enhanced how to handle an error after mio_dev_make() failure by implementing fail_after_make callbacks in pipe, thr, pro

decided to not call pthread_cancel in thr.c because it's too difficult to get it right
This commit is contained in:
hyung-hwan 2020-07-31 15:07:28 +00:00
parent a2e3195417
commit c230c92249
6 changed files with 720 additions and 57 deletions

574
mio/bin/t06.c Normal file
View File

@ -0,0 +1,574 @@
#include <mio-sck.h>
#include <mio-http.h>
#include <mio-utl.h>
#include <stdio.h>
#include <string.h>
#include <pthread.h>
#include <unistd.h>
#include <signal.h>
#include <errno.h>
#include <assert.h>
#define NUM_THRS 8
static int g_reuse_port = 0;
static mio_svc_htts_t* g_htts[NUM_THRS];
static int g_htts_no = 0;
static pthread_mutex_t g_htts_mutex = PTHREAD_MUTEX_INITIALIZER;
static int print_qparam (mio_bcs_t* key, mio_bcs_t* val, void* ctx)
{
key->len = mio_perdec_http_bcs(key, key->ptr, MIO_NULL);
val->len = mio_perdec_http_bcs(val, val->ptr, MIO_NULL);
fprintf ((FILE*)ctx, "\t[%.*s] = [%.*s]\n", (int)key->len, key->ptr, (int)val->len, val->ptr);
return 0;
}
static void on_htts_thr_request (mio_t* mio, mio_dev_thr_iopair_t* iop, mio_svc_htts_thr_func_info_t* tfi, void* ctx)
{
FILE* fp;
int i;
if (tfi->req_method != MIO_HTTP_GET)
{
write (iop->wfd, "Status: 405\r\n\r\n", 15); /* method not allowed */
return;
}
fp = fdopen(iop->wfd, "w");
if (!fp)
{
write (iop->wfd, "Status: 500\r\n\r\n", 15); /* internal server error */
return;
}
fprintf (fp, "Status: 200\r\n");
fprintf (fp, "Content-Type: text/html\r\n\r\n");
fprintf (fp, "request path = %s\n", tfi->req_path);
if (tfi->req_param)
{
fprintf (fp, "request params:\n");
mio_scan_http_qparam (tfi->req_param, print_qparam, fp);
}
for (i = 0; i < 100; i++) fprintf (fp, "%d * %d => %d\n", i, i, i * i);
/* invalid iop->wfd to mark that this function closed this file descriptor.
* no invalidation will lead to double closes on the same file descriptor. */
iop->wfd = MIO_SYSHND_INVALID;
fclose (fp);
}
static void on_htts_thr2_request (mio_t* mio, mio_dev_thr_iopair_t* iop, mio_svc_htts_thr_func_info_t* tfi, void* ctx)
{
FILE* fp, * sf;
if (tfi->req_method != MIO_HTTP_GET)
{
write (iop->wfd, "Status: 405\r\n\r\n", 15); /* method not allowed */
return;
}
fp = fdopen(iop->wfd, "w");
if (!fp)
{
write (iop->wfd, "Status: 500\r\n\r\n", 15); /* internal server error */
return;
}
sf = fopen(&tfi->req_path[5], "r");
if (!sf)
{
fprintf (fp, "Status: 404\r\n\r\n");
}
else
{
char buf[4096];
fprintf (fp, "Status: 200\r\n");
fprintf (fp, "Content-Type: text/html\r\n\r\n");
while (!feof(sf))
{
size_t n;
n = fread(buf, 1, sizeof(buf), sf);
if (n > 0) fwrite (buf, 1, n, fp);
}
fclose (sf);
}
/* invalid iop->wfd to mark that this function closed this file descriptor.
* no invalidation will lead to double closes on the same file descriptor. */
iop->wfd = MIO_SYSHND_INVALID;
fclose (fp);
}
/* ========================================================================= */
int process_http_request (mio_svc_htts_t* htts, mio_dev_sck_t* csck, mio_htre_t* req)
{
mio_t* mio = mio_svc_htts_getmio(htts);
// mio_svc_htts_cli_t* cli = mio_dev_sck_getxtn(csck);
mio_http_method_t mth;
/* percent-decode the query path to the original buffer
* since i'm not going to need it in the original form
* any more. once it's decoded in the peek mode,
* the decoded query path is made available in the
* non-peek mode as well */
MIO_DEBUG2 (mio, "[RAW-REQ] %s %s\n", mio_htre_getqmethodname(req), mio_htre_getqpath(req));
mio_htre_perdecqpath(req);
/* TODO: proper request logging */
MIO_DEBUG2 (mio, "[REQ] %s %s\n", mio_htre_getqmethodname(req), mio_htre_getqpath(req));
#if 0
mio_printf (MIO_T("================================\n"));
mio_printf (MIO_T("[%lu] %hs REQUEST ==> [%hs] version[%d.%d %hs] method[%hs]\n"),
(unsigned long)time(NULL),
(peek? MIO_MT("PEEK"): MIO_MT("HANDLE")),
mio_htre_getqpath(req),
mio_htre_getmajorversion(req),
mio_htre_getminorversion(req),
mio_htre_getverstr(req),
mio_htre_getqmethodname(req)
);
if (mio_htre_getqparam(req))
mio_printf (MIO_T("PARAMS ==> [%hs]\n"), mio_htre_getqparam(req));
mio_htb_walk (&req->hdrtab, walk, MIO_NULL);
if (mio_htre_getcontentlen(req) > 0)
{
mio_printf (MIO_T("CONTENT [%.*S]\n"), (int)mio_htre_getcontentlen(req), mio_htre_getcontentptr(req));
}
#endif
mth = mio_htre_getqmethodtype(req);
/* determine what to do once the header fields are all received.
* i don't want to delay this until the contents are received.
* if you don't like this behavior, you must implement your own
* callback function for request handling. */
#if 0
/* TODO support X-HTTP-Method-Override */
if (data.method == MIO_HTTP_POST)
{
tmp = mio_htre_getheaderval(req, MIO_MT("X-HTTP-Method-Override"));
if (tmp)
{
/*while (tmp->next) tmp = tmp->next;*/ /* get the last value */
data.method = mio_mbstohttpmethod (tmp->ptr);
}
}
#endif
#if 0
if (mth == MIO_HTTP_CONNECT)
{
/* CONNECT method must not have content set.
* however, arrange to discard it if so.
*
* NOTE: CONNECT is implemented to ignore many headers like
* 'Expect: 100-continue' and 'Connection: keep-alive'. */
mio_htre_discardcontent (req);
}
else
{
/* this part can be checked in actual mio_svc_htts_doXXX() functions.
* some doXXX handlers may not require length for POST.
* it may be able to simply accept till EOF? or treat as if CONTENT_LENGTH is 0*/
if (mth == MIO_HTTP_POST && !(req->flags & (MIO_HTRE_ATTR_LENGTH | MIO_HTRE_ATTR_CHUNKED)))
{
/* POST without Content-Length nor not chunked */
mio_htre_discardcontent (req);
/* 411 Length Required - can't keep alive. Force disconnect */
req->flags &= ~MIO_HTRE_ATTR_KEEPALIVE; /* to cause sendstatus() to close */
if (mio_svc_htts_sendstatus(htts, csck, req, 411, MIO_NULL) <= -1) goto oops;
}
else
{
#endif
const mio_bch_t* qpath = mio_htre_getqpath(req);
int x;
if (mio_comp_bcstr_limited(qpath, "/thr/", 5, 1) == 0)
x = mio_svc_htts_dothr(htts, csck, req, on_htts_thr_request, MIO_NULL);
else if (mio_comp_bcstr_limited(qpath, "/thr2/", 6, 1) == 0)
x = mio_svc_htts_dothr(htts, csck, req, on_htts_thr2_request, MIO_NULL);
else if (mio_comp_bcstr_limited(qpath, "/txt/", 5, 1) == 0)
x = mio_svc_htts_dotxt(htts, csck, req, 200, "text/plain", qpath);
else if (mio_comp_bcstr_limited(qpath, "/cgi/", 5, 1) == 0)
x = mio_svc_htts_docgi(htts, csck, req, "", mio_htre_getqpath(req));
else
x = mio_svc_htts_dofile(htts, csck, req, "", mio_htre_getqpath(req), "text/plain");
if (x <= -1) goto oops;
return 0;
oops:
mio_dev_sck_halt (csck);
return -1;
}
void* thr_func (void* arg)
{
mio_t* mio = MIO_NULL;
mio_svc_htts_t* htts = MIO_NULL;
mio_dev_sck_bind_t htts_bind_info;
mio = mio_open(MIO_NULL, 0, MIO_NULL, 512, MIO_NULL);
if (!mio)
{
printf ("Cannot open mio\n");
goto oops;
}
memset (&htts_bind_info, 0, MIO_SIZEOF(htts_bind_info));
if (g_reuse_port)
{
mio_bcstrtoskad (mio, "0.0.0.0:9987", &htts_bind_info.localaddr);
htts_bind_info.options = MIO_DEV_SCK_BIND_REUSEADDR | MIO_DEV_SCK_BIND_REUSEPORT | MIO_DEV_SCK_BIND_IGNERR;
//htts_bind_info.options |= MIO_DEV_SCK_BIND_SSL;
htts_bind_info.ssl_certfile = "localhost.crt";
htts_bind_info.ssl_keyfile = "localhost.key";
}
htts = mio_svc_htts_start(mio, &htts_bind_info, process_http_request);
if (!htts)
{
printf ("Unable to start htts\n");
goto oops;
}
pthread_mutex_lock (&g_htts_mutex);
g_htts[g_htts_no] = htts;
printf ("starting the loop for %d\n", g_htts_no);
g_htts_no = (g_htts_no + 1) % MIO_COUNTOF(g_htts);
pthread_mutex_unlock (&g_htts_mutex);
mio_loop (mio);
oops:
if (htts) mio_svc_htts_stop (htts);
if (mio) mio_close (mio);
pthread_exit (MIO_NULL);
return MIO_NULL;
}
/* ========================================================================= */
static void tcp_sck_on_disconnect (mio_dev_sck_t* tcp)
{
switch (MIO_DEV_SCK_GET_PROGRESS(tcp))
{
case MIO_DEV_SCK_CONNECTING:
MIO_INFO1 (tcp->mio, "OUTGOING SESSION DISCONNECTED - FAILED TO CONNECT (%d) TO REMOTE SERVER\n", (int)tcp->hnd);
break;
case MIO_DEV_SCK_CONNECTING_SSL:
MIO_INFO1 (tcp->mio, "OUTGOING SESSION DISCONNECTED - FAILED TO SSL-CONNECT (%d) TO REMOTE SERVER\n", (int)tcp->hnd);
break;
case MIO_DEV_SCK_LISTENING:
MIO_INFO1 (tcp->mio, "SHUTTING DOWN THE SERVER SOCKET(%d)...\n", (int)tcp->hnd);
break;
case MIO_DEV_SCK_CONNECTED:
MIO_INFO1 (tcp->mio, "OUTGOING CLIENT CONNECTION GOT TORN DOWN(%d).......\n", (int)tcp->hnd);
break;
case MIO_DEV_SCK_ACCEPTING_SSL:
MIO_INFO1 (tcp->mio, "INCOMING SSL-ACCEPT GOT DISCONNECTED(%d) ....\n", (int)tcp->hnd);
break;
case MIO_DEV_SCK_ACCEPTED:
MIO_INFO1 (tcp->mio, "INCOMING CLIENT BEING SERVED GOT DISCONNECTED(%d).......\n", (int)tcp->hnd);
break;
default:
MIO_INFO2 (tcp->mio, "SOCKET DEVICE DISCONNECTED (%d - %x)\n", (int)tcp->hnd, (unsigned int)tcp->state);
break;
}
}
static void tcp_sck_on_connect (mio_dev_sck_t* tcp)
{
mio_bch_t buf1[128], buf2[128];
mio_skadtobcstr (tcp->mio, &tcp->localaddr, buf1, MIO_COUNTOF(buf1), MIO_SKAD_TO_BCSTR_ADDR | MIO_SKAD_TO_BCSTR_PORT);
mio_skadtobcstr (tcp->mio, &tcp->remoteaddr, buf2, MIO_COUNTOF(buf2), MIO_SKAD_TO_BCSTR_ADDR | MIO_SKAD_TO_BCSTR_PORT);
if (tcp->state & MIO_DEV_SCK_CONNECTED)
{
MIO_INFO3 (tcp->mio, "DEVICE connected to a remote server... LOCAL %hs REMOTE %hs SCK: %d\n", buf1, buf2, tcp->hnd);
}
else if (tcp->state & MIO_DEV_SCK_ACCEPTED)
{
/* TODO: pass it to distributor??? */
/* THIS PART WON"T BE CALLED FOR tcp_sck_on_raw_accept.. */
}
}
static mio_tmridx_t xx_tmridx;
static int try_to_accept (mio_dev_sck_t* sck, mio_dev_sck_qxmsg_t* qxmsg, int in_mq);
typedef struct xx_mq_t xx_mq_t;
struct xx_mq_t
{
xx_mq_t* q_next;
xx_mq_t* q_prev;
mio_dev_sck_qxmsg_t msg;
};
#define XX_MQ_INIT(mq) ((mq)->q_next = (mq)->q_prev = (mq))
#define XX_MQ_TAIL(mq) ((mq)->q_prev)
#define XX_MQ_HEAD(mq) ((mq)->q_next)
#define XX_MQ_IS_EMPTY(mq) (XX_MQ_HEAD(mq) == (mq))
#define XX_MQ_IS_NODE(mq,x) ((mq) != (x))
#define XX_MQ_IS_HEAD(mq,x) (XX_MQ_HEAD(mq) == (x))
#define XX_MQ_IS_TAIL(mq,x) (XX_MQ_TAIL(mq) == (x))
#define XX_MQ_NEXT(x) ((x)->q_next)
#define XX_MQ_PREV(x) ((x)->q_prev)
#define XX_MQ_LINK(p,x,n) MIO_Q_LINK((mio_q_t*)p,(mio_q_t*)x,(mio_q_t*)n)
#define XX_MQ_UNLINK(x) MIO_Q_UNLINK((mio_q_t*)x)
#define XX_MQ_REPL(o,n) MIO_Q_REPL(o,n);
#define XX_MQ_ENQ(mq,x) XX_MQ_LINK(XX_MQ_TAIL(mq), (mio_q_t*)x, mq)
#define XX_MQ_DEQ(mq) XX_MQ_UNLINK(XX_MQ_HEAD(mq))
static xx_mq_t xx_mq;
static int schedule_timer_job_at (mio_dev_sck_t* dev, const mio_ntime_t* fire_at, mio_tmrjob_handler_t handler, mio_tmridx_t* tmridx)
{
mio_tmrjob_t tmrjob;
memset (&tmrjob, 0, MIO_SIZEOF(tmrjob));
tmrjob.ctx = dev;
if (fire_at) tmrjob.when = *fire_at;
tmrjob.handler = handler;
tmrjob.idxptr = tmridx;
return mio_instmrjob(dev->mio, &tmrjob);
}
static void enable_accept (mio_t* mio, const mio_ntime_t* now, mio_tmrjob_t* job)
{
mio_dev_sck_t* rdev = (mio_dev_sck_t*)job->ctx;
while (!XX_MQ_IS_EMPTY(&xx_mq))
{
xx_mq_t* mq;
mq = XX_MQ_HEAD(&xx_mq);
if (try_to_accept(rdev, &mq->msg, 1) == 0) return; /* EAGAIN situation */
XX_MQ_UNLINK (mq);
mio_freemem (mio, mq);
}
assert (XX_MQ_IS_EMPTY(&xx_mq));
if (mio_dev_sck_read(rdev, 1) <= -1) // it's a disaster if this fails. the acceptor will get stalled if it happens
{
printf ("DISASTER.... UNABLE TO ENABLE READ ON ACCEPTOR\n");
}
}
static int try_to_accept (mio_dev_sck_t* sck, mio_dev_sck_qxmsg_t* qxmsg, int in_mq)
{
mio_t* mio = sck->mio;
mio_svc_htts_t* htts;
pthread_mutex_lock (&g_htts_mutex);
htts = g_htts[g_htts_no];
g_htts_no = (g_htts_no + 1) % MIO_COUNTOF(g_htts);
pthread_mutex_unlock (&g_htts_mutex);
if (mio_svc_htts_writetosidechan(htts, qxmsg, MIO_SIZEOF(*qxmsg)) <= -1)
{
mio_bch_t buf[128];
if (errno == EAGAIN)
{
//printf ("sidechannel retrying %s\n", strerror(errno));
if (mio_dev_sck_read(sck, 0) <= -1) goto sidechan_write_error;
if (!in_mq)
{
xx_mq_t* mq;
mq = mio_allocmem(mio, MIO_SIZEOF(*mq));
if (MIO_UNLIKELY(!mq)) goto sidechan_write_error;
mq->msg = *qxmsg;
XX_MQ_ENQ (&xx_mq, mq);
}
if (xx_tmridx == MIO_TMRIDX_INVALID)
schedule_timer_job_at (sck, MIO_NULL, enable_accept, &xx_tmridx);
return 0; /* enqueued for later writing */
}
else
{
sidechan_write_error:
printf ("sidechannel write error %s\n", strerror(errno));
mio_skadtobcstr (mio, &qxmsg->remoteaddr, buf, MIO_COUNTOF(buf), MIO_SKAD_TO_BCSTR_ADDR | MIO_SKAD_TO_BCSTR_PORT);
MIO_INFO2 (mio, "unable to handle the accepted connection %ld from %hs\n", (long int)qxmsg->syshnd, buf);
const char* msg = "HTTP/1.0 503 Service unavailable\r\nConnection: close\r\nContent-Length: 0\r\n\r\n";
write (qxmsg->syshnd, msg, strlen(msg));
printf ("close %d\n", qxmsg->syshnd);
close (qxmsg->syshnd);
return -1; /* failed to accept */
}
}
/************************************
{
static int sc = 0;
printf ("sc => %d\n", sc++);
}
************************************/
return 1; /* full success */
}
static void tcp_sck_on_raw_accept (mio_dev_sck_t* sck, mio_syshnd_t syshnd, mio_skad_t* remoteaddr)
{
/*mio_t* mio = sck->mio;*/
/* inform the worker of this accepted syshnd */
mio_dev_sck_qxmsg_t qxmsg;
memset (&qxmsg, 0, MIO_SIZEOF(qxmsg));
qxmsg.cmd = MIO_DEV_SCK_QXMSG_NEWCONN;
qxmsg.scktype = sck->type;
qxmsg.syshnd = syshnd;
qxmsg.remoteaddr = *remoteaddr;
//printf ("A %d\n", qxmsg.syshnd);
try_to_accept (sck, &qxmsg, 0);
}
static int tcp_sck_on_write (mio_dev_sck_t* tcp, mio_iolen_t wrlen, void* wrctx, const mio_skad_t* dstaddr)
{
/* won't be invoked */
return 0;
}
static int tcp_sck_on_read (mio_dev_sck_t* tcp, const void* buf, mio_iolen_t len, const mio_skad_t* srcaddr)
{
/* won't be invoked */
return 0;
}
static int add_listener (mio_t* mio, mio_bch_t* addrstr)
{
mio_dev_sck_make_t mi;
mio_dev_sck_t* tcp;
mio_dev_sck_bind_t bi;
mio_dev_sck_listen_t li;
memset (&bi, 0, MIO_SIZEOF(bi));
if (mio_bcstrtoskad(mio, addrstr, &bi.localaddr) <= -1)
{
MIO_INFO1 (mio, "invalid listening address - %hs\n", addrstr);
return -1;
}
bi.options = MIO_DEV_SCK_BIND_REUSEADDR /*| MIO_DEV_SCK_BIND_REUSEPORT |*/;
#if defined(USE_SSL)
bi.options |= MIO_DEV_SCK_BIND_SSL;
bi.ssl_certfile = "localhost.crt";
bi.ssl_keyfile = "localhost.key";
#endif
memset (&mi, 0, MIO_SIZEOF(mi));
mi.type = (mio_skad_family(&bi.localaddr) == MIO_AF_INET? MIO_DEV_SCK_TCP4: MIO_DEV_SCK_TCP6);
mi.on_write = tcp_sck_on_write;
mi.on_read = tcp_sck_on_read;
mi.on_connect = tcp_sck_on_connect; /* this is invoked on a client accept as well */
mi.on_disconnect = tcp_sck_on_disconnect;
mi.on_raw_accept = tcp_sck_on_raw_accept;
tcp = mio_dev_sck_make(mio, 0, &mi);
if (!tcp)
{
MIO_INFO1 (mio, "Cannot make tcp - %js\n", mio_geterrmsg(mio));
return -1;
}
if (!g_reuse_port)
{
if (mio_dev_sck_bind(tcp, &bi) <= -1)
{
MIO_INFO1 (mio, "tcp mio_dev_sck_bind() failed - %js\n", mio_geterrmsg(mio));
return -1;
}
}
memset (&li, 0, MIO_SIZEOF(li));
li.backlogs = 4096;
MIO_INIT_NTIME (&li.accept_tmout, 5, 1);
if (mio_dev_sck_listen(tcp, &li) <= -1)
{
MIO_INFO1 (mio, "tcp[2] mio_dev_sck_listen() failed - %js\n", mio_geterrmsg(mio));
return -1;
}
return 0;
}
int main (int argc, char* argv[])
{
mio_t* mio = MIO_NULL;
pthread_t t[NUM_THRS];
mio_oow_t i;
struct sigaction sigact;
if (argc >= 2 && strcmp(argv[1], "-r") == 0)
{
g_reuse_port = 1;
}
memset (&sigact, 0, MIO_SIZEOF(sigact));
sigact.sa_handler = SIG_IGN;
sigaction (SIGPIPE, &sigact, MIO_NULL);
XX_MQ_INIT (&xx_mq);
xx_tmridx = MIO_TMRIDX_INVALID;
mio = mio_open(MIO_NULL, 0, MIO_NULL, 512, MIO_NULL);
if (!mio)
{
printf ("Cannot open mio\n");
goto oops;
}
for (i = 0; i < MIO_COUNTOF(t); i++)
pthread_create (&t[i], MIO_NULL, thr_func, mio);
sleep (1); /* TODO: use pthread_cond_wait()/pthread_cond_signal() or a varialble to see if all threads are up */
/* TODO: wait until all threads are ready to serve... */
if (add_listener(mio, "[::]:9987") <= -1 ||
add_listener(mio, "0.0.0.0:9987") <= -1) goto oops;
printf ("starting the main loop\n");
mio_loop (mio);
/* close all threaded mios here */
printf ("TERMINATING..NORMALLY \n");
mio_close (mio);
return 0;
oops:
printf ("TERMINATING..ABNORMALLY \n");
if (mio) mio_close (mio);
return -1;
}

View File

@ -81,10 +81,8 @@ struct thr_state_t
mio_dev_sck_on_write_t client_org_on_write; mio_dev_sck_on_write_t client_org_on_write;
mio_dev_sck_on_disconnect_t client_org_on_disconnect; mio_dev_sck_on_disconnect_t client_org_on_disconnect;
mio_htrd_recbs_t client_htrd_org_recbs; mio_htrd_recbs_t client_htrd_org_recbs;
}; };
typedef struct thr_state_t thr_state_t; typedef struct thr_state_t thr_state_t;
struct thr_peer_xtn_t struct thr_peer_xtn_t
@ -100,7 +98,6 @@ static void thr_state_halt_participating_devices (thr_state_t* thr_state)
MIO_DEBUG4 (thr_state->client->htts->mio, "HTTS(%p) - Halting participating devices in thr state %p(client=%p,peer=%p)\n", thr_state->client->htts, thr_state, thr_state->client->sck, thr_state->peer); MIO_DEBUG4 (thr_state->client->htts->mio, "HTTS(%p) - Halting participating devices in thr state %p(client=%p,peer=%p)\n", thr_state->client->htts, thr_state, thr_state->client->sck, thr_state->peer);
mio_dev_sck_halt (thr_state->client->sck); mio_dev_sck_halt (thr_state->client->sck);
/* check for peer as it may not have been started */ /* check for peer as it may not have been started */
if (thr_state->peer) mio_dev_thr_halt (thr_state->peer); if (thr_state->peer) mio_dev_thr_halt (thr_state->peer);

View File

@ -58,8 +58,19 @@ static int dev_pipe_make_master (mio_dev_t* dev, void* ctx)
/* TODO: support a named pipe. use mkfifo()? /* TODO: support a named pipe. use mkfifo()?
* support socketpair */ * support socketpair */
#if defined(HAVE_PIPE2) && defined(O_CLOEXEC) && defined(O_NONBLOCK)
if (pipe2(pfds, O_CLOEXEC | O_NONBLOCK) == -1)
{
if (errno != ENOSYS) goto pipe_error;
}
else goto pipe_done;
#endif
if (pipe(pfds) == -1) if (pipe(pfds) == -1)
{ {
#if defined(HAVE_PIPE2) && defined(O_CLOEXEC) && defined(O_NONBLOCK)
pipe_error:
#endif
mio_seterrwithsyserr (mio, 0, errno); mio_seterrwithsyserr (mio, 0, errno);
goto oops; goto oops;
} }
@ -67,11 +78,17 @@ static int dev_pipe_make_master (mio_dev_t* dev, void* ctx)
if (mio_makesyshndasync(mio, pfds[0]) <= -1 || if (mio_makesyshndasync(mio, pfds[0]) <= -1 ||
mio_makesyshndasync(mio, pfds[1]) <= -1) goto oops; mio_makesyshndasync(mio, pfds[1]) <= -1) goto oops;
if (mio_makesyshndcloexec(mio, pfds[0]) <= -1 ||
mio_makesyshndcloexec(mio, pfds[1]) <= -1) goto oops;
#if defined(HAVE_PIPE2) && defined(O_CLOEXEC) && defined(O_NONBLOCK)
pipe_done:
#endif
si.mi = info; si.mi = info;
si.pfd = pfds[0]; si.pfd = pfds[0];
si.dev_cap = MIO_DEV_CAP_IN | MIO_DEV_CAP_STREAM; si.dev_cap = MIO_DEV_CAP_IN | MIO_DEV_CAP_STREAM;
si.id = MIO_DEV_PIPE_IN; si.id = MIO_DEV_PIPE_IN;
pfds[0] = MIO_SYSHND_INVALID;
rdev->slave[MIO_DEV_PIPE_IN] = make_slave(mio, &si); rdev->slave[MIO_DEV_PIPE_IN] = make_slave(mio, &si);
if (!rdev->slave[MIO_DEV_PIPE_IN]) goto oops; if (!rdev->slave[MIO_DEV_PIPE_IN]) goto oops;
rdev->slave_count++; rdev->slave_count++;
@ -80,7 +97,7 @@ static int dev_pipe_make_master (mio_dev_t* dev, void* ctx)
si.pfd = pfds[1]; si.pfd = pfds[1];
si.dev_cap = MIO_DEV_CAP_OUT | MIO_DEV_CAP_STREAM; si.dev_cap = MIO_DEV_CAP_OUT | MIO_DEV_CAP_STREAM;
si.id = MIO_DEV_PIPE_OUT; si.id = MIO_DEV_PIPE_OUT;
pfds[1] = MIO_SYSHND_INVALID;
rdev->slave[MIO_DEV_PIPE_OUT] = make_slave(mio, &si); rdev->slave[MIO_DEV_PIPE_OUT] = make_slave(mio, &si);
if (!rdev->slave[MIO_DEV_PIPE_OUT]) goto oops; if (!rdev->slave[MIO_DEV_PIPE_OUT]) goto oops;
rdev->slave_count++; rdev->slave_count++;
@ -97,20 +114,21 @@ static int dev_pipe_make_master (mio_dev_t* dev, void* ctx)
return 0; return 0;
oops: oops:
for (i = 0; i < MIO_COUNTOF(rdev->slave); i++) if (pfds[0] != MIO_SYSHND_INVALID) close (pfds[0]);
{ if (pfds[1] != MIO_SYSHND_INVALID) close (pfds[0]);
if (rdev->slave[i])
{
mio_dev_kill ((mio_dev_t*)rdev->slave[i]);
rdev->slave[i] = MIO_NULL;
}
else if (pfds[i] != MIO_SYSHND_INVALID)
{
close (pfds[i]);
}
}
rdev->slave_count = 0;
if (rdev->slave[0])
{
mio_dev_kill ((mio_dev_t*)rdev->slave[0]);
rdev->slave[0] = MIO_NULL;
}
if (rdev->slave[1])
{
mio_dev_kill ((mio_dev_t*)rdev->slave[1]);
rdev->slave[1] = MIO_NULL;
}
rdev->slave_count = 0;
return -1; return -1;
} }
@ -204,6 +222,12 @@ static int dev_pipe_kill_slave (mio_dev_t* dev, int force)
return 0; return 0;
} }
static void dev_pipe_fail_before_make_slave (void* ctx)
{
slave_info_t* si = (slave_info_t*)ctx;
close (si->pfd);
}
static int dev_pipe_read_slave (mio_dev_t* dev, void* buf, mio_iolen_t* len, mio_devaddr_t* srcaddr) static int dev_pipe_read_slave (mio_dev_t* dev, void* buf, mio_iolen_t* len, mio_devaddr_t* srcaddr)
{ {
mio_dev_pipe_slave_t* pipe = (mio_dev_pipe_slave_t*)dev; mio_dev_pipe_slave_t* pipe = (mio_dev_pipe_slave_t*)dev;
@ -242,7 +266,7 @@ static int dev_pipe_write_slave (mio_dev_t* dev, const void* data, mio_iolen_t*
if (MIO_UNLIKELY(*len <= 0)) if (MIO_UNLIKELY(*len <= 0))
{ {
/* this is an EOF indicator */ /* this is an EOF indicator */
//mio_dev_halt (dev); /* halt this slave device to indicate EOF on the lower-level handle */* /*mio_dev_halt (dev);*/ /* halt this slave device to indicate EOF on the lower-level handle */
if (MIO_LIKELY(pipe->pfd != MIO_SYSHND_INVALID)) /* halt() doesn't close the pipe immediately. so close the underlying pipe */ if (MIO_LIKELY(pipe->pfd != MIO_SYSHND_INVALID)) /* halt() doesn't close the pipe immediately. so close the underlying pipe */
{ {
mio_dev_watch (dev, MIO_DEV_WATCH_STOP, 0); mio_dev_watch (dev, MIO_DEV_WATCH_STOP, 0);
@ -363,7 +387,7 @@ static mio_dev_mth_t dev_pipe_methods_slave =
{ {
dev_pipe_make_slave, dev_pipe_make_slave,
dev_pipe_kill_slave, dev_pipe_kill_slave,
MIO_NULL, dev_pipe_fail_before_make_slave,
dev_pipe_getsyshnd_slave, dev_pipe_getsyshnd_slave,
dev_pipe_read_slave, dev_pipe_read_slave,

View File

@ -346,10 +346,10 @@ static int dev_pro_make_master (mio_dev_t* dev, void* ctx)
si.dev_cap = MIO_DEV_CAP_OUT | MIO_DEV_CAP_STREAM; si.dev_cap = MIO_DEV_CAP_OUT | MIO_DEV_CAP_STREAM;
si.id = MIO_DEV_PRO_IN; si.id = MIO_DEV_PRO_IN;
pfds[1] = MIO_SYSHND_INVALID;
rdev->slave[MIO_DEV_PRO_IN] = make_slave(mio, &si); rdev->slave[MIO_DEV_PRO_IN] = make_slave(mio, &si);
if (!rdev->slave[MIO_DEV_PRO_IN]) goto oops; if (!rdev->slave[MIO_DEV_PRO_IN]) goto oops;
pfds[1] = MIO_SYSHND_INVALID;
rdev->slave_count++; rdev->slave_count++;
} }
@ -363,10 +363,10 @@ static int dev_pro_make_master (mio_dev_t* dev, void* ctx)
si.dev_cap = MIO_DEV_CAP_IN | MIO_DEV_CAP_STREAM; si.dev_cap = MIO_DEV_CAP_IN | MIO_DEV_CAP_STREAM;
si.id = MIO_DEV_PRO_OUT; si.id = MIO_DEV_PRO_OUT;
pfds[2] = MIO_SYSHND_INVALID;
rdev->slave[MIO_DEV_PRO_OUT] = make_slave(mio, &si); rdev->slave[MIO_DEV_PRO_OUT] = make_slave(mio, &si);
if (!rdev->slave[MIO_DEV_PRO_OUT]) goto oops; if (!rdev->slave[MIO_DEV_PRO_OUT]) goto oops;
pfds[2] = MIO_SYSHND_INVALID;
rdev->slave_count++; rdev->slave_count++;
} }
@ -380,10 +380,10 @@ static int dev_pro_make_master (mio_dev_t* dev, void* ctx)
si.dev_cap = MIO_DEV_CAP_IN | MIO_DEV_CAP_STREAM; si.dev_cap = MIO_DEV_CAP_IN | MIO_DEV_CAP_STREAM;
si.id = MIO_DEV_PRO_ERR; si.id = MIO_DEV_PRO_ERR;
pfds[4] = MIO_SYSHND_INVALID;
rdev->slave[MIO_DEV_PRO_ERR] = make_slave(mio, &si); rdev->slave[MIO_DEV_PRO_ERR] = make_slave(mio, &si);
if (!rdev->slave[MIO_DEV_PRO_ERR]) goto oops; if (!rdev->slave[MIO_DEV_PRO_ERR]) goto oops;
pfds[4] = MIO_SYSHND_INVALID;
rdev->slave_count++; rdev->slave_count++;
} }
@ -425,20 +425,6 @@ oops:
return -1; return -1;
} }
static int dev_pro_make_slave (mio_dev_t* dev, void* ctx)
{
mio_dev_pro_slave_t* rdev = (mio_dev_pro_slave_t*)dev;
slave_info_t* si = (slave_info_t*)ctx;
rdev->dev_cap = si->dev_cap;
rdev->id = si->id;
rdev->pfd = si->pfd;
/* keep rdev->master to MIO_NULL. it's set to the right master
* device in dev_pro_make() */
return 0;
}
static int dev_pro_kill_master (mio_dev_t* dev, int force) static int dev_pro_kill_master (mio_dev_t* dev, int force)
{ {
mio_t* mio = dev->mio; mio_t* mio = dev->mio;
@ -507,6 +493,20 @@ static int dev_pro_kill_master (mio_dev_t* dev, int force)
return 0; return 0;
} }
static int dev_pro_make_slave (mio_dev_t* dev, void* ctx)
{
mio_dev_pro_slave_t* rdev = (mio_dev_pro_slave_t*)dev;
slave_info_t* si = (slave_info_t*)ctx;
rdev->dev_cap = si->dev_cap;
rdev->id = si->id;
rdev->pfd = si->pfd;
/* keep rdev->master to MIO_NULL. it's set to the right master
* device in dev_pro_make() */
return 0;
}
static int dev_pro_kill_slave (mio_dev_t* dev, int force) static int dev_pro_kill_slave (mio_dev_t* dev, int force)
{ {
mio_t* mio = dev->mio; mio_t* mio = dev->mio;
@ -554,6 +554,12 @@ static int dev_pro_kill_slave (mio_dev_t* dev, int force)
return 0; return 0;
} }
static void dev_pro_fail_before_make_slave (void* ctx)
{
slave_info_t* si = (slave_info_t*)ctx;
close (si->pfd);
}
static int dev_pro_read_slave (mio_dev_t* dev, void* buf, mio_iolen_t* len, mio_devaddr_t* srcaddr) static int dev_pro_read_slave (mio_dev_t* dev, void* buf, mio_iolen_t* len, mio_devaddr_t* srcaddr)
{ {
mio_dev_pro_slave_t* pro = (mio_dev_pro_slave_t*)dev; mio_dev_pro_slave_t* pro = (mio_dev_pro_slave_t*)dev;

View File

@ -381,8 +381,14 @@ static int dev_sck_make (mio_dev_t* dev, void* ctx)
return 0; return 0;
oops: oops:
if (hnd != MIO_SYSHND_INVALID) close (hnd); if (hnd != MIO_SYSHND_INVALID)
if (side_chan != MIO_SYSHND_INVALID) close (side_chan); {
close (hnd);
}
if (side_chan != MIO_SYSHND_INVALID)
{
close (side_chan);
}
return -1; return -1;
} }

View File

@ -32,6 +32,7 @@
#include <sys/uio.h> #include <sys/uio.h>
#include <pthread.h> #include <pthread.h>
#include <stdio.h>
/* ========================================================================= */ /* ========================================================================= */
struct mio_dev_thr_info_t struct mio_dev_thr_info_t
@ -64,13 +65,21 @@ static void free_thr_info_resources (mio_t* mio, mio_dev_thr_info_t* ti)
{ {
if (ti->thr_iop.rfd != MIO_SYSHND_INVALID) if (ti->thr_iop.rfd != MIO_SYSHND_INVALID)
{ {
close (ti->thr_iop.rfd); /* this function is called at the end of run_thr_func() and
* close() can be a thread cancellation point.
*
* i must invalidate ti->thr_iop.rfd calling close() with it.
* if resetting is done after close() and close() becomes a cancellation point,
* the invalidation operation gets skipped. */
mio_syshnd_t tmp = ti->thr_iop.rfd;
ti->thr_iop.rfd = MIO_SYSHND_INVALID; ti->thr_iop.rfd = MIO_SYSHND_INVALID;
close (tmp);
} }
if (ti->thr_iop.wfd != MIO_SYSHND_INVALID) if (ti->thr_iop.wfd != MIO_SYSHND_INVALID)
{ {
close (ti->thr_iop.wfd); mio_syshnd_t tmp = ti->thr_iop.wfd;
ti->thr_iop.wfd = MIO_SYSHND_INVALID; ti->thr_iop.wfd = MIO_SYSHND_INVALID;
close (tmp);
} }
} }
@ -119,9 +128,7 @@ static void* run_thr_func (void* ctx)
ti->thr_func (ti->mio, &ti->thr_iop, ti->thr_ctx); ti->thr_func (ti->mio, &ti->thr_iop, ti->thr_ctx);
/* This part may get partially executed or not executed if the thread is cancelled */ free_thr_info_resources (ti->mio, ti);
free_thr_info_resources (ti->mio, ti); /* TODO: check if the close() call inside this call completes when it becomes a cancellation point. if so, the code must get changed */
/* ---------------------------------------------------------- */
pthread_cleanup_pop (1); pthread_cleanup_pop (1);
pthread_exit (MIO_NULL); pthread_exit (MIO_NULL);
@ -137,8 +144,20 @@ static int dev_thr_make_master (mio_dev_t* dev, void* ctx)
slave_info_t si; slave_info_t si;
int i; int i;
#if defined(HAVE_PIPE2) && defined(O_CLOEXEC) && defined(O_NONBLOCK)
if (pipe2(&pfds[0], O_CLOEXEC | O_NONBLOCK) == -1 ||
pipe2(&pfds[2], O_CLOEXEC | O_NONBLOCK) == -1)
{
if (errno != ENOSYS) goto pipe_error;
}
else goto pipe_done;
#endif
if (pipe(&pfds[0]) == -1 || pipe(&pfds[2]) == -1) if (pipe(&pfds[0]) == -1 || pipe(&pfds[2]) == -1)
{ {
#if defined(HAVE_PIPE2) && defined(O_CLOEXEC) && defined(O_NONBLOCK)
pipe_error:
#endif
mio_seterrwithsyserr (mio, 0, errno); mio_seterrwithsyserr (mio, 0, errno);
goto oops; goto oops;
} }
@ -146,26 +165,36 @@ static int dev_thr_make_master (mio_dev_t* dev, void* ctx)
if (mio_makesyshndasync(mio, pfds[1]) <= -1 || if (mio_makesyshndasync(mio, pfds[1]) <= -1 ||
mio_makesyshndasync(mio, pfds[2]) <= -1) goto oops; mio_makesyshndasync(mio, pfds[2]) <= -1) goto oops;
if (mio_makesyshndcloexec(mio, pfds[0]) <= -1 ||
mio_makesyshndcloexec(mio, pfds[1]) <= -1 ||
mio_makesyshndcloexec(mio, pfds[2]) <= -1 ||
mio_makesyshndcloexec(mio, pfds[1]) <= -1) goto oops;
#if defined(HAVE_PIPE2) && defined(O_CLOEXEC) && defined(O_NONBLOCK)
pipe_done:
#endif
si.mi = info; si.mi = info;
si.pfd = pfds[1]; si.pfd = pfds[1];
si.dev_cap = MIO_DEV_CAP_OUT | MIO_DEV_CAP_STREAM; si.dev_cap = MIO_DEV_CAP_OUT | MIO_DEV_CAP_STREAM;
si.id = MIO_DEV_THR_IN; si.id = MIO_DEV_THR_IN;
/* invalidate pfds[1] before calling make_slave() because when it fails, the
* fail_before_make(dev_thr_fail_before_make_slave) and kill(dev_thr_kill_slave) callbacks close si.pfd */
pfds[1] = MIO_SYSHND_INVALID;
rdev->slave[MIO_DEV_THR_IN] = make_slave(mio, &si); rdev->slave[MIO_DEV_THR_IN] = make_slave(mio, &si);
if (!rdev->slave[MIO_DEV_THR_IN]) goto oops; if (!rdev->slave[MIO_DEV_THR_IN]) goto oops;
pfds[1] = MIO_SYSHND_INVALID;
rdev->slave_count++; rdev->slave_count++;
si.mi = info; si.mi = info;
si.pfd = pfds[2]; si.pfd = pfds[2];
si.dev_cap = MIO_DEV_CAP_IN | MIO_DEV_CAP_STREAM; si.dev_cap = MIO_DEV_CAP_IN | MIO_DEV_CAP_STREAM;
si.id = MIO_DEV_THR_OUT; si.id = MIO_DEV_THR_OUT;
/* invalidate pfds[2] before calling make_slave() because when it fails, the
* fail_before_make(dev_thr_fail_before_make_slave) and kill(dev_thr_kill_slave) callbacks close si.pfd */
pfds[2] = MIO_SYSHND_INVALID;
rdev->slave[MIO_DEV_THR_OUT] = make_slave(mio, &si); rdev->slave[MIO_DEV_THR_OUT] = make_slave(mio, &si);
if (!rdev->slave[MIO_DEV_THR_OUT]) goto oops; if (!rdev->slave[MIO_DEV_THR_OUT]) goto oops;
pfds[2] = MIO_SYSHND_INVALID;
rdev->slave_count++; rdev->slave_count++;
for (i = 0; i < MIO_COUNTOF(rdev->slave); i++) for (i = 0; i < MIO_COUNTOF(rdev->slave); i++)
@ -200,6 +229,10 @@ static int dev_thr_make_master (mio_dev_t* dev, void* ctx)
mio_freemem (mio, ti); mio_freemem (mio, ti);
goto oops; goto oops;
} }
/* the thread function is in charge of these two file descriptors */
pfds[0] = MIO_SYSHND_INVALID;
pfds[3] = MIO_SYSHND_INVALID;
} }
/* ---------------------------------------------------------- */ /* ---------------------------------------------------------- */
@ -209,7 +242,10 @@ static int dev_thr_make_master (mio_dev_t* dev, void* ctx)
oops: oops:
for (i = 0; i < MIO_COUNTOF(pfds); i++) for (i = 0; i < MIO_COUNTOF(pfds); i++)
{ {
if (pfds[i] != MIO_SYSHND_INVALID) close (pfds[i]); if (pfds[i] != MIO_SYSHND_INVALID)
{
close (pfds[i]);
}
} }
for (i = MIO_COUNTOF(rdev->slave); i > 0; ) for (i = MIO_COUNTOF(rdev->slave); i > 0; )
@ -248,7 +284,10 @@ static int dev_thr_kill_master (mio_dev_t* dev, int force)
int i; int i;
ti = rdev->thr_info; ti = rdev->thr_info;
pthread_cancel (ti->thr_hnd); /* pthread_cancel() seems to create some dangling file descriptors not closed properly.
* i don't seem to get it working correctly as of now. proper cancellation point management
* is very difficult. without pthread_cancel() here, higher pressure on cfmb is expected */
/*pthread_cancel (ti->thr_hnd); */
if (rdev->slave_count > 0) if (rdev->slave_count > 0)
{ {
@ -269,17 +308,26 @@ static int dev_thr_kill_master (mio_dev_t* dev, int force)
} }
} }
rdev->thr_info = MIO_NULL;
if (ti->thr_done) if (ti->thr_done)
{ {
pthread_detach (ti->thr_hnd); /* pthread_join() may be blocking. */ pthread_detach (ti->thr_hnd); /* pthread_join() may be blocking. detach the thread instead */
free_thr_info_resources (mio, ti); free_thr_info_resources (mio, ti);
mio_freemem (mio, ti); mio_freemem (mio, ti);
} }
else else
{ {
#if 0
/* since pthread_join can be blocking, i'd schedule a resource destroyer with mio_addcfmb().
* see after #else */
pthread_join (ti->thr_hnd, MIO_NULL);
free_thr_info_resources (mio, ti);
mio_freemem (mio, ti);
#else
/* schedule a resource destroyer */
mio_addcfmb (mio, ti, ready_to_free_thr_info); mio_addcfmb (mio, ti, ready_to_free_thr_info);
#endif
} }
rdev->thr_info = MIO_NULL;
if (rdev->on_close) rdev->on_close (rdev, MIO_DEV_THR_MASTER); if (rdev->on_close) rdev->on_close (rdev, MIO_DEV_THR_MASTER);
return 0; return 0;
@ -332,6 +380,14 @@ static int dev_thr_kill_slave (mio_dev_t* dev, int force)
return 0; return 0;
} }
static void dev_thr_fail_before_make_slave (void* ctx)
{
slave_info_t* si = (slave_info_t*)ctx;
/* mio_dev_make() failed before it called the make() callback.
* i will close the pipe fd here instead of in the caller of mio_dev_make() */
close (si->pfd);
}
static int dev_thr_read_slave (mio_dev_t* dev, void* buf, mio_iolen_t* len, mio_devaddr_t* srcaddr) static int dev_thr_read_slave (mio_dev_t* dev, void* buf, mio_iolen_t* len, mio_devaddr_t* srcaddr)
{ {
mio_dev_thr_slave_t* thr = (mio_dev_thr_slave_t*)dev; mio_dev_thr_slave_t* thr = (mio_dev_thr_slave_t*)dev;
@ -377,7 +433,7 @@ static int dev_thr_write_slave (mio_dev_t* dev, const void* data, mio_iolen_t* l
if (MIO_UNLIKELY(*len <= 0)) if (MIO_UNLIKELY(*len <= 0))
{ {
/* this is an EOF indicator */ /* this is an EOF indicator */
/* It isn't apthrpriate to call mio_dev_halt(thr) or mio_dev_thr_close(thr->master, MIO_DEV_THR_IN) /* It isn't appropriate to call mio_dev_halt(thr) or mio_dev_thr_close(thr->master, MIO_DEV_THR_IN)
* as those functions destroy the device itself */ * as those functions destroy the device itself */
if (MIO_LIKELY(thr->pfd != MIO_SYSHND_INVALID)) if (MIO_LIKELY(thr->pfd != MIO_SYSHND_INVALID))
{ {
@ -523,7 +579,7 @@ static mio_dev_mth_t dev_thr_methods_slave =
{ {
dev_thr_make_slave, dev_thr_make_slave,
dev_thr_kill_slave, dev_thr_kill_slave,
MIO_NULL, dev_thr_fail_before_make_slave,
dev_thr_getsyshnd_slave, dev_thr_getsyshnd_slave,
dev_thr_read_slave, dev_thr_read_slave,