From: Benjamin Marzinski <bmarzins@redhat.com>
To: mwilck@suse.com
Cc: lixiaokeng@huawei.com, dm-devel@redhat.com
Subject: Re: [dm-devel] [PATCH v2 28/29] multipathd: sanitize uxsock_listen()
Date: Mon, 19 Oct 2020 18:33:52 -0500 [thread overview]
Message-ID: <20201019233352.GJ3384@octiron.msp.redhat.com> (raw)
In-Reply-To: <20201016104501.8700-29-mwilck@suse.com>
On Fri, Oct 16, 2020 at 12:45:00PM +0200, mwilck@suse.com wrote:
> From: Martin Wilck <mwilck@suse.com>
>
> We were allocating 1025 poll fds, which is not optimal. Fix it, and make this
> more easily customizable in general. Use POLLFDS_BASE rather than the
> hard-coded "2" for the number of fds we poll besides client connections.
> Introduce a maximum number of clients that can connect. When this number is
> reached, we simply stop polling the accept socket, so that new connections
> aren't accepted any more. Don't attempt to realloc() the pollfd array if the
> number of clients decreases. It's unlikely to ever be more than one or two
> pages. Finally, there's no need to wake up every 5s. Our signal handling is
> robust. Just sleep forever in ppoll() if nothing happens.
>
> Signed-off-by: Martin Wilck <mwilck@suse.com>
> ---
> multipathd/uxlsnr.c | 70 ++++++++++++++++++++++++++++-----------------
> 1 file changed, 43 insertions(+), 27 deletions(-)
>
> diff --git a/multipathd/uxlsnr.c b/multipathd/uxlsnr.c
> index ce2b680..cd462b6 100644
> --- a/multipathd/uxlsnr.c
> +++ b/multipathd/uxlsnr.c
> @@ -41,14 +41,25 @@
> #include "cli.h"
> #include "uxlsnr.h"
>
> -static struct timespec sleep_time = {5, 0};
> -
> struct client {
> struct list_head node;
> int fd;
> };
>
> -#define MIN_POLLS 1023
> +/* The number of fds we poll on, other than individual client connections */
> +#define POLLFDS_BASE 2
> +#define POLLFD_CHUNK (4096 / sizeof(struct pollfd))
> +/* Minimum mumber of pollfds to reserve for clients */
> +#define MIN_POLLS (POLLFD_CHUNK - POLLFDS_BASE)
Reviewed-by: Benjamin Marzinski <bmarzins@redhat.com>
I have one nitpick. This code looks like it's pretending to allocate
pages of memory, when it's not. Malloc's bookeeping space means that
this memory chunk will be larger than a page. Even if it was page sized,
unless userspace is specifically asking for page-aligned memory, it most
like won't get it. Since AFAIK there is no benefit to mallocing memory
in a specific size increment, it doesn't seem woirth adding any
complexity to make sure our mallocs do that.
-Ben
> +/*
> + * Max number of client connections allowed
> + * During coldplug, there may be a large number of "multipath -u"
> + * processes connecting.
> + */
> +#define MAX_CLIENTS (16384 - POLLFDS_BASE)
> +
> +/* Compile-time error if POLLFD_CHUNK is too small */
> +static __attribute__((unused)) char ___a[-(MIN_POLLS <= 0)];
>
> static LIST_HEAD(clients);
> static pthread_mutex_t client_lock = PTHREAD_MUTEX_INITIALIZER;
> @@ -282,13 +293,13 @@ void * uxsock_listen(uxsock_trigger_fn uxsock_trigger, long ux_sock,
> char *inbuf;
> char *reply;
> sigset_t mask;
> - int old_clients = MIN_POLLS;
> + int max_pfds = MIN_POLLS + POLLFDS_BASE;
> /* conf->sequence_nr will be 1 when uxsock_listen is first called */
> unsigned int sequence_nr = 0;
> struct watch_descriptors wds = { .conf_wd = -1, .dir_wd = -1 };
>
> condlog(3, "uxsock: startup listener");
> - polls = (struct pollfd *)MALLOC((MIN_POLLS + 2) * sizeof(struct pollfd));
> + polls = MALLOC(max_pfds * sizeof(*polls));
> if (!polls) {
> condlog(0, "uxsock: failed to allocate poll fds");
> exit_daemon();
> @@ -312,28 +323,33 @@ void * uxsock_listen(uxsock_trigger_fn uxsock_trigger, long ux_sock,
> list_for_each_entry(c, &clients, node) {
> num_clients++;
> }
> - if (num_clients != old_clients) {
> + if (num_clients + POLLFDS_BASE > max_pfds) {
> struct pollfd *new;
> - if (num_clients <= MIN_POLLS && old_clients > MIN_POLLS) {
> - new = REALLOC(polls, (2 + MIN_POLLS) *
> - sizeof(struct pollfd));
> - } else if (num_clients <= MIN_POLLS && old_clients <= MIN_POLLS) {
> - new = polls;
> - } else {
> - new = REALLOC(polls, (2 + num_clients) *
> - sizeof(struct pollfd));
> - }
> - if (!new) {
> - condlog(0, "%s: failed to realloc %d poll fds",
> - "uxsock", 2 + num_clients);
> - num_clients = old_clients;
> - } else {
> - old_clients = num_clients;
> + int n_new = max_pfds + POLLFD_CHUNK;
> +
> + new = REALLOC(polls, n_new * sizeof(*polls));
> + if (new) {
> + max_pfds = n_new;
> polls = new;
> + } else {
> + condlog(1, "%s: realloc failure, %d clients not served",
> + __func__,
> + num_clients + POLLFDS_BASE - max_pfds);
> + num_clients = max_pfds - POLLFDS_BASE;
> }
> }
> - polls[0].fd = ux_sock;
> - polls[0].events = POLLIN;
> + if (num_clients < MAX_CLIENTS) {
> + polls[0].fd = ux_sock;
> + polls[0].events = POLLIN;
> + } else {
> + /*
> + * New clients can't connect, num_clients won't grow
> + * to MAX_CLIENTS or higher
> + */
> + condlog(1, "%s: max client connections reached, pausing polling",
> + __func__);
> + polls[0].fd = -1;
> + }
>
> reset_watch(notify_fd, &wds, &sequence_nr);
> if (notify_fd == -1 || (wds.conf_wd == -1 && wds.dir_wd == -1))
> @@ -343,19 +359,19 @@ void * uxsock_listen(uxsock_trigger_fn uxsock_trigger, long ux_sock,
> polls[1].events = POLLIN;
>
> /* setup the clients */
> - i = 2;
> + i = POLLFDS_BASE;
> list_for_each_entry(c, &clients, node) {
> polls[i].fd = c->fd;
> polls[i].events = POLLIN;
> i++;
> - if (i >= 2 + num_clients)
> + if (i >= max_pfds)
> break;
> }
> n_pfds = i;
> pthread_cleanup_pop(1);
>
> /* most of our life is spent in this call */
> - poll_count = ppoll(polls, n_pfds, &sleep_time, &mask);
> + poll_count = ppoll(polls, n_pfds, NULL, &mask);
>
> handle_signals(false);
> if (poll_count == -1) {
> @@ -388,7 +404,7 @@ void * uxsock_listen(uxsock_trigger_fn uxsock_trigger, long ux_sock,
> }
>
> /* see if a client wants to speak to us */
> - for (i = 2; i < n_pfds; i++) {
> + for (i = POLLFDS_BASE; i < n_pfds; i++) {
> if (polls[i].revents & POLLIN) {
> struct timespec start_time;
>
> --
> 2.28.0
--
dm-devel mailing list
dm-devel@redhat.com
https://www.redhat.com/mailman/listinfo/dm-devel
next prev parent reply other threads:[~2020-10-19 23:34 UTC|newest]
Thread overview: 51+ messages / expand[flat|nested] mbox.gz Atom feed top
2020-10-16 10:44 [dm-devel] [PATCH v2 00/29] libmultipath: improve cleanup on exit mwilck
2020-10-16 10:44 ` [dm-devel] [PATCH v2 01/29] multipathd: uxlsnr: avoid deadlock " mwilck
2020-10-20 19:04 ` Benjamin Marzinski
2020-10-16 10:44 ` [dm-devel] [PATCH v2 02/29] multipathd: Fix liburcu memory leak mwilck
2020-10-16 10:44 ` [dm-devel] [PATCH v2 03/29] multipathd: move handling of io_err_stat_attr into libmultipath mwilck
2020-10-16 10:44 ` [dm-devel] [PATCH v2 04/29] multipathd: move vecs desctruction into cleanup function mwilck
2020-10-16 10:44 ` [dm-devel] [PATCH v2 05/29] multipathd: make some globals static mwilck
2020-10-16 10:44 ` [dm-devel] [PATCH v2 06/29] multipathd: move threads destruction into separate function mwilck
2020-10-16 10:44 ` [dm-devel] [PATCH v2 07/29] multipathd: move conf " mwilck
2020-10-19 18:56 ` Benjamin Marzinski
2020-10-16 10:44 ` [dm-devel] [PATCH v2 08/29] multipathd: move pid " mwilck
2020-10-16 10:44 ` [dm-devel] [PATCH v2 09/29] multipathd: close pidfile on exit mwilck
2020-10-16 10:44 ` [dm-devel] [PATCH v2 10/29] multipathd: add helper for systemd notification at exit mwilck
2020-10-16 10:44 ` [dm-devel] [PATCH v2 11/29] multipathd: child(): call cleanups in failure case, too mwilck
2020-10-16 10:44 ` [dm-devel] [PATCH v2 12/29] multipathd: unwatch_all_dmevents: check if waiter is initialized mwilck
2020-10-16 10:44 ` [dm-devel] [PATCH v2 13/29] multipathd: print error message if config can't be loaded mwilck
2020-10-16 10:44 ` [dm-devel] [PATCH v2 14/29] libmultipath: add libmp_dm_exit() mwilck
2020-10-19 19:07 ` Benjamin Marzinski
2020-10-16 10:44 ` [dm-devel] [PATCH v2 15/29] multipathd: fixup libdm deinitialization mwilck
2020-10-16 10:44 ` [dm-devel] [PATCH v2 16/29] libmultipath: log_thread_stop(): check if logarea is initialized mwilck
2020-10-16 10:44 ` [dm-devel] [PATCH v2 17/29] multipathd: add cleanup_child() exit handler mwilck
2020-10-16 10:44 ` [dm-devel] [PATCH v2 18/29] libmultipath: fix log_thread startup and teardown mwilck
2020-10-19 20:00 ` Benjamin Marzinski
2020-10-26 13:58 ` Martin Wilck
2020-10-16 10:44 ` [dm-devel] [PATCH v2 19/29] multipathd: move cleanup_{prio, checkers, foreign} to libmultipath_exit mwilck
2020-10-16 10:44 ` [dm-devel] [PATCH v2 20/29] multipath: use atexit() for cleanup handlers mwilck
2020-10-16 10:44 ` [dm-devel] [PATCH v2 21/29] mpathpersist: " mwilck
2020-10-16 10:44 ` [dm-devel] [PATCH v2 22/29] multipath: fix leaks in check_path_valid() mwilck
2020-12-16 17:34 ` Martin Wilck
2020-12-16 22:41 ` Benjamin Marzinski
2020-10-16 10:44 ` [dm-devel] [PATCH v2 23/29] multipath-tools: mpath-tools.supp: file with valgrind suppressions mwilck
2020-10-19 20:01 ` Benjamin Marzinski
2020-10-16 10:44 ` [dm-devel] [PATCH v2 24/29] libmultipath: use libmp_verbosity to track verbosity mwilck
2020-10-19 20:38 ` Benjamin Marzinski
2020-10-26 14:47 ` Martin Wilck
2020-10-16 10:44 ` [dm-devel] [PATCH v2 25/29] libmultipath: introduce symbolic values for logsink mwilck
2020-10-16 20:13 ` Benjamin Marzinski
2020-10-16 10:44 ` [dm-devel] [PATCH v2 26/29] libmultipath: simplify dlog() mwilck
2020-10-19 21:07 ` Benjamin Marzinski
2020-10-16 10:44 ` [dm-devel] [PATCH v2 27/29] multipathd: common code for "-k" and command args mwilck
2020-10-19 21:51 ` Benjamin Marzinski
2020-10-16 10:45 ` [dm-devel] [PATCH v2 28/29] multipathd: sanitize uxsock_listen() mwilck
2020-10-19 23:33 ` Benjamin Marzinski [this message]
2020-10-26 13:54 ` Martin Wilck
2020-10-16 10:45 ` [dm-devel] [PATCH v2 29/29] libmultipath: fix race between log_safe and log_thread_stop() mwilck
2020-10-20 2:20 ` Benjamin Marzinski
2020-10-26 16:22 ` Martin Wilck
2020-10-26 17:24 ` Martin Wilck
2020-11-03 0:11 ` Benjamin Marzinski
2020-11-04 12:36 ` Martin Wilck
2020-11-04 15:46 ` Benjamin Marzinski
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=20201019233352.GJ3384@octiron.msp.redhat.com \
--to=bmarzins@redhat.com \
--cc=dm-devel@redhat.com \
--cc=lixiaokeng@huawei.com \
--cc=mwilck@suse.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).