From: "Wiles, Keith" <keith.wiles@intel.com>
To: "Burakov, Anatoly" <anatoly.burakov@intel.com>
Cc: "dev@dpdk.org" <dev@dpdk.org>, "Tan, Jianfeng" <jianfeng.tan@intel.com>
Subject: Re: [dpdk-dev] [PATCH v3 3/5] eal: use locks to determine if secondary process is active
Date: Wed, 28 Feb 2018 04:17:30 +0000 [thread overview]
Message-ID: <CE4D7AF2-A1A3-4E2C-B231-EF42553701B3@intel.com> (raw)
In-Reply-To: <eec1a7d6c2afc3813cbc334e887887fecb57fcb6.1519740527.git.anatoly.burakov@intel.com>
> On Feb 27, 2018, at 8:35 AM, Anatoly Burakov <anatoly.burakov@intel.com> wrote:
>
> Previously, IPC would remove sockets it considers to be "inactive"
> based on whether they have responded. Change this to create lock
> files in addition to socket files, so that we can determine if
> secondary process is active before attempting to communicate with
> it. That way, we can distinguish secondaries that are alive but
> are not responding, from those that have already died.
>
> Signed-off-by: Anatoly Burakov <anatoly.burakov@intel.com>
> ---
>
> Notes:
> v3: no changes
>
> v2: no changes
>
> lib/librte_eal/common/eal_common_proc.c | 204 +++++++++++++++++++++++++++-----
> 1 file changed, 175 insertions(+), 29 deletions(-)
>
> diff --git a/lib/librte_eal/common/eal_common_proc.c b/lib/librte_eal/common/eal_common_proc.c
> index a6e24e6..7c87971 100644
> --- a/lib/librte_eal/common/eal_common_proc.c
> +++ b/lib/librte_eal/common/eal_common_proc.c
> @@ -13,6 +13,7 @@
> #include <stdio.h>
> #include <stdlib.h>
> #include <string.h>
> +#include <sys/file.h>
> #include <sys/time.h>
> #include <sys/types.h>
> #include <sys/socket.h>
> @@ -32,6 +33,7 @@
> #include "eal_internal_cfg.h"
>
> static int mp_fd = -1;
> +static int lock_fd = -1;
I did not find where lock_fd is closed in this code, should it be closed at some point?
> static char mp_filter[PATH_MAX]; /* Filter for secondary process sockets */
> static char mp_dir_path[PATH_MAX]; /* The directory path for all mp sockets */
> static pthread_mutex_t mp_mutex_action = PTHREAD_MUTEX_INITIALIZER;
> @@ -104,6 +106,46 @@ find_sync_request(const char *dst, const char *act_name)
> return r;
> }
>
> +static void
> +create_socket_path(const char *name, char *buf, int len)
> +{
> + const char *prefix = eal_mp_socket_path();
I thought we needed a space here after the variables, no?
> + if (strlen(name) > 0)
> + snprintf(buf, len, "%s_%s", prefix, name);
> + else
> + snprintf(buf, len, "%s", prefix);
> +}
> +
> +static void
> +create_lockfile_path(const char *name, char *buf, int len)
> +{
> + const char *prefix = eal_mp_socket_path();
Same here
> + if (strlen(name) > 1)
> + snprintf(buf, len, "%slock_%s", prefix, name);
> + else
> + snprintf(buf, len, "%slock", prefix);
> +}
> +
> +static const char *
> +get_peer_name(const char *socket_full_path)
> +{
> + char buf[PATH_MAX] = {0};
> + int len;
> +
> + /* primary process has no peer name */
> + if (strcmp(socket_full_path, eal_mp_socket_path()) == 0)
> + return NULL;
> +
> + /* construct dummy socket file name - make it one character long so that
> + * we hit the code path where underscores are added
> + */
> + create_socket_path("a", buf, sizeof(buf));
> +
> + /* we want to get everything after /path/.rte_unix_, so discard 'a' */
> + len = strlen(buf) - 1;
> + return &socket_full_path[len];
> +}
> +
> int
> rte_eal_primary_proc_alive(const char *config_file_path)
> {
> @@ -332,8 +374,29 @@ mp_handle(void *arg __rte_unused)
> static int
> open_socket_fd(void)
> {
> + char peer_name[PATH_MAX] = {0};
> + char lockfile[PATH_MAX] = {0};
> struct sockaddr_un un;
> - const char *prefix = eal_mp_socket_path();
> +
> + if (rte_eal_process_type() == RTE_PROC_SECONDARY)
> + snprintf(peer_name, sizeof(peer_name), "%d_%"PRIx64,
> + getpid(), rte_rdtsc());
> +
> + /* try to create lockfile */
> + create_lockfile_path(peer_name, lockfile, sizeof(lockfile));
> +
> + lock_fd = open(lockfile, O_CREAT | O_RDWR);
> + if (lock_fd < 0) {
> + RTE_LOG(ERR, EAL, "failed to open '%s': %s\n", lockfile,
> + strerror(errno));
> + return -1;
> + }
> + if (flock(lock_fd, LOCK_EX | LOCK_NB)) {
> + RTE_LOG(ERR, EAL, "failed to lock '%s': %s\n", lockfile,
> + strerror(errno));
Did not close the lock_fd here.
> + return -1;
> + }
> + /* no need to downgrade to shared lock */
>
> mp_fd = socket(AF_UNIX, SOCK_DGRAM, 0);
> if (mp_fd < 0) {
> @@ -343,13 +406,11 @@ open_socket_fd(void)
>
> memset(&un, 0, sizeof(un));
> un.sun_family = AF_UNIX;
> - if (rte_eal_process_type() == RTE_PROC_PRIMARY)
> - snprintf(un.sun_path, sizeof(un.sun_path), "%s", prefix);
> - else {
> - snprintf(un.sun_path, sizeof(un.sun_path), "%s_%d_%"PRIx64,
> - prefix, getpid(), rte_rdtsc());
> - }
> +
> + create_socket_path(peer_name, un.sun_path, sizeof(un.sun_path));
> +
> unlink(un.sun_path); /* May still exist since last run */
> +
> if (bind(mp_fd, (struct sockaddr *)&un, sizeof(un)) < 0) {
> RTE_LOG(ERR, EAL, "failed to bind %s: %s\n",
> un.sun_path, strerror(errno));
> @@ -361,6 +422,44 @@ open_socket_fd(void)
> return mp_fd;
> }
>
> +/* find corresponding lock file and try to lock it */
> +static int
> +socket_is_active(const char *peer_name)
> +{
> + char lockfile[PATH_MAX] = {0};
> + int fd, ret = -1;
> +
> + /* construct lockfile filename */
> + create_lockfile_path(peer_name, lockfile, sizeof(lockfile));
> +
> + /* try to lock it */
> + fd = open(lockfile, O_CREAT | O_RDWR);
> + if (fd < 0) {
> + RTE_LOG(ERR, EAL, "Cannot open '%s': %s\n", lockfile,
> + strerror(errno));
> + return -1;
> + }
> + ret = flock(fd, LOCK_EX | LOCK_NB);
> + if (ret < 0) {
> + if (errno == EWOULDBLOCK) {
> + /* file is locked */
> + ret = 1;
> + } else {
> + RTE_LOG(ERR, EAL, "Cannot lock '%s': %s\n", lockfile,
> + strerror(errno));
> + ret = -1;
> + }
> + } else {
> + ret = 0;
> + /* unlink lockfile automatically */
> + unlink(lockfile);
> + flock(fd, LOCK_UN);
> + }
> + close(fd);
> +
> + return ret;
> +}
> +
> static int
> unlink_sockets(const char *filter)
> {
> @@ -376,28 +475,33 @@ unlink_sockets(const char *filter)
> dir_fd = dirfd(mp_dir);
>
> while ((ent = readdir(mp_dir))) {
> - if (fnmatch(filter, ent->d_name, 0) == 0)
> + if (fnmatch(filter, ent->d_name, 0) == 0) {
> + const char *peer_name;
> + char path[PATH_MAX];
> + int ret;
> +
> + snprintf(path, sizeof(path), "%s/%s", mp_dir_path,
> + ent->d_name);
> + peer_name = get_peer_name(path);
> +
> + ret = socket_is_active(peer_name);
> + if (ret < 0) {
> + RTE_LOG(ERR, EAL, "Error getting socket active status\n”);
No close for mp_dir?
> + return -1;
> + } else if (ret == 1) {
> + RTE_LOG(ERR, EAL, "Socket is active (old secondary process still running?)\n”);
And here
> + return -1;
> + }
> + RTE_LOG(DEBUG, EAL, "Removing stale socket file '%s'\n",
> + ent->d_name);
> unlinkat(dir_fd, ent->d_name, 0);
Does unlinkat() close dir_fd?
> + }
> }
>
> closedir(mp_dir);
> return 0;
> }
>
> -static void
> -unlink_socket_by_path(const char *path)
> -{
> - char *filename;
> - char *fullpath = strdup(path);
> -
> - if (!fullpath)
> - return;
> - filename = basename(fullpath);
> - unlink_sockets(filename);
> - free(fullpath);
> - RTE_LOG(INFO, EAL, "Remove socket %s\n", path);
> -}
> -
> int
> rte_mp_channel_init(void)
> {
> @@ -487,10 +591,25 @@ send_msg(const char *dst_path, struct rte_mp_msg *msg, int type)
> rte_errno = errno;
> /* Check if it caused by peer process exits */
> if (errno == ECONNREFUSED) {
> - /* We don't unlink the primary's socket here */
> - if (rte_eal_process_type() == RTE_PROC_PRIMARY)
> - unlink_socket_by_path(dst_path);
> - return 0;
> + const char *peer_name = get_peer_name(dst_path);
> + int active, ret = 0;
Here we have variables in the middle of a block again.
> +
> + active = rte_eal_process_type() == RTE_PROC_PRIMARY ?
> + socket_is_active(peer_name) :
> + rte_eal_primary_proc_alive(NULL);
> +
> + if (active > 0) {
> + RTE_LOG(ERR, EAL, "Couldn't communicate with active peer\n");
> + } else if (active < 0) {
> + RTE_LOG(ERR, EAL, "Couldn't get peer status\n");
> + ret = -1;
> + } else if (rte_eal_process_type() == RTE_PROC_PRIMARY) {
> + /* peer isn't active anymore, so unlink its
> + * socket.
> + */
> + unlink(dst_path);
> + }
> + return ret;
> }
> if (errno == ENOBUFS) {
> RTE_LOG(ERR, EAL, "Peer cannot receive message %s\n",
> @@ -508,7 +627,7 @@ send_msg(const char *dst_path, struct rte_mp_msg *msg, int type)
> static int
> mp_send(struct rte_mp_msg *msg, const char *peer, int type)
> {
> - int ret = 0;
> + int dir_fd, ret = 0;
> DIR *mp_dir;
> struct dirent *ent;
>
> @@ -530,15 +649,28 @@ mp_send(struct rte_mp_msg *msg, const char *peer, int type)
> rte_errno = errno;
> return -1;
> }
> + dir_fd = dirfd(mp_dir);
when is dir_fd closed?
> while ((ent = readdir(mp_dir))) {
> char path[PATH_MAX];
> + const char *peer_name;
> + int active;
>
> if (fnmatch(mp_filter, ent->d_name, 0) != 0)
> continue;
>
> snprintf(path, sizeof(path), "%s/%s", mp_dir_path,
> ent->d_name);
> - if (send_msg(path, msg, type) < 0)
> + peer_name = get_peer_name(path);
> +
> + /* only send if we can expect to receive a reply, otherwise
> + * remove the socket.
> + */
> + active = socket_is_active(peer_name);
> + if (active < 0)
> + ret = -1;
> + else if (active == 0)
> + unlinkat(dir_fd, ent->d_name, 0);
> + else if (active > 0 && send_msg(path, msg, type) < 0)
> ret = -1;
> }
>
> @@ -663,7 +795,7 @@ int __rte_experimental
> rte_mp_request(struct rte_mp_msg *req, struct rte_mp_reply *reply,
> const struct timespec *ts)
> {
> - int ret = 0;
> + int dir_fd, ret = 0;
> DIR *mp_dir;
> struct dirent *ent;
> struct timeval now;
> @@ -698,15 +830,29 @@ rte_mp_request(struct rte_mp_msg *req, struct rte_mp_reply *reply,
> rte_errno = errno;
> return -1;
> }
> + dir_fd = dirfd(mp_dir);
When is dir_fd closed?
>
> while ((ent = readdir(mp_dir))) {
> + const char *peer_name;
> char path[PATH_MAX];
> + int active;
>
> if (fnmatch(mp_filter, ent->d_name, 0) != 0)
> continue;
>
> snprintf(path, sizeof(path), "%s/%s", mp_dir_path,
> ent->d_name);
> + peer_name = get_peer_name(path);
> +
> + active = socket_is_active(peer_name);
> +
> + if (active < 0) {
> + ret = -1;
> + break;
> + } else if (active == 0) {
> + unlinkat(dir_fd, ent->d_name, 0);
> + continue;
> + }
>
> if (mp_request_one(path, req, reply, &end))
> ret = -1;
> --
> 2.7.4
Regards,
Keith
next prev parent reply other threads:[~2018-02-28 4:17 UTC|newest]
Thread overview: 57+ messages / expand[flat|nested] mbox.gz Atom feed top
2018-02-22 18:21 [dpdk-dev] [PATCH 1/3] eal: add internal flag indicating init has completed Anatoly Burakov
2018-02-22 18:21 ` [dpdk-dev] [PATCH 2/3] eal: don't process IPC messages before init finished Anatoly Burakov
2018-02-22 18:21 ` [dpdk-dev] [PATCH 3/3] eal: use locks to determine if secondary process is active Anatoly Burakov
2018-03-02 15:14 ` [dpdk-dev] [PATCH v4 1/5] eal: add internal flag indicating init has completed Anatoly Burakov
2018-03-07 16:56 ` [dpdk-dev] [PATCH v5 0/6] Improvements for DPDK IPC Anatoly Burakov
2018-03-13 17:42 ` [dpdk-dev] [PATCH v6 " Anatoly Burakov
2018-03-21 17:43 ` Thomas Monjalon
2018-03-13 17:42 ` [dpdk-dev] [PATCH v6 1/6] eal: add internal flag indicating init has completed Anatoly Burakov
2018-03-13 20:59 ` Bruce Richardson
2018-03-13 17:42 ` [dpdk-dev] [PATCH v6 2/6] eal: abstract away IPC socket path generation Anatoly Burakov
2018-03-13 17:42 ` [dpdk-dev] [PATCH v6 3/6] eal: don't hardcode socket filter value in IPC Anatoly Burakov
2018-03-13 17:42 ` [dpdk-dev] [PATCH v6 4/6] eal: lock IPC directory on init and send Anatoly Burakov
2018-03-13 17:42 ` [dpdk-dev] [PATCH v6 5/6] eal: simplify IPC sync request timeout code Anatoly Burakov
2018-03-13 17:42 ` [dpdk-dev] [PATCH v6 6/6] eal: ignore messages until init is complete Anatoly Burakov
2018-03-07 16:56 ` [dpdk-dev] [PATCH v5 1/6] eal: add internal flag indicating init has completed Anatoly Burakov
2018-03-07 16:56 ` [dpdk-dev] [PATCH v5 2/6] eal: abstract away IPC socket path generation Anatoly Burakov
2018-03-11 9:02 ` Tan, Jianfeng
2018-03-07 16:56 ` [dpdk-dev] [PATCH v5 3/6] eal: don't hardcode socket filter value in IPC Anatoly Burakov
2018-03-07 16:56 ` [dpdk-dev] [PATCH v5 4/6] eal: lock IPC directory on init and send Anatoly Burakov
2018-03-11 9:14 ` Tan, Jianfeng
2018-03-07 16:56 ` [dpdk-dev] [PATCH v5 5/6] eal: simplify IPC sync request timeout code Anatoly Burakov
2018-03-11 9:25 ` Tan, Jianfeng
2018-03-07 16:56 ` [dpdk-dev] [PATCH v5 6/6] eal: ignore messages until init is complete Anatoly Burakov
2018-03-12 1:42 ` Tan, Jianfeng
2018-03-12 8:58 ` Burakov, Anatoly
2018-03-02 15:14 ` [dpdk-dev] [PATCH v4 2/5] eal: use file to check if secondary process is ready Anatoly Burakov
2018-03-06 11:03 ` Burakov, Anatoly
2018-03-02 15:14 ` [dpdk-dev] [PATCH v4 3/5] eal: prevent secondary process init while sending messages Anatoly Burakov
2018-03-02 15:14 ` [dpdk-dev] [PATCH v4 4/5] eal: don't hardcode socket filter value in IPC Anatoly Burakov
2018-03-02 15:14 ` [dpdk-dev] [PATCH v4 5/5] eal: simplify IPC sync request timeout code Anatoly Burakov
2018-02-22 18:32 ` [dpdk-dev] [PATCH 1/3] eal: add internal flag indicating init has completed Burakov, Anatoly
2018-02-27 13:23 ` [dpdk-dev] [PATCH v2 1/5] " Anatoly Burakov
2018-02-27 14:35 ` [dpdk-dev] [PATCH v3 " Anatoly Burakov
2018-02-28 2:12 ` Tan, Jianfeng
2018-02-28 9:43 ` Burakov, Anatoly
2018-02-27 14:35 ` [dpdk-dev] [PATCH v3 2/5] eal: don't process IPC messages before init finished Anatoly Burakov
2018-02-28 1:09 ` Tan, Jianfeng
2018-02-28 9:45 ` Burakov, Anatoly
2018-02-28 4:00 ` Wiles, Keith
2018-02-28 9:47 ` Burakov, Anatoly
2018-02-27 14:35 ` [dpdk-dev] [PATCH v3 3/5] eal: use locks to determine if secondary process is active Anatoly Burakov
2018-02-28 1:26 ` Tan, Jianfeng
2018-02-28 10:15 ` Burakov, Anatoly
2018-02-28 4:17 ` Wiles, Keith [this message]
2018-02-28 10:17 ` Burakov, Anatoly
2018-02-27 14:35 ` [dpdk-dev] [PATCH v3 4/5] eal: prevent secondary process init while sending messages Anatoly Burakov
2018-02-28 1:58 ` Tan, Jianfeng
2018-02-28 10:19 ` Burakov, Anatoly
2018-02-28 15:49 ` Tan, Jianfeng
2018-02-27 14:35 ` [dpdk-dev] [PATCH v3 5/5] eal: don't hardcode socket filter value in IPC Anatoly Burakov
2018-02-28 1:52 ` Tan, Jianfeng
2018-02-28 10:21 ` Burakov, Anatoly
2018-02-28 15:01 ` Tan, Jianfeng
2018-02-27 13:23 ` [dpdk-dev] [PATCH v2 2/5] eal: don't process IPC messages before init finished Anatoly Burakov
2018-02-27 13:23 ` [dpdk-dev] [PATCH v2 3/5] eal: use locks to determine if secondary process is active Anatoly Burakov
2018-02-27 13:23 ` [dpdk-dev] [PATCH v2 4/5] eal: prevent secondary process init while sending messages Anatoly Burakov
2018-02-27 13:23 ` [dpdk-dev] [PATCH v2 5/5] eal: don't hardcode socket filter value in IPC Anatoly Burakov
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=CE4D7AF2-A1A3-4E2C-B231-EF42553701B3@intel.com \
--to=keith.wiles@intel.com \
--cc=anatoly.burakov@intel.com \
--cc=dev@dpdk.org \
--cc=jianfeng.tan@intel.com \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).