[lustre-devel] [PATCH 02/25] lustre: lnet: Allow min stats to be reset in peers and nis
NeilBrown
neilb at suse.com
Wed Sep 26 16:59:59 PDT 2018
On Tue, Sep 25 2018, James Simmons wrote:
> From: Doug Oucharek <dougso at me.com>
>
> Allow writes to the peers and nis LNet procfs files to
> reset the mininum stat columns.
>
> Signed-off-by: Doug Oucharek <dougso at me.com>
> WC-bug-id: https://jira.whamcloud.com/browse/LU-7214
> Reviewed-on: https://review.whamcloud.com/20470
> Reviewed-by: Olaf Weber <olaf.weber at hpe.com>
> Reviewed-by: Amir Shehata <ashehata at whamcloud.com>
> Reviewed-by: Oleg Drokin <green at whamcloud.com>
> Signed-off-by: James Simmons <jsimmons at infradead.org>
> ---
> drivers/staging/lustre/lnet/lnet/router_proc.c | 69 +++++++++++++++++++++++---
> 1 file changed, 62 insertions(+), 7 deletions(-)
>
> diff --git a/drivers/staging/lustre/lnet/lnet/router_proc.c b/drivers/staging/lustre/lnet/lnet/router_proc.c
> index a887ca4..4ddd35b 100644
> --- a/drivers/staging/lustre/lnet/lnet/router_proc.c
> +++ b/drivers/staging/lustre/lnet/lnet/router_proc.c
> @@ -393,7 +393,7 @@ static int proc_lnet_peers(struct ctl_table *table, int write,
> {
> const int tmpsiz = 256;
> struct lnet_peer_table *ptable;
> - char *tmpstr;
> + char *tmpstr = NULL;
> char *s;
> int cpt = LNET_PROC_CPT_GET(*ppos);
> int ver = LNET_PROC_VER_GET(*ppos);
> @@ -402,12 +402,33 @@ static int proc_lnet_peers(struct ctl_table *table, int write,
> int rc = 0;
> int len;
>
> - BUILD_BUG_ON(LNET_PROC_HASH_BITS < LNET_PEER_HASH_BITS);
> - LASSERT(!write);
> + if (write) {
> + struct lnet_peer_ni *peer;
> + int i;
> +
> + cfs_percpt_for_each(ptable, i, the_lnet.ln_peer_tables) {
> + lnet_net_lock(i);
> + for (hash = 0; hash < LNET_PEER_HASH_SIZE; hash++) {
> + list_for_each_entry(peer,
> + &ptable->pt_hash[hash],
> + lpni_hashlist) {
> + peer->lpni_mintxcredits =
> + peer->lpni_txcredits;
> + peer->lpni_minrtrcredits =
> + peer->lpni_rtrcredits;
> + }
> + }
> + lnet_net_unlock(i);
> + }
> + *ppos += *lenp;
> + return 0;
> + }
>
> if (!*lenp)
> return 0;
>
> + BUILD_BUG_ON(LNET_PROC_HASH_BITS < LNET_PEER_HASH_BITS);
> +
> if (cpt >= LNET_CPT_NUMBER) {
> *lenp = 0;
> return 0;
> @@ -627,11 +648,45 @@ static int proc_lnet_nis(struct ctl_table *table, int write,
> char *s;
> int len;
>
> - LASSERT(!write);
> -
> if (!*lenp)
> return 0;
>
> + if (write) {
> + /* Just reset the min stat. */
> + struct lnet_net *net;
> + struct lnet_ni *ni;
> +
> + lnet_net_lock(0);
Presumably we hold lnet_net_lock(0) across the whole process to stop an
EXCLUSIVE lock from being taken while the update progresses. That makes
sense.
So why doesn't proc_lnet_peers() above do the same thing?
Is it OK if I add that to the patch?
NeilBrown
> +
> + list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
> + list_for_each_entry(ni, &net->net_ni_list, ni_netlist) {
> + struct lnet_tx_queue *tq;
> + int i;
> + int j;
> +
> + cfs_percpt_for_each(tq, i, ni->ni_tx_queues) {
> + for (j = 0; ni->ni_cpts &&
> + j < ni->ni_ncpts; j++) {
> + if (i == ni->ni_cpts[j])
> + break;
> + }
> +
> + if (j == ni->ni_ncpts)
> + continue;
> +
> + if (i != 0)
> + lnet_net_lock(i);
> + tq->tq_credits_min = tq->tq_credits;
> + if (i != 0)
> + lnet_net_unlock(i);
> + }
> + }
> + }
> + lnet_net_unlock(0);
> + *ppos += *lenp;
> + return 0;
> + }
> +
> tmpstr = kvmalloc(tmpsiz, GFP_KERNEL);
> if (!tmpstr)
> return -ENOMEM;
> @@ -847,7 +902,7 @@ static int proc_lnet_portal_rotor(struct ctl_table *table, int write,
> },
> {
> .procname = "peers",
> - .mode = 0444,
> + .mode = 0644,
> .proc_handler = &proc_lnet_peers,
> },
> {
> @@ -857,7 +912,7 @@ static int proc_lnet_portal_rotor(struct ctl_table *table, int write,
> },
> {
> .procname = "nis",
> - .mode = 0444,
> + .mode = 0644,
> .proc_handler = &proc_lnet_nis,
> },
> {
> --
> 1.8.3.1
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 832 bytes
Desc: not available
URL: <http://lists.lustre.org/pipermail/lustre-devel-lustre.org/attachments/20180927/8674b8f0/attachment.sig>
More information about the lustre-devel
mailing list