[linux-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_matching_downcall(struct orangefs_kernel_op_s *, bool);
static void orangefs_clean_up_interrupted_operation(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));
		set_op_state_purged(op);
	}
	spin_unlock(&orangefs_request_list_lock);
}

/*
 * 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 */
	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);

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

	/* queue up the operation */
	spin_lock(&orangefs_request_list_lock);
	spin_lock(&op->lock);
	set_op_state_waiting(op);
	if (flags & ORANGEFS_OP_PRIORITY)
		list_add(&op->list, &orangefs_request_list);
	else
		list_add_tail(&op->list, &orangefs_request_list);
	spin_unlock(&op->lock);
	wake_up_interruptible(&orangefs_request_list_waitq);
	if (!__is_daemon_in_service()) {
		/*
		 * 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.\n",
			     __func__);
		op->attempts++;
	}
	spin_unlock(&orangefs_request_list_lock);

	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;

	ret = wait_for_matching_downcall(op, flags & ORANGEFS_OP_INTERRUPTIBLE);

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

	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) {
			WARN_ON(1);
			/*
			 * 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;
}

bool orangefs_cancel_op_in_progress(struct orangefs_kernel_op_s *op)
{
	u64 tag = op->tag;
	if (!op_state_in_progress(op))
		return false;

	op->slot_to_free = op->upcall.req.io.buf_index;
	memset(&op->upcall, 0, sizeof(op->upcall));
	memset(&op->downcall, 0, sizeof(op->downcall));
	op->upcall.type = ORANGEFS_VFS_OP_CANCEL;
	op->upcall.req.cancel.op_tag = tag;
	op->downcall.type = ORANGEFS_VFS_OP_INVALID;
	op->downcall.status = -1;
	orangefs_new_tag(op);

	spin_lock(&orangefs_request_list_lock);
	/* orangefs_request_list_lock is enough of a barrier here */
	if (!__is_daemon_in_service()) {
		spin_unlock(&orangefs_request_list_lock);
		return false;
	}
	spin_lock(&op->lock);
	set_op_state_waiting(op);
	list_add(&op->list, &orangefs_request_list);
	spin_unlock(&op->lock);
	spin_unlock(&orangefs_request_list_lock);

	gossip_debug(GOSSIP_UTILS_DEBUG,
		     "Attempting ORANGEFS operation cancellation of tag %llu\n",
		     llu(tag));
	return true;
}

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);
	}
	reinit_completion(&op->waitq);
}

/*
 * 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.
 *
 * Returns with op->lock taken.
 */
static int wait_for_matching_downcall(struct orangefs_kernel_op_s *op,
				      bool interruptible)
{
	long timeout, n;

	timeout = op->attempts ? op_timeout_secs * HZ : MAX_SCHEDULE_TIMEOUT;

	if (interruptible)
		n = wait_for_completion_interruptible_timeout(&op->waitq, timeout);
	else
		n = wait_for_completion_killable_timeout(&op->waitq, timeout);

	spin_lock(&op->lock);

	if (op_state_serviced(op))
		return 0;

	if (unlikely(n < 0)) {
		gossip_debug(GOSSIP_WAIT_DEBUG,
			     "*** %s:"
			     " operation interrupted by a signal (tag "
			     "%llu, op %p)\n",
			     __func__,
			     llu(op->tag),
			     op);
		return -EINTR;
	}
	op->attempts++;
	if (op_state_purged(op)) {
		gossip_debug(GOSSIP_WAIT_DEBUG,
			     "*** %s:"
			     " operation purged (tag "
			     "%llu, %p, att %d)\n",
			     __func__,
			     llu(op->tag),
			     op,
			     op->attempts);
		return (op->attempts < ORANGEFS_PURGE_RETRY_COUNT) ?
			 -EAGAIN :
			 -EIO;
	}
	/* must have timed out, then... */
	gossip_debug(GOSSIP_WAIT_DEBUG,
		     "*** %s:"
		     " operation timed out (tag"
		     " %llu, %p, att %d)\n",
		     __func__,
		     llu(op->tag),
		     op,
		     op->attempts);
	return -ETIMEDOUT;
}
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
c72f15b7	19	static int wait_for_matching_downcall(struct orangefs_kernel_op_s *, bool);
d2d87a3b	20	static void orangefs_clean_up_interrupted_operation(struct orangefs_kernel_op_s *);
ade3d781	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));
1182fca3	39	set_op_state_purged(op);
1182fca3	40	}
8bb8aefd	41	spin_unlock(&orangefs_request_list_lock);
1182fca3 MM	42	}
	43
	44	/*
8bb8aefd	45	* submits a ORANGEFS operation and waits for it to complete
1182fca3 MM	46	*
	47	* Note op->downcall.status will contain the status of the operation (in
	48	* errno format), whether provided by pvfs2-client or a result of failure to
	49	* service the operation. If the caller wishes to distinguish, then
	50	* op->state can be checked to see if it was serviced or not.
	51	*
	52	* Returns contents of op->downcall.status for convenience
	53	*/
8bb8aefd	54	int service_operation(struct orangefs_kernel_op_s *op,
1182fca3 MM	55	const char *op_name,
	56	int flags)
	57	{
	58	/* flags to modify behavior */
1182fca3 MM	59	int ret = 0;
1182fca3 MM	60
ce6c414e	61	DEFINE_WAIT(wait_entry);
1182fca3 MM	62
	63	op->upcall.tgid = current->tgid;
	64	op->upcall.pid = current->pid;
	65
	66	retry_servicing:
	67	op->downcall.status = 0;
	68	gossip_debug(GOSSIP_WAIT_DEBUG,
8bb8aefd	69	"orangefs: service_operation: %s %p\n",
1182fca3 MM	70	op_name,
	71	op);
	72	gossip_debug(GOSSIP_WAIT_DEBUG,
8bb8aefd	73	"orangefs: operation posted by process: %s, pid: %i\n",
1182fca3 MM	74	current->comm,
	75	current->pid);
	76
8bb8aefd	77	if (!(flags & ORANGEFS_OP_NO_SEMAPHORE)) {
c72f15b7 AV	78	if (flags & ORANGEFS_OP_INTERRUPTIBLE)
	79	ret = mutex_lock_interruptible(&request_mutex);
	80	else
	81	ret = mutex_lock_killable(&request_mutex);
1182fca3 MM	82	/*
	83	* check to see if we were interrupted while waiting for
	84	* semaphore
	85	*/
	86	if (ret < 0) {
1182fca3 MM	87	op->downcall.status = ret;
1182fca3 MM	88	gossip_debug(GOSSIP_WAIT_DEBUG,
8bb8aefd	89	"orangefs: service_operation interrupted.\n");
1182fca3 MM	90	return ret;
	91	}
	92	}
	93
98815ade AV	94	/* queue up the operation */
	95	spin_lock(&orangefs_request_list_lock);
	96	spin_lock(&op->lock);
	97	set_op_state_waiting(op);
	98	if (flags & ORANGEFS_OP_PRIORITY)
	99	list_add(&op->list, &orangefs_request_list);
	100	else
	101	list_add_tail(&op->list, &orangefs_request_list);
	102	spin_unlock(&op->lock);
	103	wake_up_interruptible(&orangefs_request_list_waitq);
	104	if (!__is_daemon_in_service()) {
1182fca3 MM	105	/*
	106	* By incrementing the per-operation attempt counter, we
	107	* directly go into the timeout logic while waiting for
	108	* the matching downcall to be read
	109	*/
	110	gossip_debug(GOSSIP_WAIT_DEBUG,
98815ade	111	"%s:client core is NOT in service.\n",
1182fca3	112	__func__);
98815ade	113	op->attempts++;
1182fca3	114	}
98815ade	115	spin_unlock(&orangefs_request_list_lock);
1182fca3	116
8bb8aefd	117	if (!(flags & ORANGEFS_OP_NO_SEMAPHORE))
1182fca3 MM	118	mutex_unlock(&request_mutex);
	119
	120	/*
	121	* If we are asked to service an asynchronous operation from
	122	* VFS perspective, we are done.
	123	*/
8bb8aefd	124	if (flags & ORANGEFS_OP_ASYNC)
1182fca3 MM	125	return 0;
1182fca3 MM	126
c72f15b7	127	ret = wait_for_matching_downcall(op, flags & ORANGEFS_OP_INTERRUPTIBLE);
1182fca3 MM	128
	129	if (ret < 0) {
	130	/* failed to get matching downcall */
	131	if (ret == -ETIMEDOUT) {
8bb8aefd	132	gossip_err("orangefs: %s -- wait timed out; aborting attempt.\n",
1182fca3 MM	133	op_name);
1182fca3 MM	134	}
d2d87a3b	135	orangefs_clean_up_interrupted_operation(op);
1182fca3 MM	136	op->downcall.status = ret;
1182fca3 MM	137	} else {
d2d87a3b	138	spin_unlock(&op->lock);
1182fca3 MM	139	/* got matching downcall; make sure status is in errno format */
1182fca3 MM	140	op->downcall.status =
8bb8aefd	141	orangefs_normalize_to_errno(op->downcall.status);
1182fca3 MM	142	ret = op->downcall.status;
	143	}
	144
1182fca3 MM	145	BUG_ON(ret != op->downcall.status);
	146	/* retry if operation has not been serviced and if requested */
	147	if (!op_state_serviced(op) && op->downcall.status == -EAGAIN) {
	148	gossip_debug(GOSSIP_WAIT_DEBUG,
8bb8aefd	149	"orangefs: tag %llu (%s)"
1182fca3 MM	150	" -- operation to be retried (%d attempt)\n",
	151	llu(op->tag),
	152	op_name,
	153	op->attempts + 1);
	154
	155	if (!op->uses_shared_memory)
	156	/*
	157	* this operation doesn't use the shared memory
	158	* system
	159	*/
	160	goto retry_servicing;
	161
	162	/* op uses shared memory */
7d221485	163	if (orangefs_get_bufmap_init() == 0) {
6ebcc3fc	164	WARN_ON(1);
1182fca3 MM	165	/*
	166	* This operation uses the shared memory system AND
	167	* the system is not yet ready. This situation occurs
	168	* when the client-core is restarted AND there were
	169	* operations waiting to be processed or were already
	170	* in process.
	171	*/
	172	gossip_debug(GOSSIP_WAIT_DEBUG,
	173	"uses_shared_memory is true.\n");
	174	gossip_debug(GOSSIP_WAIT_DEBUG,
	175	"Client core in-service status(%d).\n",
	176	is_daemon_in_service());
	177	gossip_debug(GOSSIP_WAIT_DEBUG, "bufmap_init:%d.\n",
7d221485	178	orangefs_get_bufmap_init());
1182fca3 MM	179	gossip_debug(GOSSIP_WAIT_DEBUG,
	180	"operation's status is 0x%0x.\n",
	181	op->op_state);
	182
	183	/*
	184	* let process sleep for a few seconds so shared
	185	* memory system can be initialized.
	186	*/
ce6c414e MM	187	prepare_to_wait(&orangefs_bufmap_init_waitq,
	188	&wait_entry,
	189	TASK_INTERRUPTIBLE);
1182fca3	190
1182fca3	191	/*
8bb8aefd	192	* Wait for orangefs_bufmap_initialize() to wake me up
1182fca3 MM	193	* within the allotted time.
1182fca3 MM	194	*/
727cbfea AV	195	ret = schedule_timeout(
727cbfea AV	196	ORANGEFS_BUFMAP_WAIT_TIMEOUT_SECS * HZ);
1182fca3 MM	197
	198	gossip_debug(GOSSIP_WAIT_DEBUG,
	199	"Value returned from schedule_timeout:"
	200	"%d.\n",
	201	ret);
	202	gossip_debug(GOSSIP_WAIT_DEBUG,
	203	"Is shared memory available? (%d).\n",
7d221485	204	orangefs_get_bufmap_init());
1182fca3	205
ce6c414e	206	finish_wait(&orangefs_bufmap_init_waitq, &wait_entry);
1182fca3	207
7d221485	208	if (orangefs_get_bufmap_init() == 0) {
1182fca3 MM	209	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	210	__func__,
8bb8aefd	211	ORANGEFS_BUFMAP_WAIT_TIMEOUT_SECS,
1182fca3 MM	212	get_opname_string(op));
	213	return -EIO;
	214	}
	215
	216	/*
	217	* Return to the calling function and re-populate a
	218	* shared memory buffer.
	219	*/
	220	return -EAGAIN;
	221	}
	222	}
	223
	224	gossip_debug(GOSSIP_WAIT_DEBUG,
8bb8aefd	225	"orangefs: service_operation %s returning: %d for %p.\n",
1182fca3 MM	226	op_name,
	227	ret,
	228	op);
	229	return ret;
	230	}
	231
78699e29 AV	232	bool orangefs_cancel_op_in_progress(struct orangefs_kernel_op_s *op)
	233	{
	234	u64 tag = op->tag;
	235	if (!op_state_in_progress(op))
	236	return false;
	237
	238	op->slot_to_free = op->upcall.req.io.buf_index;
	239	memset(&op->upcall, 0, sizeof(op->upcall));
	240	memset(&op->downcall, 0, sizeof(op->downcall));
	241	op->upcall.type = ORANGEFS_VFS_OP_CANCEL;
	242	op->upcall.req.cancel.op_tag = tag;
	243	op->downcall.type = ORANGEFS_VFS_OP_INVALID;
	244	op->downcall.status = -1;
	245	orangefs_new_tag(op);
	246
	247	spin_lock(&orangefs_request_list_lock);
	248	/* orangefs_request_list_lock is enough of a barrier here */
	249	if (!__is_daemon_in_service()) {
	250	spin_unlock(&orangefs_request_list_lock);
	251	return false;
	252	}
98815ade AV	253	spin_lock(&op->lock);
	254	set_op_state_waiting(op);
	255	list_add(&op->list, &orangefs_request_list);
	256	spin_unlock(&op->lock);
78699e29 AV	257	spin_unlock(&orangefs_request_list_lock);
	258
	259	gossip_debug(GOSSIP_UTILS_DEBUG,
	260	"Attempting ORANGEFS operation cancellation of tag %llu\n",
	261	llu(tag));
	262	return true;
	263	}
	264
e07db0a2	265	static void orangefs_clean_up_interrupted_operation(struct orangefs_kernel_op_s *op)
1182fca3 MM	266	{
	267	/*
	268	* handle interrupted cases depending on what state we were in when
	269	* the interruption is detected. there is a coarse grained lock
	270	* across the operation.
	271	*
eab9b389	272	* Called with op->lock held.
1182fca3	273	*/
ed42fe05	274	op->op_state \|= OP_VFS_STATE_GIVEN_UP;
1182fca3 MM	275
	276	if (op_state_waiting(op)) {
	277	/*
	278	* upcall hasn't been read; remove op from upcall request
	279	* list.
	280	*/
	281	spin_unlock(&op->lock);
ed42fe05 AV	282	spin_lock(&orangefs_request_list_lock);
	283	list_del(&op->list);
	284	spin_unlock(&orangefs_request_list_lock);
1182fca3 MM	285	gossip_debug(GOSSIP_WAIT_DEBUG,
	286	"Interrupted: Removed op %p from request_list\n",
	287	op);
	288	} else if (op_state_in_progress(op)) {
	289	/* op must be removed from the in progress htable */
	290	spin_unlock(&op->lock);
	291	spin_lock(&htable_ops_in_progress_lock);
	292	list_del(&op->list);
	293	spin_unlock(&htable_ops_in_progress_lock);
	294	gossip_debug(GOSSIP_WAIT_DEBUG,
	295	"Interrupted: Removed op %p"
	296	" from htable_ops_in_progress\n",
	297	op);
	298	} else if (!op_state_serviced(op)) {
	299	spin_unlock(&op->lock);
	300	gossip_err("interrupted operation is in a weird state 0x%x\n",
	301	op->op_state);
84d02150 MM	302	} else {
	303	/*
	304	* It is not intended for execution to flow here,
	305	* but having this unlock here makes sparse happy.
	306	*/
	307	gossip_err("%s: can't get here.\n", __func__);
	308	spin_unlock(&op->lock);
1182fca3	309	}
d2d87a3b	310	reinit_completion(&op->waitq);
1182fca3 MM	311	}
	312
	313	/*
	314	* sleeps on waitqueue waiting for matching downcall.
	315	* if client-core finishes servicing, then we are good to go.
	316	* else if client-core exits, we get woken up here, and retry with a timeout
	317	*
	318	* Post when this call returns to the caller, the specified op will no
	319	* longer be on any list or htable.
	320	*
	321	* Returns 0 on success and -errno on failure
	322	* Errors are:
	323	* EAGAIN in case we want the caller to requeue and try again..
	324	* EINTR/EIO/ETIMEDOUT indicating we are done trying to service this
	325	* operation since client-core seems to be exiting too often
	326	* or if we were interrupted.
d2d87a3b AV	327	*
d2d87a3b AV	328	* Returns with op->lock taken.
1182fca3	329	*/
c72f15b7 AV	330	static int wait_for_matching_downcall(struct orangefs_kernel_op_s *op,
c72f15b7 AV	331	bool interruptible)
1182fca3	332	{
d2d87a3b	333	long timeout, n;
1182fca3	334
d2d87a3b	335	timeout = op->attempts ? op_timeout_secs * HZ : MAX_SCHEDULE_TIMEOUT;
c72f15b7 AV	336
	337	if (interruptible)
	338	n = wait_for_completion_interruptible_timeout(&op->waitq, timeout);
	339	else
	340	n = wait_for_completion_killable_timeout(&op->waitq, timeout);
	341
d2d87a3b	342	spin_lock(&op->lock);
1182fca3	343
d2d87a3b AV	344	if (op_state_serviced(op))
d2d87a3b AV	345	return 0;
70c6ea26	346
d2d87a3b AV	347	if (unlikely(n < 0)) {
	348	gossip_debug(GOSSIP_WAIT_DEBUG,
	349	"*** %s:"
	350	" operation interrupted by a signal (tag "
	351	"%llu, op %p)\n",
	352	__func__,
	353	llu(op->tag),
	354	op);
	355	return -EINTR;
1182fca3	356	}
d2d87a3b AV	357	op->attempts++;
	358	if (op_state_purged(op)) {
	359	gossip_debug(GOSSIP_WAIT_DEBUG,
	360	"*** %s:"
	361	" operation purged (tag "
	362	"%llu, %p, att %d)\n",
	363	__func__,
	364	llu(op->tag),
	365	op,
	366	op->attempts);
	367	return (op->attempts < ORANGEFS_PURGE_RETRY_COUNT) ?
	368	-EAGAIN :
	369	-EIO;
	370	}
	371	/* must have timed out, then... */
	372	gossip_debug(GOSSIP_WAIT_DEBUG,
	373	"*** %s:"
	374	" operation timed out (tag"
	375	" %llu, %p, att %d)\n",
	376	__func__,
	377	llu(op->tag),
	378	op,
	379	op->attempts);
	380	return -ETIMEDOUT;
1182fca3	381	}