Mirrors/libwebsockets
1
0
Fork 0
mirror of https://github.com/warmcat/libwebsockets.git synced 2025-03-09 00:00:04 +01:00
Code Issues Releases Wiki Activity

smp: adopt: deal with load balancing init window

Browse source 
With SMP as soon as we add the new sockfd to the fds table, in the
case we load-balanced the fd on to a different pt, service on it
becomes live immediately and concurrently.  This can lead to the
unexpected situation that while we think we're still initing the
new wsi from our thread, it can have lived out its whole life
concurrently from another service thread.

Add a volatile flag to inform the owning pt that if it wants to
service the wsi during this init window, it must wait and retry
next time around the event loop.
This commit is contained in:
Andy Green 2019-02-23 05:41:30 +08:00
parent 3a31c47fcd
commit a57d13cb1b
4 changed files with 36 additions and 1 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

21
lib/core-net/adopt.c
View file

@ -132,6 +132,12 @@ lws_adopt_descriptor_vhost(struct lws_vhost *vh, lws_adoption_type type,
}
#endif
/*
* Notice that in SMP case, the wsi may being being created on an
* entirely different pt / tsi for load balancing. In that case as
* we initialize it, it may become "live" concurrently unexpectedly...
*/
n = -1;
if (parent)
n = parent->tsi;
@ -194,6 +200,15 @@ lws_adopt_descriptor_vhost(struct lws_vhost *vh, lws_adoption_type type,
if (context->event_loop_ops->accept(new_wsi))
goto fail;
#if LWS_MAX_SMP > 1
/*
* Caution: after this point the wsi is live on its service thread
* which may be concurrent to this. We mark the wsi as still undergoing
* init in another pt so the assigned pt leaves it alone.
*/
new_wsi->undergoing_init_from_other_pt = 1;
#endif
if (!(type & LWS_ADOPT_ALLOW_SSL)) {
lws_pt_lock(pt, __func__);
if (__insert_wsi_socket_into_fds(context, new_wsi)) {
@ -224,6 +239,12 @@ lws_adopt_descriptor_vhost(struct lws_vhost *vh, lws_adoption_type type,
lws_role_call_adoption_bind(new_wsi, type | _LWS_ADOPT_FINISH,
vh_prot_name);
#if LWS_MAX_SMP > 1
/* its actual pt can service it now */
new_wsi->undergoing_init_from_other_pt = 0;
#endif
lws_cancel_service_pt(new_wsi);
return new_wsi;

4
lib/core-net/private.h
View file

@ -623,6 +623,10 @@ struct lws {
/* volatile to make sure code is aware other thread can change */
volatile char handling_pollout;
volatile char leave_pollout_active;
#if LWS_MAX_SMP > 1
volatile char undergoing_init_from_other_pt;
#endif
};
#define lws_is_flowcontrolled(w) (!!(wsi->rxflow_bitmap))

9
lib/core-net/service.c
View file

@ -960,6 +960,15 @@ lws_service_fd_tsi(struct lws_context *context, struct lws_pollfd *pollfd,
/* not lws connection ... leave revents alone and return */
return 0;
#if LWS_MAX_SMP > 1
if (wsi->undergoing_init_from_other_pt)
/*
* Temporary situation that other service thread is initializing
* this wsi right now for use on our service thread.
*/
return 0;
#endif
/*
* so that caller can tell we handled, past here we need to
* zero down pollfd->revents after handling

3
lib/roles/listen/ops-listen.c
View file

@ -136,7 +136,7 @@ rops_handle_POLLIN_listen(struct lws_context_per_thread *pt, struct lws *wsi,
/* already closed cleanly as necessary */
return LWS_HPI_RET_WSI_ALREADY_DIED;
}
/*
if (lws_server_socket_service_ssl(cwsi, accept_fd)) {
lws_close_free_wsi(cwsi, LWS_CLOSE_STATUS_NOSTATUS,
"listen svc fail");
@ -145,6 +145,7 @@ rops_handle_POLLIN_listen(struct lws_context_per_thread *pt, struct lws *wsi,
lwsl_info("%s: new wsi %p: wsistate 0x%x, role_ops %s\n",
__func__, cwsi, cwsi->wsistate, cwsi->role_ops->name);
*/
} while (pt->fds_count < context->fd_limit_per_thread - 1 &&
wsi->position_in_fds_table != LWS_NO_FDS_POS &&