Linux CIFS filesystem development
 help / color / mirror / Atom feed
* [PATCH 00/10] improve smbdirect_mr_io lifetime
@ 2025-10-12 19:10 Stefan Metzmacher
  2025-10-12 19:10 ` [PATCH 01/10] smb: smbdirect: introduce smbdirect_mr_io.{kref,mutex} and SMBDIRECT_MR_DISABLED Stefan Metzmacher
                   ` (10 more replies)
  0 siblings, 11 replies; 14+ messages in thread
From: Stefan Metzmacher @ 2025-10-12 19:10 UTC (permalink / raw)
  To: linux-cifs, samba-technical
  Cc: metze, Steve French, Tom Talpey, Long Li, Namjae Jeon

Hi,

these patches improve and simplify our handling of
smbdirect_mr_io structures and their lifetime.

smbd_register_mr() returns a pointer to struct smbdirect_mr_io
and smbd_deregister_mr() gives the pointer back.

But currently the memory itself is managed by the connection
(struct smbdirect_socket) and smbd_destroy() has a strange
wait loop in order to wait for smbd_deregister_mr() being
called. It means code in smbd_destroy() is aware of
the server mutex in the generic smb client handling above
the transport layer.

These patches do some cleanups and fixes before changing
the logic to use a kref and a mutex in order to allow
smbd_deregister_mr() being called after smbd_destroy()
as the memory of smbdirect_mr_io will stay in memory
but will be detached from the connection.

This makes the code independent of cifs_server_[un]lock()
and will allow us to move more smbdirect code into common
functions (shared between client and server).

I think these should go into 6.18.

Stefan Metzmacher (10):
  smb: smbdirect: introduce smbdirect_mr_io.{kref,mutex} and
    SMBDIRECT_MR_DISABLED
  smb: client: change smbd_deregister_mr() to return void
  smb: client: let destroy_mr_list() call list_del(&mr->list)
  smb: client: let destroy_mr_list() remove locked from the list
  smb: client: improve logic in allocate_mr_list()
  smb: client: improve logic in smbd_register_mr()
  smb: client: improve logic in smbd_deregister_mr()
  smb: client: call ib_dma_unmap_sg if mr->sgt.nents is not 0
  smb: client: let destroy_mr_list() call ib_dereg_mr() before
    ib_dma_unmap_sg()
  smb: client: let destroy_mr_list() keep smbdirect_mr_io memory if
    registered

 fs/smb/client/smbdirect.c                  | 312 ++++++++++++++-------
 fs/smb/client/smbdirect.h                  |   2 +-
 fs/smb/common/smbdirect/smbdirect_socket.h |  11 +-
 3 files changed, 224 insertions(+), 101 deletions(-)

-- 
2.43.0


^ permalink raw reply	[flat|nested] 14+ messages in thread

end of thread, other threads:[~2025-10-15 12:47 UTC | newest]

Thread overview: 14+ messages (download: mbox.gz follow: Atom feed
-- links below jump to the message on this page --
2025-10-12 19:10 [PATCH 00/10] improve smbdirect_mr_io lifetime Stefan Metzmacher
2025-10-12 19:10 ` [PATCH 01/10] smb: smbdirect: introduce smbdirect_mr_io.{kref,mutex} and SMBDIRECT_MR_DISABLED Stefan Metzmacher
2025-10-12 19:10 ` [PATCH 02/10] smb: client: change smbd_deregister_mr() to return void Stefan Metzmacher
2025-10-12 19:10 ` [PATCH 03/10] smb: client: let destroy_mr_list() call list_del(&mr->list) Stefan Metzmacher
2025-10-12 19:10 ` [PATCH 04/10] smb: client: let destroy_mr_list() remove locked from the list Stefan Metzmacher
2025-10-12 19:10 ` [PATCH 05/10] smb: client: improve logic in allocate_mr_list() Stefan Metzmacher
2025-10-12 19:10 ` [PATCH 06/10] smb: client: improve logic in smbd_register_mr() Stefan Metzmacher
2025-10-12 19:10 ` [PATCH 07/10] smb: client: improve logic in smbd_deregister_mr() Stefan Metzmacher
2025-10-12 19:10 ` [PATCH 08/10] smb: client: call ib_dma_unmap_sg if mr->sgt.nents is not 0 Stefan Metzmacher
2025-10-12 19:10 ` [PATCH 09/10] smb: client: let destroy_mr_list() call ib_dereg_mr() before ib_dma_unmap_sg() Stefan Metzmacher
2025-10-12 19:10 ` [PATCH 10/10] smb: client: let destroy_mr_list() keep smbdirect_mr_io memory if registered Stefan Metzmacher
2025-10-15  7:20   ` Stefan Metzmacher
2025-10-15 12:47     ` Steve French
2025-10-13  3:05 ` [PATCH 00/10] improve smbdirect_mr_io lifetime Steve French

This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox