Skip to content

Instantly share code, notes, and snippets.

@Wowfunhappy

Wowfunhappy/ipc_mqueue

Last active March 5, 2022 01:31
Show Gist options
  • Star 0 You must be signed in to star a gist
  • Fork 0 You must be signed in to fork a gist
  • Save Wowfunhappy/a214411bb52d39bcee9497363a97558d to your computer and use it in GitHub Desktop.
Save Wowfunhappy/a214411bb52d39bcee9497363a97558d to your computer and use it in GitHub Desktop.
Download ZIP
Modified version of kern_event.c to stop XNU 2422.115.4 to panic when running Chromium Legacy. See: https://github.com/blueboxd/chromium-legacy/issues/44
Raw
ipc_mqueue
/*
* Copyright (c) 2000-2007 Apple Inc. All rights reserved.
*
* @APPLE_OSREFERENCE_LICENSE_HEADER_START@
*
* This file contains Original Code and/or Modifications of Original Code
* as defined in and that are subject to the Apple Public Source License
* Version 2.0 (the 'License'). You may not use this file except in
* compliance with the License. The rights granted to you under the License
* may not be used to create, or enable the creation or redistribution of,
* unlawful or unlicensed copies of an Apple operating system, or to
* circumvent, violate, or enable the circumvention or violation of, any
* terms of an Apple operating system software license agreement.
*
* Please obtain a copy of the License at
* http://www.opensource.apple.com/apsl/ and read it before using this file.
*
* The Original Code and all software distributed under the License are
* distributed on an 'AS IS' basis, WITHOUT WARRANTY OF ANY KIND, EITHER
* EXPRESS OR IMPLIED, AND APPLE HEREBY DISCLAIMS ALL SUCH WARRANTIES,
* INCLUDING WITHOUT LIMITATION, ANY WARRANTIES OF MERCHANTABILITY,
* FITNESS FOR A PARTICULAR PURPOSE, QUIET ENJOYMENT OR NON-INFRINGEMENT.
* Please see the License for the specific language governing rights and
* limitations under the License.
*
* @APPLE_OSREFERENCE_LICENSE_HEADER_END@
*/
/*
* @OSF_FREE_COPYRIGHT@
*/
/*
* Mach Operating System
* Copyright (c) 1991,1990,1989 Carnegie Mellon University
* All Rights Reserved.
*
* Permission to use, copy, modify and distribute this software and its
* documentation is hereby granted, provided that both the copyright
* notice and this permission notice appear in all copies of the
* software, derivative works or modified versions, and any portions
* thereof, and that both notices appear in supporting documentation.
*
* CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS"
* CONDITION. CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND FOR
* ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE.
*
* Carnegie Mellon requests users of this software to return to
*
* Software Distribution Coordinator or Software.Distribution@CS.CMU.EDU
* School of Computer Science
* Carnegie Mellon University
* Pittsburgh PA 15213-3890
*
* any improvements or extensions that they make and grant Carnegie Mellon
* the rights to redistribute these changes.
*/
/*
*/
/*
* File: ipc/ipc_mqueue.c
* Author: Rich Draves
* Date: 1989
*
* Functions to manipulate IPC message queues.
*/
/*
* NOTICE: This file was modified by SPARTA, Inc. in 2006 to introduce
* support for mandatory and extensible security protections. This notice
* is included in support of clause 2.2 (b) of the Apple Public License,
* Version 2.0.
*/
#include <mach/port.h>
#include <mach/message.h>
#include <mach/sync_policy.h>
#include <kern/assert.h>
#include <kern/counters.h>
#include <kern/sched_prim.h>
#include <kern/ipc_kobject.h>
#include <kern/ipc_mig.h> /* XXX - for mach_msg_receive_continue */
#include <kern/misc_protos.h>
#include <kern/task.h>
#include <kern/thread.h>
#include <kern/wait_queue.h>
#include <ipc/ipc_mqueue.h>
#include <ipc/ipc_kmsg.h>
#include <ipc/ipc_port.h>
#include <ipc/ipc_pset.h>
#include <ipc/ipc_space.h>
#ifdef __LP64__
#include <vm/vm_map.h>
#endif
#if CONFIG_MACF_MACH
#include <security/mac_mach_internal.h>
#endif
int ipc_mqueue_full; /* address is event for queue space */
int ipc_mqueue_rcv; /* address is event for message arrival */
/* forward declarations */
void ipc_mqueue_receive_results(wait_result_t result);
/*
* Routine: ipc_mqueue_init
* Purpose:
* Initialize a newly-allocated message queue.
*/
void
ipc_mqueue_init(
ipc_mqueue_t mqueue,
boolean_t is_set)
{
if (is_set) {
wait_queue_set_init(&mqueue->imq_set_queue, SYNC_POLICY_FIFO|SYNC_POLICY_PREPOST);
} else {
wait_queue_init(&mqueue->imq_wait_queue, SYNC_POLICY_FIFO);
ipc_kmsg_queue_init(&mqueue->imq_messages);
mqueue->imq_seqno = 0;
mqueue->imq_msgcount = 0;
mqueue->imq_qlimit = MACH_PORT_QLIMIT_DEFAULT;
mqueue->imq_fullwaiters = FALSE;
}
}
/*
* Routine: ipc_mqueue_member
* Purpose:
* Indicate whether the (port) mqueue is a member of
* this portset's mqueue. We do this by checking
* whether the portset mqueue's waitq is an member of
* the port's mqueue waitq.
* Conditions:
* the portset's mqueue is not already a member
* this may block while allocating linkage structures.
*/
boolean_t
ipc_mqueue_member(
ipc_mqueue_t port_mqueue,
ipc_mqueue_t set_mqueue)
{
wait_queue_t port_waitq = &port_mqueue->imq_wait_queue;
wait_queue_set_t set_waitq = &set_mqueue->imq_set_queue;
return (wait_queue_member(port_waitq, set_waitq));
}
/*
* Routine: ipc_mqueue_remove
* Purpose:
* Remove the association between the queue and the specified
* set message queue.
*/
kern_return_t
ipc_mqueue_remove(
ipc_mqueue_t mqueue,
ipc_mqueue_t set_mqueue,
wait_queue_link_t *wqlp)
{
wait_queue_t mq_waitq = &mqueue->imq_wait_queue;
wait_queue_set_t set_waitq = &set_mqueue->imq_set_queue;
return wait_queue_unlink_nofree(mq_waitq, set_waitq, wqlp);
}
/*
* Routine: ipc_mqueue_remove_from_all
* Purpose:
* Remove the mqueue from all the sets it is a member of
* Conditions:
* Nothing locked.
*/
void
ipc_mqueue_remove_from_all(
ipc_mqueue_t mqueue,
queue_t links)
{
wait_queue_t mq_waitq = &mqueue->imq_wait_queue;
wait_queue_unlink_all_nofree(mq_waitq, links);
return;
}
/*
* Routine: ipc_mqueue_remove_all
* Purpose:
* Remove all the member queues from the specified set.
* Conditions:
* Nothing locked.
*/
void
ipc_mqueue_remove_all(
ipc_mqueue_t mqueue,
queue_t links)
{
wait_queue_set_t mq_setq = &mqueue->imq_set_queue;
wait_queue_set_unlink_all_nofree(mq_setq, links);
return;
}
/*
* Routine: ipc_mqueue_add
* Purpose:
* Associate the portset's mqueue with the port's mqueue.
* This has to be done so that posting the port will wakeup
* a portset waiter. If there are waiters on the portset
* mqueue and messages on the port mqueue, try to match them
* up now.
* Conditions:
* May block.
*/
kern_return_t
ipc_mqueue_add(
ipc_mqueue_t port_mqueue,
ipc_mqueue_t set_mqueue,
wait_queue_link_t wql)
{
wait_queue_t port_waitq = &port_mqueue->imq_wait_queue;
wait_queue_set_t set_waitq = &set_mqueue->imq_set_queue;
ipc_kmsg_queue_t kmsgq;
ipc_kmsg_t kmsg, next;
kern_return_t kr;
spl_t s;
kr = wait_queue_link_noalloc(port_waitq, set_waitq, wql);
if (kr != KERN_SUCCESS)
return kr;
/*
* Now that the set has been added to the port, there may be
* messages queued on the port and threads waiting on the set
* waitq. Lets get them together.
*/
s = splsched();
imq_lock(port_mqueue);
kmsgq = &port_mqueue->imq_messages;
for (kmsg = ipc_kmsg_queue_first(kmsgq);
kmsg != IKM_NULL;
kmsg = next) {
next = ipc_kmsg_queue_next(kmsgq, kmsg);
for (;;) {
thread_t th;
mach_msg_size_t msize;
th = wait_queue_wakeup64_identity_locked(
port_waitq,
IPC_MQUEUE_RECEIVE,
THREAD_AWAKENED,
FALSE);
/* waitq/mqueue still locked, thread locked */
if (th == THREAD_NULL)
goto leave;
/*
* If the receiver waited with a facility not directly
* related to Mach messaging, then it isn't prepared to get
* handed the message directly. Just set it running, and
* go look for another thread that can.
*/
if (th->ith_state != MACH_RCV_IN_PROGRESS) {
thread_unlock(th);
continue;
}
/*
* Found a receiver. see if they can handle the message
* correctly (the message is not too large for them, or
* they didn't care to be informed that the message was
* too large). If they can't handle it, take them off
* the list and let them go back and figure it out and
* just move onto the next.
*/
msize = ipc_kmsg_copyout_size(kmsg, th->map);
if (th->ith_msize <
(msize + REQUESTED_TRAILER_SIZE(thread_is_64bit(th), th->ith_option))) {
th->ith_state = MACH_RCV_TOO_LARGE;
th->ith_msize = msize;
if (th->ith_option & MACH_RCV_LARGE) {
/*
* let him go without message
*/
th->ith_receiver_name = port_mqueue->imq_receiver_name;
th->ith_kmsg = IKM_NULL;
th->ith_seqno = 0;
thread_unlock(th);
continue; /* find another thread */
}
} else {
th->ith_state = MACH_MSG_SUCCESS;
}
/*
* This thread is going to take this message,
* so give it to him.
*/
ipc_kmsg_rmqueue(kmsgq, kmsg);
ipc_mqueue_release_msgcount(port_mqueue);
th->ith_kmsg = kmsg;
th->ith_seqno = port_mqueue->imq_seqno++;
thread_unlock(th);
break; /* go to next message */
}
}
leave:
imq_unlock(port_mqueue);
splx(s);
return KERN_SUCCESS;
}
/*
* Routine: ipc_mqueue_changed
* Purpose:
* Wake up receivers waiting in a message queue.
* Conditions:
* The message queue is locked.
*/
void
ipc_mqueue_changed(
ipc_mqueue_t mqueue)
{
printf("\nAbout to call wait_queue_wakeup64_all_locked from ipc_mqueue_changed\n");
wait_queue_wakeup64_all_locked(
&mqueue->imq_wait_queue,
IPC_MQUEUE_RECEIVE,
THREAD_RESTART,
FALSE); /* unlock waitq? */
}
/*
* Routine: ipc_mqueue_send
* Purpose:
* Send a message to a message queue. The message holds a reference
* for the destination port for this message queue in the
* msgh_remote_port field.
*
* If unsuccessful, the caller still has possession of
* the message and must do something with it. If successful,
* the message is queued, given to a receiver, or destroyed.
* Conditions:
* mqueue is locked.
* Returns:
* MACH_MSG_SUCCESS The message was accepted.
* MACH_SEND_TIMED_OUT Caller still has message.
* MACH_SEND_INTERRUPTED Caller still has message.
*/
mach_msg_return_t
ipc_mqueue_send(
ipc_mqueue_t mqueue,
ipc_kmsg_t kmsg,
mach_msg_option_t option,
mach_msg_timeout_t send_timeout,
spl_t s)
{
int wresult;
/*
* Don't block if:
* 1) We're under the queue limit.
* 2) Caller used the MACH_SEND_ALWAYS internal option.
* 3) Message is sent to a send-once right.
*/
if (!imq_full(mqueue) ||
(!imq_full_kernel(mqueue) &&
((option & MACH_SEND_ALWAYS) ||
(MACH_MSGH_BITS_REMOTE(kmsg->ikm_header->msgh_bits) ==
MACH_MSG_TYPE_PORT_SEND_ONCE)))) {
mqueue->imq_msgcount++;
assert(mqueue->imq_msgcount > 0);
imq_unlock(mqueue);
splx(s);
} else {
thread_t cur_thread = current_thread();
uint64_t deadline;
/*
* We have to wait for space to be granted to us.
*/
if ((option & MACH_SEND_TIMEOUT) && (send_timeout == 0)) {
imq_unlock(mqueue);
splx(s);
return MACH_SEND_TIMED_OUT;
}
if (imq_full_kernel(mqueue)) {
imq_unlock(mqueue);
splx(s);
return MACH_SEND_NO_BUFFER;
}
mqueue->imq_fullwaiters = TRUE;
thread_lock(cur_thread);
if (option & MACH_SEND_TIMEOUT)
clock_interval_to_deadline(send_timeout, 1000*NSEC_PER_USEC, &deadline);
else
deadline = 0;
wresult = wait_queue_assert_wait64_locked(
&mqueue->imq_wait_queue,
IPC_MQUEUE_FULL,
THREAD_ABORTSAFE,
TIMEOUT_URGENCY_USER_NORMAL,
deadline, 0,
cur_thread);
thread_unlock(cur_thread);
imq_unlock(mqueue);
splx(s);
if (wresult == THREAD_WAITING) {
wresult = thread_block(THREAD_CONTINUE_NULL);
counter(c_ipc_mqueue_send_block++);
}
switch (wresult) {
case THREAD_TIMED_OUT:
assert(option & MACH_SEND_TIMEOUT);
return MACH_SEND_TIMED_OUT;
case THREAD_AWAKENED:
/* we can proceed - inherited msgcount from waker */
assert(mqueue->imq_msgcount > 0);
break;
case THREAD_INTERRUPTED:
return MACH_SEND_INTERRUPTED;
case THREAD_RESTART:
/* mqueue is being destroyed */
return MACH_SEND_INVALID_DEST;
default:
panic("ipc_mqueue_send");
}
}
ipc_mqueue_post(mqueue, kmsg);
return MACH_MSG_SUCCESS;
}
/*
* Routine: ipc_mqueue_release_msgcount
* Purpose:
* Release a message queue reference in the case where we
* found a waiter.
*
* Conditions:
* The message queue is locked.
* The message corresponding to this reference is off the queue.
*/
void
ipc_mqueue_release_msgcount(
ipc_mqueue_t mqueue)
{
assert(imq_held(mqueue));
assert(mqueue->imq_msgcount > 1 || ipc_kmsg_queue_empty(&mqueue->imq_messages));
mqueue->imq_msgcount--;
if (!imq_full(mqueue) && mqueue->imq_fullwaiters) {
if (wait_queue_wakeup64_one_locked(
&mqueue->imq_wait_queue,
IPC_MQUEUE_FULL,
THREAD_AWAKENED,
FALSE) != KERN_SUCCESS) {
mqueue->imq_fullwaiters = FALSE;
} else {
/* gave away our slot - add reference back */
mqueue->imq_msgcount++;
}
}
}
/*
* Routine: ipc_mqueue_post
* Purpose:
* Post a message to a waiting receiver or enqueue it. If a
* receiver is waiting, we can release our reserved space in
* the message queue.
*
* Conditions:
* If we need to queue, our space in the message queue is reserved.
*/
void
ipc_mqueue_post(
register ipc_mqueue_t mqueue,
register ipc_kmsg_t kmsg)
{
spl_t s;
/*
* While the msg queue is locked, we have control of the
* kmsg, so the ref in it for the port is still good.
*
* Check for a receiver for the message.
*/
s = splsched();
imq_lock(mqueue);
for (;;) {
wait_queue_t waitq = &mqueue->imq_wait_queue;
thread_t receiver;
mach_msg_size_t msize;
receiver = wait_queue_wakeup64_identity_locked(
waitq,
IPC_MQUEUE_RECEIVE,
THREAD_AWAKENED,
FALSE);
/* waitq still locked, thread locked */
if (receiver == THREAD_NULL) {
/*
* no receivers; queue kmsg
*/
assert(mqueue->imq_msgcount > 0);
ipc_kmsg_enqueue_macro(&mqueue->imq_messages, kmsg);
break;
}
/*
* If the receiver waited with a facility not directly
* related to Mach messaging, then it isn't prepared to get
* handed the message directly. Just set it running, and
* go look for another thread that can.
*/
if (receiver->ith_state != MACH_RCV_IN_PROGRESS) {
thread_unlock(receiver);
continue;
}
/*
* We found a waiting thread.
* If the message is too large or the scatter list is too small
* the thread we wake up will get that as its status.
*/
msize = ipc_kmsg_copyout_size(kmsg, receiver->map);
if (receiver->ith_msize <
(msize + REQUESTED_TRAILER_SIZE(thread_is_64bit(receiver), receiver->ith_option))) {
receiver->ith_msize = msize;
receiver->ith_state = MACH_RCV_TOO_LARGE;
} else {
receiver->ith_state = MACH_MSG_SUCCESS;
}
/*
* If there is no problem with the upcoming receive, or the
* receiver thread didn't specifically ask for special too
* large error condition, go ahead and select it anyway.
*/
if ((receiver->ith_state == MACH_MSG_SUCCESS) ||
!(receiver->ith_option & MACH_RCV_LARGE)) {
receiver->ith_kmsg = kmsg;
receiver->ith_seqno = mqueue->imq_seqno++;
thread_unlock(receiver);
/* we didn't need our reserved spot in the queue */
ipc_mqueue_release_msgcount(mqueue);
break;
}
/*
* Otherwise, this thread needs to be released to run
* and handle its error without getting the message. We
* need to go back and pick another one.
*/
receiver->ith_receiver_name = mqueue->imq_receiver_name;
receiver->ith_kmsg = IKM_NULL;
receiver->ith_seqno = 0;
thread_unlock(receiver);
}
imq_unlock(mqueue);
splx(s);
current_task()->messages_sent++;
return;
}
/* static */ void
ipc_mqueue_receive_results(wait_result_t saved_wait_result)
{
thread_t self = current_thread();
mach_msg_option_t option = self->ith_option;
/*
* why did we wake up?
*/
switch (saved_wait_result) {
case THREAD_TIMED_OUT:
self->ith_state = MACH_RCV_TIMED_OUT;
return;
case THREAD_INTERRUPTED:
self->ith_state = MACH_RCV_INTERRUPTED;
return;
case THREAD_RESTART:
/* something bad happened to the port/set */
self->ith_state = MACH_RCV_PORT_CHANGED;
return;
case THREAD_AWAKENED:
/*
* We do not need to go select a message, somebody
* handed us one (or a too-large indication).
*/
switch (self->ith_state) {
case MACH_RCV_SCATTER_SMALL:
case MACH_RCV_TOO_LARGE:
/*
* Somebody tried to give us a too large
* message. If we indicated that we cared,
* then they only gave us the indication,
* otherwise they gave us the indication
* AND the message anyway.
*/
if (option & MACH_RCV_LARGE) {
return;
}
case MACH_MSG_SUCCESS:
return;
default:
panic("ipc_mqueue_receive_results: strange ith_state");
}
default:
panic("ipc_mqueue_receive_results: strange wait_result");
}
}
void
ipc_mqueue_receive_continue(
__unused void *param,
wait_result_t wresult)
{
ipc_mqueue_receive_results(wresult);
mach_msg_receive_continue(); /* hard-coded for now */
}
/*
* Routine: ipc_mqueue_receive
* Purpose:
* Receive a message from a message queue.
*
* If continuation is non-zero, then we might discard
* our kernel stack when we block. We will continue
* after unblocking by executing continuation.
*
* If resume is true, then we are resuming a receive
* operation after a blocked receive discarded our stack.
* Conditions:
* Our caller must hold a reference for the port or port set
* to which this queue belongs, to keep the queue
* from being deallocated.
*
* The kmsg is returned with clean header fields
* and with the circular bit turned off.
* Returns:
* MACH_MSG_SUCCESS Message returned in kmsgp.
* MACH_RCV_TOO_LARGE Message size returned in kmsgp.
* MACH_RCV_TIMED_OUT No message obtained.
* MACH_RCV_INTERRUPTED No message obtained.
* MACH_RCV_PORT_DIED Port/set died; no message.
* MACH_RCV_PORT_CHANGED Port moved into set; no msg.
*
*/
void
ipc_mqueue_receive(
ipc_mqueue_t mqueue,
mach_msg_option_t option,
mach_msg_size_t max_size,
mach_msg_timeout_t rcv_timeout,
int interruptible)
{
wait_result_t wresult;
thread_t self = current_thread();
wresult = ipc_mqueue_receive_on_thread(mqueue, option, max_size,
rcv_timeout, interruptible,
self);
if (wresult == THREAD_NOT_WAITING)
return;
if (wresult == THREAD_WAITING) {
counter((interruptible == THREAD_ABORTSAFE) ?
c_ipc_mqueue_receive_block_user++ :
c_ipc_mqueue_receive_block_kernel++);
if (self->ith_continuation)
thread_block(ipc_mqueue_receive_continue);
/* NOTREACHED */
wresult = thread_block(THREAD_CONTINUE_NULL);
}
ipc_mqueue_receive_results(wresult);
}
wait_result_t
ipc_mqueue_receive_on_thread(
ipc_mqueue_t mqueue,
mach_msg_option_t option,
mach_msg_size_t max_size,
mach_msg_timeout_t rcv_timeout,
int interruptible,
thread_t thread)
{
ipc_kmsg_queue_t kmsgs;
wait_result_t wresult;
uint64_t deadline;
spl_t s;
#if CONFIG_MACF_MACH
ipc_labelh_t lh;
task_t task;
int rc;
#endif
s = splsched();
imq_lock(mqueue);
if (imq_is_set(mqueue)) {
queue_t q;
q = &mqueue->imq_preposts;
/*
* If we are waiting on a portset mqueue, we need to see if
* any of the member ports have work for us. Ports that
* have (or recently had) messages will be linked in the
* prepost queue for the portset. By holding the portset's
* mqueue lock during the search, we tie up any attempts by
* mqueue_deliver or portset membership changes that may
* cross our path.
*/
search_set:
while(!queue_empty(q)) {
wait_queue_link_t wql;
ipc_mqueue_t port_mq;
queue_remove_first(q, wql, wait_queue_link_t, wql_preposts);
assert(!wql_is_preposted(wql));
/*
* This is a lock order violation, so we have to do it
* "softly," putting the link back on the prepost list
* if it fails (at the tail is fine since the order of
* handling messages from different sources in a set is
* not guaranteed and we'd like to skip to the next source
* if one is available).
*/
port_mq = (ipc_mqueue_t)wql->wql_queue;
if (!imq_lock_try(port_mq)) {
queue_enter(q, wql, wait_queue_link_t, wql_preposts);
imq_unlock(mqueue);
splx(s);
mutex_pause(0);
s = splsched();
imq_lock(mqueue);
goto search_set; /* start again at beginning - SMP */
}
/*
* If there are no messages on this queue, just skip it
* (we already removed the link from the set's prepost queue).
*/
kmsgs = &port_mq->imq_messages;
if (ipc_kmsg_queue_first(kmsgs) == IKM_NULL) {
imq_unlock(port_mq);
continue;
}
/*
* There are messages, so reinsert the link back
* at the tail of the preposted queue (for fairness)
* while we still have the portset mqueue locked.
*/
queue_enter(q, wql, wait_queue_link_t, wql_preposts);
imq_unlock(mqueue);
/*
* Continue on to handling the message with just
* the port mqueue locked.
*/
ipc_mqueue_select_on_thread(port_mq, option, max_size, thread);
imq_unlock(port_mq);
#if CONFIG_MACF_MACH
if (thread->task != TASK_NULL &&
thread->ith_kmsg != NULL &&
thread->ith_kmsg->ikm_sender != NULL) {
lh = thread->ith_kmsg->ikm_sender->label;
tasklabel_lock(thread->task);
ip_lock(lh->lh_port);
rc = mac_port_check_receive(&thread->task->maclabel,
&lh->lh_label);
ip_unlock(lh->lh_port);
tasklabel_unlock(thread->task);
if (rc)
thread->ith_state = MACH_RCV_INVALID_DATA;
}
#endif
splx(s);
return THREAD_NOT_WAITING;
}
} else {
/*
* Receive on a single port. Just try to get the messages.
*/
kmsgs = &mqueue->imq_messages;
if (ipc_kmsg_queue_first(kmsgs) != IKM_NULL) {
ipc_mqueue_select_on_thread(mqueue, option, max_size, thread);
imq_unlock(mqueue);
#if CONFIG_MACF_MACH
if (thread->task != TASK_NULL &&
thread->ith_kmsg != NULL &&
thread->ith_kmsg->ikm_sender != NULL) {
lh = thread->ith_kmsg->ikm_sender->label;
tasklabel_lock(thread->task);
ip_lock(lh->lh_port);
rc = mac_port_check_receive(&thread->task->maclabel,
&lh->lh_label);
ip_unlock(lh->lh_port);
tasklabel_unlock(thread->task);
if (rc)
thread->ith_state = MACH_RCV_INVALID_DATA;
}
#endif
splx(s);
return THREAD_NOT_WAITING;
}
}
/*
* Looks like we'll have to block. The mqueue we will
* block on (whether the set's or the local port's) is
* still locked.
*/
if (option & MACH_RCV_TIMEOUT) {
if (rcv_timeout == 0) {
imq_unlock(mqueue);
splx(s);
thread->ith_state = MACH_RCV_TIMED_OUT;
return THREAD_NOT_WAITING;
}
}
thread_lock(thread);
thread->ith_state = MACH_RCV_IN_PROGRESS;
thread->ith_option = option;
thread->ith_msize = max_size;
if (option & MACH_RCV_TIMEOUT)
clock_interval_to_deadline(rcv_timeout, 1000*NSEC_PER_USEC, &deadline);
else
deadline = 0;
wresult = wait_queue_assert_wait64_locked(&mqueue->imq_wait_queue,
IPC_MQUEUE_RECEIVE,
interruptible,
TIMEOUT_URGENCY_USER_NORMAL,
deadline, 0,
thread);
/* preposts should be detected above, not here */
if (wresult == THREAD_AWAKENED)
panic("ipc_mqueue_receive_on_thread: sleep walking");
thread_unlock(thread);
imq_unlock(mqueue);
splx(s);
return wresult;
}
/*
* Routine: ipc_mqueue_select_on_thread
* Purpose:
* A receiver discovered that there was a message on the queue
* before he had to block. Pick the message off the queue and
* "post" it to thread.
* Conditions:
* mqueue locked.
* thread not locked.
* There is a message.
* Returns:
* MACH_MSG_SUCCESS Actually selected a message for ourselves.
* MACH_RCV_TOO_LARGE May or may not have pull it, but it is large
*/
void
ipc_mqueue_select_on_thread(
ipc_mqueue_t mqueue,
mach_msg_option_t option,
mach_msg_size_t max_size,
thread_t thread)
{
ipc_kmsg_t kmsg;
mach_msg_return_t mr = MACH_MSG_SUCCESS;
mach_msg_size_t rcv_size;
/*
* Do some sanity checking of our ability to receive
* before pulling the message off the queue.
*/
kmsg = ipc_kmsg_queue_first(&mqueue->imq_messages);
assert(kmsg != IKM_NULL);
/*
* If we really can't receive it, but we had the
* MACH_RCV_LARGE option set, then don't take it off
* the queue, instead return the appropriate error
* (and size needed).
*/
rcv_size = ipc_kmsg_copyout_size(kmsg, thread->map);
if (rcv_size + REQUESTED_TRAILER_SIZE(thread_is_64bit(thread), option) > max_size) {
mr = MACH_RCV_TOO_LARGE;
if (option & MACH_RCV_LARGE) {
thread->ith_receiver_name = mqueue->imq_receiver_name;
thread->ith_kmsg = IKM_NULL;
thread->ith_msize = rcv_size;
thread->ith_seqno = 0;
thread->ith_state = mr;
return;
}
}
ipc_kmsg_rmqueue_first_macro(&mqueue->imq_messages, kmsg);
ipc_mqueue_release_msgcount(mqueue);
thread->ith_seqno = mqueue->imq_seqno++;
thread->ith_kmsg = kmsg;
thread->ith_state = mr;
current_task()->messages_received++;
return;
}
/*
* Routine: ipc_mqueue_peek
* Purpose:
* Peek at a (non-set) message queue to see if it has a message
* matching the sequence number provided (if zero, then the
* first message in the queue) and return vital info about the
* message.
*
* Conditions:
* Locks may be held by callers, so this routine cannot block.
* Caller holds reference on the message queue.
*/
unsigned
ipc_mqueue_peek(ipc_mqueue_t mq,
mach_port_seqno_t *seqnop,
mach_msg_size_t *msg_sizep,
mach_msg_id_t *msg_idp,
mach_msg_max_trailer_t *msg_trailerp)
{
ipc_kmsg_queue_t kmsgq;
ipc_kmsg_t kmsg;
mach_port_seqno_t seqno, msgoff;
int res = 0;
spl_t s;
assert(!imq_is_set(mq));
s = splsched();
imq_lock(mq);
seqno = (seqnop != NULL) ? seqno = *seqnop : 0;
if (seqno == 0) {
seqno = mq->imq_seqno;
msgoff = 0;
} else if (seqno >= mq->imq_seqno &&
seqno < mq->imq_seqno + mq->imq_msgcount) {
msgoff = seqno - mq->imq_seqno;
} else
goto out;
/* look for the message that would match that seqno */
kmsgq = &mq->imq_messages;
kmsg = ipc_kmsg_queue_first(kmsgq);
while (msgoff-- && kmsg != IKM_NULL) {
kmsg = ipc_kmsg_queue_next(kmsgq, kmsg);
}
if (kmsg == IKM_NULL)
goto out;
/* found one - return the requested info */
if (seqnop != NULL)
*seqnop = seqno;
if (msg_sizep != NULL)
*msg_sizep = kmsg->ikm_header->msgh_size;
if (msg_idp != NULL)
*msg_idp = kmsg->ikm_header->msgh_id;
if (msg_trailerp != NULL)
memcpy(msg_trailerp,
(mach_msg_max_trailer_t *)((vm_offset_t)kmsg->ikm_header +
round_msg(kmsg->ikm_header->msgh_size)),
sizeof(mach_msg_max_trailer_t));
res = 1;
out:
imq_unlock(mq);
splx(s);
return res;
}
/*
* Routine: ipc_mqueue_set_peek
* Purpose:
* Peek at a message queue set to see if it has any ports
* with messages.
*
* Conditions:
* Locks may be held by callers, so this routine cannot block.
* Caller holds reference on the message queue.
*/
unsigned
ipc_mqueue_set_peek(ipc_mqueue_t mq)
{
wait_queue_link_t wql;
queue_t q;
spl_t s;
int res;
assert(imq_is_set(mq));
s = splsched();
imq_lock(mq);
/*
* peek at the contained port message queues, return as soon as
* we spot a message on one of the message queues linked on the
* prepost list. No need to lock each message queue, as only the
* head of each queue is checked. If a message wasn't there before
* we entered here, no need to find it (if we do, great).
*/
res = 0;
q = &mq->imq_preposts;
queue_iterate(q, wql, wait_queue_link_t, wql_preposts) {
ipc_mqueue_t port_mq = (ipc_mqueue_t)wql->wql_queue;
ipc_kmsg_queue_t kmsgs = &port_mq->imq_messages;
if (ipc_kmsg_queue_first(kmsgs) != IKM_NULL) {
res = 1;
break;
}
}
imq_unlock(mq);
splx(s);
return res;
}
/*
* Routine: ipc_mqueue_set_gather_member_names
* Purpose:
* Iterate a message queue set to identify the member port
* names. Actual returned names is limited to maxnames entries,
* but we keep counting the actual number of members to let
* the caller decide to retry if necessary.
*
* Conditions:
* Locks may be held by callers, so this routine cannot block.
* Caller holds reference on the message queue.
*/
void
ipc_mqueue_set_gather_member_names(
ipc_mqueue_t mq,
ipc_entry_num_t maxnames,
mach_port_name_t *names,
ipc_entry_num_t *actualp)
{
wait_queue_link_t wql;
queue_t q;
spl_t s;
ipc_entry_num_t actual = 0;
assert(imq_is_set(mq));
s = splsched();
imq_lock(mq);
/*
* Iterate over the member ports through the mqueue set links
* capturing as many names as we can.
*/
q = &mq->imq_setlinks;
queue_iterate(q, wql, wait_queue_link_t, wql_setlinks) {
ipc_mqueue_t port_mq = (ipc_mqueue_t)wql->wql_queue;
if (actual < maxnames)
names[actual] = port_mq->imq_receiver_name;
actual++;
}
imq_unlock(mq);
splx(s);
*actualp = actual;
}
/*
* Routine: ipc_mqueue_destroy
* Purpose:
* Destroy a (non-set) message queue.
* Set any blocked senders running.
* Destroy the kmsgs in the queue.
* Conditions:
* Nothing locked.
* Receivers were removed when the receive right was "changed"
*/
void
ipc_mqueue_destroy(
ipc_mqueue_t mqueue)
{
ipc_kmsg_queue_t kmqueue;
ipc_kmsg_t kmsg;
boolean_t reap = FALSE;
spl_t s;
s = splsched();
imq_lock(mqueue);
/*
* rouse all blocked senders
*/
mqueue->imq_fullwaiters = FALSE;
printf("\nAbout to call wait_queue_wakeup64_all_locked from ipc_mqueue_destroy\n");
wait_queue_wakeup64_all_locked(
&mqueue->imq_wait_queue,
IPC_MQUEUE_FULL,
THREAD_RESTART,
FALSE);
/*
* Move messages from the specified queue to the per-thread
* clean/drain queue while we have the mqueue lock.
*/
kmqueue = &mqueue->imq_messages;
while ((kmsg = ipc_kmsg_dequeue(kmqueue)) != IKM_NULL) {
boolean_t first;
first = ipc_kmsg_delayed_destroy(kmsg);
if (first)
reap = first;
}
imq_unlock(mqueue);
splx(s);
/*
* Destroy the messages we enqueued if we aren't nested
* inside some other attempt to drain the same queue.
*/
if (reap)
ipc_kmsg_reap_delayed();
}
/*
* Routine: ipc_mqueue_set_qlimit
* Purpose:
* Changes a message queue limit; the maximum number
* of messages which may be queued.
* Conditions:
* Nothing locked.
*/
void
ipc_mqueue_set_qlimit(
ipc_mqueue_t mqueue,
mach_port_msgcount_t qlimit)
{
spl_t s;
assert(qlimit <= MACH_PORT_QLIMIT_MAX);
/* wake up senders allowed by the new qlimit */
s = splsched();
imq_lock(mqueue);
if (qlimit > mqueue->imq_qlimit) {
mach_port_msgcount_t i, wakeup;
/* caution: wakeup, qlimit are unsigned */
wakeup = qlimit - mqueue->imq_qlimit;
for (i = 0; i < wakeup; i++) {
if (wait_queue_wakeup64_one_locked(
&mqueue->imq_wait_queue,
IPC_MQUEUE_FULL,
THREAD_AWAKENED,
FALSE) == KERN_NOT_WAITING) {
mqueue->imq_fullwaiters = FALSE;
break;
}
mqueue->imq_msgcount++; /* give it to the awakened thread */
}
}
mqueue->imq_qlimit = qlimit;
imq_unlock(mqueue);
splx(s);
}
/*
* Routine: ipc_mqueue_set_seqno
* Purpose:
* Changes an mqueue's sequence number.
* Conditions:
* Caller holds a reference to the queue's containing object.
*/
void
ipc_mqueue_set_seqno(
ipc_mqueue_t mqueue,
mach_port_seqno_t seqno)
{
spl_t s;
s = splsched();
imq_lock(mqueue);
mqueue->imq_seqno = seqno;
imq_unlock(mqueue);
splx(s);
}
/*
* Routine: ipc_mqueue_copyin
* Purpose:
* Convert a name in a space to a message queue.
* Conditions:
* Nothing locked. If successful, the caller gets a ref for
* for the object. This ref ensures the continued existence of
* the queue.
* Returns:
* MACH_MSG_SUCCESS Found a message queue.
* MACH_RCV_INVALID_NAME The space is dead.
* MACH_RCV_INVALID_NAME The name doesn't denote a right.
* MACH_RCV_INVALID_NAME
* The denoted right is not receive or port set.
* MACH_RCV_IN_SET Receive right is a member of a set.
*/
mach_msg_return_t
ipc_mqueue_copyin(
ipc_space_t space,
mach_port_name_t name,
ipc_mqueue_t *mqueuep,
ipc_object_t *objectp)
{
ipc_entry_t entry;
ipc_object_t object;
ipc_mqueue_t mqueue;
is_read_lock(space);
if (!is_active(space)) {
is_read_unlock(space);
return MACH_RCV_INVALID_NAME;
}
entry = ipc_entry_lookup(space, name);
if (entry == IE_NULL) {
is_read_unlock(space);
return MACH_RCV_INVALID_NAME;
}
object = entry->ie_object;
if (entry->ie_bits & MACH_PORT_TYPE_RECEIVE) {
ipc_port_t port;
port = (ipc_port_t) object;
assert(port != IP_NULL);
ip_lock(port);
assert(ip_active(port));
assert(port->ip_receiver_name == name);
assert(port->ip_receiver == space);
is_read_unlock(space);
mqueue = &port->ip_messages;
} else if (entry->ie_bits & MACH_PORT_TYPE_PORT_SET) {
ipc_pset_t pset;
pset = (ipc_pset_t) object;
assert(pset != IPS_NULL);
ips_lock(pset);
assert(ips_active(pset));
assert(pset->ips_local_name == name);
is_read_unlock(space);
mqueue = &pset->ips_messages;
} else {
is_read_unlock(space);
return MACH_RCV_INVALID_NAME;
}
/*
* At this point, the object is locked and active,
* the space is unlocked, and mqueue is initialized.
*/
io_reference(object);
io_unlock(object);
*objectp = object;
*mqueuep = mqueue;
return MACH_MSG_SUCCESS;
}
Raw
kern_event.c
/*
* Copyright (c) 2000-2013 Apple Inc. All rights reserved.
*
* @APPLE_OSREFERENCE_LICENSE_HEADER_START@
*
* This file contains Original Code and/or Modifications of Original Code
* as defined in and that are subject to the Apple Public Source License
* Version 2.0 (the 'License'). You may not use this file except in
* compliance with the License. The rights granted to you under the License
* may not be used to create, or enable the creation or redistribution of,
* unlawful or unlicensed copies of an Apple operating system, or to
* circumvent, violate, or enable the circumvention or violation of, any
* terms of an Apple operating system software license agreement.
*
* Please obtain a copy of the License at
* http://www.opensource.apple.com/apsl/ and read it before using this file.
*
* The Original Code and all software distributed under the License are
* distributed on an 'AS IS' basis, WITHOUT WARRANTY OF ANY KIND, EITHER
* EXPRESS OR IMPLIED, AND APPLE HEREBY DISCLAIMS ALL SUCH WARRANTIES,
* INCLUDING WITHOUT LIMITATION, ANY WARRANTIES OF MERCHANTABILITY,
* FITNESS FOR A PARTICULAR PURPOSE, QUIET ENJOYMENT OR NON-INFRINGEMENT.
* Please see the License for the specific language governing rights and
* limitations under the License.
*
* @APPLE_OSREFERENCE_LICENSE_HEADER_END@
*
*/
/*-
* Copyright (c) 1999,2000,2001 Jonathan Lemon <jlemon@FreeBSD.org>
* All rights reserved.
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions
* are met:
* 1. Redistributions of source code must retain the above copyright
* notice, this list of conditions and the following disclaimer.
* 2. Redistributions in binary form must reproduce the above copyright
* notice, this list of conditions and the following disclaimer in the
* documentation and/or other materials provided with the distribution.
*
* THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
* ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
* IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
* ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
* FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
* DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
* OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
* HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
* LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
* OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
* SUCH DAMAGE.
*/
/*
* @(#)kern_event.c 1.0 (3/31/2000)
*/
#include <stdint.h>
#include <sys/param.h>
#include <sys/systm.h>
#include <sys/filedesc.h>
#include <sys/kernel.h>
#include <sys/proc_internal.h>
#include <sys/kauth.h>
#include <sys/malloc.h>
#include <sys/unistd.h>
#include <sys/file_internal.h>
#include <sys/fcntl.h>
#include <sys/select.h>
#include <sys/queue.h>
#include <sys/event.h>
#include <sys/eventvar.h>
#include <sys/protosw.h>
#include <sys/socket.h>
#include <sys/socketvar.h>
#include <sys/stat.h>
#include <sys/sysctl.h>
#include <sys/uio.h>
#include <sys/sysproto.h>
#include <sys/user.h>
#include <sys/vnode_internal.h>
#include <string.h>
#include <sys/proc_info.h>
#include <sys/codesign.h>
#include <kern/lock.h>
#include <kern/clock.h>
#include <kern/thread_call.h>
#include <kern/sched_prim.h>
#include <kern/zalloc.h>
#include <kern/assert.h>
#include <libkern/libkern.h>
#include "net/net_str_id.h"
#include <mach/task.h>
#if VM_PRESSURE_EVENTS
#include <kern/vm_pressure.h>
#endif
#if CONFIG_MEMORYSTATUS
#include <sys/kern_memorystatus.h>
#endif
MALLOC_DEFINE(M_KQUEUE, "kqueue", "memory for kqueue system");
#define KQ_EVENT NULL
/*Plumbing needed for wait queue logging.*/
#define EVENT_MASK_BITS
typedef struct my_wait_queue { // happy little wait queue
unsigned long int
/* boolean_t */ wq_type:2, /* only public field */
wq_fifo:1, /* fifo wakeup policy? */
wq_prepost:1, /* waitq supports prepost? set only */
wq_eventmask:((sizeof(long) * 8) - 4);
} myWaitQueue;
#define _WAIT_QUEUE_inited 0x2
#define wait_queue_is_valid(wq) \
(((wq)->wq_type & ~1) == _WAIT_QUEUE_inited)
static inline void kqlock(struct kqueue *kq);
static inline void kqunlock(struct kqueue *kq);
static int kqlock2knoteuse(struct kqueue *kq, struct knote *kn);
static int kqlock2knoteusewait(struct kqueue *kq, struct knote *kn);
static int kqlock2knotedrop(struct kqueue *kq, struct knote *kn);
static int knoteuse2kqlock(struct kqueue *kq, struct knote *kn);
static void kqueue_wakeup(struct kqueue *kq, int closed);
static int kqueue_read(struct fileproc *fp, struct uio *uio,
int flags, vfs_context_t ctx);
static int kqueue_write(struct fileproc *fp, struct uio *uio,
int flags, vfs_context_t ctx);
static int kqueue_ioctl(struct fileproc *fp, u_long com, caddr_t data,
vfs_context_t ctx);
static int kqueue_select(struct fileproc *fp, int which, void *wql,
vfs_context_t ctx);
static int kqueue_close(struct fileglob *fg, vfs_context_t ctx);
static int kqueue_kqfilter(struct fileproc *fp, struct knote *kn,
vfs_context_t ctx);
static int kqueue_drain(struct fileproc *fp, vfs_context_t ctx);
extern int kqueue_stat(struct fileproc *fp, void *ub, int isstat64,
vfs_context_t ctx);
static const struct fileops kqueueops = {
.fo_type = DTYPE_KQUEUE,
.fo_read = kqueue_read,
.fo_write = kqueue_write,
.fo_ioctl = kqueue_ioctl,
.fo_select = kqueue_select,
.fo_close = kqueue_close,
.fo_kqfilter = kqueue_kqfilter,
.fo_drain = kqueue_drain,
};
static int kevent_internal(struct proc *p, int iskev64, user_addr_t changelist,
int nchanges, user_addr_t eventlist, int nevents, int fd,
user_addr_t utimeout, unsigned int flags, int32_t *retval);
static int kevent_copyin(user_addr_t *addrp, struct kevent64_s *kevp,
struct proc *p, int iskev64);
static int kevent_copyout(struct kevent64_s *kevp, user_addr_t *addrp,
struct proc *p, int iskev64);
char * kevent_description(struct kevent64_s *kevp, char *s, size_t n);
static int kevent_callback(struct kqueue *kq, struct kevent64_s *kevp,
void *data);
static void kevent_continue(struct kqueue *kq, void *data, int error);
static void kqueue_scan_continue(void *contp, wait_result_t wait_result);
static int kqueue_process(struct kqueue *kq, kevent_callback_t callback,
void *data, int *countp, struct proc *p);
static int kqueue_begin_processing(struct kqueue *kq);
static void kqueue_end_processing(struct kqueue *kq);
static int knote_process(struct knote *kn, kevent_callback_t callback,
void *data, struct kqtailq *inprocessp, struct proc *p);
static void knote_put(struct knote *kn);
static int knote_fdpattach(struct knote *kn, struct filedesc *fdp,
struct proc *p);
static void knote_drop(struct knote *kn, struct proc *p);
static void knote_activate(struct knote *kn, int);
static void knote_deactivate(struct knote *kn);
static void knote_enqueue(struct knote *kn);
static void knote_dequeue(struct knote *kn);
static struct knote *knote_alloc(void);
static void knote_free(struct knote *kn);
static int filt_fileattach(struct knote *kn);
static struct filterops file_filtops = {
.f_isfd = 1,
.f_attach = filt_fileattach,
};
static void filt_kqdetach(struct knote *kn);
static int filt_kqueue(struct knote *kn, long hint);
static struct filterops kqread_filtops = {
.f_isfd = 1,
.f_detach = filt_kqdetach,
.f_event = filt_kqueue,
};
/* placeholder for not-yet-implemented filters */
static int filt_badattach(struct knote *kn);
static struct filterops bad_filtops = {
.f_attach = filt_badattach,
};
static int filt_procattach(struct knote *kn);
static void filt_procdetach(struct knote *kn);
static int filt_proc(struct knote *kn, long hint);
static struct filterops proc_filtops = {
.f_attach = filt_procattach,
.f_detach = filt_procdetach,
.f_event = filt_proc,
};
#if VM_PRESSURE_EVENTS
static int filt_vmattach(struct knote *kn);
static void filt_vmdetach(struct knote *kn);
static int filt_vm(struct knote *kn, long hint);
static struct filterops vm_filtops = {
.f_attach = filt_vmattach,
.f_detach = filt_vmdetach,
.f_event = filt_vm,
};
#endif /* VM_PRESSURE_EVENTS */
#if CONFIG_MEMORYSTATUS
extern struct filterops memorystatus_filtops;
#endif /* CONFIG_MEMORYSTATUS */
extern struct filterops fs_filtops;
extern struct filterops sig_filtops;
/* Timer filter */
static int filt_timerattach(struct knote *kn);
static void filt_timerdetach(struct knote *kn);
static int filt_timer(struct knote *kn, long hint);
static void filt_timertouch(struct knote *kn, struct kevent64_s *kev,
long type);
static struct filterops timer_filtops = {
.f_attach = filt_timerattach,
.f_detach = filt_timerdetach,
.f_event = filt_timer,
.f_touch = filt_timertouch,
};
/* Helpers */
static void filt_timerexpire(void *knx, void *param1);
static int filt_timervalidate(struct knote *kn);
static void filt_timerupdate(struct knote *kn);
static void filt_timercancel(struct knote *kn);
#define TIMER_RUNNING 0x1
#define TIMER_CANCELWAIT 0x2
static lck_mtx_t _filt_timerlock;
static void filt_timerlock(void);
static void filt_timerunlock(void);
static zone_t knote_zone;
#define KN_HASH(val, mask) (((val) ^ (val >> 8)) & (mask))
#if 0
extern struct filterops aio_filtops;
#endif
/* Mach portset filter */
extern struct filterops machport_filtops;
/* User filter */
static int filt_userattach(struct knote *kn);
static void filt_userdetach(struct knote *kn);
static int filt_user(struct knote *kn, long hint);
static void filt_usertouch(struct knote *kn, struct kevent64_s *kev,
long type);
static struct filterops user_filtops = {
.f_attach = filt_userattach,
.f_detach = filt_userdetach,
.f_event = filt_user,
.f_touch = filt_usertouch,
};
/*
* Table for all system-defined filters.
*/
static struct filterops *sysfilt_ops[] = {
&file_filtops, /* EVFILT_READ */
&file_filtops, /* EVFILT_WRITE */
#if 0
&aio_filtops, /* EVFILT_AIO */
#else
&bad_filtops, /* EVFILT_AIO */
#endif
&file_filtops, /* EVFILT_VNODE */
&proc_filtops, /* EVFILT_PROC */
&sig_filtops, /* EVFILT_SIGNAL */
&timer_filtops, /* EVFILT_TIMER */
&machport_filtops, /* EVFILT_MACHPORT */
&fs_filtops, /* EVFILT_FS */
&user_filtops, /* EVFILT_USER */
&bad_filtops, /* unused */
#if VM_PRESSURE_EVENTS
&vm_filtops, /* EVFILT_VM */
#else
&bad_filtops, /* EVFILT_VM */
#endif
&file_filtops, /* EVFILT_SOCK */
#if CONFIG_MEMORYSTATUS
&memorystatus_filtops, /* EVFILT_MEMORYSTATUS */
#else
&bad_filtops, /* EVFILT_MEMORYSTATUS */
#endif
};
/*
* kqueue/note lock attributes and implementations
*
* kqueues have locks, while knotes have use counts
* Most of the knote state is guarded by the object lock.
* the knote "inuse" count and status use the kqueue lock.
*/
lck_grp_attr_t * kq_lck_grp_attr;
lck_grp_t * kq_lck_grp;
lck_attr_t * kq_lck_attr;
static inline void
kqlock(struct kqueue *kq)
{
lck_spin_lock(&kq->kq_lock);
}
static inline void
kqunlock(struct kqueue *kq)
{
lck_spin_unlock(&kq->kq_lock);
}
/*
* Convert a kq lock to a knote use referece.
*
* If the knote is being dropped, we can't get
* a use reference, so just return with it
* still locked.
* - kq locked at entry
* - unlock on exit if we get the use reference
*/
static int
kqlock2knoteuse(struct kqueue *kq, struct knote *kn)
{
if (kn->kn_status & KN_DROPPING)
return (0);
kn->kn_inuse++;
kqunlock(kq);
return (1);
}
/*
* Convert a kq lock to a knote use referece,
* but wait for attach and drop events to complete.
*
* If the knote is being dropped, we can't get
* a use reference, so just return with it
* still locked.
* - kq locked at entry
* - kq always unlocked on exit
*/
static int
kqlock2knoteusewait(struct kqueue *kq, struct knote *kn)
{
if ((kn->kn_status & (KN_DROPPING | KN_ATTACHING)) != 0) {
kn->kn_status |= KN_USEWAIT;
if (!wait_queue_is_valid((myWaitQueue*)kq->kq_wqs)) {
printf("\nAbout to call wait_queue_assert_wait with invalid wait queue from kqlock2knoteusewait.\n");
}
wait_queue_assert_wait((wait_queue_t)kq->kq_wqs,
&kn->kn_status, THREAD_UNINT, 0);
kqunlock(kq);
thread_block(THREAD_CONTINUE_NULL);
return (0);
}
kn->kn_inuse++;
kqunlock(kq);
return (1);
}
/*
* Convert from a knote use reference back to kq lock.
*
* Drop a use reference and wake any waiters if
* this is the last one.
*
* The exit return indicates if the knote is
* still alive - but the kqueue lock is taken
* unconditionally.
*/
static int
knoteuse2kqlock(struct kqueue *kq, struct knote *kn)
{
kqlock(kq);
if (--kn->kn_inuse == 0) {
if ((kn->kn_status & KN_ATTACHING) != 0) {
kn->kn_status &= ~KN_ATTACHING;
}
if ((kn->kn_status & KN_USEWAIT) != 0) {
kn->kn_status &= ~KN_USEWAIT;
wait_queue_wakeup_all((wait_queue_t)kq->kq_wqs,
&kn->kn_status, THREAD_AWAKENED);
}
}
return ((kn->kn_status & KN_DROPPING) == 0);
}
/*
* Convert a kq lock to a knote drop reference.
*
* If the knote is in use, wait for the use count
* to subside. We first mark our intention to drop
* it - keeping other users from "piling on."
* If we are too late, we have to wait for the
* other drop to complete.
*
* - kq locked at entry
* - always unlocked on exit.
* - caller can't hold any locks that would prevent
* the other dropper from completing.
*/
static int
kqlock2knotedrop(struct kqueue *kq, struct knote *kn)
{
int oktodrop;
oktodrop = ((kn->kn_status & (KN_DROPPING | KN_ATTACHING)) == 0);
kn->kn_status |= KN_DROPPING;
if (oktodrop) {
if (kn->kn_inuse == 0) {
kqunlock(kq);
return (oktodrop);
}
}
kn->kn_status |= KN_USEWAIT;
if (!wait_queue_is_valid((myWaitQueue*)kq->kq_wqs)) {
printf("\nAbout to call wait_queue_assert_wait with invalid wait queue from kqlock2knotedrop.\n");
}
wait_queue_assert_wait((wait_queue_t)kq->kq_wqs, &kn->kn_status,
THREAD_UNINT, 0);
kqunlock(kq);
thread_block(THREAD_CONTINUE_NULL);
return (oktodrop);
}
/*
* Release a knote use count reference.
*/
static void
knote_put(struct knote *kn)
{
struct kqueue *kq = kn->kn_kq;
kqlock(kq);
if (--kn->kn_inuse == 0) {
if ((kn->kn_status & KN_USEWAIT) != 0) {
kn->kn_status &= ~KN_USEWAIT;
wait_queue_wakeup_all((wait_queue_t)kq->kq_wqs,
&kn->kn_status, THREAD_AWAKENED);
}
}
kqunlock(kq);
}
static int
filt_fileattach(struct knote *kn)
{
return (fo_kqfilter(kn->kn_fp, kn, vfs_context_current()));
}
#define f_flag f_fglob->fg_flag
#define f_msgcount f_fglob->fg_msgcount
#define f_cred f_fglob->fg_cred
#define f_ops f_fglob->fg_ops
#define f_offset f_fglob->fg_offset
#define f_data f_fglob->fg_data
static void
filt_kqdetach(struct knote *kn)
{
struct kqueue *kq = (struct kqueue *)kn->kn_fp->f_data;
kqlock(kq);
KNOTE_DETACH(&kq->kq_sel.si_note, kn);
kqunlock(kq);
}
/*ARGSUSED*/
static int
filt_kqueue(struct knote *kn, __unused long hint)
{
struct kqueue *kq = (struct kqueue *)kn->kn_fp->f_data;
kn->kn_data = kq->kq_count;
return (kn->kn_data > 0);
}
static int
filt_procattach(struct knote *kn)
{
struct proc *p;
assert(PID_MAX < NOTE_PDATAMASK);
if ((kn->kn_sfflags & (NOTE_TRACK | NOTE_TRACKERR | NOTE_CHILD)) != 0)
return (ENOTSUP);
p = proc_find(kn->kn_id);
if (p == NULL) {
return (ESRCH);
}
const int NoteExitStatusBits = NOTE_EXIT | NOTE_EXITSTATUS;
if ((kn->kn_sfflags & NoteExitStatusBits) == NoteExitStatusBits)
do {
pid_t selfpid = proc_selfpid();
if (p->p_ppid == selfpid)
break; /* parent => ok */
if ((p->p_lflag & P_LTRACED) != 0 &&
(p->p_oppid == selfpid))
break; /* parent-in-waiting => ok */
proc_rele(p);
return (EACCES);
} while (0);
proc_klist_lock();
kn->kn_flags |= EV_CLEAR; /* automatically set */
kn->kn_ptr.p_proc = p; /* store the proc handle */
KNOTE_ATTACH(&p->p_klist, kn);
proc_klist_unlock();
proc_rele(p);
return (0);
}
/*
* The knote may be attached to a different process, which may exit,
* leaving nothing for the knote to be attached to. In that case,
* the pointer to the process will have already been nulled out.
*/
static void
filt_procdetach(struct knote *kn)
{
struct proc *p;
proc_klist_lock();
p = kn->kn_ptr.p_proc;
if (p != PROC_NULL) {
kn->kn_ptr.p_proc = PROC_NULL;
KNOTE_DETACH(&p->p_klist, kn);
}
proc_klist_unlock();
}
static int
filt_proc(struct knote *kn, long hint)
{
/*
* Note: a lot of bits in hint may be obtained from the knote
* To free some of those bits, see <rdar://problem/12592988> Freeing up
* bits in hint for filt_proc
*/
/* hint is 0 when called from above */
if (hint != 0) {
u_int event;
/* ALWAYS CALLED WITH proc_klist_lock when (hint != 0) */
/*
* mask off extra data
*/
event = (u_int)hint & NOTE_PCTRLMASK;
/*
* termination lifecycle events can happen while a debugger
* has reparented a process, in which case notifications
* should be quashed except to the tracing parent. When
* the debugger reaps the child (either via wait4(2) or
* process exit), the child will be reparented to the original
* parent and these knotes re-fired.
*/
if (event & NOTE_EXIT) {
if ((kn->kn_ptr.p_proc->p_oppid != 0)
&& (kn->kn_kq->kq_p->p_pid != kn->kn_ptr.p_proc->p_ppid)) {
/*
* This knote is not for the current ptrace(2) parent, ignore.
*/
return 0;
}
}
/*
* if the user is interested in this event, record it.
*/
if (kn->kn_sfflags & event)
kn->kn_fflags |= event;
#pragma clang diagnostic push
#pragma clang diagnostic ignored "-Wdeprecated-declarations"
if ((event == NOTE_REAP) || ((event == NOTE_EXIT) && !(kn->kn_sfflags & NOTE_REAP))) {
kn->kn_flags |= (EV_EOF | EV_ONESHOT);
}
#pragma clang diagnostic pop
if (event == NOTE_EXIT) {
kn->kn_data = 0;
if ((kn->kn_sfflags & NOTE_EXITSTATUS) != 0) {
kn->kn_fflags |= NOTE_EXITSTATUS;
kn->kn_data |= (hint & NOTE_PDATAMASK);
}
if ((kn->kn_sfflags & NOTE_EXIT_DETAIL) != 0) {
kn->kn_fflags |= NOTE_EXIT_DETAIL;
if ((kn->kn_ptr.p_proc->p_lflag &
P_LTERM_DECRYPTFAIL) != 0) {
kn->kn_data |= NOTE_EXIT_DECRYPTFAIL;
}
if ((kn->kn_ptr.p_proc->p_lflag &
P_LTERM_JETSAM) != 0) {
kn->kn_data |= NOTE_EXIT_MEMORY;
switch (kn->kn_ptr.p_proc->p_lflag &
P_JETSAM_MASK) {
case P_JETSAM_VMPAGESHORTAGE:
kn->kn_data |= NOTE_EXIT_MEMORY_VMPAGESHORTAGE;
break;
case P_JETSAM_VMTHRASHING:
kn->kn_data |= NOTE_EXIT_MEMORY_VMTHRASHING;
break;
case P_JETSAM_VNODE:
kn->kn_data |= NOTE_EXIT_MEMORY_VNODE;
break;
case P_JETSAM_HIWAT:
kn->kn_data |= NOTE_EXIT_MEMORY_HIWAT;
break;
case P_JETSAM_PID:
kn->kn_data |= NOTE_EXIT_MEMORY_PID;
break;
case P_JETSAM_IDLEEXIT:
kn->kn_data |= NOTE_EXIT_MEMORY_IDLE;
break;
}
}
if ((kn->kn_ptr.p_proc->p_csflags &
CS_KILLED) != 0) {
kn->kn_data |= NOTE_EXIT_CSERROR;
}
}
}
}
/* atomic check, no locking need when called from above */
return (kn->kn_fflags != 0);
}
#if VM_PRESSURE_EVENTS
/*
* Virtual memory kevents
*
* author: Matt Jacobson [matthew_jacobson@apple.com]
*/
static int
filt_vmattach(struct knote *kn)
{
/*
* The note will be cleared once the information has been flushed to
* the client. If there is still pressure, we will be re-alerted.
*/
kn->kn_flags |= EV_CLEAR;
return (vm_knote_register(kn));
}
static void
filt_vmdetach(struct knote *kn)
{
vm_knote_unregister(kn);
}
static int
filt_vm(struct knote *kn, long hint)
{
/* hint == 0 means this is just an alive? check (always true) */
if (hint != 0) {
const pid_t pid = (pid_t)hint;
if ((kn->kn_sfflags & NOTE_VM_PRESSURE) &&
(kn->kn_kq->kq_p->p_pid == pid)) {
kn->kn_fflags |= NOTE_VM_PRESSURE;
}
}
return (kn->kn_fflags != 0);
}
#endif /* VM_PRESSURE_EVENTS */
/*
* filt_timervalidate - process data from user
*
* Converts to either interval or deadline format.
*
* The saved-data field in the knote contains the
* time value. The saved filter-flags indicates
* the unit of measurement.
*
* After validation, either the saved-data field
* contains the interval in absolute time, or ext[0]
* contains the expected deadline. If that deadline
* is in the past, ext[0] is 0.
*
* Returns EINVAL for unrecognized units of time.
*
* Timer filter lock is held.
*
*/
static int
filt_timervalidate(struct knote *kn)
{
uint64_t multiplier;
uint64_t raw = 0;
switch (kn->kn_sfflags & (NOTE_SECONDS|NOTE_USECONDS|NOTE_NSECONDS)) {
case NOTE_SECONDS:
multiplier = NSEC_PER_SEC;
break;
case NOTE_USECONDS:
multiplier = NSEC_PER_USEC;
break;
case NOTE_NSECONDS:
multiplier = 1;
break;
case 0: /* milliseconds (default) */
multiplier = NSEC_PER_SEC / 1000;
break;
default:
return (EINVAL);
}
/* transform the slop delta(leeway) in kn_ext[1] if passed to same time scale */
if(kn->kn_sfflags & NOTE_LEEWAY){
nanoseconds_to_absolutetime((uint64_t)kn->kn_ext[1] * multiplier, &raw);
kn->kn_ext[1] = raw;
}
nanoseconds_to_absolutetime((uint64_t)kn->kn_sdata * multiplier, &raw);
kn->kn_ext[0] = 0;
kn->kn_sdata = 0;
if (kn->kn_sfflags & NOTE_ABSOLUTE) {
clock_sec_t seconds;
clock_nsec_t nanoseconds;
uint64_t now;
clock_get_calendar_nanotime(&seconds, &nanoseconds);
nanoseconds_to_absolutetime((uint64_t)seconds * NSEC_PER_SEC +
nanoseconds, &now);
if (raw < now) {
/* time has already passed */
kn->kn_ext[0] = 0;
} else {
raw -= now;
clock_absolutetime_interval_to_deadline(raw,
&kn->kn_ext[0]);
}
} else {
kn->kn_sdata = raw;
}
return (0);
}
/*
* filt_timerupdate - compute the next deadline
*
* Repeating timers store their interval in kn_sdata. Absolute
* timers have already calculated the deadline, stored in ext[0].
*
* On return, the next deadline (or zero if no deadline is needed)
* is stored in kn_ext[0].
*
* Timer filter lock is held.
*/
static void
filt_timerupdate(struct knote *kn)
{
/* if there's no interval, deadline is just in kn_ext[0] */
if (kn->kn_sdata == 0)
return;
/* if timer hasn't fired before, fire in interval nsecs */
if (kn->kn_ext[0] == 0) {
clock_absolutetime_interval_to_deadline(kn->kn_sdata,
&kn->kn_ext[0]);
} else {
/*
* If timer has fired before, schedule the next pop
* relative to the last intended deadline.
*
* We could check for whether the deadline has expired,
* but the thread call layer can handle that.
*/
kn->kn_ext[0] += kn->kn_sdata;
}
}
/*
* filt_timerexpire - the timer callout routine
*
* Just propagate the timer event into the knote
* filter routine (by going through the knote
* synchronization point). Pass a hint to
* indicate this is a real event, not just a
* query from above.
*/
static void
filt_timerexpire(void *knx, __unused void *spare)
{
struct klist timer_list;
struct knote *kn = knx;
filt_timerlock();
kn->kn_hookid &= ~TIMER_RUNNING;
/* no "object" for timers, so fake a list */
SLIST_INIT(&timer_list);
SLIST_INSERT_HEAD(&timer_list, kn, kn_selnext);
KNOTE(&timer_list, 1);
/* if someone is waiting for timer to pop */
if (kn->kn_hookid & TIMER_CANCELWAIT) {
struct kqueue *kq = kn->kn_kq;
wait_queue_wakeup_all((wait_queue_t)kq->kq_wqs, &kn->kn_hook,
THREAD_AWAKENED);
}
filt_timerunlock();
}
/*
* Cancel a running timer (or wait for the pop).
* Timer filter lock is held.
*/
static void
filt_timercancel(struct knote *kn)
{
struct kqueue *kq = kn->kn_kq;
thread_call_t callout = kn->kn_hook;
boolean_t cancelled;
if (kn->kn_hookid & TIMER_RUNNING) {
/* cancel the callout if we can */
cancelled = thread_call_cancel(callout);
if (cancelled) {
kn->kn_hookid &= ~TIMER_RUNNING;
} else {
/* we have to wait for the expire routine. */
kn->kn_hookid |= TIMER_CANCELWAIT;
if (!wait_queue_is_valid((myWaitQueue*)kq->kq_wqs)) {
printf("\nAbout to call wait_queue_assert_wait with invalid wait queue from filt_timercancel.\n");
}
wait_queue_assert_wait((wait_queue_t)kq->kq_wqs,
&kn->kn_hook, THREAD_UNINT, 0);
filt_timerunlock();
thread_block(THREAD_CONTINUE_NULL);
filt_timerlock();
assert((kn->kn_hookid & TIMER_RUNNING) == 0);
}
}
}
/*
* Allocate a thread call for the knote's lifetime, and kick off the timer.
*/
static int
filt_timerattach(struct knote *kn)
{
thread_call_t callout;
int error;
callout = thread_call_allocate(filt_timerexpire, kn);
if (NULL == callout)
return (ENOMEM);
filt_timerlock();
error = filt_timervalidate(kn);
if (error != 0) {
filt_timerunlock();
return (error);
}
kn->kn_hook = (void*)callout;
kn->kn_hookid = 0;
/* absolute=EV_ONESHOT */
if (kn->kn_sfflags & NOTE_ABSOLUTE)
kn->kn_flags |= EV_ONESHOT;
filt_timerupdate(kn);
if (kn->kn_ext[0]) {
kn->kn_flags |= EV_CLEAR;
unsigned int timer_flags = 0;
if (kn->kn_sfflags & NOTE_CRITICAL)
timer_flags |= THREAD_CALL_DELAY_USER_CRITICAL;
else if (kn->kn_sfflags & NOTE_BACKGROUND)
timer_flags |= THREAD_CALL_DELAY_USER_BACKGROUND;
else
timer_flags |= THREAD_CALL_DELAY_USER_NORMAL;
if (kn->kn_sfflags & NOTE_LEEWAY)
timer_flags |= THREAD_CALL_DELAY_LEEWAY;
thread_call_enter_delayed_with_leeway(callout, NULL,
kn->kn_ext[0], kn->kn_ext[1], timer_flags);
kn->kn_hookid |= TIMER_RUNNING;
} else {
/* fake immediate */
kn->kn_data = 1;
}
filt_timerunlock();
return (0);
}
/*
* Shut down the timer if it's running, and free the callout.
*/
static void
filt_timerdetach(struct knote *kn)
{
thread_call_t callout;
filt_timerlock();
callout = (thread_call_t)kn->kn_hook;
filt_timercancel(kn);
filt_timerunlock();
thread_call_free(callout);
}
static int
filt_timer(struct knote *kn, long hint)
{
int result;
if (hint) {
/* real timer pop -- timer lock held by filt_timerexpire */
kn->kn_data++;
if (((kn->kn_hookid & TIMER_CANCELWAIT) == 0) &&
((kn->kn_flags & EV_ONESHOT) == 0)) {
/* evaluate next time to fire */
filt_timerupdate(kn);
if (kn->kn_ext[0]) {
unsigned int timer_flags = 0;
/* keep the callout and re-arm */
if (kn->kn_sfflags & NOTE_CRITICAL)
timer_flags |= THREAD_CALL_DELAY_USER_CRITICAL;
else if (kn->kn_sfflags & NOTE_BACKGROUND)
timer_flags |= THREAD_CALL_DELAY_USER_BACKGROUND;
else
timer_flags |= THREAD_CALL_DELAY_USER_NORMAL;
if (kn->kn_sfflags & NOTE_LEEWAY)
timer_flags |= THREAD_CALL_DELAY_LEEWAY;
thread_call_enter_delayed_with_leeway(kn->kn_hook, NULL,
kn->kn_ext[0], kn->kn_ext[1], timer_flags);
kn->kn_hookid |= TIMER_RUNNING;
}
}
return (1);
}
/* user-query */
filt_timerlock();
result = (kn->kn_data != 0);
filt_timerunlock();
return (result);
}
/*
* filt_timertouch - update knote with new user input
*
* Cancel and restart the timer based on new user data. When
* the user picks up a knote, clear the count of how many timer
* pops have gone off (in kn_data).
*/
static void
filt_timertouch(struct knote *kn, struct kevent64_s *kev, long type)
{
int error;
filt_timerlock();
switch (type) {
case EVENT_REGISTER:
/* cancel current call */
filt_timercancel(kn);
/* recalculate deadline */
kn->kn_sdata = kev->data;
kn->kn_sfflags = kev->fflags;
kn->kn_ext[0] = kev->ext[0];
kn->kn_ext[1] = kev->ext[1];
error = filt_timervalidate(kn);
if (error) {
/* no way to report error, so mark it in the knote */
kn->kn_flags |= EV_ERROR;
kn->kn_data = error;
break;
}
/* start timer if necessary */
filt_timerupdate(kn);
if (kn->kn_ext[0]) {
unsigned int timer_flags = 0;
if (kn->kn_sfflags & NOTE_CRITICAL)
timer_flags |= THREAD_CALL_DELAY_USER_CRITICAL;
else if (kn->kn_sfflags & NOTE_BACKGROUND)
timer_flags |= THREAD_CALL_DELAY_USER_BACKGROUND;
else
timer_flags |= THREAD_CALL_DELAY_USER_NORMAL;
if (kn->kn_sfflags & NOTE_LEEWAY)
timer_flags |= THREAD_CALL_DELAY_LEEWAY;
thread_call_enter_delayed_with_leeway(kn->kn_hook, NULL,
kn->kn_ext[0], kn->kn_ext[1], timer_flags);
kn->kn_hookid |= TIMER_RUNNING;
} else {
/* pretend the timer has fired */
kn->kn_data = 1;
}
break;
case EVENT_PROCESS:
/* reset the timer pop count in kn_data */
*kev = kn->kn_kevent;
kev->ext[0] = 0;
kn->kn_data = 0;
if (kn->kn_flags & EV_CLEAR)
kn->kn_fflags = 0;
break;
default:
panic("%s: - invalid type (%ld)", __func__, type);
break;
}
filt_timerunlock();
}
static void
filt_timerlock(void)
{
lck_mtx_lock(&_filt_timerlock);
}
static void
filt_timerunlock(void)
{
lck_mtx_unlock(&_filt_timerlock);
}
static int
filt_userattach(struct knote *kn)
{
/* EVFILT_USER knotes are not attached to anything in the kernel */
kn->kn_hook = NULL;
if (kn->kn_fflags & NOTE_TRIGGER) {
kn->kn_hookid = 1;
} else {
kn->kn_hookid = 0;
}
return (0);
}
static void
filt_userdetach(__unused struct knote *kn)
{
/* EVFILT_USER knotes are not attached to anything in the kernel */
}
static int
filt_user(struct knote *kn, __unused long hint)
{
return (kn->kn_hookid);
}
static void
filt_usertouch(struct knote *kn, struct kevent64_s *kev, long type)
{
uint32_t ffctrl;
switch (type) {
case EVENT_REGISTER:
if (kev->fflags & NOTE_TRIGGER) {
kn->kn_hookid = 1;
}
ffctrl = kev->fflags & NOTE_FFCTRLMASK;
kev->fflags &= NOTE_FFLAGSMASK;
switch (ffctrl) {
case NOTE_FFNOP:
break;
case NOTE_FFAND:
OSBitAndAtomic(kev->fflags, &kn->kn_sfflags);
break;
case NOTE_FFOR:
OSBitOrAtomic(kev->fflags, &kn->kn_sfflags);
break;
case NOTE_FFCOPY:
kn->kn_sfflags = kev->fflags;
break;
}
kn->kn_sdata = kev->data;
break;
case EVENT_PROCESS:
*kev = kn->kn_kevent;
kev->fflags = (volatile UInt32)kn->kn_sfflags;
kev->data = kn->kn_sdata;
if (kn->kn_flags & EV_CLEAR) {
kn->kn_hookid = 0;
kn->kn_data = 0;
kn->kn_fflags = 0;
}
break;
default:
panic("%s: - invalid type (%ld)", __func__, type);
break;
}
}
/*
* JMM - placeholder for not-yet-implemented filters
*/
static int
filt_badattach(__unused struct knote *kn)
{
return (ENOTSUP);
}
struct kqueue *
kqueue_alloc(struct proc *p)
{
struct filedesc *fdp = p->p_fd;
struct kqueue *kq;
MALLOC_ZONE(kq, struct kqueue *, sizeof (struct kqueue), M_KQUEUE,
M_WAITOK);
if (kq != NULL) {
wait_queue_set_t wqs;
wqs = wait_queue_set_alloc(SYNC_POLICY_FIFO |
SYNC_POLICY_PREPOST);
if (wqs != NULL) {
bzero(kq, sizeof (struct kqueue));
lck_spin_init(&kq->kq_lock, kq_lck_grp, kq_lck_attr);
TAILQ_INIT(&kq->kq_head);
kq->kq_wqs = wqs;
kq->kq_p = p;
} else {
FREE_ZONE(kq, sizeof (struct kqueue), M_KQUEUE);
}
}
if (fdp->fd_knlistsize < 0) {
proc_fdlock(p);
if (fdp->fd_knlistsize < 0)
fdp->fd_knlistsize = 0; /* this process has had a kq */
proc_fdunlock(p);
}
return (kq);
}
/*
* kqueue_dealloc - detach all knotes from a kqueue and free it
*
* We walk each list looking for knotes referencing this
* this kqueue. If we find one, we try to drop it. But
* if we fail to get a drop reference, that will wait
* until it is dropped. So, we can just restart again
* safe in the assumption that the list will eventually
* not contain any more references to this kqueue (either
* we dropped them all, or someone else did).
*
* Assumes no new events are being added to the kqueue.
* Nothing locked on entry or exit.
*/
void
kqueue_dealloc(struct kqueue *kq)
{
struct proc *p = kq->kq_p;
struct filedesc *fdp = p->p_fd;
struct knote *kn;
int i;
proc_fdlock(p);
for (i = 0; i < fdp->fd_knlistsize; i++) {
kn = SLIST_FIRST(&fdp->fd_knlist[i]);
while (kn != NULL) {
if (kq == kn->kn_kq) {
kqlock(kq);
proc_fdunlock(p);
/* drop it ourselves or wait */
if (kqlock2knotedrop(kq, kn)) {
kn->kn_fop->f_detach(kn);
knote_drop(kn, p);
}
proc_fdlock(p);
/* start over at beginning of list */
kn = SLIST_FIRST(&fdp->fd_knlist[i]);
continue;
}
kn = SLIST_NEXT(kn, kn_link);
}
}
if (fdp->fd_knhashmask != 0) {
for (i = 0; i < (int)fdp->fd_knhashmask + 1; i++) {
kn = SLIST_FIRST(&fdp->fd_knhash[i]);
while (kn != NULL) {
if (kq == kn->kn_kq) {
kqlock(kq);
proc_fdunlock(p);
/* drop it ourselves or wait */
if (kqlock2knotedrop(kq, kn)) {
kn->kn_fop->f_detach(kn);
knote_drop(kn, p);
}
proc_fdlock(p);
/* start over at beginning of list */
kn = SLIST_FIRST(&fdp->fd_knhash[i]);
continue;
}
kn = SLIST_NEXT(kn, kn_link);
}
}
}
proc_fdunlock(p);
/*
* before freeing the wait queue set for this kqueue,
* make sure it is unlinked from all its containing (select) sets.
*/
wait_queue_unlink_all((wait_queue_t)kq->kq_wqs);
wait_queue_set_free(kq->kq_wqs);
lck_spin_destroy(&kq->kq_lock, kq_lck_grp);
FREE_ZONE(kq, sizeof (struct kqueue), M_KQUEUE);
}
int
kqueue_body(struct proc *p, fp_allocfn_t fp_zalloc, void *cra, int32_t *retval)
{
struct kqueue *kq;
struct fileproc *fp;
int fd, error;
error = falloc_withalloc(p,
&fp, &fd, vfs_context_current(), fp_zalloc, cra);
if (error) {
return (error);
}
kq = kqueue_alloc(p);
if (kq == NULL) {
fp_free(p, fd, fp);
return (ENOMEM);
}
fp->f_flag = FREAD | FWRITE;
fp->f_ops = &kqueueops;
fp->f_data = kq;
proc_fdlock(p);
*fdflags(p, fd) |= UF_EXCLOSE;
procfdtbl_releasefd(p, fd, NULL);
fp_drop(p, fd, fp, 1);
proc_fdunlock(p);
*retval = fd;
return (error);
}
int
kqueue(struct proc *p, __unused struct kqueue_args *uap, int32_t *retval)
{
return (kqueue_body(p, fileproc_alloc_init, NULL, retval));
}
static int
kevent_copyin(user_addr_t *addrp, struct kevent64_s *kevp, struct proc *p,
int iskev64)
{
int advance;
int error;
if (iskev64) {
advance = sizeof (struct kevent64_s);
error = copyin(*addrp, (caddr_t)kevp, advance);
} else if (IS_64BIT_PROCESS(p)) {
struct user64_kevent kev64;
bzero(kevp, sizeof (struct kevent64_s));
advance = sizeof (kev64);
error = copyin(*addrp, (caddr_t)&kev64, advance);
if (error)
return (error);
kevp->ident = kev64.ident;
kevp->filter = kev64.filter;
kevp->flags = kev64.flags;
kevp->fflags = kev64.fflags;
kevp->data = kev64.data;
kevp->udata = kev64.udata;
} else {
struct user32_kevent kev32;
bzero(kevp, sizeof (struct kevent64_s));
advance = sizeof (kev32);
error = copyin(*addrp, (caddr_t)&kev32, advance);
if (error)
return (error);
kevp->ident = (uintptr_t)kev32.ident;
kevp->filter = kev32.filter;
kevp->flags = kev32.flags;
kevp->fflags = kev32.fflags;
kevp->data = (intptr_t)kev32.data;
kevp->udata = CAST_USER_ADDR_T(kev32.udata);
}
if (!error)
*addrp += advance;
return (error);
}
static int
kevent_copyout(struct kevent64_s *kevp, user_addr_t *addrp, struct proc *p,
int iskev64)
{
int advance;
int error;
if (iskev64) {
advance = sizeof (struct kevent64_s);
error = copyout((caddr_t)kevp, *addrp, advance);
} else if (IS_64BIT_PROCESS(p)) {
struct user64_kevent kev64;
/*
* deal with the special case of a user-supplied
* value of (uintptr_t)-1.
*/
kev64.ident = (kevp->ident == (uintptr_t)-1) ?
(uint64_t)-1LL : (uint64_t)kevp->ident;
kev64.filter = kevp->filter;
kev64.flags = kevp->flags;
kev64.fflags = kevp->fflags;
kev64.data = (int64_t) kevp->data;
kev64.udata = kevp->udata;
advance = sizeof (kev64);
error = copyout((caddr_t)&kev64, *addrp, advance);
} else {
struct user32_kevent kev32;
kev32.ident = (uint32_t)kevp->ident;
kev32.filter = kevp->filter;
kev32.flags = kevp->flags;
kev32.fflags = kevp->fflags;
kev32.data = (int32_t)kevp->data;
kev32.udata = kevp->udata;
advance = sizeof (kev32);
error = copyout((caddr_t)&kev32, *addrp, advance);
}
if (!error)
*addrp += advance;
return (error);
}
/*
* kevent_continue - continue a kevent syscall after blocking
*
* assume we inherit a use count on the kq fileglob.
*/
static void
kevent_continue(__unused struct kqueue *kq, void *data, int error)
{
struct _kevent *cont_args;
struct fileproc *fp;
int32_t *retval;
int noutputs;
int fd;
struct proc *p = current_proc();
cont_args = (struct _kevent *)data;
noutputs = cont_args->eventout;
retval = cont_args->retval;
fd = cont_args->fd;
fp = cont_args->fp;
fp_drop(p, fd, fp, 0);
/* don't restart after signals... */
if (error == ERESTART)
error = EINTR;
else if (error == EWOULDBLOCK)
error = 0;
if (error == 0)
*retval = noutputs;
unix_syscall_return(error);
}
/*
* kevent - [syscall] register and wait for kernel events
*
*/
int
kevent(struct proc *p, struct kevent_args *uap, int32_t *retval)
{
return (kevent_internal(p,
0,
uap->changelist,
uap->nchanges,
uap->eventlist,
uap->nevents,
uap->fd,
uap->timeout,
0, /* no flags from old kevent() call */
retval));
}
int
kevent64(struct proc *p, struct kevent64_args *uap, int32_t *retval)
{
return (kevent_internal(p,
1,
uap->changelist,
uap->nchanges,
uap->eventlist,
uap->nevents,
uap->fd,
uap->timeout,
uap->flags,
retval));
}
static int
kevent_internal(struct proc *p, int iskev64, user_addr_t changelist,
int nchanges, user_addr_t ueventlist, int nevents, int fd,
user_addr_t utimeout, __unused unsigned int flags,
int32_t *retval)
{
struct _kevent *cont_args;
uthread_t ut;
struct kqueue *kq;
struct fileproc *fp;
struct kevent64_s kev;
int error, noutputs;
struct timeval atv;
/* convert timeout to absolute - if we have one */
if (utimeout != USER_ADDR_NULL) {
struct timeval rtv;
if (IS_64BIT_PROCESS(p)) {
struct user64_timespec ts;
error = copyin(utimeout, &ts, sizeof(ts));
if ((ts.tv_sec & 0xFFFFFFFF00000000ull) != 0)
error = EINVAL;
else
TIMESPEC_TO_TIMEVAL(&rtv, &ts);
} else {
struct user32_timespec ts;
error = copyin(utimeout, &ts, sizeof(ts));
TIMESPEC_TO_TIMEVAL(&rtv, &ts);
}
if (error)
return (error);
if (itimerfix(&rtv))
return (EINVAL);
getmicrouptime(&atv);
timevaladd(&atv, &rtv);
} else {
atv.tv_sec = 0;
atv.tv_usec = 0;
}
/* get a usecount for the kq itself */
if ((error = fp_getfkq(p, fd, &fp, &kq)) != 0)
return (error);
/* each kq should only be used for events of one type */
kqlock(kq);
if (kq->kq_state & (KQ_KEV32 | KQ_KEV64)) {
if (((iskev64 && (kq->kq_state & KQ_KEV32)) ||
(!iskev64 && (kq->kq_state & KQ_KEV64)))) {
error = EINVAL;
kqunlock(kq);
goto errorout;
}
} else {
kq->kq_state |= (iskev64 ? KQ_KEV64 : KQ_KEV32);
}
kqunlock(kq);
/* register all the change requests the user provided... */
noutputs = 0;
while (nchanges > 0 && error == 0) {
error = kevent_copyin(&changelist, &kev, p, iskev64);
if (error)
break;
kev.flags &= ~EV_SYSFLAGS;
error = kevent_register(kq, &kev, p);
if ((error || (kev.flags & EV_RECEIPT)) && nevents > 0) {
kev.flags = EV_ERROR;
kev.data = error;
error = kevent_copyout(&kev, &ueventlist, p, iskev64);
if (error == 0) {
nevents--;
noutputs++;
}
}
nchanges--;
}
/* store the continuation/completion data in the uthread */
ut = (uthread_t)get_bsdthread_info(current_thread());
cont_args = &ut->uu_kevent.ss_kevent;
cont_args->fp = fp;
cont_args->fd = fd;
cont_args->retval = retval;
cont_args->eventlist = ueventlist;
cont_args->eventcount = nevents;
cont_args->eventout = noutputs;
cont_args->eventsize = iskev64;
if (nevents > 0 && noutputs == 0 && error == 0)
error = kqueue_scan(kq, kevent_callback,
kevent_continue, cont_args,
&atv, p);
kevent_continue(kq, cont_args, error);
errorout:
fp_drop(p, fd, fp, 0);
return (error);
}
/*
* kevent_callback - callback for each individual event
*
* called with nothing locked
* caller holds a reference on the kqueue
*/
static int
kevent_callback(__unused struct kqueue *kq, struct kevent64_s *kevp,
void *data)
{
struct _kevent *cont_args;
int error;
int iskev64;
cont_args = (struct _kevent *)data;
assert(cont_args->eventout < cont_args->eventcount);
iskev64 = cont_args->eventsize;
/*
* Copy out the appropriate amount of event data for this user.
*/
error = kevent_copyout(kevp, &cont_args->eventlist, current_proc(),
iskev64);
/*
* If there isn't space for additional events, return
* a harmless error to stop the processing here
*/
if (error == 0 && ++cont_args->eventout == cont_args->eventcount)
error = EWOULDBLOCK;
return (error);
}
/*
* kevent_description - format a description of a kevent for diagnostic output
*
* called with a 128-byte string buffer
*/
char *
kevent_description(struct kevent64_s *kevp, char *s, size_t n)
{
snprintf(s, n,
"kevent="
"{.ident=%#llx, .filter=%d, .flags=%#x, .fflags=%#x, .data=%#llx, .udata=%#llx, .ext[0]=%#llx, .ext[1]=%#llx}",
kevp->ident,
kevp->filter,
kevp->flags,
kevp->fflags,
kevp->data,
kevp->udata,
kevp->ext[0],
kevp->ext[1]);
return (s);
}
/*
* kevent_register - add a new event to a kqueue
*
* Creates a mapping between the event source and
* the kqueue via a knote data structure.
*
* Because many/most the event sources are file
* descriptor related, the knote is linked off
* the filedescriptor table for quick access.
*
* called with nothing locked
* caller holds a reference on the kqueue
*/
int
kevent_register(struct kqueue *kq, struct kevent64_s *kev,
__unused struct proc *ctxp)
{
struct proc *p = kq->kq_p;
struct filedesc *fdp = p->p_fd;
struct filterops *fops;
struct fileproc *fp = NULL;
struct knote *kn = NULL;
int error = 0;
if (kev->filter < 0) {
if (kev->filter + EVFILT_SYSCOUNT < 0)
return (EINVAL);
fops = sysfilt_ops[~kev->filter]; /* to 0-base index */
} else {
/*
* XXX
* filter attach routine is responsible for insuring that
* the identifier can be attached to it.
*/
printf("unknown filter: %d\n", kev->filter);
return (EINVAL);
}
restart:
/* this iocount needs to be dropped if it is not registered */
proc_fdlock(p);
if (fops->f_isfd && (error = fp_lookup(p, kev->ident, &fp, 1)) != 0) {
proc_fdunlock(p);
return (error);
}
if (fops->f_isfd) {
/* fd-based knotes are linked off the fd table */
if (kev->ident < (u_int)fdp->fd_knlistsize) {
SLIST_FOREACH(kn, &fdp->fd_knlist[kev->ident], kn_link)
if (kq == kn->kn_kq &&
kev->filter == kn->kn_filter)
break;
}
} else {
/* hash non-fd knotes here too */
if (fdp->fd_knhashmask != 0) {
struct klist *list;
list = &fdp->fd_knhash[
KN_HASH((u_long)kev->ident, fdp->fd_knhashmask)];
SLIST_FOREACH(kn, list, kn_link)
if (kev->ident == kn->kn_id &&
kq == kn->kn_kq &&
kev->filter == kn->kn_filter)
break;
}
}
/*
* kn now contains the matching knote, or NULL if no match
*/
if (kn == NULL) {
if ((kev->flags & (EV_ADD|EV_DELETE)) == EV_ADD) {
kn = knote_alloc();
if (kn == NULL) {
proc_fdunlock(p);
error = ENOMEM;
goto done;
}
kn->kn_fp = fp;
kn->kn_kq = kq;
kn->kn_tq = &kq->kq_head;
kn->kn_fop = fops;
kn->kn_sfflags = kev->fflags;
kn->kn_sdata = kev->data;
kev->fflags = 0;
kev->data = 0;
kn->kn_kevent = *kev;
kn->kn_inuse = 1; /* for f_attach() */
kn->kn_status = KN_ATTACHING;
/* before anyone can find it */
if (kev->flags & EV_DISABLE)
kn->kn_status |= KN_DISABLED;
error = knote_fdpattach(kn, fdp, p);
proc_fdunlock(p);
if (error) {
knote_free(kn);
goto done;
}
/*
* apply reference count to knote structure, and
* do not release it at the end of this routine.
*/
fp = NULL;
error = fops->f_attach(kn);
kqlock(kq);
if (error != 0) {
/*
* Failed to attach correctly, so drop.
* All other possible users/droppers
* have deferred to us.
*/
kn->kn_status |= KN_DROPPING;
kqunlock(kq);
knote_drop(kn, p);
goto done;
} else if (kn->kn_status & KN_DROPPING) {
/*
* Attach succeeded, but someone else
* deferred their drop - now we have
* to do it for them (after detaching).
*/
kqunlock(kq);
kn->kn_fop->f_detach(kn);
knote_drop(kn, p);
goto done;
}
kn->kn_status &= ~KN_ATTACHING;
kqunlock(kq);
} else {
proc_fdunlock(p);
error = ENOENT;
goto done;
}
} else {
/* existing knote - get kqueue lock */
kqlock(kq);
proc_fdunlock(p);
if (kev->flags & EV_DELETE) {
knote_dequeue(kn);
kn->kn_status |= KN_DISABLED;
if (kqlock2knotedrop(kq, kn)) {
kn->kn_fop->f_detach(kn);
knote_drop(kn, p);
}
goto done;
}
/* update status flags for existing knote */
if (kev->flags & EV_DISABLE) {
knote_dequeue(kn);
kn->kn_status |= KN_DISABLED;
} else if (kev->flags & EV_ENABLE) {
kn->kn_status &= ~KN_DISABLED;
if (kn->kn_status & KN_ACTIVE)
knote_enqueue(kn);
}
/*
* The user may change some filter values after the
* initial EV_ADD, but doing so will not reset any
* filter which have already been triggered.
*/
kn->kn_kevent.udata = kev->udata;
if (fops->f_isfd || fops->f_touch == NULL) {
kn->kn_sfflags = kev->fflags;
kn->kn_sdata = kev->data;
}
/*
* If somebody is in the middle of dropping this
* knote - go find/insert a new one. But we have
* wait for this one to go away first. Attaches
* running in parallel may also drop/modify the
* knote. Wait for those to complete as well and
* then start over if we encounter one.
*/
if (!kqlock2knoteusewait(kq, kn)) {
/* kqueue, proc_fdlock both unlocked */
goto restart;
}
/*
* Call touch routine to notify filter of changes
* in filter values.
*/
if (!fops->f_isfd && fops->f_touch != NULL)
fops->f_touch(kn, kev, EVENT_REGISTER);
}
/* still have use ref on knote */
/*
* If the knote is not marked to always stay enqueued,
* invoke the filter routine to see if it should be
* enqueued now.
*/
if ((kn->kn_status & KN_STAYQUEUED) == 0 && kn->kn_fop->f_event(kn, 0)) {
if (knoteuse2kqlock(kq, kn))
knote_activate(kn, 1);
kqunlock(kq);
} else {
knote_put(kn);
}
done:
if (fp != NULL)
fp_drop(p, kev->ident, fp, 0);
return (error);
}
/*
* knote_process - process a triggered event
*
* Validate that it is really still a triggered event
* by calling the filter routines (if necessary). Hold
* a use reference on the knote to avoid it being detached.
* If it is still considered triggered, invoke the callback
* routine provided and move it to the provided inprocess
* queue.
*
* caller holds a reference on the kqueue.
* kqueue locked on entry and exit - but may be dropped
*/
static int
knote_process(struct knote *kn,
kevent_callback_t callback,
void *data,
struct kqtailq *inprocessp,
struct proc *p)
{
struct kqueue *kq = kn->kn_kq;
struct kevent64_s kev;
int touch;
int result;
int error;
/*
* Determine the kevent state we want to return.
*
* Some event states need to be revalidated before returning
* them, others we take the snapshot at the time the event
* was enqueued.
*
* Events with non-NULL f_touch operations must be touched.
* Triggered events must fill in kev for the callback.
*
* Convert our lock to a use-count and call the event's
* filter routine(s) to update.
*/
if ((kn->kn_status & KN_DISABLED) != 0) {
result = 0;
touch = 0;
} else {
int revalidate;
result = 1;
revalidate = ((kn->kn_status & KN_STAYQUEUED) != 0 ||
(kn->kn_flags & EV_ONESHOT) == 0);
touch = (!kn->kn_fop->f_isfd && kn->kn_fop->f_touch != NULL);
if (revalidate || touch) {
if (revalidate)
knote_deactivate(kn);
/* call the filter/touch routines with just a ref */
if (kqlock2knoteuse(kq, kn)) {
/* if we have to revalidate, call the filter */
if (revalidate) {
result = kn->kn_fop->f_event(kn, 0);
}
/*
* capture the kevent data - using touch if
* specified
*/
if (result && touch) {
kn->kn_fop->f_touch(kn, &kev,
EVENT_PROCESS);
}
/*
* convert back to a kqlock - bail if the knote
* went away
*/
if (!knoteuse2kqlock(kq, kn)) {
return (EJUSTRETURN);
} else if (result) {
/*
* if revalidated as alive, make sure
* it's active
*/
if (!(kn->kn_status & KN_ACTIVE)) {
knote_activate(kn, 0);
}
/*
* capture all events that occurred
* during filter
*/
if (!touch) {
kev = kn->kn_kevent;
}
} else if ((kn->kn_status & KN_STAYQUEUED) == 0) {
/*
* was already dequeued, so just bail on
* this one
*/
return (EJUSTRETURN);
}
} else {
return (EJUSTRETURN);
}
} else {
kev = kn->kn_kevent;
}
}
/* move knote onto inprocess queue */
assert(kn->kn_tq == &kq->kq_head);
TAILQ_REMOVE(&kq->kq_head, kn, kn_tqe);
kn->kn_tq = inprocessp;
TAILQ_INSERT_TAIL(inprocessp, kn, kn_tqe);
/*
* Determine how to dispatch the knote for future event handling.
* not-fired: just return (do not callout).
* One-shot: deactivate it.
* Clear: deactivate and clear the state.
* Dispatch: don't clear state, just deactivate it and mark it disabled.
* All others: just leave where they are.
*/
if (result == 0) {
return (EJUSTRETURN);
} else if ((kn->kn_flags & EV_ONESHOT) != 0) {
knote_deactivate(kn);
if (kqlock2knotedrop(kq, kn)) {
kn->kn_fop->f_detach(kn);
knote_drop(kn, p);
}
} else if ((kn->kn_flags & (EV_CLEAR | EV_DISPATCH)) != 0) {
if ((kn->kn_flags & EV_DISPATCH) != 0) {
/* deactivate and disable all dispatch knotes */
knote_deactivate(kn);
kn->kn_status |= KN_DISABLED;
} else if (!touch || kn->kn_fflags == 0) {
/* only deactivate if nothing since the touch */
knote_deactivate(kn);
}
if (!touch && (kn->kn_flags & EV_CLEAR) != 0) {
/* manually clear non-touch knotes */
kn->kn_data = 0;
kn->kn_fflags = 0;
}
kqunlock(kq);
} else {
/*
* leave on inprocess queue. We'll
* move all the remaining ones back
* the kq queue and wakeup any
* waiters when we are done.
*/
kqunlock(kq);
}
/* callback to handle each event as we find it */
error = (callback)(kq, &kev, data);
kqlock(kq);
return (error);
}
/*
* Return 0 to indicate that processing should proceed,
* -1 if there is nothing to process.
*
* Called with kqueue locked and returns the same way,
* but may drop lock temporarily.
*/
static int
kqueue_begin_processing(struct kqueue *kq)
{
for (;;) {
if (kq->kq_count == 0) {
return (-1);
}
/* if someone else is processing the queue, wait */
if (kq->kq_nprocess != 0) {
if (!wait_queue_is_valid((myWaitQueue*)kq->kq_wqs)) {
printf("\nAbout to call wait_queue_assert_wait with invalid wait queue from kqueue_begin_processing.\n");
}
wait_queue_assert_wait((wait_queue_t)kq->kq_wqs,
&kq->kq_nprocess, THREAD_UNINT, 0);
kq->kq_state |= KQ_PROCWAIT;
kqunlock(kq);
thread_block(THREAD_CONTINUE_NULL);
kqlock(kq);
} else {
kq->kq_nprocess = 1;
return (0);
}
}
}
/*
* Called with kqueue lock held.
*/
static void
kqueue_end_processing(struct kqueue *kq)
{
kq->kq_nprocess = 0;
if (kq->kq_state & KQ_PROCWAIT) {
kq->kq_state &= ~KQ_PROCWAIT;
wait_queue_wakeup_all((wait_queue_t)kq->kq_wqs,
&kq->kq_nprocess, THREAD_AWAKENED);
}
}
/*
* kqueue_process - process the triggered events in a kqueue
*
* Walk the queued knotes and validate that they are
* really still triggered events by calling the filter
* routines (if necessary). Hold a use reference on
* the knote to avoid it being detached. For each event
* that is still considered triggered, invoke the
* callback routine provided.
*
* caller holds a reference on the kqueue.
* kqueue locked on entry and exit - but may be dropped
* kqueue list locked (held for duration of call)
*/
static int
kqueue_process(struct kqueue *kq,
kevent_callback_t callback,
void *data,
int *countp,
struct proc *p)
{
struct kqtailq inprocess;
struct knote *kn;
int nevents;
int error;
TAILQ_INIT(&inprocess);
if (kqueue_begin_processing(kq) == -1) {
*countp = 0;
/* Nothing to process */
return (0);
}
/*
* Clear any pre-posted status from previous runs, so we
* only detect events that occur during this run.
*/
wait_queue_sub_clearrefs(kq->kq_wqs);
/*
* loop through the enqueued knotes, processing each one and
* revalidating those that need it. As they are processed,
* they get moved to the inprocess queue (so the loop can end).
*/
error = 0;
nevents = 0;
while (error == 0 &&
(kn = TAILQ_FIRST(&kq->kq_head)) != NULL) {
error = knote_process(kn, callback, data, &inprocess, p);
if (error == EJUSTRETURN)
error = 0;
else
nevents++;
}
/*
* With the kqueue still locked, move any knotes
* remaining on the inprocess queue back to the
* kq's queue and wake up any waiters.
*/
while ((kn = TAILQ_FIRST(&inprocess)) != NULL) {
assert(kn->kn_tq == &inprocess);
TAILQ_REMOVE(&inprocess, kn, kn_tqe);
kn->kn_tq = &kq->kq_head;
TAILQ_INSERT_TAIL(&kq->kq_head, kn, kn_tqe);
}
kqueue_end_processing(kq);
*countp = nevents;
return (error);
}
static void
kqueue_scan_continue(void *data, wait_result_t wait_result)
{
thread_t self = current_thread();
uthread_t ut = (uthread_t)get_bsdthread_info(self);
struct _kqueue_scan * cont_args = &ut->uu_kevent.ss_kqueue_scan;
struct kqueue *kq = (struct kqueue *)data;
int error;
int count;
/* convert the (previous) wait_result to a proper error */
switch (wait_result) {
case THREAD_AWAKENED:
kqlock(kq);
error = kqueue_process(kq, cont_args->call, cont_args, &count,
current_proc());
if (error == 0 && count == 0) {
if (!wait_queue_is_valid((myWaitQueue*)kq->kq_wqs)) {
printf("\nAbout to call wait_queue_assert_wait with invalid wait queue from kqueue_scan_continue.\n");
}
wait_queue_assert_wait((wait_queue_t)kq->kq_wqs,
KQ_EVENT, THREAD_ABORTSAFE, cont_args->deadline);
kq->kq_state |= KQ_SLEEP;
kqunlock(kq);
thread_block_parameter(kqueue_scan_continue, kq);
/* NOTREACHED */
}
kqunlock(kq);
break;
case THREAD_TIMED_OUT:
error = EWOULDBLOCK;
break;
case THREAD_INTERRUPTED:
error = EINTR;
break;
case THREAD_RESTART:
printf("\nkqueue_scan_continue was called with a wait_result of THREAD_RESTART. A vanilla 2422 XNU kernel would have panicked!\n");
error = EBADF;
break;
default:
panic("%s: - invalid wait_result (%d)", __func__,
wait_result);
error = 0;
}
/* call the continuation with the results */
assert(cont_args->cont != NULL);
(cont_args->cont)(kq, cont_args->data, error);
}
/*
* kqueue_scan - scan and wait for events in a kqueue
*
* Process the triggered events in a kqueue.
*
* If there are no events triggered arrange to
* wait for them. If the caller provided a
* continuation routine, then kevent_scan will
* also.
*
* The callback routine must be valid.
* The caller must hold a use-count reference on the kq.
*/
int
kqueue_scan(struct kqueue *kq,
kevent_callback_t callback,
kqueue_continue_t continuation,
void *data,
struct timeval *atvp,
struct proc *p)
{
thread_continue_t cont = THREAD_CONTINUE_NULL;
uint64_t deadline;
int error;
int first;
assert(callback != NULL);
first = 1;
for (;;) {
wait_result_t wait_result;
int count;
/*
* Make a pass through the kq to find events already
* triggered.
*/
kqlock(kq);
error = kqueue_process(kq, callback, data, &count, p);
if (error || count)
break; /* lock still held */
/* looks like we have to consider blocking */
if (first) {
first = 0;
/* convert the timeout to a deadline once */
if (atvp->tv_sec || atvp->tv_usec) {
uint64_t now;
clock_get_uptime(&now);
nanoseconds_to_absolutetime((uint64_t)atvp->tv_sec * NSEC_PER_SEC +
atvp->tv_usec * (long)NSEC_PER_USEC,
&deadline);
if (now >= deadline) {
/* non-blocking call */
error = EWOULDBLOCK;
break; /* lock still held */
}
deadline -= now;
clock_absolutetime_interval_to_deadline(deadline, &deadline);
} else {
deadline = 0; /* block forever */
}
if (continuation) {
uthread_t ut = (uthread_t)get_bsdthread_info(current_thread());
struct _kqueue_scan *cont_args = &ut->uu_kevent.ss_kqueue_scan;
cont_args->call = callback;
cont_args->cont = continuation;
cont_args->deadline = deadline;
cont_args->data = data;
cont = kqueue_scan_continue;
}
}
/* go ahead and wait */
if (!wait_queue_is_valid((myWaitQueue*)kq->kq_wqs)) {
printf("\nAbout to call wait_queue_assert_wait_with_leeway with invalid wait queue from kqueue_scan.\n");
}
wait_queue_assert_wait_with_leeway((wait_queue_t)kq->kq_wqs,
KQ_EVENT, THREAD_ABORTSAFE, TIMEOUT_URGENCY_USER_NORMAL,
deadline, 0);
kq->kq_state |= KQ_SLEEP;
kqunlock(kq);
wait_result = thread_block_parameter(cont, kq);
/* NOTREACHED if (continuation != NULL) */
switch (wait_result) {
case THREAD_AWAKENED:
continue;
case THREAD_TIMED_OUT:
return (EWOULDBLOCK);
case THREAD_INTERRUPTED:
return (EINTR);
default:
panic("%s: - bad wait_result (%d)", __func__,
wait_result);
error = 0;
}
}
kqunlock(kq);
return (error);
}
/*
* XXX
* This could be expanded to call kqueue_scan, if desired.
*/
/*ARGSUSED*/
static int
kqueue_read(__unused struct fileproc *fp,
__unused struct uio *uio,
__unused int flags,
__unused vfs_context_t ctx)
{
return (ENXIO);
}
/*ARGSUSED*/
static int
kqueue_write(__unused struct fileproc *fp,
__unused struct uio *uio,
__unused int flags,
__unused vfs_context_t ctx)
{
return (ENXIO);
}
/*ARGSUSED*/
static int
kqueue_ioctl(__unused struct fileproc *fp,
__unused u_long com,
__unused caddr_t data,
__unused vfs_context_t ctx)
{
return (ENOTTY);
}
/*ARGSUSED*/
static int
kqueue_select(struct fileproc *fp, int which, void *wql,
__unused vfs_context_t ctx)
{
struct kqueue *kq = (struct kqueue *)fp->f_data;
struct knote *kn;
struct kqtailq inprocessq;
int retnum = 0;
if (which != FREAD)
return (0);
TAILQ_INIT(&inprocessq);
kqlock(kq);
/*
* If this is the first pass, link the wait queue associated with the
* the kqueue onto the wait queue set for the select(). Normally we
* use selrecord() for this, but it uses the wait queue within the
* selinfo structure and we need to use the main one for the kqueue to
* catch events from KN_STAYQUEUED sources. So we do the linkage manually.
* (The select() call will unlink them when it ends).
*/
if (wql != NULL) {
thread_t cur_act = current_thread();
struct uthread * ut = get_bsdthread_info(cur_act);
kq->kq_state |= KQ_SEL;
wait_queue_link_noalloc((wait_queue_t)kq->kq_wqs, ut->uu_wqset,
(wait_queue_link_t)wql);
}
if (kqueue_begin_processing(kq) == -1) {
kqunlock(kq);
return (0);
}
if (kq->kq_count != 0) {
/*
* there is something queued - but it might be a
* KN_STAYQUEUED knote, which may or may not have
* any events pending. So, we have to walk the
* list of knotes to see, and peek at the stay-
* queued ones to be really sure.
*/
while ((kn = (struct knote *)TAILQ_FIRST(&kq->kq_head)) != NULL) {
if ((kn->kn_status & KN_STAYQUEUED) == 0) {
retnum = 1;
goto out;
}
TAILQ_REMOVE(&kq->kq_head, kn, kn_tqe);
TAILQ_INSERT_TAIL(&inprocessq, kn, kn_tqe);
if (kqlock2knoteuse(kq, kn)) {
unsigned peek;
peek = kn->kn_fop->f_peek(kn);
if (knoteuse2kqlock(kq, kn)) {
if (peek > 0) {
retnum = 1;
goto out;
}
} else {
retnum = 0;
}
}
}
}
out:
/* Return knotes to active queue */
while ((kn = TAILQ_FIRST(&inprocessq)) != NULL) {
TAILQ_REMOVE(&inprocessq, kn, kn_tqe);
kn->kn_tq = &kq->kq_head;
TAILQ_INSERT_TAIL(&kq->kq_head, kn, kn_tqe);
}
kqueue_end_processing(kq);
kqunlock(kq);
return (retnum);
}
/*
* kqueue_close -
*/
/*ARGSUSED*/
static int
kqueue_close(struct fileglob *fg, __unused vfs_context_t ctx)
{
struct kqueue *kq = (struct kqueue *)fg->fg_data;
kqueue_dealloc(kq);
fg->fg_data = NULL;
return (0);
}
/*ARGSUSED*/
/*
* The callers has taken a use-count reference on this kqueue and will donate it
* to the kqueue we are being added to. This keeps the kqueue from closing until
* that relationship is torn down.
*/
static int
kqueue_kqfilter(__unused struct fileproc *fp, struct knote *kn, __unused vfs_context_t ctx)
{
struct kqueue *kq = (struct kqueue *)kn->kn_fp->f_data;
struct kqueue *parentkq = kn->kn_kq;
if (parentkq == kq ||
kn->kn_filter != EVFILT_READ)
return (1);
/*
* We have to avoid creating a cycle when nesting kqueues
* inside another. Rather than trying to walk the whole
* potential DAG of nested kqueues, we just use a simple
* ceiling protocol. When a kqueue is inserted into another,
* we check that the (future) parent is not already nested
* into another kqueue at a lower level than the potenial
* child (because it could indicate a cycle). If that test
* passes, we just mark the nesting levels accordingly.
*/
kqlock(parentkq);
if (parentkq->kq_level > 0 &&
parentkq->kq_level < kq->kq_level)
{
kqunlock(parentkq);
return (1);
} else {
/* set parent level appropriately */
if (parentkq->kq_level == 0)
parentkq->kq_level = 2;
if (parentkq->kq_level < kq->kq_level + 1)
parentkq->kq_level = kq->kq_level + 1;
kqunlock(parentkq);
kn->kn_fop = &kqread_filtops;
kqlock(kq);
KNOTE_ATTACH(&kq->kq_sel.si_note, kn);
/* indicate nesting in child, if needed */
if (kq->kq_level == 0)
kq->kq_level = 1;
kqunlock(kq);
return (0);
}
}
/*
* kqueue_drain - called when kq is closed
*/
/*ARGSUSED*/
static int
kqueue_drain(struct fileproc *fp, __unused vfs_context_t ctx)
{
struct kqueue *kq = (struct kqueue *)fp->f_fglob->fg_data;
kqlock(kq);
kqueue_wakeup(kq, 1);
kqunlock(kq);
return (0);
}
/*ARGSUSED*/
int
kqueue_stat(struct fileproc *fp, void *ub, int isstat64, __unused vfs_context_t ctx)
{
struct kqueue *kq = (struct kqueue *)fp->f_data;
if (isstat64 != 0) {
struct stat64 *sb64 = (struct stat64 *)ub;
bzero((void *)sb64, sizeof(*sb64));
sb64->st_size = kq->kq_count;
if (kq->kq_state & KQ_KEV64)
sb64->st_blksize = sizeof(struct kevent64_s);
else
sb64->st_blksize = sizeof(struct kevent);
sb64->st_mode = S_IFIFO;
} else {
struct stat *sb = (struct stat *)ub;
bzero((void *)sb, sizeof(*sb));
sb->st_size = kq->kq_count;
if (kq->kq_state & KQ_KEV64)
sb->st_blksize = sizeof(struct kevent64_s);
else
sb->st_blksize = sizeof(struct kevent);
sb->st_mode = S_IFIFO;
}
return (0);
}
/*
* Called with the kqueue locked
*/
static void
kqueue_wakeup(struct kqueue *kq, int closed)
{
if ((kq->kq_state & (KQ_SLEEP | KQ_SEL)) != 0 || kq->kq_nprocess > 0) {
kq->kq_state &= ~(KQ_SLEEP | KQ_SEL);
wait_queue_wakeup_all((wait_queue_t)kq->kq_wqs, KQ_EVENT,
(closed) ? THREAD_INTERRUPTED : THREAD_AWAKENED);
}
}
void
klist_init(struct klist *list)
{
SLIST_INIT(list);
}
/*
* Query/Post each knote in the object's list
*
* The object lock protects the list. It is assumed
* that the filter/event routine for the object can
* determine that the object is already locked (via
* the hint) and not deadlock itself.
*
* The object lock should also hold off pending
* detach/drop operations. But we'll prevent it here
* too - just in case.
*/
void
knote(struct klist *list, long hint)
{
struct knote *kn;
SLIST_FOREACH(kn, list, kn_selnext) {
struct kqueue *kq = kn->kn_kq;
kqlock(kq);
if (kqlock2knoteuse(kq, kn)) {
int result;
/* call the event with only a use count */
result = kn->kn_fop->f_event(kn, hint);
/* if its not going away and triggered */
if (knoteuse2kqlock(kq, kn) && result)
knote_activate(kn, 1);
/* lock held again */
}
kqunlock(kq);
}
}
/*
* attach a knote to the specified list. Return true if this is the first entry.
* The list is protected by whatever lock the object it is associated with uses.
*/
int
knote_attach(struct klist *list, struct knote *kn)
{
int ret = SLIST_EMPTY(list);
SLIST_INSERT_HEAD(list, kn, kn_selnext);
return (ret);
}
/*
* detach a knote from the specified list. Return true if that was the last entry.
* The list is protected by whatever lock the object it is associated with uses.
*/
int
knote_detach(struct klist *list, struct knote *kn)
{
SLIST_REMOVE(list, kn, knote, kn_selnext);
return (SLIST_EMPTY(list));
}
/*
* For a given knote, link a provided wait queue directly with the kqueue.
* Wakeups will happen via recursive wait queue support. But nothing will move
* the knote to the active list at wakeup (nothing calls knote()). Instead,
* we permanently enqueue them here.
*
* kqueue and knote references are held by caller.
*
* caller provides the wait queue link structure.
*/
int
knote_link_wait_queue(struct knote *kn, struct wait_queue *wq, wait_queue_link_t wql)
{
struct kqueue *kq = kn->kn_kq;
kern_return_t kr;
kr = wait_queue_link_noalloc(wq, kq->kq_wqs, wql);
if (kr == KERN_SUCCESS) {
knote_markstayqueued(kn);
return (0);
} else {
return (EINVAL);
}
}
/*
* Unlink the provided wait queue from the kqueue associated with a knote.
* Also remove it from the magic list of directly attached knotes.
*
* Note that the unlink may have already happened from the other side, so
* ignore any failures to unlink and just remove it from the kqueue list.
*
* On success, caller is responsible for the link structure
*/
int
knote_unlink_wait_queue(struct knote *kn, struct wait_queue *wq, wait_queue_link_t *wqlp)
{
struct kqueue *kq = kn->kn_kq;
kern_return_t kr;
kr = wait_queue_unlink_nofree(wq, kq->kq_wqs, wqlp);
kqlock(kq);
kn->kn_status &= ~KN_STAYQUEUED;
knote_dequeue(kn);
kqunlock(kq);
return ((kr != KERN_SUCCESS) ? EINVAL : 0);
}
/*
* remove all knotes referencing a specified fd
*
* Essentially an inlined knote_remove & knote_drop
* when we know for sure that the thing is a file
*
* Entered with the proc_fd lock already held.
* It returns the same way, but may drop it temporarily.
*/
void
knote_fdclose(struct proc *p, int fd)
{
struct filedesc *fdp = p->p_fd;
struct klist *list;
struct knote *kn;
list = &fdp->fd_knlist[fd];
while ((kn = SLIST_FIRST(list)) != NULL) {
struct kqueue *kq = kn->kn_kq;
if (kq->kq_p != p)
panic("%s: proc mismatch (kq->kq_p=%p != p=%p)",
__func__, kq->kq_p, p);
kqlock(kq);
proc_fdunlock(p);
/*
* Convert the lock to a drop ref.
* If we get it, go ahead and drop it.
* Otherwise, we waited for it to
* be dropped by the other guy, so
* it is safe to move on in the list.
*/
if (kqlock2knotedrop(kq, kn)) {
kn->kn_fop->f_detach(kn);
knote_drop(kn, p);
}
proc_fdlock(p);
/* the fd tables may have changed - start over */
list = &fdp->fd_knlist[fd];
}
}
/* proc_fdlock held on entry (and exit) */
static int
knote_fdpattach(struct knote *kn, struct filedesc *fdp, struct proc *p)
{
struct klist *list = NULL;
if (! kn->kn_fop->f_isfd) {
if (fdp->fd_knhashmask == 0)
fdp->fd_knhash = hashinit(CONFIG_KN_HASHSIZE, M_KQUEUE,
&fdp->fd_knhashmask);
list = &fdp->fd_knhash[KN_HASH(kn->kn_id, fdp->fd_knhashmask)];
} else {
if ((u_int)fdp->fd_knlistsize <= kn->kn_id) {
u_int size = 0;
if (kn->kn_id >= (uint64_t)p->p_rlimit[RLIMIT_NOFILE].rlim_cur
|| kn->kn_id >= (uint64_t)maxfiles)
return (EINVAL);
/* have to grow the fd_knlist */
size = fdp->fd_knlistsize;
while (size <= kn->kn_id)
size += KQEXTENT;
if (size >= (UINT_MAX/sizeof(struct klist *)))
return (EINVAL);
MALLOC(list, struct klist *,
size * sizeof(struct klist *), M_KQUEUE, M_WAITOK);
if (list == NULL)
return (ENOMEM);
bcopy((caddr_t)fdp->fd_knlist, (caddr_t)list,
fdp->fd_knlistsize * sizeof(struct klist *));
bzero((caddr_t)list +
fdp->fd_knlistsize * sizeof(struct klist *),
(size - fdp->fd_knlistsize) * sizeof(struct klist *));
FREE(fdp->fd_knlist, M_KQUEUE);
fdp->fd_knlist = list;
fdp->fd_knlistsize = size;
}
list = &fdp->fd_knlist[kn->kn_id];
}
SLIST_INSERT_HEAD(list, kn, kn_link);
return (0);
}
/*
* should be called at spl == 0, since we don't want to hold spl
* while calling fdrop and free.
*/
static void
knote_drop(struct knote *kn, __unused struct proc *ctxp)
{
struct kqueue *kq = kn->kn_kq;
struct proc *p = kq->kq_p;
struct filedesc *fdp = p->p_fd;
struct klist *list;
int needswakeup;
proc_fdlock(p);
if (kn->kn_fop->f_isfd)
list = &fdp->fd_knlist[kn->kn_id];
else
list = &fdp->fd_knhash[KN_HASH(kn->kn_id, fdp->fd_knhashmask)];
SLIST_REMOVE(list, kn, knote, kn_link);
kqlock(kq);
knote_dequeue(kn);
needswakeup = (kn->kn_status & KN_USEWAIT);
kqunlock(kq);
proc_fdunlock(p);
if (needswakeup)
wait_queue_wakeup_all((wait_queue_t)kq->kq_wqs, &kn->kn_status,
THREAD_AWAKENED);
if (kn->kn_fop->f_isfd)
fp_drop(p, kn->kn_id, kn->kn_fp, 0);
knote_free(kn);
}
/* called with kqueue lock held */
static void
knote_activate(struct knote *kn, int propagate)
{
struct kqueue *kq = kn->kn_kq;
kn->kn_status |= KN_ACTIVE;
knote_enqueue(kn);
kqueue_wakeup(kq, 0);
/* this is a real event: wake up the parent kq, too */
if (propagate)
KNOTE(&kq->kq_sel.si_note, 0);
}
/* called with kqueue lock held */
static void
knote_deactivate(struct knote *kn)
{
kn->kn_status &= ~KN_ACTIVE;
knote_dequeue(kn);
}
/* called with kqueue lock held */
static void
knote_enqueue(struct knote *kn)
{
if ((kn->kn_status & (KN_QUEUED | KN_STAYQUEUED)) == KN_STAYQUEUED ||
(kn->kn_status & (KN_QUEUED | KN_STAYQUEUED | KN_DISABLED)) == 0) {
struct kqtailq *tq = kn->kn_tq;
struct kqueue *kq = kn->kn_kq;
TAILQ_INSERT_TAIL(tq, kn, kn_tqe);
kn->kn_status |= KN_QUEUED;
kq->kq_count++;
}
}
/* called with kqueue lock held */
static void
knote_dequeue(struct knote *kn)
{
struct kqueue *kq = kn->kn_kq;
if ((kn->kn_status & (KN_QUEUED | KN_STAYQUEUED)) == KN_QUEUED) {
struct kqtailq *tq = kn->kn_tq;
TAILQ_REMOVE(tq, kn, kn_tqe);
kn->kn_tq = &kq->kq_head;
kn->kn_status &= ~KN_QUEUED;
kq->kq_count--;
}
}
void
knote_init(void)
{
knote_zone = zinit(sizeof(struct knote), 8192*sizeof(struct knote),
8192, "knote zone");
/* allocate kq lock group attribute and group */
kq_lck_grp_attr = lck_grp_attr_alloc_init();
kq_lck_grp = lck_grp_alloc_init("kqueue", kq_lck_grp_attr);
/* Allocate kq lock attribute */
kq_lck_attr = lck_attr_alloc_init();
/* Initialize the timer filter lock */
lck_mtx_init(&_filt_timerlock, kq_lck_grp, kq_lck_attr);
#if VM_PRESSURE_EVENTS
/* Initialize the vm pressure list lock */
vm_pressure_init(kq_lck_grp, kq_lck_attr);
#endif
#if CONFIG_MEMORYSTATUS
/* Initialize the memorystatus list lock */
memorystatus_kevent_init(kq_lck_grp, kq_lck_attr);
#endif
}
SYSINIT(knote, SI_SUB_PSEUDO, SI_ORDER_ANY, knote_init, NULL)
static struct knote *
knote_alloc(void)
{
return ((struct knote *)zalloc(knote_zone));
}
static void
knote_free(struct knote *kn)
{
zfree(knote_zone, kn);
}
#if SOCKETS
#include <sys/param.h>
#include <sys/socket.h>
#include <sys/protosw.h>
#include <sys/domain.h>
#include <sys/mbuf.h>
#include <sys/kern_event.h>
#include <sys/malloc.h>
#include <sys/sys_domain.h>
#include <sys/syslog.h>
static lck_grp_attr_t *kev_lck_grp_attr;
static lck_attr_t *kev_lck_attr;
static lck_grp_t *kev_lck_grp;
static decl_lck_rw_data(,kev_lck_data);
static lck_rw_t *kev_rwlock = &kev_lck_data;
static int kev_attach(struct socket *so, int proto, struct proc *p);
static int kev_detach(struct socket *so);
static int kev_control(struct socket *so, u_long cmd, caddr_t data,
struct ifnet *ifp, struct proc *p);
static lck_mtx_t * event_getlock(struct socket *, int);
static int event_lock(struct socket *, int, void *);
static int event_unlock(struct socket *, int, void *);
static int event_sofreelastref(struct socket *);
static void kev_delete(struct kern_event_pcb *);
static struct pr_usrreqs event_usrreqs = {
.pru_attach = kev_attach,
.pru_control = kev_control,
.pru_detach = kev_detach,
.pru_soreceive = soreceive,
};
static struct protosw eventsw[] = {
{
.pr_type = SOCK_RAW,
.pr_protocol = SYSPROTO_EVENT,
.pr_flags = PR_ATOMIC,
.pr_usrreqs = &event_usrreqs,
.pr_lock = event_lock,
.pr_unlock = event_unlock,
.pr_getlock = event_getlock,
}
};
static lck_mtx_t *
event_getlock(struct socket *so, int locktype)
{
#pragma unused(locktype)
struct kern_event_pcb *ev_pcb = (struct kern_event_pcb *)so->so_pcb;
if (so->so_pcb != NULL) {
if (so->so_usecount < 0)
panic("%s: so=%p usecount=%d lrh= %s\n", __func__,
so, so->so_usecount, solockhistory_nr(so));
/* NOTREACHED */
} else {
panic("%s: so=%p NULL NO so_pcb %s\n", __func__,
so, solockhistory_nr(so));
/* NOTREACHED */
}
return (&ev_pcb->evp_mtx);
}
static int
event_lock(struct socket *so, int refcount, void *lr)
{
void *lr_saved;
if (lr == NULL)
lr_saved = __builtin_return_address(0);
else
lr_saved = lr;
if (so->so_pcb != NULL) {
lck_mtx_lock(&((struct kern_event_pcb *)so->so_pcb)->evp_mtx);
} else {
panic("%s: so=%p NO PCB! lr=%p lrh= %s\n", __func__,
so, lr_saved, solockhistory_nr(so));
/* NOTREACHED */
}
if (so->so_usecount < 0) {
panic("%s: so=%p so_pcb=%p lr=%p ref=%d lrh= %s\n", __func__,
so, so->so_pcb, lr_saved, so->so_usecount,
solockhistory_nr(so));
/* NOTREACHED */
}
if (refcount)
so->so_usecount++;
so->lock_lr[so->next_lock_lr] = lr_saved;
so->next_lock_lr = (so->next_lock_lr+1) % SO_LCKDBG_MAX;
return (0);
}
static int
event_unlock(struct socket *so, int refcount, void *lr)
{
void *lr_saved;
lck_mtx_t *mutex_held;
if (lr == NULL)
lr_saved = __builtin_return_address(0);
else
lr_saved = lr;
if (refcount)
so->so_usecount--;
if (so->so_usecount < 0) {
panic("%s: so=%p usecount=%d lrh= %s\n", __func__,
so, so->so_usecount, solockhistory_nr(so));
/* NOTREACHED */
}
if (so->so_pcb == NULL) {
panic("%s: so=%p NO PCB usecount=%d lr=%p lrh= %s\n", __func__,
so, so->so_usecount, (void *)lr_saved,
solockhistory_nr(so));
/* NOTREACHED */
}
mutex_held = (&((struct kern_event_pcb *)so->so_pcb)->evp_mtx);
lck_mtx_assert(mutex_held, LCK_MTX_ASSERT_OWNED);
so->unlock_lr[so->next_unlock_lr] = lr_saved;
so->next_unlock_lr = (so->next_unlock_lr+1) % SO_LCKDBG_MAX;
if (so->so_usecount == 0) {
VERIFY(so->so_flags & SOF_PCBCLEARING);
event_sofreelastref(so);
} else {
lck_mtx_unlock(mutex_held);
}
return (0);
}
static int
event_sofreelastref(struct socket *so)
{
struct kern_event_pcb *ev_pcb = (struct kern_event_pcb *)so->so_pcb;
lck_mtx_assert(&(ev_pcb->evp_mtx), LCK_MTX_ASSERT_OWNED);
so->so_pcb = NULL;
/*
* Disable upcall in the event another thread is in kev_post_msg()
* appending record to the receive socket buffer, since sbwakeup()
* may release the socket lock otherwise.
*/
so->so_rcv.sb_flags &= ~SB_UPCALL;
so->so_snd.sb_flags &= ~SB_UPCALL;
so->so_event = NULL;
lck_mtx_unlock(&(ev_pcb->evp_mtx));
lck_mtx_assert(&(ev_pcb->evp_mtx), LCK_MTX_ASSERT_NOTOWNED);
lck_rw_lock_exclusive(kev_rwlock);
LIST_REMOVE(ev_pcb, evp_link);
lck_rw_done(kev_rwlock);
kev_delete(ev_pcb);
sofreelastref(so, 1);
return (0);
}
static int event_proto_count = (sizeof (eventsw) / sizeof (struct protosw));
static
struct kern_event_head kern_event_head;
static u_int32_t static_event_id = 0;
#define EVPCB_ZONE_MAX 65536
#define EVPCB_ZONE_NAME "kerneventpcb"
static struct zone *ev_pcb_zone;
/*
* Install the protosw's for the NKE manager. Invoked at extension load time
*/
void
kern_event_init(struct domain *dp)
{
struct protosw *pr;
int i;
VERIFY(!(dp->dom_flags & DOM_INITIALIZED));
VERIFY(dp == systemdomain);
kev_lck_grp_attr = lck_grp_attr_alloc_init();
if (kev_lck_grp_attr == NULL) {
panic("%s: lck_grp_attr_alloc_init failed\n", __func__);
/* NOTREACHED */
}
kev_lck_grp = lck_grp_alloc_init("Kernel Event Protocol",
kev_lck_grp_attr);
if (kev_lck_grp == NULL) {
panic("%s: lck_grp_alloc_init failed\n", __func__);
/* NOTREACHED */
}
kev_lck_attr = lck_attr_alloc_init();
if (kev_lck_attr == NULL) {
panic("%s: lck_attr_alloc_init failed\n", __func__);
/* NOTREACHED */
}
lck_rw_init(kev_rwlock, kev_lck_grp, kev_lck_attr);
if (kev_rwlock == NULL) {
panic("%s: lck_mtx_alloc_init failed\n", __func__);
/* NOTREACHED */
}
for (i = 0, pr = &eventsw[0]; i < event_proto_count; i++, pr++)
net_add_proto(pr, dp, 1);
ev_pcb_zone = zinit(sizeof(struct kern_event_pcb),
EVPCB_ZONE_MAX * sizeof(struct kern_event_pcb), 0, EVPCB_ZONE_NAME);
if (ev_pcb_zone == NULL) {
panic("%s: failed allocating ev_pcb_zone", __func__);
/* NOTREACHED */
}
zone_change(ev_pcb_zone, Z_EXPAND, TRUE);
zone_change(ev_pcb_zone, Z_CALLERACCT, TRUE);
}
static int
kev_attach(struct socket *so, __unused int proto, __unused struct proc *p)
{
int error = 0;
struct kern_event_pcb *ev_pcb;
error = soreserve(so, KEV_SNDSPACE, KEV_RECVSPACE);
if (error != 0)
return (error);
if ((ev_pcb = (struct kern_event_pcb *)zalloc(ev_pcb_zone)) == NULL) {
return (ENOBUFS);
}
bzero(ev_pcb, sizeof(struct kern_event_pcb));
lck_mtx_init(&ev_pcb->evp_mtx, kev_lck_grp, kev_lck_attr);
ev_pcb->evp_socket = so;
ev_pcb->evp_vendor_code_filter = 0xffffffff;
so->so_pcb = (caddr_t) ev_pcb;
lck_rw_lock_exclusive(kev_rwlock);
LIST_INSERT_HEAD(&kern_event_head, ev_pcb, evp_link);
lck_rw_done(kev_rwlock);
return (error);
}
static void
kev_delete(struct kern_event_pcb *ev_pcb)
{
VERIFY(ev_pcb != NULL);
lck_mtx_destroy(&ev_pcb->evp_mtx, kev_lck_grp);
zfree(ev_pcb_zone, ev_pcb);
}
static int
kev_detach(struct socket *so)
{
struct kern_event_pcb *ev_pcb = (struct kern_event_pcb *) so->so_pcb;
if (ev_pcb != NULL) {
soisdisconnected(so);
so->so_flags |= SOF_PCBCLEARING;
}
return (0);
}
/*
* For now, kev_vendor_code and mbuf_tags use the same
* mechanism.
*/
errno_t kev_vendor_code_find(
const char *string,
u_int32_t *out_vendor_code)
{
if (strlen(string) >= KEV_VENDOR_CODE_MAX_STR_LEN) {
return (EINVAL);
}
return (net_str_id_find_internal(string, out_vendor_code,
NSI_VENDOR_CODE, 1));
}
errno_t
kev_msg_post(struct kev_msg *event_msg)
{
mbuf_tag_id_t min_vendor, max_vendor;
net_str_id_first_last(&min_vendor, &max_vendor, NSI_VENDOR_CODE);
if (event_msg == NULL)
return (EINVAL);
/*
* Limit third parties to posting events for registered vendor codes
* only
*/
if (event_msg->vendor_code < min_vendor ||
event_msg->vendor_code > max_vendor)
return (EINVAL);
return (kev_post_msg(event_msg));
}
int
kev_post_msg(struct kev_msg *event_msg)
{
struct mbuf *m, *m2;
struct kern_event_pcb *ev_pcb;
struct kern_event_msg *ev;
char *tmp;
u_int32_t total_size;
int i;
/* Verify the message is small enough to fit in one mbuf w/o cluster */
total_size = KEV_MSG_HEADER_SIZE;
for (i = 0; i < 5; i++) {
if (event_msg->dv[i].data_length == 0)
break;
total_size += event_msg->dv[i].data_length;
}
if (total_size > MLEN) {
return (EMSGSIZE);
}
m = m_get(M_DONTWAIT, MT_DATA);
if (m == 0)
return (ENOBUFS);
ev = mtod(m, struct kern_event_msg *);
total_size = KEV_MSG_HEADER_SIZE;
tmp = (char *) &ev->event_data[0];
for (i = 0; i < 5; i++) {
if (event_msg->dv[i].data_length == 0)
break;
total_size += event_msg->dv[i].data_length;
bcopy(event_msg->dv[i].data_ptr, tmp,
event_msg->dv[i].data_length);
tmp += event_msg->dv[i].data_length;
}
ev->id = ++static_event_id;
ev->total_size = total_size;
ev->vendor_code = event_msg->vendor_code;
ev->kev_class = event_msg->kev_class;
ev->kev_subclass = event_msg->kev_subclass;
ev->event_code = event_msg->event_code;
m->m_len = total_size;
lck_rw_lock_shared(kev_rwlock);
for (ev_pcb = LIST_FIRST(&kern_event_head);
ev_pcb;
ev_pcb = LIST_NEXT(ev_pcb, evp_link)) {
lck_mtx_lock(&ev_pcb->evp_mtx);
if (ev_pcb->evp_socket->so_pcb == NULL) {
lck_mtx_unlock(&ev_pcb->evp_mtx);
continue;
}
if (ev_pcb->evp_vendor_code_filter != KEV_ANY_VENDOR) {
if (ev_pcb->evp_vendor_code_filter != ev->vendor_code) {
lck_mtx_unlock(&ev_pcb->evp_mtx);
continue;
}
if (ev_pcb->evp_class_filter != KEV_ANY_CLASS) {
if (ev_pcb->evp_class_filter != ev->kev_class) {
lck_mtx_unlock(&ev_pcb->evp_mtx);
continue;
}
if ((ev_pcb->evp_subclass_filter != KEV_ANY_SUBCLASS) &&
(ev_pcb->evp_subclass_filter != ev->kev_subclass)) {
lck_mtx_unlock(&ev_pcb->evp_mtx);
continue;
}
}
}
m2 = m_copym(m, 0, m->m_len, M_NOWAIT);
if (m2 == 0) {
m_free(m);
lck_mtx_unlock(&ev_pcb->evp_mtx);
lck_rw_done(kev_rwlock);
return (ENOBUFS);
}
if (sbappendrecord(&ev_pcb->evp_socket->so_rcv, m2))
sorwakeup(ev_pcb->evp_socket);
lck_mtx_unlock(&ev_pcb->evp_mtx);
}
m_free(m);
lck_rw_done(kev_rwlock);
return (0);
}
static int
kev_control(struct socket *so,
u_long cmd,
caddr_t data,
__unused struct ifnet *ifp,
__unused struct proc *p)
{
struct kev_request *kev_req = (struct kev_request *) data;
struct kern_event_pcb *ev_pcb;
struct kev_vendor_code *kev_vendor;
u_int32_t *id_value = (u_int32_t *) data;
switch (cmd) {
case SIOCGKEVID:
*id_value = static_event_id;
break;
case SIOCSKEVFILT:
ev_pcb = (struct kern_event_pcb *) so->so_pcb;
ev_pcb->evp_vendor_code_filter = kev_req->vendor_code;
ev_pcb->evp_class_filter = kev_req->kev_class;
ev_pcb->evp_subclass_filter = kev_req->kev_subclass;
break;
case SIOCGKEVFILT:
ev_pcb = (struct kern_event_pcb *) so->so_pcb;
kev_req->vendor_code = ev_pcb->evp_vendor_code_filter;
kev_req->kev_class = ev_pcb->evp_class_filter;
kev_req->kev_subclass = ev_pcb->evp_subclass_filter;
break;
case SIOCGKEVVENDOR:
kev_vendor = (struct kev_vendor_code *)data;
/* Make sure string is NULL terminated */
kev_vendor->vendor_string[KEV_VENDOR_CODE_MAX_STR_LEN-1] = 0;
return (net_str_id_find_internal(kev_vendor->vendor_string,
&kev_vendor->vendor_code, NSI_VENDOR_CODE, 0));
default:
return (ENOTSUP);
}
return (0);
}
#endif /* SOCKETS */
int
fill_kqueueinfo(struct kqueue *kq, struct kqueue_info * kinfo)
{
struct vinfo_stat * st;
/* No need for the funnel as fd is kept alive */
st = &kinfo->kq_stat;
st->vst_size = kq->kq_count;
if (kq->kq_state & KQ_KEV64)
st->vst_blksize = sizeof(struct kevent64_s);
else
st->vst_blksize = sizeof(struct kevent);
st->vst_mode = S_IFIFO;
if (kq->kq_state & KQ_SEL)
kinfo->kq_state |= PROC_KQUEUE_SELECT;
if (kq->kq_state & KQ_SLEEP)
kinfo->kq_state |= PROC_KQUEUE_SLEEP;
return (0);
}
void
knote_markstayqueued(struct knote *kn)
{
kqlock(kn->kn_kq);
kn->kn_status |= KN_STAYQUEUED;
knote_enqueue(kn);
kqunlock(kn->kn_kq);
}
Raw
wait_queue.c
/*
* Copyright (c) 2000-2009 Apple Inc. All rights reserved.
*
* @APPLE_OSREFERENCE_LICENSE_HEADER_START@
*
* This file contains Original Code and/or Modifications of Original Code
* as defined in and that are subject to the Apple Public Source License
* Version 2.0 (the 'License'). You may not use this file except in
* compliance with the License. The rights granted to you under the License
* may not be used to create, or enable the creation or redistribution of,
* unlawful or unlicensed copies of an Apple operating system, or to
* circumvent, violate, or enable the circumvention or violation of, any
* terms of an Apple operating system software license agreement.
*
* Please obtain a copy of the License at
* http://www.opensource.apple.com/apsl/ and read it before using this file.
*
* The Original Code and all software distributed under the License are
* distributed on an 'AS IS' basis, WITHOUT WARRANTY OF ANY KIND, EITHER
* EXPRESS OR IMPLIED, AND APPLE HEREBY DISCLAIMS ALL SUCH WARRANTIES,
* INCLUDING WITHOUT LIMITATION, ANY WARRANTIES OF MERCHANTABILITY,
* FITNESS FOR A PARTICULAR PURPOSE, QUIET ENJOYMENT OR NON-INFRINGEMENT.
* Please see the License for the specific language governing rights and
* limitations under the License.
*
* @APPLE_OSREFERENCE_LICENSE_HEADER_END@
*/
/*
* @OSF_FREE_COPYRIGHT@
*/
/*
* Mach Operating System
* Copyright (c) 1991,1990,1989,1988,1987 Carnegie Mellon University
* All Rights Reserved.
*
* Permission to use, copy, modify and distribute this software and its
* documentation is hereby granted, provided that both the copyright
* notice and this permission notice appear in all copies of the
* software, derivative works or modified versions, and any portions
* thereof, and that both notices appear in supporting documentation.
*
* CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS"
* CONDITION. CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND FOR
* ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE.
*
* Carnegie Mellon requests users of this software to return to
*
* Software Distribution Coordinator or Software.Distribution@CS.CMU.EDU
* School of Computer Science
* Carnegie Mellon University
* Pittsburgh PA 15213-3890
*
* any improvements or extensions that they make and grant Carnegie Mellon
* the rights to redistribute these changes.
*/
/*
*/
/*
* File: wait_queue.c (adapted from sched_prim.c)
* Author: Avadis Tevanian, Jr.
* Date: 1986
*
* Primitives for manipulating wait queues: either global
* ones from sched_prim.c, or private ones associated with
* particular structures(pots, semaphores, etc..).
*/
#include <kern/kern_types.h>
#include <kern/simple_lock.h>
#include <kern/zalloc.h>
#include <kern/queue.h>
#include <kern/spl.h>
#include <mach/sync_policy.h>
#include <kern/mach_param.h>
#include <kern/sched_prim.h>
#include <kern/wait_queue.h>
#include <vm/vm_kern.h>
/* forward declarations */
static boolean_t wait_queue_member_locked(
wait_queue_t wq,
wait_queue_set_t wq_set);
static void wait_queues_init(void);
#define WAIT_QUEUE_MAX thread_max
#define WAIT_QUEUE_SET_MAX task_max * 3
#define WAIT_QUEUE_LINK_MAX PORT_MAX / 2 + (WAIT_QUEUE_MAX * WAIT_QUEUE_SET_MAX) / 64
static zone_t _wait_queue_link_zone;
static zone_t _wait_queue_set_zone;
static zone_t _wait_queue_zone;
/* see rdar://6737748&5561610; we need an unshadowed
* definition of a WaitQueueLink for debugging,
* but it needs to be used somewhere to wind up in
* the dSYM file. */
volatile WaitQueueLink *unused_except_for_debugging;
/*
* Waiting protocols and implementation:
*
* Each thread may be waiting for exactly one event; this event
* is set using assert_wait(). That thread may be awakened either
* by performing a thread_wakeup_prim() on its event,
* or by directly waking that thread up with clear_wait().
*
* The implementation of wait events uses a hash table. Each
* bucket is queue of threads having the same hash function
* value; the chain for the queue (linked list) is the run queue
* field. [It is not possible to be waiting and runnable at the
* same time.]
*
* Locks on both the thread and on the hash buckets govern the
* wait event field and the queue chain field. Because wakeup
* operations only have the event as an argument, the event hash
* bucket must be locked before any thread.
*
* Scheduling operations may also occur at interrupt level; therefore,
* interrupts below splsched() must be prevented when holding
* thread or hash bucket locks.
*
* The wait event hash table declarations are as follows:
*/
struct wait_queue boot_wait_queue[1];
__private_extern__ struct wait_queue *wait_queues = &boot_wait_queue[0];
__private_extern__ uint32_t num_wait_queues = 1;
#define P2ROUNDUP(x, align) (-(-((uint32_t)(x)) & -(align)))
#define ROUNDDOWN(x,y) (((x)/(y))*(y))
static uint32_t
compute_wait_hash_size(void)
{
uint32_t hsize, queues;
if (PE_parse_boot_argn("wqsize", &hsize, sizeof(hsize)))
return (hsize);
queues = thread_max / 11;
hsize = P2ROUNDUP(queues * sizeof(struct wait_queue), PAGE_SIZE);
return hsize;
}
static void
wait_queues_init(void)
{
uint32_t i, whsize, qsz;
kern_return_t kret;
/*
* Determine the amount of memory we're willing to reserve for
* the waitqueue hash table
*/
whsize = compute_wait_hash_size();
/* Determine the number of waitqueues we can fit. */
qsz = sizeof (struct wait_queue);
whsize = ROUNDDOWN(whsize, qsz);
num_wait_queues = whsize / qsz;
/*
* The hash algorithm requires that this be a power of 2, so we
* just mask off all the low-order bits.
*/
for (i = 0; i < 31; i++) {
uint32_t bit = (1 << i);
if ((num_wait_queues & bit) == num_wait_queues)
break;
num_wait_queues &= ~bit;
}
assert(num_wait_queues > 0);
/* Now determine how much memory we really need. */
whsize = P2ROUNDUP(num_wait_queues * qsz, PAGE_SIZE);
kret = kernel_memory_allocate(kernel_map, (vm_offset_t *) &wait_queues,
whsize, 0, KMA_KOBJECT|KMA_NOPAGEWAIT);
if (kret != KERN_SUCCESS || wait_queues == NULL)
panic("kernel_memory_allocate() failed to allocate wait queues, error: %d, whsize: 0x%x", kret, whsize);
for (i = 0; i < num_wait_queues; i++) {
wait_queue_init(&wait_queues[i], SYNC_POLICY_FIFO);
}
}
void
wait_queue_bootstrap(void)
{
wait_queues_init();
_wait_queue_zone = zinit(sizeof(struct wait_queue),
WAIT_QUEUE_MAX * sizeof(struct wait_queue),
sizeof(struct wait_queue),
"wait queues");
zone_change(_wait_queue_zone, Z_NOENCRYPT, TRUE);
_wait_queue_set_zone = zinit(sizeof(struct wait_queue_set),
WAIT_QUEUE_SET_MAX * sizeof(struct wait_queue_set),
sizeof(struct wait_queue_set),
"wait queue sets");
zone_change(_wait_queue_set_zone, Z_NOENCRYPT, TRUE);
_wait_queue_link_zone = zinit(sizeof(struct _wait_queue_link),
WAIT_QUEUE_LINK_MAX * sizeof(struct _wait_queue_link),
sizeof(struct _wait_queue_link),
"wait queue links");
zone_change(_wait_queue_link_zone, Z_NOENCRYPT, TRUE);
}
/*
* Routine: wait_queue_init
* Purpose:
* Initialize a previously allocated wait queue.
* Returns:
* KERN_SUCCESS - The wait_queue_t was initialized
* KERN_INVALID_ARGUMENT - The policy parameter was invalid
*/
kern_return_t
wait_queue_init(
wait_queue_t wq,
int policy)
{
/* only FIFO and LIFO for now */
if ((policy & SYNC_POLICY_FIXED_PRIORITY) != 0)
return KERN_INVALID_ARGUMENT;
wq->wq_fifo = ((policy & SYNC_POLICY_REVERSED) == 0);
wq->wq_type = _WAIT_QUEUE_inited;
wq->wq_eventmask = 0;
queue_init(&wq->wq_queue);
hw_lock_init(&wq->wq_interlock);
return KERN_SUCCESS;
}
/*
* Routine: wait_queue_alloc
* Purpose:
* Allocate and initialize a wait queue for use outside of
* of the mach part of the kernel.
* Conditions:
* Nothing locked - can block.
* Returns:
* The allocated and initialized wait queue
* WAIT_QUEUE_NULL if there is a resource shortage
*/
wait_queue_t
wait_queue_alloc(
int policy)
{
wait_queue_t wq;
kern_return_t ret;
wq = (wait_queue_t) zalloc(_wait_queue_zone);
if (wq != WAIT_QUEUE_NULL) {
ret = wait_queue_init(wq, policy);
if (ret != KERN_SUCCESS) {
zfree(_wait_queue_zone, wq);
wq = WAIT_QUEUE_NULL;
}
}
return wq;
}
/*
* Routine: wait_queue_free
* Purpose:
* Free an allocated wait queue.
* Conditions:
* May block.
*/
kern_return_t
wait_queue_free(
wait_queue_t wq)
{
if (!wait_queue_is_queue(wq))
return KERN_INVALID_ARGUMENT;
if (!queue_empty(&wq->wq_queue))
return KERN_FAILURE;
zfree(_wait_queue_zone, wq);
return KERN_SUCCESS;
}
/*
* Routine: wait_queue_set_init
* Purpose:
* Initialize a previously allocated wait queue set.
* Returns:
* KERN_SUCCESS - The wait_queue_set_t was initialized
* KERN_INVALID_ARGUMENT - The policy parameter was invalid
*/
kern_return_t
wait_queue_set_init(
wait_queue_set_t wqset,
int policy)
{
kern_return_t ret;
ret = wait_queue_init(&wqset->wqs_wait_queue, policy);
if (ret != KERN_SUCCESS)
return ret;
wqset->wqs_wait_queue.wq_type = _WAIT_QUEUE_SET_inited;
if (policy & SYNC_POLICY_PREPOST)
wqset->wqs_wait_queue.wq_prepost = TRUE;
else
wqset->wqs_wait_queue.wq_prepost = FALSE;
queue_init(&wqset->wqs_setlinks);
queue_init(&wqset->wqs_preposts);
return KERN_SUCCESS;
}
kern_return_t
wait_queue_sub_init(
wait_queue_set_t wqset,
int policy)
{
return wait_queue_set_init(wqset, policy);
}
kern_return_t
wait_queue_sub_clearrefs(
wait_queue_set_t wq_set)
{
wait_queue_link_t wql;
queue_t q;
spl_t s;
if (!wait_queue_is_set(wq_set))
return KERN_INVALID_ARGUMENT;
s = splsched();
wqs_lock(wq_set);
q = &wq_set->wqs_preposts;
while (!queue_empty(q)) {
queue_remove_first(q, wql, wait_queue_link_t, wql_preposts);
assert(!wql_is_preposted(wql));
}
wqs_unlock(wq_set);
splx(s);
return KERN_SUCCESS;
}
/*
* Routine: wait_queue_set_alloc
* Purpose:
* Allocate and initialize a wait queue set for
* use outside of the mach part of the kernel.
* Conditions:
* May block.
* Returns:
* The allocated and initialized wait queue set
* WAIT_QUEUE_SET_NULL if there is a resource shortage
*/
wait_queue_set_t
wait_queue_set_alloc(
int policy)
{
wait_queue_set_t wq_set;
wq_set = (wait_queue_set_t) zalloc(_wait_queue_set_zone);
if (wq_set != WAIT_QUEUE_SET_NULL) {
kern_return_t ret;
ret = wait_queue_set_init(wq_set, policy);
if (ret != KERN_SUCCESS) {
zfree(_wait_queue_set_zone, wq_set);
wq_set = WAIT_QUEUE_SET_NULL;
}
}
return wq_set;
}
/*
* Routine: wait_queue_set_free
* Purpose:
* Free an allocated wait queue set
* Conditions:
* May block.
*/
kern_return_t
wait_queue_set_free(
wait_queue_set_t wq_set)
{
if (!wait_queue_is_set(wq_set))
return KERN_INVALID_ARGUMENT;
if (!queue_empty(&wq_set->wqs_wait_queue.wq_queue))
return KERN_FAILURE;
zfree(_wait_queue_set_zone, wq_set);
return KERN_SUCCESS;
}
/*
*
* Routine: wait_queue_set_size
* Routine: wait_queue_link_size
* Purpose:
* Return the size of opaque wait queue structures
*/
unsigned int wait_queue_set_size(void) { return sizeof(WaitQueueSet); }
unsigned int wait_queue_link_size(void) { return sizeof(WaitQueueLink); }
/* declare a unique type for wait queue link structures */
static unsigned int _wait_queue_link;
static unsigned int _wait_queue_link_noalloc;
static unsigned int _wait_queue_unlinked;
#define WAIT_QUEUE_LINK ((void *)&_wait_queue_link)
#define WAIT_QUEUE_LINK_NOALLOC ((void *)&_wait_queue_link_noalloc)
#define WAIT_QUEUE_UNLINKED ((void *)&_wait_queue_unlinked)
#define WAIT_QUEUE_ELEMENT_CHECK(wq, wqe) \
WQASSERT(((wqe)->wqe_queue == (wq) && \
queue_next(queue_prev((queue_t) (wqe))) == (queue_t)(wqe)), \
"wait queue element list corruption: wq=%#x, wqe=%#x", \
(wq), (wqe))
#define WQSPREV(wqs, wql) ((wait_queue_link_t)queue_prev( \
((&(wqs)->wqs_setlinks == (queue_t)(wql)) ? \
(queue_t)(wql) : &(wql)->wql_setlinks)))
#define WQSNEXT(wqs, wql) ((wait_queue_link_t)queue_next( \
((&(wqs)->wqs_setlinks == (queue_t)(wql)) ? \
(queue_t)(wql) : &(wql)->wql_setlinks)))
#define WAIT_QUEUE_SET_LINK_CHECK(wqs, wql) \
WQASSERT(((((wql)->wql_type == WAIT_QUEUE_LINK) || \
((wql)->wql_type == WAIT_QUEUE_LINK_NOALLOC)) && \
((wql)->wql_setqueue == (wqs)) && \
(((wql)->wql_queue->wq_type == _WAIT_QUEUE_inited) || \
((wql)->wql_queue->wq_type == _WAIT_QUEUE_SET_inited)) && \
(WQSNEXT((wqs), WQSPREV((wqs),(wql))) == (wql))), \
"wait queue set links corruption: wqs=%#x, wql=%#x", \
(wqs), (wql))
#if defined(_WAIT_QUEUE_DEBUG_)
#define WQASSERT(e, s, p0, p1) ((e) ? 0 : panic(s, p0, p1))
#define WAIT_QUEUE_CHECK(wq) \
MACRO_BEGIN \
queue_t q2 = &(wq)->wq_queue; \
wait_queue_element_t wqe2 = (wait_queue_element_t) queue_first(q2); \
while (!queue_end(q2, (queue_entry_t)wqe2)) { \
WAIT_QUEUE_ELEMENT_CHECK((wq), wqe2); \
wqe2 = (wait_queue_element_t) queue_next((queue_t) wqe2); \
} \
MACRO_END
#define WAIT_QUEUE_SET_CHECK(wqs) \
MACRO_BEGIN \
queue_t q2 = &(wqs)->wqs_setlinks; \
wait_queue_link_t wql2 = (wait_queue_link_t) queue_first(q2); \
while (!queue_end(q2, (queue_entry_t)wql2)) { \
WAIT_QUEUE_SET_LINK_CHECK((wqs), wql2); \
wql2 = (wait_queue_link_t) wql2->wql_setlinks.next; \
} \
MACRO_END
#else /* !_WAIT_QUEUE_DEBUG_ */
#define WQASSERT(e, s, p0, p1) assert(e)
#define WAIT_QUEUE_CHECK(wq)
#define WAIT_QUEUE_SET_CHECK(wqs)
#endif /* !_WAIT_QUEUE_DEBUG_ */
/*
* Routine: wait_queue_global
* Purpose:
* Indicate if this wait queue is a global wait queue or not.
*/
static boolean_t
wait_queue_global(
wait_queue_t wq)
{
if ((wq >= wait_queues) && (wq <= (wait_queues + num_wait_queues))) {
return TRUE;
}
return FALSE;
}
/*
* Routine: wait_queue_member_locked
* Purpose:
* Indicate if this set queue is a member of the queue
* Conditions:
* The wait queue is locked
* The set queue is just that, a set queue
*/
static boolean_t
wait_queue_member_locked(
wait_queue_t wq,
wait_queue_set_t wq_set)
{
wait_queue_element_t wq_element;
queue_t q;
assert(wait_queue_held(wq));
assert(wait_queue_is_set(wq_set));
q = &wq->wq_queue;
wq_element = (wait_queue_element_t) queue_first(q);
while (!queue_end(q, (queue_entry_t)wq_element)) {
WAIT_QUEUE_ELEMENT_CHECK(wq, wq_element);
if ((wq_element->wqe_type == WAIT_QUEUE_LINK) ||
(wq_element->wqe_type == WAIT_QUEUE_LINK_NOALLOC)) {
wait_queue_link_t wql = (wait_queue_link_t)wq_element;
if (wql->wql_setqueue == wq_set)
return TRUE;
}
wq_element = (wait_queue_element_t)
queue_next((queue_t) wq_element);
}
return FALSE;
}
/*
* Routine: wait_queue_member
* Purpose:
* Indicate if this set queue is a member of the queue
* Conditions:
* The set queue is just that, a set queue
*/
boolean_t
wait_queue_member(
wait_queue_t wq,
wait_queue_set_t wq_set)
{
boolean_t ret;
spl_t s;
if (!wait_queue_is_set(wq_set))
return FALSE;
s = splsched();
wait_queue_lock(wq);
ret = wait_queue_member_locked(wq, wq_set);
wait_queue_unlock(wq);
splx(s);
return ret;
}
/*
* Routine: wait_queue_link_internal
* Purpose:
* Insert a set wait queue into a wait queue. This
* requires us to link the two together using a wait_queue_link
* structure that was provided.
* Conditions:
* The wait queue being inserted must be inited as a set queue
* The wait_queue_link structure must already be properly typed
*/
static
kern_return_t
wait_queue_link_internal(
wait_queue_t wq,
wait_queue_set_t wq_set,
wait_queue_link_t wql)
{
wait_queue_element_t wq_element;
queue_t q;
spl_t s;
if (!wait_queue_is_valid(wq) || !wait_queue_is_set(wq_set))
return KERN_INVALID_ARGUMENT;
/*
* There are probably fewer threads and sets associated with
* the wait queue than there are wait queues associated with
* the set. So let's validate it that way.
*/
s = splsched();
wait_queue_lock(wq);
q = &wq->wq_queue;
wq_element = (wait_queue_element_t) queue_first(q);
while (!queue_end(q, (queue_entry_t)wq_element)) {
WAIT_QUEUE_ELEMENT_CHECK(wq, wq_element);
if ((wq_element->wqe_type == WAIT_QUEUE_LINK ||
wq_element->wqe_type == WAIT_QUEUE_LINK_NOALLOC) &&
((wait_queue_link_t)wq_element)->wql_setqueue == wq_set) {
wait_queue_unlock(wq);
splx(s);
return KERN_ALREADY_IN_SET;
}
wq_element = (wait_queue_element_t)
queue_next((queue_t) wq_element);
}
/*
* Not already a member, so we can add it.
*/
wqs_lock(wq_set);
WAIT_QUEUE_SET_CHECK(wq_set);
assert(wql->wql_type == WAIT_QUEUE_LINK ||
wql->wql_type == WAIT_QUEUE_LINK_NOALLOC);
wql->wql_queue = wq;
wql_clear_prepost(wql);
queue_enter(&wq->wq_queue, wql, wait_queue_link_t, wql_links);
wql->wql_setqueue = wq_set;
queue_enter(&wq_set->wqs_setlinks, wql, wait_queue_link_t, wql_setlinks);
wqs_unlock(wq_set);
wait_queue_unlock(wq);
splx(s);
return KERN_SUCCESS;
}
/*
* Routine: wait_queue_link_noalloc
* Purpose:
* Insert a set wait queue into a wait queue. This
* requires us to link the two together using a wait_queue_link
* structure that we allocate.
* Conditions:
* The wait queue being inserted must be inited as a set queue
*/
kern_return_t
wait_queue_link_noalloc(
wait_queue_t wq,
wait_queue_set_t wq_set,
wait_queue_link_t wql)
{
wql->wql_type = WAIT_QUEUE_LINK_NOALLOC;
return wait_queue_link_internal(wq, wq_set, wql);
}
/*
* Routine: wait_queue_link
* Purpose:
* Insert a set wait queue into a wait queue. This
* requires us to link the two together using a wait_queue_link
* structure that we allocate.
* Conditions:
* The wait queue being inserted must be inited as a set queue
*/
kern_return_t
wait_queue_link(
wait_queue_t wq,
wait_queue_set_t wq_set)
{
wait_queue_link_t wql;
kern_return_t ret;
wql = (wait_queue_link_t) zalloc(_wait_queue_link_zone);
if (wql == WAIT_QUEUE_LINK_NULL)
return KERN_RESOURCE_SHORTAGE;
wql->wql_type = WAIT_QUEUE_LINK;
ret = wait_queue_link_internal(wq, wq_set, wql);
if (ret != KERN_SUCCESS)
zfree(_wait_queue_link_zone, wql);
return ret;
}
wait_queue_link_t
wait_queue_link_allocate(void)
{
wait_queue_link_t wql;
wql = zalloc(_wait_queue_link_zone); /* Can't fail */
bzero(wql, sizeof(*wql));
wql->wql_type = WAIT_QUEUE_UNLINKED;
return wql;
}
kern_return_t
wait_queue_link_free(wait_queue_link_t wql)
{
zfree(_wait_queue_link_zone, wql);
return KERN_SUCCESS;
}
/*
* Routine: wait_queue_unlink_locked
* Purpose:
* Undo the linkage between a wait queue and a set.
*/
static void
wait_queue_unlink_locked(
wait_queue_t wq,
wait_queue_set_t wq_set,
wait_queue_link_t wql)
{
assert(wait_queue_held(wq));
assert(wait_queue_held(&wq_set->wqs_wait_queue));
wql->wql_queue = WAIT_QUEUE_NULL;
queue_remove(&wq->wq_queue, wql, wait_queue_link_t, wql_links);
wql->wql_setqueue = WAIT_QUEUE_SET_NULL;
queue_remove(&wq_set->wqs_setlinks, wql, wait_queue_link_t, wql_setlinks);
if (wql_is_preposted(wql)) {
queue_t ppq = &wq_set->wqs_preposts;
queue_remove(ppq, wql, wait_queue_link_t, wql_preposts);
}
wql->wql_type = WAIT_QUEUE_UNLINKED;
WAIT_QUEUE_CHECK(wq);
WAIT_QUEUE_SET_CHECK(wq_set);
}
/*
* Routine: wait_queue_unlink_nofree
* Purpose:
* Remove the linkage between a wait queue and a set,
* returning the linkage structure to the caller to
* free later.
* Conditions:
* The wait queue being must be a member set queue
*/
kern_return_t
wait_queue_unlink_nofree(
wait_queue_t wq,
wait_queue_set_t wq_set,
wait_queue_link_t *wqlp)
{
wait_queue_element_t wq_element;
wait_queue_link_t wql;
queue_t q;
spl_t s;
if (!wait_queue_is_valid(wq) || !wait_queue_is_set(wq_set)) {
return KERN_INVALID_ARGUMENT;
}
s = splsched();
wait_queue_lock(wq);
q = &wq->wq_queue;
wq_element = (wait_queue_element_t) queue_first(q);
while (!queue_end(q, (queue_entry_t)wq_element)) {
WAIT_QUEUE_ELEMENT_CHECK(wq, wq_element);
if (wq_element->wqe_type == WAIT_QUEUE_LINK ||
wq_element->wqe_type == WAIT_QUEUE_LINK_NOALLOC) {
wql = (wait_queue_link_t)wq_element;
if (wql->wql_setqueue == wq_set) {
wqs_lock(wq_set);
wait_queue_unlink_locked(wq, wq_set, wql);
wqs_unlock(wq_set);
wait_queue_unlock(wq);
splx(s);
*wqlp = wql;
return KERN_SUCCESS;
}
}
wq_element = (wait_queue_element_t)
queue_next((queue_t) wq_element);
}
wait_queue_unlock(wq);
splx(s);
return KERN_NOT_IN_SET;
}
/*
* Routine: wait_queue_unlink
* Purpose:
* Remove the linkage between a wait queue and a set,
* freeing the linkage structure.
* Conditions:
* The wait queue being must be a member set queue
*/
kern_return_t
wait_queue_unlink(
wait_queue_t wq,
wait_queue_set_t wq_set)
{
wait_queue_element_t wq_element;
wait_queue_link_t wql;
queue_t q;
spl_t s;
if (!wait_queue_is_valid(wq) || !wait_queue_is_set(wq_set)) {
return KERN_INVALID_ARGUMENT;
}
s = splsched();
wait_queue_lock(wq);
q = &wq->wq_queue;
wq_element = (wait_queue_element_t) queue_first(q);
while (!queue_end(q, (queue_entry_t)wq_element)) {
WAIT_QUEUE_ELEMENT_CHECK(wq, wq_element);
if (wq_element->wqe_type == WAIT_QUEUE_LINK ||
wq_element->wqe_type == WAIT_QUEUE_LINK_NOALLOC) {
wql = (wait_queue_link_t)wq_element;
if (wql->wql_setqueue == wq_set) {
boolean_t alloced;
alloced = (wql->wql_type == WAIT_QUEUE_LINK);
wqs_lock(wq_set);
wait_queue_unlink_locked(wq, wq_set, wql);
wqs_unlock(wq_set);
wait_queue_unlock(wq);
splx(s);
if (alloced)
zfree(_wait_queue_link_zone, wql);
return KERN_SUCCESS;
}
}
wq_element = (wait_queue_element_t)
queue_next((queue_t) wq_element);
}
wait_queue_unlock(wq);
splx(s);
return KERN_NOT_IN_SET;
}
/*
* Routine: wait_queue_unlink_all_nofree_locked
* Purpose:
* Remove the linkage between a wait queue and all its sets.
* All the linkage structures are returned to the caller for
* later freeing.
* Conditions:
* Wait queue locked.
*/
static void
wait_queue_unlink_all_nofree_locked(
wait_queue_t wq,
queue_t links)
{
wait_queue_element_t wq_element;
wait_queue_element_t wq_next_element;
wait_queue_set_t wq_set;
wait_queue_link_t wql;
queue_t q;
q = &wq->wq_queue;
wq_element = (wait_queue_element_t) queue_first(q);
while (!queue_end(q, (queue_entry_t)wq_element)) {
WAIT_QUEUE_ELEMENT_CHECK(wq, wq_element);
wq_next_element = (wait_queue_element_t)
queue_next((queue_t) wq_element);
if (wq_element->wqe_type == WAIT_QUEUE_LINK ||
wq_element->wqe_type == WAIT_QUEUE_LINK_NOALLOC) {
wql = (wait_queue_link_t)wq_element;
wq_set = wql->wql_setqueue;
wqs_lock(wq_set);
wait_queue_unlink_locked(wq, wq_set, wql);
wqs_unlock(wq_set);
enqueue(links, &wql->wql_links);
}
wq_element = wq_next_element;
}
}
/*
* Routine: wait_queue_unlink_all_nofree
* Purpose:
* Remove the linkage between a wait queue and all its sets.
* All the linkage structures are returned to the caller for
* later freeing.
* Conditions:
* Nothing of interest locked.
*/
kern_return_t
wait_queue_unlink_all_nofree(
wait_queue_t wq,
queue_t links)
{
spl_t s;
if (!wait_queue_is_valid(wq)) {
printf("\nReturning KERN_INVALID_ARGUMENT from wait_queue_unlink_all_nofree\n");
return KERN_INVALID_ARGUMENT;
}
s = splsched();
wait_queue_lock(wq);
wait_queue_unlink_all_nofree_locked(wq, links);
wait_queue_unlock(wq);
splx(s);
return(KERN_SUCCESS);
}
/*
* Routine: wait_queue_unlink_all_locked
* Purpose:
* Remove the linkage between a locked wait queue and all its
* sets and enqueue the allocated ones onto the links queue
* provided.
* Conditions:
* Wait queue locked.
*/
static void
wait_queue_unlink_all_locked(
wait_queue_t wq,
queue_t links)
{
wait_queue_element_t wq_element;
wait_queue_element_t wq_next_element;
wait_queue_set_t wq_set;
wait_queue_link_t wql;
queue_t q;
q = &wq->wq_queue;
wq_element = (wait_queue_element_t) queue_first(q);
while (!queue_end(q, (queue_entry_t)wq_element)) {
boolean_t alloced;
WAIT_QUEUE_ELEMENT_CHECK(wq, wq_element);
wq_next_element = (wait_queue_element_t)
queue_next((queue_t) wq_element);
alloced = (wq_element->wqe_type == WAIT_QUEUE_LINK);
if (alloced || wq_element->wqe_type == WAIT_QUEUE_LINK_NOALLOC) {
wql = (wait_queue_link_t)wq_element;
wq_set = wql->wql_setqueue;
wqs_lock(wq_set);
wait_queue_unlink_locked(wq, wq_set, wql);
wqs_unlock(wq_set);
if (alloced)
enqueue(links, &wql->wql_links);
}
wq_element = wq_next_element;
}
}
/*
* Routine: wait_queue_unlink_all
* Purpose:
* Remove the linkage between a wait queue and all its sets.
* All the linkage structures that were allocated internally
* are freed. The others are the caller's responsibility.
* Conditions:
* Nothing of interest locked.
*/
kern_return_t
wait_queue_unlink_all(
wait_queue_t wq)
{
wait_queue_link_t wql;
queue_head_t links_queue_head;
queue_t links = &links_queue_head;
spl_t s;
if (!wait_queue_is_valid(wq)) {
printf("\nReturning KERN_INVALID_ARGUMENT from wait_queue_unlink_all\n");
return KERN_INVALID_ARGUMENT;
}
queue_init(links);
s = splsched();
wait_queue_lock(wq);
wait_queue_unlink_all_locked(wq, links);
wait_queue_unlock(wq);
splx(s);
while(!queue_empty(links)) {
wql = (wait_queue_link_t) dequeue(links);
zfree(_wait_queue_link_zone, wql);
}
return(KERN_SUCCESS);
}
/* legacy interface naming */
kern_return_t
wait_subqueue_unlink_all(
wait_queue_set_t wq_set)
{
return wait_queue_set_unlink_all(wq_set);
}
/*
* Routine: wait_queue_set_unlink_all_nofree
* Purpose:
* Remove the linkage between a set wait queue and all its
* member wait queues and all the sets it may be a member of.
* The links structures are returned for later freeing by the
* caller.
* Conditions:
* The wait queue must be a set
*/
kern_return_t
wait_queue_set_unlink_all_nofree(
wait_queue_set_t wq_set,
queue_t links)
{
wait_queue_link_t wql;
wait_queue_t wq;
queue_t q;
spl_t s;
if (!wait_queue_is_set(wq_set)) {
return KERN_INVALID_ARGUMENT;
}
retry:
s = splsched();
wqs_lock(wq_set);
/* remove the wait queues that are members of our set */
q = &wq_set->wqs_setlinks;
wql = (wait_queue_link_t)queue_first(q);
while (!queue_end(q, (queue_entry_t)wql)) {
WAIT_QUEUE_SET_LINK_CHECK(wq_set, wql);
wq = wql->wql_queue;
if (wait_queue_lock_try(wq)) {
wait_queue_unlink_locked(wq, wq_set, wql);
wait_queue_unlock(wq);
enqueue(links, &wql->wql_links);
wql = (wait_queue_link_t)queue_first(q);
} else {
wqs_unlock(wq_set);
splx(s);
delay(1);
goto retry;
}
}
/* remove this set from sets it belongs to */
wait_queue_unlink_all_nofree_locked(&wq_set->wqs_wait_queue, links);
wqs_unlock(wq_set);
splx(s);
return(KERN_SUCCESS);
}
/*
* Routine: wait_queue_set_unlink_all
* Purpose:
* Remove the linkage between a set wait queue and all its
* member wait queues and all the sets it may be members of.
* The link structures are freed for those links which were
* dynamically allocated.
* Conditions:
* The wait queue must be a set
*/
kern_return_t
wait_queue_set_unlink_all(
wait_queue_set_t wq_set)
{
wait_queue_link_t wql;
wait_queue_t wq;
queue_t q;
queue_head_t links_queue_head;
queue_t links = &links_queue_head;
spl_t s;
if (!wait_queue_is_set(wq_set)) {
return KERN_INVALID_ARGUMENT;
}
queue_init(links);
retry:
s = splsched();
wqs_lock(wq_set);
/* remove the wait queues that are members of our set */
q = &wq_set->wqs_setlinks;
wql = (wait_queue_link_t)queue_first(q);
while (!queue_end(q, (queue_entry_t)wql)) {
WAIT_QUEUE_SET_LINK_CHECK(wq_set, wql);
wq = wql->wql_queue;
if (wait_queue_lock_try(wq)) {
boolean_t alloced;
alloced = (wql->wql_type == WAIT_QUEUE_LINK);
wait_queue_unlink_locked(wq, wq_set, wql);
wait_queue_unlock(wq);
if (alloced)
enqueue(links, &wql->wql_links);
wql = (wait_queue_link_t)queue_first(q);
} else {
wqs_unlock(wq_set);
splx(s);
delay(1);
goto retry;
}
}
/* remove this set from sets it belongs to */
wait_queue_unlink_all_locked(&wq_set->wqs_wait_queue, links);
wqs_unlock(wq_set);
splx(s);
while (!queue_empty (links)) {
wql = (wait_queue_link_t) dequeue(links);
zfree(_wait_queue_link_zone, wql);
}
return(KERN_SUCCESS);
}
kern_return_t
wait_queue_set_unlink_one(
wait_queue_set_t wq_set,
wait_queue_link_t wql)
{
wait_queue_t wq;
spl_t s;
assert(wait_queue_is_set(wq_set));
retry:
s = splsched();
wqs_lock(wq_set);
WAIT_QUEUE_SET_CHECK(wq_set);
/* Already unlinked, e.g. by selclearthread() */
if (wql->wql_type == WAIT_QUEUE_UNLINKED) {
goto out;
}
WAIT_QUEUE_SET_LINK_CHECK(wq_set, wql);
/* On a wait queue, and we hold set queue lock ... */
wq = wql->wql_queue;
if (wait_queue_lock_try(wq)) {
wait_queue_unlink_locked(wq, wq_set, wql);
wait_queue_unlock(wq);
} else {
wqs_unlock(wq_set);
splx(s);
delay(1);
goto retry;
}
out:
wqs_unlock(wq_set);
splx(s);
return KERN_SUCCESS;
}
/*
* Routine: wait_queue_assert_wait64_locked
* Purpose:
* Insert the current thread into the supplied wait queue
* waiting for a particular event to be posted to that queue.
*
* Conditions:
* The wait queue is assumed locked.
* The waiting thread is assumed locked.
*
*/
__private_extern__ wait_result_t
wait_queue_assert_wait64_locked(
wait_queue_t wq,
event64_t event,
wait_interrupt_t interruptible,
wait_timeout_urgency_t urgency,
uint64_t deadline,
uint64_t leeway,
thread_t thread)
{
wait_result_t wait_result;
boolean_t realtime;
if (!wait_queue_assert_possible(thread))
panic("wait_queue_assert_wait64_locked");
if (wq->wq_type == _WAIT_QUEUE_SET_inited) {
wait_queue_set_t wqs = (wait_queue_set_t)wq;
if (event == NO_EVENT64 && wqs_is_preposted(wqs))
return(THREAD_AWAKENED);
}
/*
* Realtime threads get priority for wait queue placements.
* This allows wait_queue_wakeup_one to prefer a waiting
* realtime thread, similar in principle to performing
* a wait_queue_wakeup_all and allowing scheduler prioritization
* to run the realtime thread, but without causing the
* lock contention of that scenario.
*/
realtime = (thread->sched_pri >= BASEPRI_REALTIME);
/*
* This is the extent to which we currently take scheduling attributes
* into account. If the thread is vm priviledged, we stick it at
* the front of the queue. Later, these queues will honor the policy
* value set at wait_queue_init time.
*/
wait_result = thread_mark_wait_locked(thread, interruptible);
if (wait_result == THREAD_WAITING) {
if (!wq->wq_fifo
|| (thread->options & TH_OPT_VMPRIV)
|| realtime)
enqueue_head(&wq->wq_queue, (queue_entry_t) thread);
else
enqueue_tail(&wq->wq_queue, (queue_entry_t) thread);
thread->wait_event = event;
thread->wait_queue = wq;
if (deadline != 0) {
if (!timer_call_enter_with_leeway(&thread->wait_timer, NULL,
deadline, leeway, urgency, FALSE))
thread->wait_timer_active++;
thread->wait_timer_is_set = TRUE;
}
if (wait_queue_global(wq)) {
wq->wq_eventmask = wq->wq_eventmask | CAST_TO_EVENT_MASK(event);
}
}
return(wait_result);
}
/*
* Routine: wait_queue_assert_wait
* Purpose:
* Insert the current thread into the supplied wait queue
* waiting for a particular event to be posted to that queue.
*
* Conditions:
* nothing of interest locked.
*/
wait_result_t
wait_queue_assert_wait(
wait_queue_t wq,
event_t event,
wait_interrupt_t interruptible,
uint64_t deadline)
{
spl_t s;
wait_result_t ret;
thread_t thread = current_thread();
/* If it is an invalid wait queue, you can't wait on it */
if (!wait_queue_is_valid(wq)) {
printf("\nReturning thread->wait_result = THREAD_RESTART from wait_queue_assert_wait\n");
return (thread->wait_result = THREAD_RESTART);
}
s = splsched();
wait_queue_lock(wq);
thread_lock(thread);
ret = wait_queue_assert_wait64_locked(wq, CAST_DOWN(event64_t,event),
interruptible,
TIMEOUT_URGENCY_SYS_NORMAL,
deadline, 0,
thread);
thread_unlock(thread);
wait_queue_unlock(wq);
splx(s);
return(ret);
}
/*
* Routine: wait_queue_assert_wait_with_leeway
* Purpose:
* Insert the current thread into the supplied wait queue
* waiting for a particular event to be posted to that queue.
* Deadline values are specified with urgency and leeway.
*
* Conditions:
* nothing of interest locked.
*/
wait_result_t
wait_queue_assert_wait_with_leeway(
wait_queue_t wq,
event_t event,
wait_interrupt_t interruptible,
wait_timeout_urgency_t urgency,
uint64_t deadline,
uint64_t leeway)
{
spl_t s;
wait_result_t ret;
thread_t thread = current_thread();
/* If it is an invalid wait queue, you can't wait on it */
if (!wait_queue_is_valid(wq)) {
printf("\nReturning thread->wait_result = THREAD_RESTART from wait_queue_assert_wait_with_leeway\n");
return (thread->wait_result = THREAD_RESTART);
}
s = splsched();
wait_queue_lock(wq);
thread_lock(thread);
ret = wait_queue_assert_wait64_locked(wq, CAST_DOWN(event64_t,event),
interruptible,
urgency, deadline, leeway,
thread);
thread_unlock(thread);
wait_queue_unlock(wq);
splx(s);
return(ret);
}
/*
* Routine: wait_queue_assert_wait64
* Purpose:
* Insert the current thread into the supplied wait queue
* waiting for a particular event to be posted to that queue.
* Conditions:
* nothing of interest locked.
*/
wait_result_t
wait_queue_assert_wait64(
wait_queue_t wq,
event64_t event,
wait_interrupt_t interruptible,
uint64_t deadline)
{
spl_t s;
wait_result_t ret;
thread_t thread = current_thread();
/* If it is an invalid wait queue, you cant wait on it */
if (!wait_queue_is_valid(wq)) {
printf("\nReturning thread->wait_result = THREAD_RESTART from wait_queue_assert_wait64\n");
return (thread->wait_result = THREAD_RESTART);
}
s = splsched();
wait_queue_lock(wq);
thread_lock(thread);
ret = wait_queue_assert_wait64_locked(wq, event, interruptible,
TIMEOUT_URGENCY_SYS_NORMAL,
deadline, 0,
thread);
thread_unlock(thread);
wait_queue_unlock(wq);
splx(s);
return(ret);
}
/*
* Routine: wait_queue_assert_wait64_with_leeway
* Purpose:
* Insert the current thread into the supplied wait queue
* waiting for a particular event to be posted to that queue.
* Deadline values are specified with urgency and leeway.
* Conditions:
* nothing of interest locked.
*/
wait_result_t
wait_queue_assert_wait64_with_leeway(
wait_queue_t wq,
event64_t event,
wait_interrupt_t interruptible,
wait_timeout_urgency_t urgency,
uint64_t deadline,
uint64_t leeway)
{
spl_t s;
wait_result_t ret;
thread_t thread = current_thread();
/* If it is an invalid wait queue, you cant wait on it */
if (!wait_queue_is_valid(wq)) {
printf("\nReturning thread->wait_result = THREAD_RESTART from wait_queue_assert_wait64_with_leeway\n");
return (thread->wait_result = THREAD_RESTART);
}
s = splsched();
wait_queue_lock(wq);
thread_lock(thread);
ret = wait_queue_assert_wait64_locked(wq, event, interruptible,
urgency, deadline, leeway,
thread);
thread_unlock(thread);
wait_queue_unlock(wq);
splx(s);
return(ret);
}
/*
* Routine: _wait_queue_select64_all
* Purpose:
* Select all threads off a wait queue that meet the
* supplied criteria.
* Conditions:
* at splsched
* wait queue locked
* wake_queue initialized and ready for insertion
* possibly recursive
* Returns:
* a queue of locked threads
*/
static void
_wait_queue_select64_all(
wait_queue_t wq,
event64_t event,
queue_t wake_queue)
{
wait_queue_element_t wq_element;
wait_queue_element_t wqe_next;
unsigned long eventmask = 0;
boolean_t is_queue_global = FALSE;
queue_t q;
is_queue_global = wait_queue_global(wq);
if (is_queue_global) {
eventmask = CAST_TO_EVENT_MASK(event);
if ((wq->wq_eventmask & eventmask) != eventmask) {
return;
}
eventmask = 0;
}
q = &wq->wq_queue;
wq_element = (wait_queue_element_t) queue_first(q);
while (!queue_end(q, (queue_entry_t)wq_element)) {
WAIT_QUEUE_ELEMENT_CHECK(wq, wq_element);
wqe_next = (wait_queue_element_t)
queue_next((queue_t) wq_element);
/*
* We may have to recurse if this is a compound wait queue.
*/
if (wq_element->wqe_type == WAIT_QUEUE_LINK ||
wq_element->wqe_type == WAIT_QUEUE_LINK_NOALLOC) {
wait_queue_link_t wql = (wait_queue_link_t)wq_element;
wait_queue_set_t set_queue = wql->wql_setqueue;
/*
* We have to check the set wait queue. If it is marked
* as pre-post, and it is the "generic event" then mark
* it pre-posted now (if not already).
*/
wqs_lock(set_queue);
if (event == NO_EVENT64 && set_queue->wqs_prepost && !wql_is_preposted(wql)) {
queue_t ppq = &set_queue->wqs_preposts;
queue_enter(ppq, wql, wait_queue_link_t, wql_preposts);
}
if (! wait_queue_empty(&set_queue->wqs_wait_queue))
_wait_queue_select64_all(&set_queue->wqs_wait_queue, event, wake_queue);
wqs_unlock(set_queue);
} else {
/*
* Otherwise, its a thread. If it is waiting on
* the event we are posting to this queue, pull
* it off the queue and stick it in out wake_queue.
*/
thread_t t = (thread_t)(void *)wq_element;
if (t->wait_event == event) {
thread_lock(t);
remqueue((queue_entry_t) t);
enqueue (wake_queue, (queue_entry_t) t);
t->wait_queue = WAIT_QUEUE_NULL;
t->wait_event = NO_EVENT64;
t->at_safe_point = FALSE;
/* returned locked */
} else {
if (is_queue_global) {
eventmask = eventmask |
CAST_TO_EVENT_MASK(t->wait_event);
}
}
}
wq_element = wqe_next;
}
/* Update event mask if global wait queue */
if (is_queue_global) {
wq->wq_eventmask = eventmask;
}
}
/*
* Routine: wait_queue_wakeup64_all_locked
* Purpose:
* Wakeup some number of threads that are in the specified
* wait queue and waiting on the specified event.
* Conditions:
* wait queue already locked (may be released).
* Returns:
* KERN_SUCCESS - Threads were woken up
* KERN_NOT_WAITING - No threads were waiting <wq,event> pair
*/
__private_extern__ kern_return_t
wait_queue_wakeup64_all_locked(
wait_queue_t wq,
event64_t event,
wait_result_t result,
boolean_t unlock)
{
queue_head_t wake_queue_head;
queue_t q = &wake_queue_head;
kern_return_t res;
// assert(wait_queue_held(wq));
// if(!wq->wq_interlock.lock_data) { /* (BRINGUP */
// panic("wait_queue_wakeup64_all_locked: lock not held on %p\n", wq); /* (BRINGUP) */
// }
queue_init(q);
/*
* Select the threads that we will wake up. The threads
* are returned to us locked and cleanly removed from the
* wait queue.
*/
_wait_queue_select64_all(wq, event, q);
if (unlock)
wait_queue_unlock(wq);
/*
* For each thread, set it running.
*/
res = KERN_NOT_WAITING;
while (!queue_empty (q)) {
thread_t thread = (thread_t)(void *) dequeue(q);
res = thread_go(thread, result);
assert(res == KERN_SUCCESS);
thread_unlock(thread);
}
return res;
}
/*
* Routine: wait_queue_wakeup_all
* Purpose:
* Wakeup some number of threads that are in the specified
* wait queue and waiting on the specified event.
* Conditions:
* Nothing locked
* Returns:
* KERN_SUCCESS - Threads were woken up
* KERN_NOT_WAITING - No threads were waiting <wq,event> pair
*/
kern_return_t
wait_queue_wakeup_all(
wait_queue_t wq,
event_t event,
wait_result_t result)
{
kern_return_t ret;
spl_t s;
if (!wait_queue_is_valid(wq)) {
printf("\nReturning KERN_INVALID_ARGUMENT from wait_queue_wakeup_all\n");
return KERN_INVALID_ARGUMENT;
}
s = splsched();
wait_queue_lock(wq);
// if(!wq->wq_interlock.lock_data) { /* (BRINGUP */
// panic("wait_queue_wakeup_all: we did not get the lock on %p\n", wq); /* (BRINGUP) */
// }
ret = wait_queue_wakeup64_all_locked(
wq, CAST_DOWN(event64_t,event),
result, TRUE);
/* lock released */
splx(s);
return ret;
}
/*
* Routine: wait_queue_wakeup64_all
* Purpose:
* Wakeup some number of threads that are in the specified
* wait queue and waiting on the specified event.
* Conditions:
* Nothing locked
* Returns:
* KERN_SUCCESS - Threads were woken up
* KERN_NOT_WAITING - No threads were waiting <wq,event> pair
*/
kern_return_t
wait_queue_wakeup64_all(
wait_queue_t wq,
event64_t event,
wait_result_t result)
{
kern_return_t ret;
spl_t s;
if (!wait_queue_is_valid(wq)) {
printf("\nReturning KERN_INVALID_ARGUMENT from wait_queue_wakeup64_all\n");
return KERN_INVALID_ARGUMENT;
}
s = splsched();
wait_queue_lock(wq);
ret = wait_queue_wakeup64_all_locked(wq, event, result, TRUE);
/* lock released */
splx(s);
return ret;
}
/*
* Routine: _wait_queue_select64_one
* Purpose:
* Select the best thread off a wait queue that meet the
* supplied criteria.
* Conditions:
* at splsched
* wait queue locked
* possibly recursive
* Returns:
* a locked thread - if one found
* Note:
* This is where the sync policy of the wait queue comes
* into effect. For now, we just assume FIFO/LIFO.
*/
static thread_t
_wait_queue_select64_one(
wait_queue_t wq,
event64_t event)
{
wait_queue_element_t wq_element;
wait_queue_element_t wqe_next;
thread_t t = THREAD_NULL;
thread_t fifo_thread = THREAD_NULL;
boolean_t is_queue_fifo = TRUE;
boolean_t is_queue_global = FALSE;
boolean_t thread_imp_donor = FALSE;
boolean_t realtime = FALSE;
unsigned long eventmask = 0;
queue_t q;
if (wait_queue_global(wq)) {
eventmask = CAST_TO_EVENT_MASK(event);
if ((wq->wq_eventmask & eventmask) != eventmask) {
return THREAD_NULL;
}
eventmask = 0;
is_queue_global = TRUE;
#if IMPORTANCE_INHERITANCE
is_queue_fifo = FALSE;
#endif /* IMPORTANCE_INHERITANCE */
}
q = &wq->wq_queue;
wq_element = (wait_queue_element_t) queue_first(q);
while (!queue_end(q, (queue_entry_t)wq_element)) {
WAIT_QUEUE_ELEMENT_CHECK(wq, wq_element);
wqe_next = (wait_queue_element_t)
queue_next((queue_t) wq_element);
/*
* We may have to recurse if this is a compound wait queue.
*/
if (wq_element->wqe_type == WAIT_QUEUE_LINK ||
wq_element->wqe_type == WAIT_QUEUE_LINK_NOALLOC) {
wait_queue_link_t wql = (wait_queue_link_t)wq_element;
wait_queue_set_t set_queue = wql->wql_setqueue;
/*
* We have to check the set wait queue. If the set
* supports pre-posting, it isn't already preposted,
* and we didn't find a thread in the set, then mark it.
*
* If we later find a thread, there may be a spurious
* pre-post here on this set. The wait side has to check
* for that either pre- or post-wait.
*/
wqs_lock(set_queue);
if (! wait_queue_empty(&set_queue->wqs_wait_queue)) {
t = _wait_queue_select64_one(&set_queue->wqs_wait_queue, event);
}
if (t != THREAD_NULL) {
wqs_unlock(set_queue);
return t;
}
if (event == NO_EVENT64 && set_queue->wqs_prepost && !wql_is_preposted(wql)) {
queue_t ppq = &set_queue->wqs_preposts;
queue_enter(ppq, wql, wait_queue_link_t, wql_preposts);
}
wqs_unlock(set_queue);
} else {
/*
* Otherwise, its a thread. If it is waiting on
* the event we are posting to this queue, pull
* it off the queue and stick it in out wake_queue.
*/
t = (thread_t)(void *)wq_element;
if (t->wait_event == event) {
if (fifo_thread == THREAD_NULL) {
fifo_thread = t;
}
#if IMPORTANCE_INHERITANCE
/*
* Checking imp donor bit does not need thread lock or
* or task lock since we have the wait queue lock and
* thread can not be removed from it without acquiring
* wait queue lock. The imp donor bit may change
* once we read its value, but it is ok to wake
* a thread while someone drops importance assertion
* on the that thread.
*/
thread_imp_donor = task_is_importance_donor(t->task);
#endif /* IMPORTANCE_INHERITANCE */
realtime = (t->sched_pri >= BASEPRI_REALTIME);
if (is_queue_fifo || thread_imp_donor || realtime ||
(t->options & TH_OPT_VMPRIV)) {
thread_lock(t);
remqueue((queue_entry_t) t);
t->wait_queue = WAIT_QUEUE_NULL;
t->wait_event = NO_EVENT64;
t->at_safe_point = FALSE;
return t; /* still locked */
}
}
if (is_queue_global) {
eventmask = eventmask | CAST_TO_EVENT_MASK(t->wait_event);
}
t = THREAD_NULL;
}
wq_element = wqe_next;
}
if (is_queue_global) {
wq->wq_eventmask = eventmask;
}
#if IMPORTANCE_INHERITANCE
if (fifo_thread != THREAD_NULL) {
thread_lock(fifo_thread);
remqueue((queue_entry_t) fifo_thread);
fifo_thread->wait_queue = WAIT_QUEUE_NULL;
fifo_thread->wait_event = NO_EVENT64;
fifo_thread->at_safe_point = FALSE;
return fifo_thread; /* still locked */
}
#endif /* IMPORTANCE_INHERITANCE */
return THREAD_NULL;
}
/*
* Routine: wait_queue_pull_thread_locked
* Purpose:
* Pull a thread off its wait queue and (possibly) unlock
* the waitq.
* Conditions:
* at splsched
* wait queue locked
* thread locked
* Returns:
* with the thread still locked.
*/
void
wait_queue_pull_thread_locked(
wait_queue_t waitq,
thread_t thread,
boolean_t unlock)
{
assert(thread->wait_queue == waitq);
remqueue((queue_entry_t)thread );
thread->wait_queue = WAIT_QUEUE_NULL;
thread->wait_event = NO_EVENT64;
thread->at_safe_point = FALSE;
if (unlock)
wait_queue_unlock(waitq);
}
/*
* Routine: wait_queue_select64_thread
* Purpose:
* Look for a thread and remove it from the queues, if
* (and only if) the thread is waiting on the supplied
* <wait_queue, event> pair.
* Conditions:
* at splsched
* wait queue locked
* possibly recursive
* Returns:
* KERN_NOT_WAITING: Thread is not waiting here.
* KERN_SUCCESS: It was, and is now removed (returned locked)
*/
static kern_return_t
_wait_queue_select64_thread(
wait_queue_t wq,
event64_t event,
thread_t thread)
{
wait_queue_element_t wq_element;
wait_queue_element_t wqe_next;
kern_return_t res = KERN_NOT_WAITING;
queue_t q = &wq->wq_queue;
thread_lock(thread);
if ((thread->wait_queue == wq) && (thread->wait_event == event)) {
remqueue((queue_entry_t) thread);
thread->at_safe_point = FALSE;
thread->wait_event = NO_EVENT64;
thread->wait_queue = WAIT_QUEUE_NULL;
/* thread still locked */
return KERN_SUCCESS;
}
thread_unlock(thread);
/*
* The wait_queue associated with the thread may be one of this
* wait queue's sets. Go see. If so, removing it from
* there is like removing it from here.
*/
wq_element = (wait_queue_element_t) queue_first(q);
while (!queue_end(q, (queue_entry_t)wq_element)) {
WAIT_QUEUE_ELEMENT_CHECK(wq, wq_element);
wqe_next = (wait_queue_element_t)
queue_next((queue_t) wq_element);
if (wq_element->wqe_type == WAIT_QUEUE_LINK ||
wq_element->wqe_type == WAIT_QUEUE_LINK_NOALLOC) {
wait_queue_link_t wql = (wait_queue_link_t)wq_element;
wait_queue_set_t set_queue = wql->wql_setqueue;
wqs_lock(set_queue);
if (! wait_queue_empty(&set_queue->wqs_wait_queue)) {
res = _wait_queue_select64_thread(&set_queue->wqs_wait_queue,
event,
thread);
}
wqs_unlock(set_queue);
if (res == KERN_SUCCESS)
return KERN_SUCCESS;
}
wq_element = wqe_next;
}
return res;
}
/*
* Routine: wait_queue_wakeup64_identity_locked
* Purpose:
* Select a single thread that is most-eligible to run and set
* set it running. But return the thread locked.
*
* Conditions:
* at splsched
* wait queue locked
* possibly recursive
* Returns:
* a pointer to the locked thread that was awakened
*/
__private_extern__ thread_t
wait_queue_wakeup64_identity_locked(
wait_queue_t wq,
event64_t event,
wait_result_t result,
boolean_t unlock)
{
kern_return_t res;
thread_t thread;
assert(wait_queue_held(wq));
thread = _wait_queue_select64_one(wq, event);
if (unlock)
wait_queue_unlock(wq);
if (thread) {
res = thread_go(thread, result);
assert(res == KERN_SUCCESS);
}
return thread; /* still locked if not NULL */
}
/*
* Routine: wait_queue_wakeup64_one_locked
* Purpose:
* Select a single thread that is most-eligible to run and set
* set it runnings.
*
* Conditions:
* at splsched
* wait queue locked
* possibly recursive
* Returns:
* KERN_SUCCESS: It was, and is, now removed.
* KERN_NOT_WAITING - No thread was waiting <wq,event> pair
*/
__private_extern__ kern_return_t
wait_queue_wakeup64_one_locked(
wait_queue_t wq,
event64_t event,
wait_result_t result,
boolean_t unlock)
{
thread_t thread;
assert(wait_queue_held(wq));
thread = _wait_queue_select64_one(wq, event);
if (unlock)
wait_queue_unlock(wq);
if (thread) {
kern_return_t res;
res = thread_go(thread, result);
assert(res == KERN_SUCCESS);
thread_unlock(thread);
return res;
}
return KERN_NOT_WAITING;
}
/*
* Routine: wait_queue_wakeup_one
* Purpose:
* Wakeup the most appropriate thread that is in the specified
* wait queue for the specified event.
* Conditions:
* Nothing locked
* Returns:
* KERN_SUCCESS - Thread was woken up
* KERN_NOT_WAITING - No thread was waiting <wq,event> pair
*/
kern_return_t
wait_queue_wakeup_one(
wait_queue_t wq,
event_t event,
wait_result_t result,
int priority)
{
thread_t thread;
spl_t s;
if (!wait_queue_is_valid(wq)) {
printf("\nReturning KERN_INVALID_ARGUMENT from wait_queue_wakeup_one\n");
return KERN_INVALID_ARGUMENT;
}
s = splsched();
wait_queue_lock(wq);
thread = _wait_queue_select64_one(wq, CAST_DOWN(event64_t,event));
wait_queue_unlock(wq);
if (thread) {
kern_return_t res;
if (thread->sched_pri < priority) {
if (priority <= MAXPRI) {
set_sched_pri(thread, priority);
thread->was_promoted_on_wakeup = 1;
thread->sched_flags |= TH_SFLAG_PROMOTED;
}
}
res = thread_go(thread, result);
assert(res == KERN_SUCCESS);
thread_unlock(thread);
splx(s);
return res;
}
splx(s);
return KERN_NOT_WAITING;
}
/*
* Routine: wait_queue_wakeup64_one
* Purpose:
* Wakeup the most appropriate thread that is in the specified
* wait queue for the specified event.
* Conditions:
* Nothing locked
* Returns:
* KERN_SUCCESS - Thread was woken up
* KERN_NOT_WAITING - No thread was waiting <wq,event> pair
*/
kern_return_t
wait_queue_wakeup64_one(
wait_queue_t wq,
event64_t event,
wait_result_t result)
{
thread_t thread;
spl_t s;
if (!wait_queue_is_valid(wq)) {
printf("\nReturning KERN_INVALID_ARGUMENT from wait_queue_wakeup64_one\n");
return KERN_INVALID_ARGUMENT;
}
s = splsched();
wait_queue_lock(wq);
thread = _wait_queue_select64_one(wq, event);
wait_queue_unlock(wq);
if (thread) {
kern_return_t res;
res = thread_go(thread, result);
assert(res == KERN_SUCCESS);
thread_unlock(thread);
splx(s);
return res;
}
splx(s);
return KERN_NOT_WAITING;
}
/*
* Routine: wait_queue_wakeup64_thread_locked
* Purpose:
* Wakeup the particular thread that was specified if and only
* it was in this wait queue (or one of it's set queues)
* and waiting on the specified event.
*
* This is much safer than just removing the thread from
* whatever wait queue it happens to be on. For instance, it
* may have already been awoken from the wait you intended to
* interrupt and waited on something else (like another
* semaphore).
* Conditions:
* at splsched
* wait queue already locked (may be released).
* Returns:
* KERN_SUCCESS - the thread was found waiting and awakened
* KERN_NOT_WAITING - the thread was not waiting here
*/
__private_extern__ kern_return_t
wait_queue_wakeup64_thread_locked(
wait_queue_t wq,
event64_t event,
thread_t thread,
wait_result_t result,
boolean_t unlock)
{
kern_return_t res;
assert(wait_queue_held(wq));
/*
* See if the thread was still waiting there. If so, it got
* dequeued and returned locked.
*/
res = _wait_queue_select64_thread(wq, event, thread);
if (unlock)
wait_queue_unlock(wq);
if (res != KERN_SUCCESS)
return KERN_NOT_WAITING;
res = thread_go(thread, result);
assert(res == KERN_SUCCESS);
thread_unlock(thread);
return res;
}
/*
* Routine: wait_queue_wakeup_thread
* Purpose:
* Wakeup the particular thread that was specified if and only
* it was in this wait queue (or one of it's set queues)
* and waiting on the specified event.
*
* This is much safer than just removing the thread from
* whatever wait queue it happens to be on. For instance, it
* may have already been awoken from the wait you intended to
* interrupt and waited on something else (like another
* semaphore).
* Conditions:
* nothing of interest locked
* we need to assume spl needs to be raised
* Returns:
* KERN_SUCCESS - the thread was found waiting and awakened
* KERN_NOT_WAITING - the thread was not waiting here
*/
kern_return_t
wait_queue_wakeup_thread(
wait_queue_t wq,
event_t event,
thread_t thread,
wait_result_t result)
{
kern_return_t res;
spl_t s;
if (!wait_queue_is_valid(wq)) {
printf("\nReturning KERN_INVALID_ARGUMENT from wait_queue_wakeup_thread\n");
return KERN_INVALID_ARGUMENT;
}
s = splsched();
wait_queue_lock(wq);
res = _wait_queue_select64_thread(wq, CAST_DOWN(event64_t,event), thread);
wait_queue_unlock(wq);
if (res == KERN_SUCCESS) {
res = thread_go(thread, result);
assert(res == KERN_SUCCESS);
thread_unlock(thread);
splx(s);
return res;
}
splx(s);
return KERN_NOT_WAITING;
}
/*
* Routine: wait_queue_wakeup64_thread
* Purpose:
* Wakeup the particular thread that was specified if and only
* it was in this wait queue (or one of it's set's queues)
* and waiting on the specified event.
*
* This is much safer than just removing the thread from
* whatever wait queue it happens to be on. For instance, it
* may have already been awoken from the wait you intended to
* interrupt and waited on something else (like another
* semaphore).
* Conditions:
* nothing of interest locked
* we need to assume spl needs to be raised
* Returns:
* KERN_SUCCESS - the thread was found waiting and awakened
* KERN_NOT_WAITING - the thread was not waiting here
*/
kern_return_t
wait_queue_wakeup64_thread(
wait_queue_t wq,
event64_t event,
thread_t thread,
wait_result_t result)
{
kern_return_t res;
spl_t s;
if (!wait_queue_is_valid(wq)) {
printf("\nReturning KERN_INVALID_ARGUMENT from wait_queue_wakeup64_thread\n");
return KERN_INVALID_ARGUMENT;
}
s = splsched();
wait_queue_lock(wq);
res = _wait_queue_select64_thread(wq, event, thread);
wait_queue_unlock(wq);
if (res == KERN_SUCCESS) {
res = thread_go(thread, result);
assert(res == KERN_SUCCESS);
thread_unlock(thread);
splx(s);
return res;
}
splx(s);
return KERN_NOT_WAITING;
}
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment