1
0
Fork 0
mirror of https://github.com/warmcat/libwebsockets.git synced 2025-03-16 00:00:07 +01:00
libwebsockets/lib/core-net/pollfd.c
Andy Green 3549a94ce6 roles: compress role ops structs
role ops are usually only sparsely filled, there are currently 20
function pointers but several roles only fill in two.  No single
role has more than 14 of the ops.  On a 32/64 bit build this part
of the ops struct takes a fixed 80 / 160 bytes then.

First reduce the type of the callback reason part from uint16_t to
uint8_t, this saves 12 bytes unconditionally.

Change to a separate function pointer array with a nybble index
array, it costs 10 bytes for the index and a pointer to the
separate array, for 32-bit the cost is

2 + (4 x ops_used)

and for 64-bit

6 + (8 x ops_used)

for 2 x ops_used it means 32-bit: 10 vs 80 / 64-bit: 22 vs 160

For a typical system with h1 (9), h2 (14), listen (2), netlink (2),
pipe (1), raw_skt (3), ws (12), == 43 ops_used out of 140, it means
the .rodata for this reduced from 32-bit: 560 -> 174 (386 byte
saving) and 64-bit: 1120 -> 350 (770 byte saving)

This doesn't account for the changed function ops calling code, two
ways were tried, a preprocessor macro and explicit functions

For an x86_64 gcc 10 build with most options, release mode,
.text + .rodata

before patch:          553282
accessor macro:        552714 (568 byte saving)
accessor functions:    553674 (392 bytes worse than without patch)

therefore we went with the macros
2020-11-28 10:58:38 +00:00

664 lines
17 KiB
C

/*
* libwebsockets - small server side websockets and web server implementation
*
* Copyright (C) 2010 - 2020 Andy Green <andy@warmcat.com>
*
* Permission is hereby granted, free of charge, to any person obtaining a copy
* of this software and associated documentation files (the "Software"), to
* deal in the Software without restriction, including without limitation the
* rights to use, copy, modify, merge, publish, distribute, sublicense, and/or
* sell copies of the Software, and to permit persons to whom the Software is
* furnished to do so, subject to the following conditions:
*
* The above copyright notice and this permission notice shall be included in
* all copies or substantial portions of the Software.
*
* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
* IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
* FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
* AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
* LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
* FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
* IN THE SOFTWARE.
*/
#include "private-lib-core.h"
int
_lws_change_pollfd(struct lws *wsi, int _and, int _or, struct lws_pollargs *pa)
{
#if !defined(LWS_WITH_EVENT_LIBS)
volatile struct lws_context_per_thread *vpt;
#endif
struct lws_context_per_thread *pt;
struct lws_context *context;
int ret = 0, pa_events;
struct lws_pollfd *pfd;
int sampled_tid, tid;
if (!wsi)
return 0;
assert(wsi->position_in_fds_table == LWS_NO_FDS_POS ||
wsi->position_in_fds_table >= 0);
if (wsi->position_in_fds_table == LWS_NO_FDS_POS)
return 0;
if (((volatile struct lws *)wsi)->handling_pollout &&
!_and && _or == LWS_POLLOUT) {
/*
* Happening alongside service thread handling POLLOUT.
* The danger is when he is finished, he will disable POLLOUT,
* countermanding what we changed here.
*
* Instead of changing the fds, inform the service thread
* what happened, and ask it to leave POLLOUT active on exit
*/
((volatile struct lws *)wsi)->leave_pollout_active = 1;
/*
* by definition service thread is not in poll wait, so no need
* to cancel service
*/
lwsl_debug("%s: using leave_pollout_active\n", __func__);
return 0;
}
context = wsi->a.context;
pt = &context->pt[(int)wsi->tsi];
assert(wsi->position_in_fds_table < (int)pt->fds_count);
#if !defined(LWS_WITH_EVENT_LIBS)
/*
* This only applies when we use the default poll() event loop.
*
* BSD can revert pa->events at any time, when the kernel decides to
* exit from poll(). We can't protect against it using locking.
*
* Therefore we must check first if the service thread is in poll()
* wait; if so, we know we must be being called from a foreign thread,
* and we must keep a strictly ordered list of changes we made instead
* of trying to apply them, since when poll() exits, which may happen
* at any time it would revert our changes.
*
* The plat code will apply them when it leaves the poll() wait
* before doing anything else.
*/
vpt = (volatile struct lws_context_per_thread *)pt;
vpt->foreign_spinlock = 1;
lws_memory_barrier();
if (vpt->inside_poll) {
struct lws_foreign_thread_pollfd *ftp, **ftp1;
/*
* We are certainly a foreign thread trying to change events
* while the service thread is in the poll() wait.
*
* Create a list of changes to be applied after poll() exit,
* instead of trying to apply them now.
*/
ftp = lws_malloc(sizeof(*ftp), "ftp");
if (!ftp) {
vpt->foreign_spinlock = 0;
lws_memory_barrier();
ret = -1;
goto bail;
}
ftp->_and = _and;
ftp->_or = _or;
ftp->fd_index = wsi->position_in_fds_table;
ftp->next = NULL;
lws_pt_lock(pt, __func__);
/* place at END of list to maintain order */
ftp1 = (struct lws_foreign_thread_pollfd **)
&vpt->foreign_pfd_list;
while (*ftp1)
ftp1 = &((*ftp1)->next);
*ftp1 = ftp;
vpt->foreign_spinlock = 0;
lws_memory_barrier();
lws_pt_unlock(pt);
lws_cancel_service_pt(wsi);
return 0;
}
vpt->foreign_spinlock = 0;
lws_memory_barrier();
#endif
#if !defined(__linux__)
/* OSX couldn't see close on stdin pipe side otherwise */
_or |= LWS_POLLHUP;
#endif
pfd = &pt->fds[wsi->position_in_fds_table];
pa->fd = wsi->desc.sockfd;
lwsl_debug("%s: wsi %p: fd %d events %d -> %d\n", __func__, wsi,
pa->fd, pfd->events, (pfd->events & ~_and) | _or);
pa->prev_events = pfd->events;
pa->events = pfd->events = (pfd->events & ~_and) | _or;
if (wsi->mux_substream)
return 0;
#if defined(LWS_WITH_EXTERNAL_POLL)
if (wsi->a.vhost &&
wsi->a.vhost->protocols[0].callback(wsi,
LWS_CALLBACK_CHANGE_MODE_POLL_FD,
wsi->user_space, (void *)pa, 0)) {
ret = -1;
goto bail;
}
#endif
if (context->event_loop_ops->io) {
if (_and & LWS_POLLIN)
context->event_loop_ops->io(wsi,
LWS_EV_STOP | LWS_EV_READ);
if (_or & LWS_POLLIN)
context->event_loop_ops->io(wsi,
LWS_EV_START | LWS_EV_READ);
if (_and & LWS_POLLOUT)
context->event_loop_ops->io(wsi,
LWS_EV_STOP | LWS_EV_WRITE);
if (_or & LWS_POLLOUT)
context->event_loop_ops->io(wsi,
LWS_EV_START | LWS_EV_WRITE);
}
/*
* if we changed something in this pollfd...
* ... and we're running in a different thread context
* than the service thread...
* ... and the service thread is waiting ...
* then cancel it to force a restart with our changed events
*/
pa_events = pa->prev_events != pa->events;
if (pa_events) {
if (lws_plat_change_pollfd(context, wsi, pfd)) {
lwsl_info("%s failed\n", __func__);
ret = -1;
goto bail;
}
sampled_tid = pt->service_tid;
if (sampled_tid && wsi->a.vhost) {
tid = wsi->a.vhost->protocols[0].callback(wsi,
LWS_CALLBACK_GET_THREAD_ID, NULL, NULL, 0);
if (tid == -1) {
ret = -1;
goto bail;
}
if (tid != sampled_tid)
lws_cancel_service_pt(wsi);
}
}
bail:
return ret;
}
#if defined(LWS_WITH_SERVER)
/*
* Enable or disable listen sockets on this pt globally...
* it's modulated according to the pt having space for a new accept.
*/
static void
lws_accept_modulation(struct lws_context *context,
struct lws_context_per_thread *pt, int allow)
{
struct lws_vhost *vh = context->vhost_list;
struct lws_pollargs pa1;
while (vh) {
if (vh->lserv_wsi) {
if (allow)
_lws_change_pollfd(vh->lserv_wsi,
0, LWS_POLLIN, &pa1);
else
_lws_change_pollfd(vh->lserv_wsi,
LWS_POLLIN, 0, &pa1);
}
vh = vh->vhost_next;
}
}
#endif
#if _LWS_ENABLED_LOGS & LLL_WARN
void
__dump_fds(struct lws_context_per_thread *pt, const char *s)
{
unsigned int n;
lwsl_warn("%s: fds_count %u, %s\n", __func__, pt->fds_count, s);
for (n = 0; n < pt->fds_count; n++) {
struct lws *wsi = wsi_from_fd(pt->context, pt->fds[n].fd);
lwsl_warn(" %d: fd %d, wsi %p, pos_in_fds: %d\n",
n + 1, pt->fds[n].fd, wsi,
wsi ? wsi->position_in_fds_table : -1);
}
}
#else
#define __dump_fds(x, y)
#endif
int
__insert_wsi_socket_into_fds(struct lws_context *context, struct lws *wsi)
{
#if defined(LWS_WITH_EXTERNAL_POLL)
struct lws_pollargs pa = { wsi->desc.sockfd, LWS_POLLIN, 0 };
#endif
struct lws_context_per_thread *pt = &context->pt[(int)wsi->tsi];
int ret = 0;
// __dump_fds(pt, "pre insert");
lws_pt_assert_lock_held(pt);
lwsl_debug("%s: %p: tsi=%d, sock=%d, pos-in-fds=%d\n",
__func__, wsi, wsi->tsi, wsi->desc.sockfd, pt->fds_count);
if ((unsigned int)pt->fds_count >= context->fd_limit_per_thread) {
lwsl_err("Too many fds (%d vs %d)\n", context->max_fds,
context->fd_limit_per_thread );
return 1;
}
#if !defined(_WIN32)
if (!wsi->a.context->max_fds_unrelated_to_ulimit &&
wsi->desc.sockfd - lws_plat_socket_offset() >= context->max_fds) {
lwsl_err("Socket fd %d is too high (%d) offset %d\n",
wsi->desc.sockfd, context->max_fds,
lws_plat_socket_offset());
return 1;
}
#endif
assert(wsi);
#if defined(LWS_WITH_NETLINK)
assert(wsi->event_pipe || wsi->a.vhost || wsi == pt->netlink);
#else
assert(wsi->event_pipe || wsi->a.vhost);
#endif
assert(lws_socket_is_valid(wsi->desc.sockfd));
#if defined(LWS_WITH_EXTERNAL_POLL)
if (wsi->a.vhost &&
wsi->a.vhost->protocols[0].callback(wsi, LWS_CALLBACK_LOCK_POLL,
wsi->user_space, (void *) &pa, 1))
return -1;
#endif
if (insert_wsi(context, wsi))
return -1;
pt->count_conns++;
wsi->position_in_fds_table = pt->fds_count;
pt->fds[wsi->position_in_fds_table].fd = wsi->desc.sockfd;
pt->fds[wsi->position_in_fds_table].events = LWS_POLLIN;
#if defined(LWS_WITH_EXTERNAL_POLL)
pa.events = pt->fds[pt->fds_count].events;
#endif
lws_plat_insert_socket_into_fds(context, wsi);
#if defined(LWS_WITH_EXTERNAL_POLL)
/* external POLL support via protocol 0 */
if (wsi->a.vhost &&
wsi->a.vhost->protocols[0].callback(wsi, LWS_CALLBACK_ADD_POLL_FD,
wsi->user_space, (void *) &pa, 0))
ret = -1;
#endif
#if defined(LWS_WITH_SERVER)
/* if no more room, defeat accepts on this service thread */
if ((unsigned int)pt->fds_count == context->fd_limit_per_thread - 1)
lws_accept_modulation(context, pt, 0);
#endif
#if defined(LWS_WITH_EXTERNAL_POLL)
if (wsi->a.vhost &&
wsi->a.vhost->protocols[0].callback(wsi, LWS_CALLBACK_UNLOCK_POLL,
wsi->user_space, (void *)&pa, 1))
ret = -1;
#endif
// __dump_fds(pt, "post insert");
return ret;
}
int
__remove_wsi_socket_from_fds(struct lws *wsi)
{
struct lws_context *context = wsi->a.context;
#if defined(LWS_WITH_EXTERNAL_POLL)
struct lws_pollargs pa = { wsi->desc.sockfd, 0, 0 };
#endif
struct lws_context_per_thread *pt = &context->pt[(int)wsi->tsi];
struct lws *end_wsi;
int v, m, ret = 0;
lws_pt_assert_lock_held(pt);
// __dump_fds(pt, "pre remove");
#if !defined(_WIN32)
if (!wsi->a.context->max_fds_unrelated_to_ulimit &&
wsi->desc.sockfd - lws_plat_socket_offset() > context->max_fds) {
lwsl_err("fd %d too high (%d)\n", wsi->desc.sockfd,
context->max_fds);
return 1;
}
#endif
#if defined(LWS_WITH_EXTERNAL_POLL)
if (wsi->a.vhost && wsi->a.vhost->protocols &&
wsi->a.vhost->protocols[0].callback(wsi, LWS_CALLBACK_LOCK_POLL,
wsi->user_space, (void *)&pa, 1))
return -1;
#endif
__lws_same_vh_protocol_remove(wsi);
/* the guy who is to be deleted's slot index in pt->fds */
m = wsi->position_in_fds_table;
/* these are the only valid possibilities for position_in_fds_table */
assert(m == LWS_NO_FDS_POS || (m >= 0 && (unsigned int)m < pt->fds_count));
if (context->event_loop_ops->io)
context->event_loop_ops->io(wsi,
LWS_EV_STOP | LWS_EV_READ | LWS_EV_WRITE |
LWS_EV_PREPARE_DELETION);
/*
lwsl_notice("%s: wsi=%p, skt=%d, fds pos=%d, end guy pos=%d, endfd=%d\n",
__func__, wsi, wsi->desc.sockfd, wsi->position_in_fds_table,
pt->fds_count, pt->fds[pt->fds_count - 1].fd); */
if (m != LWS_NO_FDS_POS) {
char fixup = 0;
assert(pt->fds_count && (unsigned int)m != pt->fds_count);
/* deletion guy's lws_lookup entry needs nuking */
delete_from_fd(context, wsi->desc.sockfd);
if ((unsigned int)m != pt->fds_count - 1) {
/* have the last guy take up the now vacant slot */
pt->fds[m] = pt->fds[pt->fds_count - 1];
fixup = 1;
}
pt->fds[pt->fds_count - 1].fd = -1;
/* this decrements pt->fds_count */
lws_plat_delete_socket_from_fds(context, wsi, m);
pt->count_conns--;
if (fixup) {
v = (int) pt->fds[m].fd;
/* old end guy's "position in fds table" is now the
* deletion guy's old one */
end_wsi = wsi_from_fd(context, v);
if (!end_wsi) {
lwsl_err("no wsi for fd %d pos %d, "
"pt->fds_count=%d\n",
(int)pt->fds[m].fd, m, pt->fds_count);
// assert(0);
} else
end_wsi->position_in_fds_table = m;
}
/* removed wsi has no position any more */
wsi->position_in_fds_table = LWS_NO_FDS_POS;
}
#if defined(LWS_WITH_EXTERNAL_POLL)
/* remove also from external POLL support via protocol 0 */
if (lws_socket_is_valid(wsi->desc.sockfd) && wsi->a.vhost &&
wsi->a.vhost->protocols[0].callback(wsi, LWS_CALLBACK_DEL_POLL_FD,
wsi->user_space, (void *) &pa, 0))
ret = -1;
#endif
#if defined(LWS_WITH_SERVER)
if (!context->being_destroyed &&
/* if this made some room, accept connects on this thread */
(unsigned int)pt->fds_count < context->fd_limit_per_thread - 1)
lws_accept_modulation(context, pt, 1);
#endif
#if defined(LWS_WITH_EXTERNAL_POLL)
if (wsi->a.vhost &&
wsi->a.vhost->protocols[0].callback(wsi, LWS_CALLBACK_UNLOCK_POLL,
wsi->user_space, (void *) &pa, 1))
ret = -1;
#endif
// __dump_fds(pt, "post remove");
return ret;
}
int
__lws_change_pollfd(struct lws *wsi, int _and, int _or)
{
struct lws_context *context;
struct lws_pollargs pa;
int ret = 0;
if (!wsi || (!wsi->a.protocol && !wsi->event_pipe) ||
wsi->position_in_fds_table == LWS_NO_FDS_POS)
return 0;
context = lws_get_context(wsi);
if (!context)
return 1;
#if defined(LWS_WITH_EXTERNAL_POLL)
if (wsi->a.vhost &&
wsi->a.vhost->protocols[0].callback(wsi, LWS_CALLBACK_LOCK_POLL,
wsi->user_space, (void *) &pa, 0))
return -1;
#endif
ret = _lws_change_pollfd(wsi, _and, _or, &pa);
#if defined(LWS_WITH_EXTERNAL_POLL)
if (wsi->a.vhost &&
wsi->a.vhost->protocols[0].callback(wsi, LWS_CALLBACK_UNLOCK_POLL,
wsi->user_space, (void *) &pa, 0))
ret = -1;
#endif
return ret;
}
int
lws_change_pollfd(struct lws *wsi, int _and, int _or)
{
struct lws_context_per_thread *pt;
int ret = 0;
pt = &wsi->a.context->pt[(int)wsi->tsi];
lws_pt_lock(pt, __func__);
ret = __lws_change_pollfd(wsi, _and, _or);
lws_pt_unlock(pt);
return ret;
}
int
lws_callback_on_writable(struct lws *wsi)
{
struct lws_context_per_thread *pt;
struct lws *w = wsi;
if (lwsi_state(wsi) == LRS_SHUTDOWN)
return 0;
if (wsi->socket_is_permanently_unusable)
return 0;
pt = &wsi->a.context->pt[(int)wsi->tsi];
#if defined(LWS_WITH_DETAILED_LATENCY)
if (!wsi->detlat.earliest_write_req)
wsi->detlat.earliest_write_req = lws_now_usecs();
#endif
lws_stats_bump(pt, LWSSTATS_C_WRITEABLE_CB_REQ, 1);
#if defined(LWS_WITH_STATS)
if (!wsi->active_writable_req_us) {
wsi->active_writable_req_us = lws_now_usecs();
lws_stats_bump(pt, LWSSTATS_C_WRITEABLE_CB_EFF_REQ, 1);
}
#endif
if (lws_rops_fidx(wsi->role_ops, LWS_ROPS_callback_on_writable)) {
int q = lws_rops_func_fidx(wsi->role_ops,
LWS_ROPS_callback_on_writable).
callback_on_writable(wsi);
//lwsl_notice("%s: rops_cow says %d\n", __func__, q);
if (q)
return 1;
w = lws_get_network_wsi(wsi);
} else
if (w->position_in_fds_table == LWS_NO_FDS_POS) {
lwsl_debug("%s: failed to find socket %d\n", __func__,
wsi->desc.sockfd);
return -1;
}
//lwsl_notice("%s: marking for POLLOUT %p (wsi %p)\n", __func__, w, wsi);
if (__lws_change_pollfd(w, 0, LWS_POLLOUT))
return -1;
return 1;
}
/*
* stitch protocol choice into the vh protocol linked list
* We always insert ourselves at the start of the list
*
* X <-> B
* X <-> pAn <-> pB
*
* Illegal to attach more than once without detach inbetween
*/
void
lws_same_vh_protocol_insert(struct lws *wsi, int n)
{
lws_vhost_lock(wsi->a.vhost);
lws_dll2_remove(&wsi->same_vh_protocol);
lws_dll2_add_head(&wsi->same_vh_protocol,
&wsi->a.vhost->same_vh_protocol_owner[n]);
wsi->bound_vhost_index = n;
lws_vhost_unlock(wsi->a.vhost);
}
void
__lws_same_vh_protocol_remove(struct lws *wsi)
{
if (wsi->a.vhost && wsi->a.vhost->same_vh_protocol_owner)
lws_dll2_remove(&wsi->same_vh_protocol);
}
void
lws_same_vh_protocol_remove(struct lws *wsi)
{
if (!wsi->a.vhost)
return;
lws_vhost_lock(wsi->a.vhost);
__lws_same_vh_protocol_remove(wsi);
lws_vhost_unlock(wsi->a.vhost);
}
int
lws_callback_on_writable_all_protocol_vhost(const struct lws_vhost *vhost,
const struct lws_protocols *protocol)
{
struct lws *wsi;
int n;
if (protocol < vhost->protocols ||
protocol >= (vhost->protocols + vhost->count_protocols)) {
lwsl_err("%s: protocol %p is not from vhost %p (%p - %p)\n",
__func__, protocol, vhost->protocols, vhost,
(vhost->protocols + vhost->count_protocols));
return -1;
}
n = (int)(protocol - vhost->protocols);
lws_start_foreach_dll_safe(struct lws_dll2 *, d, d1,
lws_dll2_get_head(&vhost->same_vh_protocol_owner[n])) {
wsi = lws_container_of(d, struct lws, same_vh_protocol);
assert(wsi->a.protocol == protocol);
lws_callback_on_writable(wsi);
} lws_end_foreach_dll_safe(d, d1);
return 0;
}
int
lws_callback_on_writable_all_protocol(const struct lws_context *context,
const struct lws_protocols *protocol)
{
struct lws_vhost *vhost;
int n;
if (!context)
return 0;
vhost = context->vhost_list;
while (vhost) {
for (n = 0; n < vhost->count_protocols; n++)
if (protocol->callback ==
vhost->protocols[n].callback &&
!strcmp(protocol->name, vhost->protocols[n].name))
break;
if (n != vhost->count_protocols)
lws_callback_on_writable_all_protocol_vhost(
vhost, &vhost->protocols[n]);
vhost = vhost->vhost_next;
}
return 0;
}