haiku/src/system/kernel/wait_for_objects.cpp

/*
 * Copyright 2007-2008, Ingo Weinhold, bonefish@cs.tu-berlin.de.
 * Copyright 2002-2008, Axel Dörfler, axeld@pinc-software.de.
 * Distributed under the terms of the MIT License.
 */

#include <fs/select_sync_pool.h>
#include <wait_for_objects.h>

#include <new>

#include <poll.h>
#include <signal.h>
#include <stdlib.h>
#include <string.h>
#include <sys/select.h>

#include <OS.h>
#include <Select.h>

#include <AutoDeleter.h>

#include <fs/fd.h>
#include <port.h>
#include <sem.h>
#include <syscalls.h>
#include <syscall_restart.h>
#include <thread.h>
#include <util/AutoLock.h>
#include <util/DoublyLinkedList.h>
#include <vfs.h>


//#define TRACE_WAIT_FOR_OBJECTS
#ifdef TRACE_WAIT_FOR_OBJECTS
#	define PRINT(x) dprintf x
#	define FUNCTION(x) dprintf x
#else
#	define PRINT(x) ;
#	define FUNCTION(x) ;
#endif


using std::nothrow;


struct select_sync_pool_entry
	: DoublyLinkedListLinkImpl<select_sync_pool_entry> {
	selectsync			*sync;
	uint16				events;
};

typedef DoublyLinkedList<select_sync_pool_entry> SelectSyncPoolEntryList;

struct select_sync_pool {
	SelectSyncPoolEntryList	entries;
};


struct select_ops {
	status_t (*select)(int32 object, struct select_info* info, bool kernel);
	status_t (*deselect)(int32 object, struct select_info* info, bool kernel);
};


static const select_ops kSelectOps[] = {
	// B_OBJECT_TYPE_FD
	{
		select_fd,
		deselect_fd
	},

	// B_OBJECT_TYPE_SEMAPHORE
	{
		select_sem,
		deselect_sem
	},

	// B_OBJECT_TYPE_PORT
	{
		select_port,
		deselect_port
	},

	// B_OBJECT_TYPE_THREAD
	{
		select_thread,
		deselect_thread
	}
};

static const uint32 kSelectOpsCount = sizeof(kSelectOps) / sizeof(select_ops);


/*!
	Clears all bits in the fd_set - since we are using variable sized
	arrays in the kernel, we can't use the FD_ZERO() macro provided by
	sys/select.h for this task.
	All other FD_xxx() macros are safe to use, though.
*/
static inline void
fd_zero(fd_set *set, int numFDs)
{
	if (set != NULL)
		memset(set, 0, _howmany(numFDs, NFDBITS) * sizeof(fd_mask));
}


static status_t
create_select_sync(int numFDs, select_sync*& _sync)
{
	// create sync structure
	select_sync* sync = new(nothrow) select_sync;
	if (sync == NULL)
		return B_NO_MEMORY;
	ObjectDeleter<select_sync> syncDeleter(sync);

	// create info set
	sync->set = new(nothrow) select_info[numFDs];
	if (sync->set == NULL)
		return B_NO_MEMORY;
	ArrayDeleter<select_info> setDeleter(sync->set);

	// create select event semaphore
	sync->sem = create_sem(0, "select");
	if (sync->sem < 0)
		return sync->sem;

	// create lock
	status_t error = benaphore_init(&sync->lock, "select sync");
	if (error != B_OK) {
		delete_sem(sync->sem);
		return error;
	}

	sync->count = numFDs;
	sync->ref_count = 1;

	for (int i = 0; i < numFDs; i++) {
		sync->set[i].next = NULL;
		sync->set[i].sync = sync;
	}

	setDeleter.Detach();
	syncDeleter.Detach();
	_sync = sync;

	return B_OK;
}


void
put_select_sync(select_sync* sync)
{
	FUNCTION(("put_select_sync(%p): -> %ld\n", sync, sync->ref_count - 1));

	if (atomic_add(&sync->ref_count, -1) == 1) {
		delete_sem(sync->sem);
		benaphore_destroy(&sync->lock);
		delete[] sync->set;
		delete sync;
	}
}


static int
common_select(int numFDs, fd_set *readSet, fd_set *writeSet, fd_set *errorSet,
	bigtime_t timeout, const sigset_t *sigMask, bool kernel)
{
	status_t status = B_OK;
	int fd;

	FUNCTION(("[%ld] common_select(%d, %p, %p, %p, %lld, %p, %d)\n",
		find_thread(NULL), numFDs, readSet, writeSet, errorSet, timeout,
		sigMask, kernel));

	// check if fds are valid before doing anything

	for (fd = 0; fd < numFDs; fd++) {
		if (((readSet && FD_ISSET(fd, readSet))
			|| (writeSet && FD_ISSET(fd, writeSet))
			|| (errorSet && FD_ISSET(fd, errorSet)))
			&& !fd_is_valid(fd, kernel))
			return B_FILE_ERROR;
	}

	// allocate sync object
	select_sync* sync;
	status = create_select_sync(numFDs, sync);
	if (status != B_OK)
		return status;

	// start selecting file descriptors

	BenaphoreLocker locker(sync->lock);

	for (fd = 0; fd < numFDs; fd++) {
		sync->set[fd].selected_events = 0;
		sync->set[fd].events = 0;

		if (readSet && FD_ISSET(fd, readSet))
			sync->set[fd].selected_events = SELECT_FLAG(B_SELECT_READ);
		if (writeSet && FD_ISSET(fd, writeSet))
			sync->set[fd].selected_events |= SELECT_FLAG(B_SELECT_WRITE);
		if (errorSet && FD_ISSET(fd, errorSet))
			sync->set[fd].selected_events |= SELECT_FLAG(B_SELECT_ERROR);

		if (sync->set[fd].selected_events != 0) {
			select_fd(fd, sync->set + fd, kernel);
				// array position is the same as the fd for select()
		}
	}

	locker.Unlock();

	// set new signal mask
	sigset_t oldSigMask;
	if (sigMask != NULL)
		sigprocmask(SIG_SETMASK, sigMask, &oldSigMask);

	// wait for something to happen
	status = acquire_sem_etc(sync->sem, 1,
		B_CAN_INTERRUPT | (timeout != -1 ? B_ABSOLUTE_TIMEOUT : 0), timeout);

	// restore the old signal mask
	if (sigMask != NULL)
		sigprocmask(SIG_SETMASK, &oldSigMask, NULL);

	PRINT(("common_select(): acquire_sem_etc() returned: %lx\n", status));

	// deselect file descriptors

	locker.Lock();

	for (fd = 0; fd < numFDs; fd++)
		deselect_fd(fd, sync->set + fd, kernel);

	PRINT(("common_select(): events deselected\n"));

	// collect the events that happened in the meantime

	int count = 0;

	if (status == B_INTERRUPTED) {
		// We must not clear the sets in this case, as applications may
		// rely on the contents of them.
		locker.Unlock();
		put_select_sync(sync);
		return B_INTERRUPTED;
	}

	// Clear sets to store the received events
	// (we can't use the macros, because we have variable sized arrays;
	// the other FD_xxx() macros are safe, though).
	fd_zero(readSet, numFDs);
	fd_zero(writeSet, numFDs);
	fd_zero(errorSet, numFDs);

	if (status == B_OK) {
		for (count = 0, fd = 0;fd < numFDs; fd++) {
			if (readSet && sync->set[fd].events & SELECT_FLAG(B_SELECT_READ)) {
				FD_SET(fd, readSet);
				count++;
			}
			if (writeSet
				&& sync->set[fd].events & SELECT_FLAG(B_SELECT_WRITE)) {
				FD_SET(fd, writeSet);
				count++;
			}
			if (errorSet
				&& sync->set[fd].events & SELECT_FLAG(B_SELECT_ERROR)) {
				FD_SET(fd, errorSet);
				count++;
			}
		}
	}

	// B_TIMED_OUT and B_WOULD_BLOCK are supposed to return 0

	locker.Unlock();
	put_select_sync(sync);

	return count;
}


static int
common_poll(struct pollfd *fds, nfds_t numFDs, bigtime_t timeout, bool kernel)
{
	status_t status = B_OK;
	int count = 0;
	uint32 i;

	// allocate sync object
	select_sync* sync;
	status = create_select_sync(numFDs, sync);
	if (status != B_OK)
		return status;

	// start polling file descriptors (by selecting them)

	BenaphoreLocker locker(sync->lock);

	for (i = 0; i < numFDs; i++) {
		int fd = fds[i].fd;

		// initialize events masks
		sync->set[i].selected_events = fds[i].events & ~POLLNVAL
			| POLLERR | POLLHUP;
		sync->set[i].events = 0;

		if (select_fd(fd, sync->set + i, kernel) == B_OK) {
			fds[i].revents = 0;
			if (sync->set[i].selected_events != 0)
				count++;
		} else
			fds[i].revents = POLLNVAL;
	}

	if (count < 1) {
		count = B_BAD_VALUE;
		goto err;
	}

	locker.Unlock();

	status = acquire_sem_etc(sync->sem, 1,
		B_CAN_INTERRUPT | (timeout != -1 ? B_ABSOLUTE_TIMEOUT : 0), timeout);

	// deselect file descriptors

	locker.Lock();

	for (i = 0; i < numFDs; i++)
		deselect_fd(fds[i].fd, sync->set + i, kernel);

	// collect the events that are happened in the meantime

	switch (status) {
		case B_OK:
			for (count = 0, i = 0; i < numFDs; i++) {
				if (fds[i].revents == POLLNVAL)
					continue;

				// POLLxxx flags and B_SELECT_xxx flags are compatible
				fds[i].revents = sync->set[i].events
					& sync->set[i].selected_events;
				if (fds[i].revents != 0)
					count++;
			}
			break;
		case B_INTERRUPTED:
			count = B_INTERRUPTED;
			break;
		default:
			// B_TIMED_OUT, and B_WOULD_BLOCK
			count = 0;
	}

err:
	locker.Unlock();
	put_select_sync(sync);

	return count;
}


static ssize_t
common_wait_for_objects(object_wait_info* infos, int numInfos, uint32 flags,
	bigtime_t timeout, bool kernel)
{
	status_t status = B_OK;

	// allocate sync object
	select_sync* sync;
	status = create_select_sync(numInfos, sync);
	if (status != B_OK)
		return status;

	// start selecting objects

	BenaphoreLocker locker(sync->lock);

	ssize_t count = 0;
	bool invalid = false;
	for (int i = 0; i < numInfos; i++) {
		uint16 type = infos[i].type;
		int32 object = infos[i].object;

		// initialize events masks
		sync->set[i].selected_events = infos[i].events
			| B_EVENT_INVALID | B_EVENT_ERROR | B_EVENT_DISCONNECTED;
		sync->set[i].events = 0;

		if (type < kSelectOpsCount
			&& kSelectOps[type].select(object, sync->set + i, kernel) == B_OK) {
			infos[i].events = 0;
			if (sync->set[i].selected_events != 0)
				count++;
		} else {
			sync->set[i].selected_events = B_EVENT_INVALID;
			infos[i].events = B_EVENT_INVALID;
			invalid = true;
		}
	}

	locker.Unlock();

	if (count < 1) {
		put_select_sync(sync);
		return B_BAD_VALUE;
	}

	if (!invalid) {
		status = acquire_sem_etc(sync->sem, 1, B_CAN_INTERRUPT | flags,
			timeout);
	}

	// deselect objects

	locker.Lock();

	for (int i = 0; i < numInfos; i++) {
		uint16 type = infos[i].type;

		if (type < kSelectOpsCount && (infos[i].events & B_EVENT_INVALID) == 0)
			kSelectOps[type].deselect(infos[i].object, sync->set + i, kernel);
	}

	// collect the events that are happened in the meantime

	switch (status) {
		case B_OK:
			count = 0;
			for (int i = 0; i < numInfos; i++) {
				infos[i].events = sync->set[i].events
					& sync->set[i].selected_events;
				if (infos[i].events != 0)
					count++;
			}
			break;
		default:
			// B_INTERRUPTED, B_TIMED_OUT, and B_WOULD_BLOCK
			count = status;
			break;
	}

	locker.Unlock();
	put_select_sync(sync);

	return count;
}


// #pragma mark - kernel private


status_t
notify_select_events(select_info* info, uint16 events)
{
	FUNCTION(("notify_select_events(%p (%p), 0x%x)\n", info, info->sync,
		events));

	if (info == NULL
		|| info->sync == NULL
		|| info->sync->sem < B_OK)
		return B_BAD_VALUE;

	info->events |= events;

	// only wake up the waiting select()/poll() call if the events
	// match one of the selected ones
	if (info->selected_events & events)
		return release_sem_etc(info->sync->sem, 1, B_DO_NOT_RESCHEDULE);

	return B_OK;
}


void
notify_select_events_list(select_info* list, uint16 events)
{
	struct select_info* info = list;
	while (info != NULL) {
		notify_select_events(info, events);
		info = info->next;
	}
}


//	#pragma mark - public kernel API


status_t
notify_select_event(struct selectsync *sync, uint8 event)
{
	return notify_select_events((select_info*)sync, SELECT_FLAG(event));
}


//	#pragma mark - private kernel exported API


static select_sync_pool_entry *
find_select_sync_pool_entry(select_sync_pool *pool, selectsync *sync)
{
	for (SelectSyncPoolEntryList::Iterator it = pool->entries.GetIterator();
		 it.HasNext();) {
		select_sync_pool_entry *entry = it.Next();
		if (entry->sync == sync)
			return entry;
	}

	return NULL;
}


static status_t
add_select_sync_pool_entry(select_sync_pool *pool, selectsync *sync,
	uint8 event)
{
	// check, whether the entry does already exist
	select_sync_pool_entry *entry = find_select_sync_pool_entry(pool, sync);
	if (!entry) {
		entry = new (std::nothrow) select_sync_pool_entry;
		if (!entry)
			return B_NO_MEMORY;

		entry->sync = sync;
		entry->events = 0;

		pool->entries.Add(entry);
	}

	entry->events |= SELECT_FLAG(event);

	return B_OK;
}


status_t
add_select_sync_pool_entry(select_sync_pool **_pool, selectsync *sync,
	uint8 event)
{
	// create the pool, if necessary
	select_sync_pool *pool = *_pool;
	if (!pool) {
		pool = new (std::nothrow) select_sync_pool;
		if (!pool)
			return B_NO_MEMORY;

		*_pool = pool;
	}

	// add the entry
	status_t error = add_select_sync_pool_entry(pool, sync, event);

	// cleanup
	if (pool->entries.IsEmpty()) {
		delete pool;
		*_pool = NULL;
	}

	return error;
}


status_t
remove_select_sync_pool_entry(select_sync_pool **_pool, selectsync *sync,
	uint8 event)
{
	select_sync_pool *pool = *_pool;
	if (!pool)
		return B_ENTRY_NOT_FOUND;

	// clear the event flag of the concerned entries
	bool found = false;
	for (SelectSyncPoolEntryList::Iterator it = pool->entries.GetIterator();
		 it.HasNext();) {
		select_sync_pool_entry *entry = it.Next();
		if (entry->sync == sync) {
			found = true;
			entry->events &= ~SELECT_FLAG(event);

			// remove the entry, if no longer needed
			if (entry->events == 0) {
				it.Remove();
				delete entry;
			}
		}
	}

	if (!found)
		return B_ENTRY_NOT_FOUND;

	// delete the pool, if no longer needed
	if (pool->entries.IsEmpty()) {
		delete pool;
		*_pool = NULL;
	}

	return B_OK;
}


void
delete_select_sync_pool(select_sync_pool *pool)
{
	if (!pool)
		return;

	while (select_sync_pool_entry *entry = pool->entries.Head()) {
		pool->entries.Remove(entry);
		delete entry;
	}

	delete pool;
}


void
notify_select_event_pool(select_sync_pool *pool, uint8 event)
{
	if (!pool)
		return;

	FUNCTION(("notify_select_event_pool(%p, %u)\n", pool, event));

	for (SelectSyncPoolEntryList::Iterator it = pool->entries.GetIterator();
		 it.HasNext();) {
		select_sync_pool_entry *entry = it.Next();
		if (entry->events & SELECT_FLAG(event))
			notify_select_event(entry->sync, event);
	}
}


//	#pragma mark - Kernel POSIX layer


ssize_t
_kern_select(int numFDs, fd_set *readSet, fd_set *writeSet, fd_set *errorSet,
	bigtime_t timeout, const sigset_t *sigMask)
{
	if (timeout >= 0)
		timeout += system_time();

	return common_select(numFDs, readSet, writeSet, errorSet, timeout,
		sigMask, true);
}


ssize_t
_kern_poll(struct pollfd *fds, int numFDs, bigtime_t timeout)
{
	if (timeout >= 0)
		timeout += system_time();

	return common_poll(fds, numFDs, timeout, true);
}


ssize_t
_kern_wait_for_objects(object_wait_info* infos, int numInfos, uint32 flags,
	bigtime_t timeout)
{
	return common_wait_for_objects(infos, numInfos, flags, timeout, true);
}


//	#pragma mark - User syscalls


ssize_t
_user_select(int numFDs, fd_set *userReadSet, fd_set *userWriteSet,
	fd_set *userErrorSet, bigtime_t timeout, const sigset_t *userSigMask)
{
	fd_set *readSet = NULL, *writeSet = NULL, *errorSet = NULL;
	uint32 bytes = _howmany(numFDs, NFDBITS) * sizeof(fd_mask);
	sigset_t sigMask;
	int result;

	syscall_restart_handle_timeout_pre(timeout);

	if (numFDs < 0)
		return B_BAD_VALUE;

	if ((userReadSet != NULL && !IS_USER_ADDRESS(userReadSet))
		|| (userWriteSet != NULL && !IS_USER_ADDRESS(userWriteSet))
		|| (userErrorSet != NULL && !IS_USER_ADDRESS(userErrorSet))
		|| (userSigMask != NULL && !IS_USER_ADDRESS(userSigMask)))
		return B_BAD_ADDRESS;

	// copy parameters

	if (userReadSet != NULL) {
		readSet = (fd_set *)malloc(bytes);
		if (readSet == NULL)
			return B_NO_MEMORY;

		if (user_memcpy(readSet, userReadSet, bytes) < B_OK) {
			result = B_BAD_ADDRESS;
			goto err;
		}
	}

	if (userWriteSet != NULL) {
		writeSet = (fd_set *)malloc(bytes);
		if (writeSet == NULL) {
			result = B_NO_MEMORY;
			goto err;
		}
		if (user_memcpy(writeSet, userWriteSet, bytes) < B_OK) {
			result = B_BAD_ADDRESS;
			goto err;
		}
	}

	if (userErrorSet != NULL) {
		errorSet = (fd_set *)malloc(bytes);
		if (errorSet == NULL) {
			result = B_NO_MEMORY;
			goto err;
		}
		if (user_memcpy(errorSet, userErrorSet, bytes) < B_OK) {
			result = B_BAD_ADDRESS;
			goto err;
		}
	}

	if (userSigMask != NULL)
		sigMask = *userSigMask;

	result = common_select(numFDs, readSet, writeSet, errorSet, timeout,
		userSigMask ? &sigMask : NULL, false);

	// copy back results

	if (result >= B_OK
		&& ((readSet != NULL
				&& user_memcpy(userReadSet, readSet, bytes) < B_OK)
			|| (writeSet != NULL
				&& user_memcpy(userWriteSet, writeSet, bytes) < B_OK)
			|| (errorSet != NULL
				&& user_memcpy(userErrorSet, errorSet, bytes) < B_OK))) {
		result = B_BAD_ADDRESS;
	} else
		syscall_restart_handle_timeout_post(result, timeout);

err:
	free(readSet);
	free(writeSet);
	free(errorSet);

	return result;
}


ssize_t
_user_poll(struct pollfd *userfds, int numFDs, bigtime_t timeout)
{
	struct pollfd *fds;
	size_t bytes;
	int result;

	syscall_restart_handle_timeout_pre(timeout);

	if (numFDs < 0)
		return B_BAD_VALUE;

	if (userfds == NULL || !IS_USER_ADDRESS(userfds))
		return B_BAD_ADDRESS;

	// copy parameters

	fds = (struct pollfd *)malloc(bytes = numFDs * sizeof(struct pollfd));
	if (fds == NULL)
		return B_NO_MEMORY;

	if (user_memcpy(fds, userfds, bytes) < B_OK) {
		result = B_BAD_ADDRESS;
		goto err;
	}

	result = common_poll(fds, numFDs, timeout, false);

	// copy back results
	if (result >= B_OK && user_memcpy(userfds, fds, bytes) < B_OK)
		result = B_BAD_ADDRESS;
	else
		syscall_restart_handle_timeout_post(result, timeout);

err:
	free(fds);

	return result;
}


ssize_t
_user_wait_for_objects(object_wait_info* userInfos, int numInfos, uint32 flags,
	bigtime_t timeout)
{
	syscall_restart_handle_timeout_pre(flags, timeout);

	if (numInfos < 0)
		return B_BAD_VALUE;

	if (userInfos == NULL || !IS_USER_ADDRESS(userInfos))
		return B_BAD_ADDRESS;

	int bytes = sizeof(object_wait_info) * numInfos;
	object_wait_info* infos = (object_wait_info*)malloc(bytes);
	if (infos == NULL)
		return B_NO_MEMORY;

	// copy parameters to kernel space, call the function, and copy the results
	// back
	ssize_t result;
	if (user_memcpy(infos, userInfos, bytes) == B_OK) {
		result = common_wait_for_objects(infos, numInfos, flags, timeout,
			false);

		if (result >= 0 && user_memcpy(userInfos, infos, bytes) != B_OK)
			result = B_BAD_ADDRESS;
		else
			syscall_restart_handle_timeout_post(result, timeout);
	} else
		result = B_BAD_ADDRESS;

	free(infos);

	return result;
}