[linux-2.6-block.git] / fs / orangefs / waitqueue.c

/*
 * (C) 2001 Clemson University and The University of Chicago
 * (C) 2011 Omnibond Systems
 *
 * Changes by Acxiom Corporation to implement generic service_operation()
 * function, Copyright Acxiom Corporation, 2005.
 *
 * See COPYING in top-level directory.
 */

/*
 *  In-kernel waitqueue operations.
 */

#include "protocol.h"
#include "orangefs-kernel.h"
#include "orangefs-bufmap.h"

static int wait_for_cancellation_downcall(struct orangefs_kernel_op_s *);
static int wait_for_matching_downcall(struct orangefs_kernel_op_s *);

/*
 * What we do in this function is to walk the list of operations that are
 * present in the request queue and mark them as purged.
 * NOTE: This is called from the device close after client-core has
 * guaranteed that no new operations could appear on the list since the
 * client-core is anyway going to exit.
 */
void purge_waiting_ops(void)
{
	struct orangefs_kernel_op_s *op;

	spin_lock(&orangefs_request_list_lock);
	list_for_each_entry(op, &orangefs_request_list, list) {
		gossip_debug(GOSSIP_WAIT_DEBUG,
			     "pvfs2-client-core: purging op tag %llu %s\n",
			     llu(op->tag),
			     get_opname_string(op));
		spin_lock(&op->lock);
		set_op_state_purged(op);
		spin_unlock(&op->lock);
	}
	spin_unlock(&orangefs_request_list_lock);
}

static inline void
add_op_to_request_list(struct orangefs_kernel_op_s *op)
{
	spin_lock(&orangefs_request_list_lock);
	spin_lock(&op->lock);
	set_op_state_waiting(op);
	list_add_tail(&op->list, &orangefs_request_list);
	spin_unlock(&orangefs_request_list_lock);
	spin_unlock(&op->lock);
	wake_up_interruptible(&orangefs_request_list_waitq);
}

static inline
void add_priority_op_to_request_list(struct orangefs_kernel_op_s *op)
{
	spin_lock(&orangefs_request_list_lock);
	spin_lock(&op->lock);
	set_op_state_waiting(op);

	list_add(&op->list, &orangefs_request_list);
	spin_unlock(&orangefs_request_list_lock);
	spin_unlock(&op->lock);
	wake_up_interruptible(&orangefs_request_list_waitq);
}

/*
 * submits a ORANGEFS operation and waits for it to complete
 *
 * Note op->downcall.status will contain the status of the operation (in
 * errno format), whether provided by pvfs2-client or a result of failure to
 * service the operation.  If the caller wishes to distinguish, then
 * op->state can be checked to see if it was serviced or not.
 *
 * Returns contents of op->downcall.status for convenience
 */
int service_operation(struct orangefs_kernel_op_s *op,
		      const char *op_name,
		      int flags)
{
	/* flags to modify behavior */
	sigset_t orig_sigset;
	int ret = 0;

	DEFINE_WAIT(wait_entry);

	op->upcall.tgid = current->tgid;
	op->upcall.pid = current->pid;

retry_servicing:
	op->downcall.status = 0;
	gossip_debug(GOSSIP_WAIT_DEBUG,
		     "orangefs: service_operation: %s %p\n",
		     op_name,
		     op);
	gossip_debug(GOSSIP_WAIT_DEBUG,
		     "orangefs: operation posted by process: %s, pid: %i\n",
		     current->comm,
		     current->pid);

	/* mask out signals if this operation is not to be interrupted */
	if (!(flags & ORANGEFS_OP_INTERRUPTIBLE))
		orangefs_block_signals(&orig_sigset);

	if (!(flags & ORANGEFS_OP_NO_SEMAPHORE)) {
		ret = mutex_lock_interruptible(&request_mutex);
		/*
		 * check to see if we were interrupted while waiting for
		 * semaphore
		 */
		if (ret < 0) {
			if (!(flags & ORANGEFS_OP_INTERRUPTIBLE))
				orangefs_set_signals(&orig_sigset);
			op->downcall.status = ret;
			gossip_debug(GOSSIP_WAIT_DEBUG,
				     "orangefs: service_operation interrupted.\n");
			return ret;
		}
	}

	gossip_debug(GOSSIP_WAIT_DEBUG,
		     "%s:About to call is_daemon_in_service().\n",
		     __func__);

	if (is_daemon_in_service() < 0) {
		/*
		 * By incrementing the per-operation attempt counter, we
		 * directly go into the timeout logic while waiting for
		 * the matching downcall to be read
		 */
		gossip_debug(GOSSIP_WAIT_DEBUG,
			     "%s:client core is NOT in service(%d).\n",
			     __func__,
			     is_daemon_in_service());
		op->attempts++;
	}

	/* queue up the operation */
	if (flags & ORANGEFS_OP_PRIORITY) {
		add_priority_op_to_request_list(op);
	} else {
		gossip_debug(GOSSIP_WAIT_DEBUG,
			     "%s:About to call add_op_to_request_list().\n",
			     __func__);
		add_op_to_request_list(op);
	}

	if (!(flags & ORANGEFS_OP_NO_SEMAPHORE))
		mutex_unlock(&request_mutex);

	/*
	 * If we are asked to service an asynchronous operation from
	 * VFS perspective, we are done.
	 */
	if (flags & ORANGEFS_OP_ASYNC)
		return 0;

	if (flags & ORANGEFS_OP_CANCELLATION) {
		gossip_debug(GOSSIP_WAIT_DEBUG,
			     "%s:"
			     "About to call wait_for_cancellation_downcall.\n",
			     __func__);
		ret = wait_for_cancellation_downcall(op);
	} else {
		ret = wait_for_matching_downcall(op);
	}

	if (ret < 0) {
		/* failed to get matching downcall */
		if (ret == -ETIMEDOUT) {
			gossip_err("orangefs: %s -- wait timed out; aborting attempt.\n",
				   op_name);
		}
		op->downcall.status = ret;
	} else {
		/* got matching downcall; make sure status is in errno format */
		op->downcall.status =
		    orangefs_normalize_to_errno(op->downcall.status);
		ret = op->downcall.status;
	}

	if (!(flags & ORANGEFS_OP_INTERRUPTIBLE))
		orangefs_set_signals(&orig_sigset);

	BUG_ON(ret != op->downcall.status);
	/* retry if operation has not been serviced and if requested */
	if (!op_state_serviced(op) && op->downcall.status == -EAGAIN) {
		gossip_debug(GOSSIP_WAIT_DEBUG,
			     "orangefs: tag %llu (%s)"
			     " -- operation to be retried (%d attempt)\n",
			     llu(op->tag),
			     op_name,
			     op->attempts + 1);

		if (!op->uses_shared_memory)
			/*
			 * this operation doesn't use the shared memory
			 * system
			 */
			goto retry_servicing;

		/* op uses shared memory */
		if (orangefs_get_bufmap_init() == 0) {
			/*
			 * This operation uses the shared memory system AND
			 * the system is not yet ready. This situation occurs
			 * when the client-core is restarted AND there were
			 * operations waiting to be processed or were already
			 * in process.
			 */
			gossip_debug(GOSSIP_WAIT_DEBUG,
				     "uses_shared_memory is true.\n");
			gossip_debug(GOSSIP_WAIT_DEBUG,
				     "Client core in-service status(%d).\n",
				     is_daemon_in_service());
			gossip_debug(GOSSIP_WAIT_DEBUG, "bufmap_init:%d.\n",
				     orangefs_get_bufmap_init());
			gossip_debug(GOSSIP_WAIT_DEBUG,
				     "operation's status is 0x%0x.\n",
				     op->op_state);

			/*
			 * let process sleep for a few seconds so shared
			 * memory system can be initialized.
			 */
			prepare_to_wait(&orangefs_bufmap_init_waitq,
					&wait_entry,
					TASK_INTERRUPTIBLE);

			/*
			 * Wait for orangefs_bufmap_initialize() to wake me up
			 * within the allotted time.
			 */
			ret = schedule_timeout(
				ORANGEFS_BUFMAP_WAIT_TIMEOUT_SECS * HZ);

			gossip_debug(GOSSIP_WAIT_DEBUG,
				     "Value returned from schedule_timeout:"
				     "%d.\n",
				     ret);
			gossip_debug(GOSSIP_WAIT_DEBUG,
				     "Is shared memory available? (%d).\n",
				     orangefs_get_bufmap_init());

			finish_wait(&orangefs_bufmap_init_waitq, &wait_entry);

			if (orangefs_get_bufmap_init() == 0) {
				gossip_err("%s:The shared memory system has not started in %d seconds after the client core restarted.  Aborting user's request(%s).\n",
					   __func__,
					   ORANGEFS_BUFMAP_WAIT_TIMEOUT_SECS,
					   get_opname_string(op));
				return -EIO;
			}

			/*
			 * Return to the calling function and re-populate a
			 * shared memory buffer.
			 */
			return -EAGAIN;
		}
	}

	gossip_debug(GOSSIP_WAIT_DEBUG,
		     "orangefs: service_operation %s returning: %d for %p.\n",
		     op_name,
		     ret,
		     op);
	return ret;
}

static void orangefs_clean_up_interrupted_operation(struct orangefs_kernel_op_s *op)
{
	/*
	 * handle interrupted cases depending on what state we were in when
	 * the interruption is detected.  there is a coarse grained lock
	 * across the operation.
	 *
	 * Called with op->lock held.
	 */
	op->op_state |= OP_VFS_STATE_GIVEN_UP;

	if (op_state_waiting(op)) {
		/*
		 * upcall hasn't been read; remove op from upcall request
		 * list.
		 */
		spin_unlock(&op->lock);
		spin_lock(&orangefs_request_list_lock);
		list_del(&op->list);
		spin_unlock(&orangefs_request_list_lock);
		gossip_debug(GOSSIP_WAIT_DEBUG,
			     "Interrupted: Removed op %p from request_list\n",
			     op);
	} else if (op_state_in_progress(op)) {
		/* op must be removed from the in progress htable */
		spin_unlock(&op->lock);
		spin_lock(&htable_ops_in_progress_lock);
		list_del(&op->list);
		spin_unlock(&htable_ops_in_progress_lock);
		gossip_debug(GOSSIP_WAIT_DEBUG,
			     "Interrupted: Removed op %p"
			     " from htable_ops_in_progress\n",
			     op);
	} else if (!op_state_serviced(op)) {
		spin_unlock(&op->lock);
		gossip_err("interrupted operation is in a weird state 0x%x\n",
			   op->op_state);
	} else {
		/*
		 * It is not intended for execution to flow here,
		 * but having this unlock here makes sparse happy.
		 */
		gossip_err("%s: can't get here.\n", __func__);
		spin_unlock(&op->lock);
	}
}

/*
 * sleeps on waitqueue waiting for matching downcall.
 * if client-core finishes servicing, then we are good to go.
 * else if client-core exits, we get woken up here, and retry with a timeout
 *
 * Post when this call returns to the caller, the specified op will no
 * longer be on any list or htable.
 *
 * Returns 0 on success and -errno on failure
 * Errors are:
 * EAGAIN in case we want the caller to requeue and try again..
 * EINTR/EIO/ETIMEDOUT indicating we are done trying to service this
 * operation since client-core seems to be exiting too often
 * or if we were interrupted.
 */
static int wait_for_matching_downcall(struct orangefs_kernel_op_s *op)
{
	int ret = -EINVAL;
	DEFINE_WAIT(wait_entry);

	while (1) {
		spin_lock(&op->lock);
		prepare_to_wait(&op->waitq, &wait_entry, TASK_INTERRUPTIBLE);
		if (op_state_serviced(op)) {
			spin_unlock(&op->lock);
			ret = 0;
			break;
		}

		if (unlikely(signal_pending(current))) {
			gossip_debug(GOSSIP_WAIT_DEBUG,
				     "*** %s:"
				     " operation interrupted by a signal (tag "
				     "%llu, op %p)\n",
				     __func__,
				     llu(op->tag),
				     op);
			orangefs_clean_up_interrupted_operation(op);
			ret = -EINTR;
			break;
		}

		/*
		 * if this was our first attempt and client-core
		 * has not purged our operation, we are happy to
		 * simply wait
		 */
		if (op->attempts == 0 && !op_state_purged(op)) {
			spin_unlock(&op->lock);
			schedule();
		} else {
			spin_unlock(&op->lock);
			/*
			 * subsequent attempts, we retry exactly once
			 * with timeouts
			 */
			if (!schedule_timeout(op_timeout_secs * HZ)) {
				gossip_debug(GOSSIP_WAIT_DEBUG,
					     "*** %s:"
					     " operation timed out (tag"
					     " %llu, %p, att %d)\n",
					     __func__,
					     llu(op->tag),
					     op,
					     op->attempts);
				ret = -ETIMEDOUT;
				spin_lock(&op->lock);
				orangefs_clean_up_interrupted_operation(op);
				break;
			}
		}
		spin_lock(&op->lock);
		op->attempts++;
		/*
		 * if the operation was purged in the meantime, it
		 * is better to requeue it afresh but ensure that
		 * we have not been purged repeatedly. This could
		 * happen if client-core crashes when an op
		 * is being serviced, so we requeue the op, client
		 * core crashes again so we requeue the op, client
		 * core starts, and so on...
		 */
		if (op_state_purged(op)) {
			ret = (op->attempts < ORANGEFS_PURGE_RETRY_COUNT) ?
				 -EAGAIN :
				 -EIO;
			gossip_debug(GOSSIP_WAIT_DEBUG,
				     "*** %s:"
				     " operation purged (tag "
				     "%llu, %p, att %d)\n",
				     __func__,
				     llu(op->tag),
				     op,
				     op->attempts);
			orangefs_clean_up_interrupted_operation(op);
			break;
		}
		spin_unlock(&op->lock);
	}

	spin_lock(&op->lock);
	finish_wait(&op->waitq, &wait_entry);
	spin_unlock(&op->lock);

	return ret;
}

/*
 * similar to wait_for_matching_downcall(), but used in the special case
 * of I/O cancellations.
 *
 * Note we need a special wait function because if this is called we already
 *      know that a signal is pending in current and need to service the
 *      cancellation upcall anyway.  the only way to exit this is to either
 *      timeout or have the cancellation be serviced properly.
 */
static int wait_for_cancellation_downcall(struct orangefs_kernel_op_s *op)
{
	int ret = -EINVAL;
	DEFINE_WAIT(wait_entry);

	while (1) {
		spin_lock(&op->lock);
		prepare_to_wait(&op->waitq, &wait_entry, TASK_INTERRUPTIBLE);
		if (op_state_serviced(op)) {
			gossip_debug(GOSSIP_WAIT_DEBUG,
				     "%s:op-state is SERVICED.\n",
				     __func__);
			spin_unlock(&op->lock);
			ret = 0;
			break;
		}

		if (signal_pending(current)) {
			gossip_debug(GOSSIP_WAIT_DEBUG,
				     "%s:operation interrupted by a signal (tag"
				     " %llu, op %p)\n",
				     __func__,
				     llu(op->tag),
				     op);
			orangefs_clean_up_interrupted_operation(op);
			ret = -EINTR;
			break;
		}

		gossip_debug(GOSSIP_WAIT_DEBUG,
			     "%s:About to call schedule_timeout.\n",
			     __func__);
		spin_unlock(&op->lock);
		ret = schedule_timeout(op_timeout_secs * HZ);

		gossip_debug(GOSSIP_WAIT_DEBUG,
			     "%s:Value returned from schedule_timeout(%d).\n",
			     __func__,
			     ret);
		if (!ret) {
			gossip_debug(GOSSIP_WAIT_DEBUG,
				     "%s:*** operation timed out: %p\n",
				     __func__,
				     op);
			spin_lock(&op->lock);
			orangefs_clean_up_interrupted_operation(op);
			ret = -ETIMEDOUT;
			break;
		}

		gossip_debug(GOSSIP_WAIT_DEBUG,
			     "%s:Breaking out of loop, regardless of value returned by schedule_timeout.\n",
			     __func__);
		ret = -ETIMEDOUT;
		break;
	}

	spin_lock(&op->lock);
	finish_wait(&op->waitq, &wait_entry);
	spin_unlock(&op->lock);

	gossip_debug(GOSSIP_WAIT_DEBUG,
		     "%s:returning ret(%d)\n",
		     __func__,
		     ret);

	return ret;
}
Commit	Line	Data
1182fca3 MM	1	/*
	2	* (C) 2001 Clemson University and The University of Chicago
	3	* (C) 2011 Omnibond Systems
	4	*
	5	* Changes by Acxiom Corporation to implement generic service_operation()
	6	* function, Copyright Acxiom Corporation, 2005.
	7	*
	8	* See COPYING in top-level directory.
	9	*/
	10
	11	/*
	12	* In-kernel waitqueue operations.
	13	*/
	14
	15	#include "protocol.h"
575e9461 MM	16	#include "orangefs-kernel.h"
575e9461 MM	17	#include "orangefs-bufmap.h"
1182fca3	18
ade3d781 AV	19	static int wait_for_cancellation_downcall(struct orangefs_kernel_op_s *);
	20	static int wait_for_matching_downcall(struct orangefs_kernel_op_s *);
	21
1182fca3 MM	22	/*
	23	* What we do in this function is to walk the list of operations that are
	24	* present in the request queue and mark them as purged.
	25	* NOTE: This is called from the device close after client-core has
	26	* guaranteed that no new operations could appear on the list since the
	27	* client-core is anyway going to exit.
	28	*/
	29	void purge_waiting_ops(void)
	30	{
8bb8aefd	31	struct orangefs_kernel_op_s *op;
1182fca3	32
8bb8aefd YL	33	spin_lock(&orangefs_request_list_lock);
8bb8aefd YL	34	list_for_each_entry(op, &orangefs_request_list, list) {
1182fca3 MM	35	gossip_debug(GOSSIP_WAIT_DEBUG,
	36	"pvfs2-client-core: purging op tag %llu %s\n",
	37	llu(op->tag),
	38	get_opname_string(op));
	39	spin_lock(&op->lock);
	40	set_op_state_purged(op);
	41	spin_unlock(&op->lock);
1182fca3	42	}
8bb8aefd	43	spin_unlock(&orangefs_request_list_lock);
1182fca3 MM	44	}
1182fca3 MM	45
fc916da5 AV	46	static inline void
	47	add_op_to_request_list(struct orangefs_kernel_op_s *op)
	48	{
	49	spin_lock(&orangefs_request_list_lock);
	50	spin_lock(&op->lock);
	51	set_op_state_waiting(op);
	52	list_add_tail(&op->list, &orangefs_request_list);
	53	spin_unlock(&orangefs_request_list_lock);
	54	spin_unlock(&op->lock);
	55	wake_up_interruptible(&orangefs_request_list_waitq);
	56	}
	57
	58	static inline
	59	void add_priority_op_to_request_list(struct orangefs_kernel_op_s *op)
	60	{
	61	spin_lock(&orangefs_request_list_lock);
	62	spin_lock(&op->lock);
	63	set_op_state_waiting(op);
	64
	65	list_add(&op->list, &orangefs_request_list);
	66	spin_unlock(&orangefs_request_list_lock);
	67	spin_unlock(&op->lock);
	68	wake_up_interruptible(&orangefs_request_list_waitq);
	69	}
	70
1182fca3	71	/*
8bb8aefd	72	* submits a ORANGEFS operation and waits for it to complete
1182fca3 MM	73	*
	74	* Note op->downcall.status will contain the status of the operation (in
	75	* errno format), whether provided by pvfs2-client or a result of failure to
	76	* service the operation. If the caller wishes to distinguish, then
	77	* op->state can be checked to see if it was serviced or not.
	78	*
	79	* Returns contents of op->downcall.status for convenience
	80	*/
8bb8aefd	81	int service_operation(struct orangefs_kernel_op_s *op,
1182fca3 MM	82	const char *op_name,
	83	int flags)
	84	{
	85	/* flags to modify behavior */
	86	sigset_t orig_sigset;
	87	int ret = 0;
	88
ce6c414e	89	DEFINE_WAIT(wait_entry);
1182fca3 MM	90
	91	op->upcall.tgid = current->tgid;
	92	op->upcall.pid = current->pid;
	93
	94	retry_servicing:
	95	op->downcall.status = 0;
	96	gossip_debug(GOSSIP_WAIT_DEBUG,
8bb8aefd	97	"orangefs: service_operation: %s %p\n",
1182fca3 MM	98	op_name,
	99	op);
	100	gossip_debug(GOSSIP_WAIT_DEBUG,
8bb8aefd	101	"orangefs: operation posted by process: %s, pid: %i\n",
1182fca3 MM	102	current->comm,
	103	current->pid);
	104
	105	/* mask out signals if this operation is not to be interrupted */
8bb8aefd	106	if (!(flags & ORANGEFS_OP_INTERRUPTIBLE))
c146c0b8	107	orangefs_block_signals(&orig_sigset);
1182fca3	108
8bb8aefd	109	if (!(flags & ORANGEFS_OP_NO_SEMAPHORE)) {
1182fca3 MM	110	ret = mutex_lock_interruptible(&request_mutex);
	111	/*
	112	* check to see if we were interrupted while waiting for
	113	* semaphore
	114	*/
	115	if (ret < 0) {
8bb8aefd	116	if (!(flags & ORANGEFS_OP_INTERRUPTIBLE))
c146c0b8	117	orangefs_set_signals(&orig_sigset);
1182fca3 MM	118	op->downcall.status = ret;
1182fca3 MM	119	gossip_debug(GOSSIP_WAIT_DEBUG,
8bb8aefd	120	"orangefs: service_operation interrupted.\n");
1182fca3 MM	121	return ret;
	122	}
	123	}
	124
	125	gossip_debug(GOSSIP_WAIT_DEBUG,
	126	"%s:About to call is_daemon_in_service().\n",
	127	__func__);
	128
	129	if (is_daemon_in_service() < 0) {
	130	/*
	131	* By incrementing the per-operation attempt counter, we
	132	* directly go into the timeout logic while waiting for
	133	* the matching downcall to be read
	134	*/
	135	gossip_debug(GOSSIP_WAIT_DEBUG,
	136	"%s:client core is NOT in service(%d).\n",
	137	__func__,
	138	is_daemon_in_service());
	139	op->attempts++;
	140	}
	141
	142	/* queue up the operation */
8bb8aefd	143	if (flags & ORANGEFS_OP_PRIORITY) {
1182fca3 MM	144	add_priority_op_to_request_list(op);
	145	} else {
	146	gossip_debug(GOSSIP_WAIT_DEBUG,
	147	"%s:About to call add_op_to_request_list().\n",
	148	__func__);
	149	add_op_to_request_list(op);
	150	}
	151
8bb8aefd	152	if (!(flags & ORANGEFS_OP_NO_SEMAPHORE))
1182fca3 MM	153	mutex_unlock(&request_mutex);
	154
	155	/*
	156	* If we are asked to service an asynchronous operation from
	157	* VFS perspective, we are done.
	158	*/
8bb8aefd	159	if (flags & ORANGEFS_OP_ASYNC)
1182fca3 MM	160	return 0;
1182fca3 MM	161
8bb8aefd	162	if (flags & ORANGEFS_OP_CANCELLATION) {
1182fca3 MM	163	gossip_debug(GOSSIP_WAIT_DEBUG,
	164	"%s:"
	165	"About to call wait_for_cancellation_downcall.\n",
	166	__func__);
	167	ret = wait_for_cancellation_downcall(op);
	168	} else {
	169	ret = wait_for_matching_downcall(op);
	170	}
	171
	172	if (ret < 0) {
	173	/* failed to get matching downcall */
	174	if (ret == -ETIMEDOUT) {
8bb8aefd	175	gossip_err("orangefs: %s -- wait timed out; aborting attempt.\n",
1182fca3 MM	176	op_name);
	177	}
	178	op->downcall.status = ret;
	179	} else {
	180	/* got matching downcall; make sure status is in errno format */
	181	op->downcall.status =
8bb8aefd	182	orangefs_normalize_to_errno(op->downcall.status);
1182fca3 MM	183	ret = op->downcall.status;
	184	}
	185
8bb8aefd	186	if (!(flags & ORANGEFS_OP_INTERRUPTIBLE))
c146c0b8	187	orangefs_set_signals(&orig_sigset);
1182fca3 MM	188
	189	BUG_ON(ret != op->downcall.status);
	190	/* retry if operation has not been serviced and if requested */
	191	if (!op_state_serviced(op) && op->downcall.status == -EAGAIN) {
	192	gossip_debug(GOSSIP_WAIT_DEBUG,
8bb8aefd	193	"orangefs: tag %llu (%s)"
1182fca3 MM	194	" -- operation to be retried (%d attempt)\n",
	195	llu(op->tag),
	196	op_name,
	197	op->attempts + 1);
	198
	199	if (!op->uses_shared_memory)
	200	/*
	201	* this operation doesn't use the shared memory
	202	* system
	203	*/
	204	goto retry_servicing;
	205
	206	/* op uses shared memory */
7d221485	207	if (orangefs_get_bufmap_init() == 0) {
1182fca3 MM	208	/*
	209	* This operation uses the shared memory system AND
	210	* the system is not yet ready. This situation occurs
	211	* when the client-core is restarted AND there were
	212	* operations waiting to be processed or were already
	213	* in process.
	214	*/
	215	gossip_debug(GOSSIP_WAIT_DEBUG,
	216	"uses_shared_memory is true.\n");
	217	gossip_debug(GOSSIP_WAIT_DEBUG,
	218	"Client core in-service status(%d).\n",
	219	is_daemon_in_service());
	220	gossip_debug(GOSSIP_WAIT_DEBUG, "bufmap_init:%d.\n",
7d221485	221	orangefs_get_bufmap_init());
1182fca3 MM	222	gossip_debug(GOSSIP_WAIT_DEBUG,
	223	"operation's status is 0x%0x.\n",
	224	op->op_state);
	225
	226	/*
	227	* let process sleep for a few seconds so shared
	228	* memory system can be initialized.
	229	*/
ce6c414e MM	230	prepare_to_wait(&orangefs_bufmap_init_waitq,
	231	&wait_entry,
	232	TASK_INTERRUPTIBLE);
1182fca3	233
1182fca3	234	/*
8bb8aefd	235	* Wait for orangefs_bufmap_initialize() to wake me up
1182fca3 MM	236	* within the allotted time.
1182fca3 MM	237	*/
727cbfea AV	238	ret = schedule_timeout(
727cbfea AV	239	ORANGEFS_BUFMAP_WAIT_TIMEOUT_SECS * HZ);
1182fca3 MM	240
	241	gossip_debug(GOSSIP_WAIT_DEBUG,
	242	"Value returned from schedule_timeout:"
	243	"%d.\n",
	244	ret);
	245	gossip_debug(GOSSIP_WAIT_DEBUG,
	246	"Is shared memory available? (%d).\n",
7d221485	247	orangefs_get_bufmap_init());
1182fca3	248
ce6c414e	249	finish_wait(&orangefs_bufmap_init_waitq, &wait_entry);
1182fca3	250
7d221485	251	if (orangefs_get_bufmap_init() == 0) {
1182fca3 MM	252	gossip_err("%s:The shared memory system has not started in %d seconds after the client core restarted. Aborting user's request(%s).\n",
1182fca3 MM	253	__func__,
8bb8aefd	254	ORANGEFS_BUFMAP_WAIT_TIMEOUT_SECS,
1182fca3 MM	255	get_opname_string(op));
	256	return -EIO;
	257	}
	258
	259	/*
	260	* Return to the calling function and re-populate a
	261	* shared memory buffer.
	262	*/
	263	return -EAGAIN;
	264	}
	265	}
	266
	267	gossip_debug(GOSSIP_WAIT_DEBUG,
8bb8aefd	268	"orangefs: service_operation %s returning: %d for %p.\n",
1182fca3 MM	269	op_name,
	270	ret,
	271	op);
	272	return ret;
	273	}
	274
e07db0a2	275	static void orangefs_clean_up_interrupted_operation(struct orangefs_kernel_op_s *op)
1182fca3 MM	276	{
	277	/*
	278	* handle interrupted cases depending on what state we were in when
	279	* the interruption is detected. there is a coarse grained lock
	280	* across the operation.
	281	*
eab9b389	282	* Called with op->lock held.
1182fca3	283	*/
ed42fe05	284	op->op_state \|= OP_VFS_STATE_GIVEN_UP;
1182fca3 MM	285
	286	if (op_state_waiting(op)) {
	287	/*
	288	* upcall hasn't been read; remove op from upcall request
	289	* list.
	290	*/
	291	spin_unlock(&op->lock);
ed42fe05 AV	292	spin_lock(&orangefs_request_list_lock);
	293	list_del(&op->list);
	294	spin_unlock(&orangefs_request_list_lock);
1182fca3 MM	295	gossip_debug(GOSSIP_WAIT_DEBUG,
	296	"Interrupted: Removed op %p from request_list\n",
	297	op);
	298	} else if (op_state_in_progress(op)) {
	299	/* op must be removed from the in progress htable */
	300	spin_unlock(&op->lock);
	301	spin_lock(&htable_ops_in_progress_lock);
	302	list_del(&op->list);
	303	spin_unlock(&htable_ops_in_progress_lock);
	304	gossip_debug(GOSSIP_WAIT_DEBUG,
	305	"Interrupted: Removed op %p"
	306	" from htable_ops_in_progress\n",
	307	op);
	308	} else if (!op_state_serviced(op)) {
	309	spin_unlock(&op->lock);
	310	gossip_err("interrupted operation is in a weird state 0x%x\n",
	311	op->op_state);
84d02150 MM	312	} else {
	313	/*
	314	* It is not intended for execution to flow here,
	315	* but having this unlock here makes sparse happy.
	316	*/
	317	gossip_err("%s: can't get here.\n", __func__);
	318	spin_unlock(&op->lock);
1182fca3 MM	319	}
	320	}
	321
	322	/*
	323	* sleeps on waitqueue waiting for matching downcall.
	324	* if client-core finishes servicing, then we are good to go.
	325	* else if client-core exits, we get woken up here, and retry with a timeout
	326	*
	327	* Post when this call returns to the caller, the specified op will no
	328	* longer be on any list or htable.
	329	*
	330	* Returns 0 on success and -errno on failure
	331	* Errors are:
	332	* EAGAIN in case we want the caller to requeue and try again..
	333	* EINTR/EIO/ETIMEDOUT indicating we are done trying to service this
	334	* operation since client-core seems to be exiting too often
	335	* or if we were interrupted.
	336	*/
b7ae37b0	337	static int wait_for_matching_downcall(struct orangefs_kernel_op_s *op)
1182fca3 MM	338	{
1182fca3 MM	339	int ret = -EINVAL;
ce6c414e	340	DEFINE_WAIT(wait_entry);
1182fca3 MM	341
1182fca3 MM	342	while (1) {
1182fca3	343	spin_lock(&op->lock);
ce6c414e	344	prepare_to_wait(&op->waitq, &wait_entry, TASK_INTERRUPTIBLE);
1182fca3 MM	345	if (op_state_serviced(op)) {
	346	spin_unlock(&op->lock);
	347	ret = 0;
	348	break;
	349	}
1182fca3	350
70c6ea26 AV	351	if (unlikely(signal_pending(current))) {
	352	gossip_debug(GOSSIP_WAIT_DEBUG,
	353	"*** %s:"
	354	" operation interrupted by a signal (tag "
	355	"%llu, op %p)\n",
	356	__func__,
	357	llu(op->tag),
	358	op);
	359	orangefs_clean_up_interrupted_operation(op);
	360	ret = -EINTR;
	361	break;
	362	}
	363
	364	/*
	365	* if this was our first attempt and client-core
	366	* has not purged our operation, we are happy to
	367	* simply wait
	368	*/
70c6ea26 AV	369	if (op->attempts == 0 && !op_state_purged(op)) {
	370	spin_unlock(&op->lock);
	371	schedule();
	372	} else {
	373	spin_unlock(&op->lock);
1182fca3	374	/*
70c6ea26 AV	375	* subsequent attempts, we retry exactly once
70c6ea26 AV	376	* with timeouts
1182fca3	377	*/
727cbfea	378	if (!schedule_timeout(op_timeout_secs * HZ)) {
1182fca3 MM	379	gossip_debug(GOSSIP_WAIT_DEBUG,
1182fca3 MM	380	"*** %s:"
70c6ea26 AV	381	" operation timed out (tag"
70c6ea26 AV	382	" %llu, %p, att %d)\n",
1182fca3 MM	383	__func__,
	384	llu(op->tag),
	385	op,
	386	op->attempts);
70c6ea26	387	ret = -ETIMEDOUT;
eab9b389	388	spin_lock(&op->lock);
8bb8aefd	389	orangefs_clean_up_interrupted_operation(op);
1182fca3 MM	390	break;
1182fca3 MM	391	}
70c6ea26 AV	392	}
	393	spin_lock(&op->lock);
	394	op->attempts++;
	395	/*
	396	* if the operation was purged in the meantime, it
	397	* is better to requeue it afresh but ensure that
	398	* we have not been purged repeatedly. This could
	399	* happen if client-core crashes when an op
	400	* is being serviced, so we requeue the op, client
	401	* core crashes again so we requeue the op, client
	402	* core starts, and so on...
	403	*/
	404	if (op_state_purged(op)) {
	405	ret = (op->attempts < ORANGEFS_PURGE_RETRY_COUNT) ?
	406	-EAGAIN :
	407	-EIO;
70c6ea26 AV	408	gossip_debug(GOSSIP_WAIT_DEBUG,
	409	"*** %s:"
	410	" operation purged (tag "
	411	"%llu, %p, att %d)\n",
	412	__func__,
	413	llu(op->tag),
	414	op,
	415	op->attempts);
	416	orangefs_clean_up_interrupted_operation(op);
	417	break;
1182fca3	418	}
70c6ea26	419	spin_unlock(&op->lock);
1182fca3 MM	420	}
1182fca3 MM	421
1182fca3	422	spin_lock(&op->lock);
ce6c414e	423	finish_wait(&op->waitq, &wait_entry);
1182fca3 MM	424	spin_unlock(&op->lock);
	425
	426	return ret;
	427	}
	428
	429	/*
	430	* similar to wait_for_matching_downcall(), but used in the special case
	431	* of I/O cancellations.
	432	*
	433	* Note we need a special wait function because if this is called we already
	434	* know that a signal is pending in current and need to service the
	435	* cancellation upcall anyway. the only way to exit this is to either
	436	* timeout or have the cancellation be serviced properly.
	437	*/
b7ae37b0	438	static int wait_for_cancellation_downcall(struct orangefs_kernel_op_s *op)
1182fca3 MM	439	{
1182fca3 MM	440	int ret = -EINVAL;
ce6c414e	441	DEFINE_WAIT(wait_entry);
1182fca3 MM	442
1182fca3 MM	443	while (1) {
1182fca3	444	spin_lock(&op->lock);
ce6c414e	445	prepare_to_wait(&op->waitq, &wait_entry, TASK_INTERRUPTIBLE);
1182fca3 MM	446	if (op_state_serviced(op)) {
	447	gossip_debug(GOSSIP_WAIT_DEBUG,
	448	"%s:op-state is SERVICED.\n",
	449	__func__);
	450	spin_unlock(&op->lock);
	451	ret = 0;
	452	break;
	453	}
1182fca3 MM	454
	455	if (signal_pending(current)) {
	456	gossip_debug(GOSSIP_WAIT_DEBUG,
	457	"%s:operation interrupted by a signal (tag"
	458	" %llu, op %p)\n",
	459	__func__,
	460	llu(op->tag),
	461	op);
8bb8aefd	462	orangefs_clean_up_interrupted_operation(op);
1182fca3 MM	463	ret = -EINTR;
	464	break;
	465	}
	466
	467	gossip_debug(GOSSIP_WAIT_DEBUG,
	468	"%s:About to call schedule_timeout.\n",
	469	__func__);
eab9b389	470	spin_unlock(&op->lock);
727cbfea	471	ret = schedule_timeout(op_timeout_secs * HZ);
1182fca3 MM	472
	473	gossip_debug(GOSSIP_WAIT_DEBUG,
	474	"%s:Value returned from schedule_timeout(%d).\n",
	475	__func__,
	476	ret);
	477	if (!ret) {
	478	gossip_debug(GOSSIP_WAIT_DEBUG,
	479	"%s:*** operation timed out: %p\n",
	480	__func__,
	481	op);
eab9b389	482	spin_lock(&op->lock);
8bb8aefd	483	orangefs_clean_up_interrupted_operation(op);
1182fca3 MM	484	ret = -ETIMEDOUT;
	485	break;
	486	}
	487
	488	gossip_debug(GOSSIP_WAIT_DEBUG,
	489	"%s:Breaking out of loop, regardless of value returned by schedule_timeout.\n",
	490	__func__);
	491	ret = -ETIMEDOUT;
	492	break;
	493	}
	494
1182fca3	495	spin_lock(&op->lock);
ce6c414e	496	finish_wait(&op->waitq, &wait_entry);
1182fca3 MM	497	spin_unlock(&op->lock);
	498
	499	gossip_debug(GOSSIP_WAIT_DEBUG,
	500	"%s:returning ret(%d)\n",
	501	__func__,
	502	ret);
	503
	504	return ret;
	505	}