1
linux/fs/afs/server_list.c
David Howells 495f2ae9e3 afs: Fix fileserver rotation
Fix the fileserver rotation so that it doesn't use RTT as the basis for
deciding which server and address to use as this doesn't necessarily give a
good indication of the best path.  Instead, use the configurable preference
list in conjunction with whatever probes have succeeded at the time of
looking.

To this end, make the following changes:

 (1) Keep an array of "server states" to track what addresses we've tried
     on each server and move the waitqueue entries there that we'll need
     for probing.

 (2) Each afs_server_state struct is made to pin the corresponding server's
     endpoint state rather than the afs_operation struct carrying a pin on
     the server we're currently looking at.

 (3) Drop the server list preference; we now always rescan the server list.

 (4) afs_wait_for_probes() now uses the server state list to guide it in
     what it waits for (and to provide the waitqueue entries) and returns
     an indication of whether we'd got a response, run out of responsive
     addresses or the endpoint state had been superseded and we need to
     restart the iteration.

 (5) Call afs_get_address_preferences*() occasionally to refresh the
     preference values.

 (6) When picking a server, scan the addresses of the servers for which we
     have as-yet untested communications, looking for the highest priority
     one and use that instead of trying all the addresses for a particular
     server in ascending-RTT order.

 (7) When a Busy or Offline state is seen across all available servers, do
     a short sleep.

 (8) If we detect that we accessed a future RO volume version whilst it is
     undergoing replication, reissue the op against the older version until
     at least half of the servers are replicated.

 (9) Whilst RO replication is ongoing, increase the frequency of Volume
     Location server checks for that volume to every ten minutes instead of
     hourly.

Also add a tracepoint to track progress through the rotation algorithm.

Signed-off-by: David Howells <dhowells@redhat.com>
cc: Marc Dionne <marc.dionne@auristor.com>
cc: linux-afs@lists.infradead.org
2024-01-01 16:37:27 +00:00

250 lines
6.2 KiB
C

// SPDX-License-Identifier: GPL-2.0-or-later
/* AFS fileserver list management.
*
* Copyright (C) 2017 Red Hat, Inc. All Rights Reserved.
* Written by David Howells (dhowells@redhat.com)
*/
#include <linux/kernel.h>
#include <linux/slab.h>
#include "internal.h"
void afs_put_serverlist(struct afs_net *net, struct afs_server_list *slist)
{
int i;
if (slist && refcount_dec_and_test(&slist->usage)) {
for (i = 0; i < slist->nr_servers; i++)
afs_unuse_server(net, slist->servers[i].server,
afs_server_trace_put_slist);
kfree_rcu(slist, rcu);
}
}
/*
* Build a server list from a VLDB record.
*/
struct afs_server_list *afs_alloc_server_list(struct afs_volume *volume,
struct key *key,
struct afs_vldb_entry *vldb)
{
struct afs_server_list *slist;
struct afs_server *server;
unsigned int type_mask = 1 << volume->type;
bool use_newrepsites = false;
int ret = -ENOMEM, nr_servers = 0, newrep = 0, i, j, usable = 0;
/* Work out if we're going to restrict to NEWREPSITE-marked servers or
* not. If at least one site is marked as NEWREPSITE, then it's likely
* that "vos release" is busy updating RO sites. We cut over from one
* to the other when >=50% of the sites have been updated. Sites that
* are in the process of being updated are marked DONTUSE.
*/
for (i = 0; i < vldb->nr_servers; i++) {
if (!(vldb->fs_mask[i] & type_mask))
continue;
nr_servers++;
if (vldb->vlsf_flags[i] & AFS_VLSF_DONTUSE)
continue;
usable++;
if (vldb->vlsf_flags[i] & AFS_VLSF_NEWREPSITE)
newrep++;
}
slist = kzalloc(struct_size(slist, servers, nr_servers), GFP_KERNEL);
if (!slist)
goto error;
if (newrep) {
if (newrep < usable / 2) {
slist->ro_replicating = AFS_RO_REPLICATING_USE_OLD;
} else {
slist->ro_replicating = AFS_RO_REPLICATING_USE_NEW;
use_newrepsites = true;
}
}
refcount_set(&slist->usage, 1);
rwlock_init(&slist->lock);
/* Make sure a records exists for each server in the list. */
for (i = 0; i < vldb->nr_servers; i++) {
unsigned long se_flags = 0;
bool newrepsite = vldb->vlsf_flags[i] & AFS_VLSF_NEWREPSITE;
if (!(vldb->fs_mask[i] & type_mask))
continue;
if (vldb->vlsf_flags[i] & AFS_VLSF_DONTUSE)
__set_bit(AFS_SE_EXCLUDED, &se_flags);
if (newrep && (newrepsite ^ use_newrepsites))
__set_bit(AFS_SE_EXCLUDED, &se_flags);
server = afs_lookup_server(volume->cell, key, &vldb->fs_server[i],
vldb->addr_version[i]);
if (IS_ERR(server)) {
ret = PTR_ERR(server);
if (ret == -ENOENT ||
ret == -ENOMEDIUM)
continue;
goto error_2;
}
/* Insertion-sort by UUID */
for (j = 0; j < slist->nr_servers; j++)
if (memcmp(&slist->servers[j].server->uuid,
&server->uuid,
sizeof(server->uuid)) >= 0)
break;
if (j < slist->nr_servers) {
if (slist->servers[j].server == server) {
afs_put_server(volume->cell->net, server,
afs_server_trace_put_slist_isort);
continue;
}
memmove(slist->servers + j + 1,
slist->servers + j,
(slist->nr_servers - j) * sizeof(struct afs_server_entry));
}
slist->servers[j].server = server;
slist->servers[j].volume = volume;
slist->servers[j].flags = se_flags;
slist->servers[j].cb_expires_at = AFS_NO_CB_PROMISE;
slist->nr_servers++;
}
if (slist->nr_servers == 0) {
ret = -EDESTADDRREQ;
goto error_2;
}
return slist;
error_2:
afs_put_serverlist(volume->cell->net, slist);
error:
return ERR_PTR(ret);
}
/*
* Copy the annotations from an old server list to its potential replacement.
*/
bool afs_annotate_server_list(struct afs_server_list *new,
struct afs_server_list *old)
{
unsigned long mask = 1UL << AFS_SE_EXCLUDED;
int i;
if (old->nr_servers != new->nr_servers ||
old->ro_replicating != new->ro_replicating)
goto changed;
for (i = 0; i < old->nr_servers; i++) {
if (old->servers[i].server != new->servers[i].server)
goto changed;
if ((old->servers[i].flags & mask) != (new->servers[i].flags & mask))
goto changed;
}
return false;
changed:
return true;
}
/*
* Attach a volume to the servers it is going to use.
*/
void afs_attach_volume_to_servers(struct afs_volume *volume, struct afs_server_list *slist)
{
struct afs_server_entry *se, *pe;
struct afs_server *server;
struct list_head *p;
unsigned int i;
down_write(&volume->cell->vs_lock);
for (i = 0; i < slist->nr_servers; i++) {
se = &slist->servers[i];
server = se->server;
list_for_each(p, &server->volumes) {
pe = list_entry(p, struct afs_server_entry, slink);
if (volume->vid <= pe->volume->vid)
break;
}
list_add_tail(&se->slink, p);
}
slist->attached = true;
up_write(&volume->cell->vs_lock);
}
/*
* Reattach a volume to the servers it is going to use when server list is
* replaced. We try to switch the attachment points to avoid rewalking the
* lists.
*/
void afs_reattach_volume_to_servers(struct afs_volume *volume, struct afs_server_list *new,
struct afs_server_list *old)
{
unsigned int n = 0, o = 0;
down_write(&volume->cell->vs_lock);
while (n < new->nr_servers || o < old->nr_servers) {
struct afs_server_entry *pn = n < new->nr_servers ? &new->servers[n] : NULL;
struct afs_server_entry *po = o < old->nr_servers ? &old->servers[o] : NULL;
struct afs_server_entry *s;
struct list_head *p;
int diff;
if (pn && po && pn->server == po->server) {
pn->cb_expires_at = po->cb_expires_at;
list_replace(&po->slink, &pn->slink);
n++;
o++;
continue;
}
if (pn && po)
diff = memcmp(&pn->server->uuid, &po->server->uuid,
sizeof(pn->server->uuid));
else
diff = pn ? -1 : 1;
if (diff < 0) {
list_for_each(p, &pn->server->volumes) {
s = list_entry(p, struct afs_server_entry, slink);
if (volume->vid <= s->volume->vid)
break;
}
list_add_tail(&pn->slink, p);
n++;
} else {
list_del(&po->slink);
o++;
}
}
up_write(&volume->cell->vs_lock);
}
/*
* Detach a volume from the servers it has been using.
*/
void afs_detach_volume_from_servers(struct afs_volume *volume, struct afs_server_list *slist)
{
unsigned int i;
if (!slist->attached)
return;
down_write(&volume->cell->vs_lock);
for (i = 0; i < slist->nr_servers; i++)
list_del(&slist->servers[i].slink);
slist->attached = false;
up_write(&volume->cell->vs_lock);
}