Hi Linus,
Could you pull this in the upcoming merge window please? The majority of
the patches are aimed at fixing and improving the AFS filesystem's rotation
over server IP addresses, but there are also some fixes from Oleg Nesterov
for the use of read_seqbegin_or_lock().
(1) Fix fileserver probe handling so that the next round of probes doesn't
break ongoing server/address rotation by clearing all the probe result
tracking. This could occasionally cause the rotation algorithm to
drop straight through, give a 'successful' result without actually
emitting any RPC calls, leaving the reply buffer in an undefined
state.
Instead, detach the probe results into a separate struct
and allocate a new one each time we start probing and update the
pointer to it. Probes are also sent in order of address preference to
try and improve the chance that the preferred one will complete first.
(2) Fix server rotation so that it uses configurable address preferences
across on the probes that have completed so far than ranking them by
RTT as the latter doesn't necessarily give the best route. The
preference list can be altered by writing into
/proc/net/afs/addr_prefs.
(3) Fix the handling of Read-Only (and Backup) volume callbacks as there
is one per volume, not one per file, so if someone performs a command
that, say, offlines the volume but doesn't change it, when it comes
back online we don't spam the server with a status fetch for every
vnode we're using. Instead, check the Creation timestamp in the
VolSync record when prompted by a callback break.
(4) Handle volume regression (ie. a RW volume being restored from a
backup) by scrubbing all cache data for that volume. This is detected
from the VolSync creation timestamp.
(5) Adjust abort handling and abort -> error mapping to match better with
what other AFS clients do.
(6) Fix offline and busy volume state handling as they only apply to
individual server instances and not entire volumes and the rotation
algorithm should go and look at other servers if available. Also make
it sleep briefly before each retry if all the volume instances are
unavailable.
Tested-by: Marc Dionne <[email protected]>
Thanks,
David
Changes
=======
ver #3)
- Return an error from afs_d_revalidate() if validation of parent dir
fails with ERESTARTSYS.
- afs_update_volume_state() must only update the recorded volume
expiration time if we obtained a callback record from the server.
- Actually call afs_update_volume_state() (this was being done in a later
patch).
ver #2)
- Drop the first two rxrpc fix patches - one has gone through the net tree
and the other needs a bit more work, but neither is necessary for this
series.
- Add a couple of missing symbol exports.
- Treat UAEIO as VIO too.
- Switch to using atomic64_t for creation & update times because 64-bit
cmpxchg isn't available on some 32-bit arches.
- Some patches went upstream separately as fixes (commit
5b7ad877e4d81f8904ce83982b1ba5c6e83deccb).
- Use atomic64_t for vnode->cb_expires_at() as 64-bit xchg() is not
univerally available.
- Use rcu_access_pointer() rather than passing an __rcu pointer directly to
kfree_rcu().
Link: https://lore.kernel.org/r/[email protected]/ # v1
Link: https://lore.kernel.org/r/[email protected]/ # v2
---
The following changes since commit 861deac3b092f37b2c5e6871732f3e11486f7082:
Linux 6.7-rc7 (2023-12-23 16:25:56 -0800)
are available in the Git repository at:
git://git.kernel.org/pub/scm/linux/kernel/git/dhowells/linux-fs.git tags/afs-fix-rotation-20240105
for you to fetch changes up to abcbd3bfbbfe97a8912d0c929d4aa18f50d9bc52:
afs: trace: Log afs_make_call(), including server address (2024-01-01 16:37:27 +0000)
----------------------------------------------------------------
AFS fileserver rotation fix
----------------------------------------------------------------
David Howells (36):
afs: Remove whitespace before most ')' from the trace header
afs: Automatically generate trace tag enums
afs: Add comments on abort handling
afs: Turn the afs_addr_list address array into an array of structs
rxrpc, afs: Allow afs to pin rxrpc_peer objects
afs: Don't skip server addresses for which we didn't get an RTT reading
afs: Rename addr_list::failed to probe_failed
afs: Handle the VIO and UAEIO aborts explicitly
afs: Use op->nr_iterations=-1 to indicate to begin fileserver iteration
afs: Wrap most op->error accesses with inline funcs
afs: Don't put afs_call in afs_wait_for_call_to_complete()
afs: Simplify error handling
afs: Add a tracepoint for struct afs_addr_list
afs: Rename some fields
afs: Use peer + service_id as call address
afs: Fold the afs_addr_cursor struct in
rxrpc: Create a procfile to display outstanding client conn bundles
afs: Add some more info to /proc/net/afs/servers
afs: Remove the unimplemented afs_cmp_addr_list()
afs: Provide a way to configure address priorities
afs: Mark address lists with configured priorities
afs: Dispatch fileserver probes in priority order
afs: Dispatch vlserver probes in priority order
afs: Keep a record of the current fileserver endpoint state
afs: Combine the endpoint state bools into a bitmask
afs: Make it possible to find the volumes that are using a server
afs: Defer volume record destruction to a workqueue
afs: Move the vnode/volume validity checking code into its own file
afs: Apply server breaks to mmap'd files in the call processor
afs: Fix comment in afs_do_lookup()
afs: Don't leave DONTUSE/NEWREPSITE servers out of server list
afs: Parse the VolSync record in the reply of a number of RPC ops
afs: Overhaul invalidation handling to better support RO volumes
afs: Fix fileserver rotation
afs: Fix offline and busy message emission
afs: trace: Log afs_make_call(), including server address
Oleg Nesterov (4):
afs: fix the usage of read_seqbegin_or_lock() in afs_lookup_volume_rcu()
afs: fix the usage of read_seqbegin_or_lock() in afs_find_server*()
afs: use read_seqbegin() in afs_check_validity() and afs_getattr()
rxrpc_find_service_conn_rcu: fix the usage of read_seqbegin_or_lock()
fs/afs/Makefile | 2 +
fs/afs/addr_list.c | 224 +++++--------
fs/afs/addr_prefs.c | 531 +++++++++++++++++++++++++++++
fs/afs/afs.h | 3 +-
fs/afs/callback.c | 141 +++++---
fs/afs/cell.c | 5 +-
fs/afs/cmservice.c | 5 +-
fs/afs/dir.c | 66 ++--
fs/afs/dir_silly.c | 2 +-
fs/afs/file.c | 20 +-
fs/afs/fs_operation.c | 85 +++--
fs/afs/fs_probe.c | 323 ++++++++++--------
fs/afs/fsclient.c | 74 +++-
fs/afs/inode.c | 204 +----------
fs/afs/internal.h | 370 +++++++++++++-------
fs/afs/main.c | 1 +
fs/afs/misc.c | 10 +-
fs/afs/proc.c | 102 +++++-
fs/afs/rotate.c | 520 +++++++++++++++++++++--------
fs/afs/rxrpc.c | 107 +++---
fs/afs/server.c | 135 ++++----
fs/afs/server_list.c | 174 ++++++++--
fs/afs/super.c | 7 +-
fs/afs/validation.c | 473 ++++++++++++++++++++++++++
fs/afs/vl_alias.c | 69 +---
fs/afs/vl_list.c | 29 +-
fs/afs/vl_probe.c | 60 ++--
fs/afs/vl_rotate.c | 215 +++++++-----
fs/afs/vlclient.c | 143 +++++---
fs/afs/volume.c | 61 ++--
fs/afs/write.c | 6 +-
fs/afs/yfsclient.c | 25 +-
include/net/af_rxrpc.h | 15 +-
include/trace/events/afs.h | 779 ++++++++++++++++++++++++-------------------
include/trace/events/rxrpc.h | 3 +
net/rxrpc/af_rxrpc.c | 62 +++-
net/rxrpc/ar-internal.h | 6 +-
net/rxrpc/call_object.c | 17 +-
net/rxrpc/conn_client.c | 10 +
net/rxrpc/conn_service.c | 3 +-
net/rxrpc/net_ns.c | 4 +
net/rxrpc/peer_object.c | 58 ++--
net/rxrpc/proc.c | 76 +++++
net/rxrpc/sendmsg.c | 11 +-
44 files changed, 3544 insertions(+), 1692 deletions(-)
create mode 100644 fs/afs/addr_prefs.c
create mode 100644 fs/afs/validation.c
The pull request you sent on Fri, 05 Jan 2024 16:23:35 +0000:
> git://git.kernel.org/pub/scm/linux/kernel/git/dhowells/linux-fs.git tags/afs-fix-rotation-20240105
has been merged into torvalds/linux.git:
https://git.kernel.org/torvalds/c/0c59ae1290741854b6cf597ef05bfa9bc811389f
Thank you!
--
Deet-doot-dot, I am a bot.
https://korg.docs.kernel.org/prtracker.html