generic_perform_write+0x253/0x7d0 [<0>] ceph_write_iter+0xafd/0x1010 [<0>] vfs_write+0x5bc/0x1030 [<0>] ksys_write+0xee/0x1c0 [<0>] do_syscall_64+0x64/0x100 [<0>] entry_SYSCALL_64_after_hwframe+0x76/0x7e bash tries to write the history file when logging out and that hangs. This apparently always happens after somebody does copy_file_range on Ceph - "cp" in my example. Triggering this is trivial for me, so if you need any more data, I can get it. Max
Attachment:
trace.br
Description: Binary data
