cluster-commits May 2015

cluster-commits@lists.stg.fedorahosted.org

1 participants
2 discussions

cluster: RHEL6 - gfs_controld: Retry uevent recv() and reconnect uevent socket on error
by jruemker 27 May '15

27 May '15

Gitweb: http://git.fedorahosted.org/git/?p=cluster.git;a=commitdiff;h=353cec33e438d… Commit: 353cec33e438d32a4be1a3eeb165abad57133864 Parent: 40b2dfa44ac95304d5e8caafe8e243b8ac75233c Author: John Ruemker <jruemker(a)redhat.com> AuthorDate: Wed May 27 16:01:39 2015 -0400 Committer: John Ruemker <jruemker(a)redhat.com> CommitterDate: Wed May 27 16:01:39 2015 -0400 gfs_controld: Retry uevent recv() and reconnect uevent socket on error When processing a uevent, on ENOBUFS retry up to three times before giving up. For other errors besides EAGAIN and EINTR, or after the maximum ENOBUFS retries, reconnect the uevent netlink socket. --- group/gfs_controld/main.c | 75 +++++++++++++++++++++++++++----------------- 1 files changed, 46 insertions(+), 29 deletions(-) diff --git a/group/gfs_controld/main.c b/group/gfs_controld/main.c index dc841da..c27b192 100644 --- a/group/gfs_controld/main.c +++ b/group/gfs_controld/main.c @@ -8,6 +8,7 @@ #define LOCKFILE_NAME "/var/run/gfs_controld.pid" #define CLIENT_NALLOC 32 #define UEVENT_BUF_SIZE 4096 +#define MAX_UEVENT_NOBUFS 3 static int client_maxi; static int client_size; @@ -291,20 +292,62 @@ static char *uevent_fsname(const char *vars[]) return (name && name[0]) ? name + 1 : NULL; } +static int setup_uevent(void) +{ + struct sockaddr_nl snl; + int s, rv; + + s = socket(AF_NETLINK, SOCK_DGRAM, NETLINK_KOBJECT_UEVENT); + if (s < 0) { + log_error("uevent netlink socket"); + return s; + } + + memset(&snl, 0, sizeof(snl)); + snl.nl_family = AF_NETLINK; + snl.nl_pid = getpid(); + snl.nl_groups = 1; + + rv = bind(s, (struct sockaddr *) &snl, sizeof(snl)); + if (rv < 0) { + log_error("uevent bind error %d errno %d", rv, errno); + close(s); + return rv; + } + + return s; +} + static void process_uevent(int ci) { char buf[UEVENT_BUF_SIZE]; const char *uevent_vals[Env_Last]; char *fsname; - int rv; + int rv, nobufs = 0; retry_recv: rv = recv(client[ci].fd, &buf, sizeof(buf), 0); if (rv < 0) { + log_error("uevent recv error %d errno %d", rv, errno); if (errno == EINTR) goto retry_recv; - if (errno != EAGAIN) - log_error("uevent recv error %d errno %d", rv, errno); + if (errno == ENOBUFS && nobufs < MAX_UEVENT_NOBUFS) { + pollfd[ci].revents &= ~POLLERR; + nobufs++; + goto retry_recv; + } + if (errno != EAGAIN) { + close(client[ci].fd); + rv = setup_uevent(); + if (rv < 0) { + /* revent error will trigger cleanup by loop */ + return; + } + client[ci].fd = rv; + pollfd[ci].fd = rv; + pollfd[ci].revents = 0; + log_debug("uevent socket reconnected"); + } return; } buf[rv] = 0; @@ -375,32 +418,6 @@ static void process_uevent(int ci) } } -static int setup_uevent(void) -{ - struct sockaddr_nl snl; - int s, rv; - - s = socket(AF_NETLINK, SOCK_DGRAM, NETLINK_KOBJECT_UEVENT); - if (s < 0) { - log_error("uevent netlink socket"); - return s; - } - - memset(&snl, 0, sizeof(snl)); - snl.nl_family = AF_NETLINK; - snl.nl_pid = getpid(); - snl.nl_groups = 1; - - rv = bind(s, (struct sockaddr *) &snl, sizeof(snl)); - if (rv < 0) { - log_error("uevent bind error %d errno %d", rv, errno); - close(s); - return rv; - } - - return s; -} - static void init_header(struct gfsc_header *h, int cmd, char *name, int result, int extra_len) {

1 0

cluster: RHEL6 - dlm_controld: Retry uevent recv() and reconnect uevent socket on error
by jruemker 27 May '15

27 May '15

Gitweb: http://git.fedorahosted.org/git/?p=cluster.git;a=commitdiff;h=40b2dfa44ac95… Commit: 40b2dfa44ac95304d5e8caafe8e243b8ac75233c Parent: 090957641013f2362f8fbb99ad30eb4b25a4f54e Author: John Ruemker <jruemker(a)redhat.com> AuthorDate: Wed May 27 10:04:55 2015 -0400 Committer: John Ruemker <jruemker(a)redhat.com> CommitterDate: Wed May 27 10:04:55 2015 -0400 dlm_controld: Retry uevent recv() and reconnect uevent socket on error When processing a uevent, on ENOBUFS retry up to three times before giving up. For other errors besides EAGAIN and EINTR, or after the maximum ENOBUFS retries, reconnect the uevent netlink socket. --- group/dlm_controld/main.c | 75 +++++++++++++++++++++++++++----------------- 1 files changed, 46 insertions(+), 29 deletions(-) diff --git a/group/dlm_controld/main.c b/group/dlm_controld/main.c index eeab77c..9b60007 100644 --- a/group/dlm_controld/main.c +++ b/group/dlm_controld/main.c @@ -10,6 +10,7 @@ #define LOCKFILE_NAME "/var/run/dlm_controld.pid" #define CLIENT_NALLOC 32 +#define MAX_UEVENT_NOBUFS 3 static int client_maxi; static int client_size = 0; @@ -290,6 +291,32 @@ const char *dlm_mode_str(int mode) return "??"; } +static int setup_uevent(void) +{ + struct sockaddr_nl snl; + int s, rv; + + s = socket(AF_NETLINK, SOCK_DGRAM, NETLINK_KOBJECT_UEVENT); + if (s < 0) { + log_error("uevent netlink socket"); + return s; + } + + memset(&snl, 0, sizeof(snl)); + snl.nl_family = AF_NETLINK; + snl.nl_pid = getpid(); + snl.nl_groups = 1; + + rv = bind(s, (struct sockaddr *) &snl, sizeof(snl)); + if (rv < 0) { + log_error("uevent bind error %d errno %d", rv, errno); + close(s); + return rv; + } + + return s; +} + /* recv "online" (join) and "offline" (leave) messages from dlm via uevents and pass them on to groupd */ @@ -298,7 +325,7 @@ static void process_uevent(int ci) struct lockspace *ls; char buf[MAXLINE]; char *argv[MAXARGS], *act, *sys; - int rv, argc = 0; + int rv, argc = 0, nobufs = 0; memset(buf, 0, sizeof(buf)); memset(argv, 0, sizeof(char *) * MAXARGS); @@ -306,10 +333,26 @@ static void process_uevent(int ci) retry_recv: rv = recv(client[ci].fd, &buf, sizeof(buf), 0); if (rv < 0) { + log_error("uevent recv error %d errno %d", rv, errno); if (errno == EINTR) goto retry_recv; - if (errno != EAGAIN) - log_error("uevent recv error %d errno %d", rv, errno); + if (errno == ENOBUFS && nobufs < MAX_UEVENT_NOBUFS) { + pollfd[ci].revents &= ~POLLERR; + nobufs++; + goto retry_recv; + } + if (errno != EAGAIN) { + close(client[ci].fd); + rv = setup_uevent(); + if (rv < 0) { + /* revent error will trigger cleanup by loop */ + return; + } + client[ci].fd = rv; + pollfd[ci].fd = rv; + pollfd[ci].revents = 0; + log_debug("uevent socket reconnected"); + } return; } @@ -374,32 +417,6 @@ static void process_uevent(int ci) act, rv, errno); } -static int setup_uevent(void) -{ - struct sockaddr_nl snl; - int s, rv; - - s = socket(AF_NETLINK, SOCK_DGRAM, NETLINK_KOBJECT_UEVENT); - if (s < 0) { - log_error("uevent netlink socket"); - return s; - } - - memset(&snl, 0, sizeof(snl)); - snl.nl_family = AF_NETLINK; - snl.nl_pid = getpid(); - snl.nl_groups = 1; - - rv = bind(s, (struct sockaddr *) &snl, sizeof(snl)); - if (rv < 0) { - log_error("uevent bind error %d errno %d", rv, errno); - close(s); - return rv; - } - - return s; -} - static void init_header(struct dlmc_header *h, int cmd, char *name, int result, int extra_len) {

1 0

2024

2023

2022

2021

2020

2019

2018

2017

2016

2015

2014

2013

2012

2011

2010

2009

cluster-commits May 2015