[mirror_ubuntu-bionic-kernel.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 *, long, 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);
		gossip_debug(GOSSIP_DEV_DEBUG,
			     "%s: op:%s: op_state:%d: process:%s:\n",
			     __func__,
			     get_opname_string(op),
			     op->op_state,
			     current->comm);
	}
	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)
{
	long timeout = MAX_SCHEDULE_TIMEOUT;
	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,
		     "%s: %s op:%p: process:%s: pid:%d:\n",
		     __func__,
		     op_name,
		     op,
		     current->comm,
		     current->pid);

	/*
	 * If ORANGEFS_OP_NO_MUTEX was set in flags, we need to avoid
	 * acquiring the request_mutex because we're servicing a
	 * high priority remount operation and the request_mutex is
	 * already taken.
	 */
	if (!(flags & ORANGEFS_OP_NO_MUTEX)) {
		if (flags & ORANGEFS_OP_INTERRUPTIBLE)
			ret = mutex_lock_interruptible(&orangefs_request_mutex);
		else
			ret = mutex_lock_killable(&orangefs_request_mutex);
		/*
		 * check to see if we were interrupted while waiting for
		 * mutex
		 */
		if (ret < 0) {
			op->downcall.status = ret;
			gossip_debug(GOSSIP_WAIT_DEBUG,
				     "%s: service_operation interrupted.\n",
				     __func__);
			return ret;
		}
	}

	/* queue up the operation */
	spin_lock(&orangefs_request_list_lock);
	spin_lock(&op->lock);
	set_op_state_waiting(op);
	gossip_debug(GOSSIP_DEV_DEBUG,
		     "%s: op:%s: op_state:%d: process:%s:\n",
		     __func__,
		     get_opname_string(op),
		     op->op_state,
		     current->comm);
	/* add high priority remount op to the front of the line. */
	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()) {
		gossip_debug(GOSSIP_WAIT_DEBUG,
			     "%s:client core is NOT in service.\n",
			     __func__);
		/*
		 * Don't wait for the userspace component to return if
		 * the filesystem is being umounted anyway.
		 */
		if (op->upcall.type == ORANGEFS_VFS_OP_FS_UMOUNT)
			timeout = 0;
		else
			timeout = op_timeout_secs * HZ;
	}
	spin_unlock(&orangefs_request_list_lock);

	if (!(flags & ORANGEFS_OP_NO_MUTEX))
		mutex_unlock(&orangefs_request_mutex);

	ret = wait_for_matching_downcall(op, timeout,
					 flags & ORANGEFS_OP_INTERRUPTIBLE);

	gossip_debug(GOSSIP_WAIT_DEBUG,
		     "%s: wait_for_matching_downcall returned %d for %p\n",
		     __func__,
		     ret,
		     op);

	/* got matching downcall; make sure status is in errno format */
	if (!ret) {
		spin_unlock(&op->lock);
		op->downcall.status =
		    orangefs_normalize_to_errno(op->downcall.status);
		ret = op->downcall.status;
		goto out;
	}

	/* failed to get matching downcall */
	if (ret == -ETIMEDOUT) {
		gossip_err("%s: %s -- wait timed out; aborting attempt.\n",
			   __func__,
			   op_name);
	}

	/*
	 * remove a waiting op from the request list or
	 * remove an in-progress op from the in-progress list.
	 */
	orangefs_clean_up_interrupted_operation(op);

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

		/*
		 * io ops (ops that use the shared memory buffer) have
		 * to be returned to their caller for a retry. Other ops
		 * can just be recycled here.
		 */
		if (!op->uses_shared_memory)
			goto retry_servicing;
	}

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

/* This can get called on an I/O op if it had a bad service_operation. */
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);
	gossip_debug(GOSSIP_DEV_DEBUG,
		     "%s: op:%s: op_state:%d: process:%s:\n",
		     __func__,
		     get_opname_string(op),
		     op->op_state,
		     current->comm);
	list_add(&op->list, &orangefs_request_list);
	spin_unlock(&op->lock);
	spin_unlock(&orangefs_request_list_lock);

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

/*
 * Change an op to the "given up" state and remove it from its list.
 */
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.
	 *
	 * Called with op->lock held.
	 */

	/*
	 * List manipulation code elsewhere will ignore ops that
	 * have been given up upon.
	 */
	op->op_state |= OP_VFS_STATE_GIVEN_UP;

	if (list_empty(&op->list)) {
		/* caught copying to/from daemon */
		BUG_ON(op_state_serviced(op));
		spin_unlock(&op->lock);
		wait_for_completion(&op->waitq);
	} else 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_init(&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(&orangefs_htable_ops_in_progress_lock);
		list_del_init(&op->list);
		spin_unlock(&orangefs_htable_ops_in_progress_lock);
		gossip_debug(GOSSIP_WAIT_DEBUG,
			     "Interrupted: Removed op %p"
			     " from htable_ops_in_progress\n",
			     op);
	} else {
		spin_unlock(&op->lock);
		gossip_err("interrupted operation is in a weird state 0x%x\n",
			   op->op_state);
	}
	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
 *
 * When this call returns to the caller, the specified op will no
 * longer be in either the in_progress hash table or on the request list.
 *
 * 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,
				      long timeout,
				      bool interruptible)
{
	long n;

	/*
	 * There's a "schedule_timeout" inside of these wait
	 * primitives, during which the op is out of the hands of the
	 * user process that needs something done and is being
	 * manipulated by the client-core process.
	 */
	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, tag %llu, %p\n",
			     __func__,
			     llu(op->tag),
			     op);
		return -EINTR;
	}
	if (op_state_purged(op)) {
		gossip_debug(GOSSIP_WAIT_DEBUG,
			     "%s: operation purged, tag %llu, %p, %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, %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
05b39a8b	19	static int wait_for_matching_downcall(struct orangefs_kernel_op_s *, long, 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);
9d9e7ba9 MM	40	gossip_debug(GOSSIP_DEV_DEBUG,
	41	"%s: op:%s: op_state:%d: process:%s:\n",
	42	__func__,
	43	get_opname_string(op),
	44	op->op_state,
	45	current->comm);
1182fca3	46	}
8bb8aefd	47	spin_unlock(&orangefs_request_list_lock);
1182fca3 MM	48	}
	49
	50	/*
8bb8aefd	51	* submits a ORANGEFS operation and waits for it to complete
1182fca3 MM	52	*
	53	* Note op->downcall.status will contain the status of the operation (in
	54	* errno format), whether provided by pvfs2-client or a result of failure to
	55	* service the operation. If the caller wishes to distinguish, then
	56	* op->state can be checked to see if it was serviced or not.
	57	*
	58	* Returns contents of op->downcall.status for convenience
	59	*/
8bb8aefd	60	int service_operation(struct orangefs_kernel_op_s *op,
1182fca3 MM	61	const char *op_name,
	62	int flags)
	63	{
05b39a8b	64	long timeout = MAX_SCHEDULE_TIMEOUT;
1182fca3 MM	65	int ret = 0;
1182fca3 MM	66
ce6c414e	67	DEFINE_WAIT(wait_entry);
1182fca3 MM	68
	69	op->upcall.tgid = current->tgid;
	70	op->upcall.pid = current->pid;
	71
	72	retry_servicing:
	73	op->downcall.status = 0;
	74	gossip_debug(GOSSIP_WAIT_DEBUG,
5253487e MM	75	"%s: %s op:%p: process:%s: pid:%d:\n",
5253487e MM	76	__func__,
1182fca3	77	op_name,
5253487e	78	op,
1182fca3 MM	79	current->comm,
	80	current->pid);
	81
adcf34a2 MM	82	/*
adcf34a2 MM	83	* If ORANGEFS_OP_NO_MUTEX was set in flags, we need to avoid
ca9f518e	84	* acquiring the request_mutex because we're servicing a
adcf34a2 MM	85	* high priority remount operation and the request_mutex is
	86	* already taken.
	87	*/
	88	if (!(flags & ORANGEFS_OP_NO_MUTEX)) {
c72f15b7	89	if (flags & ORANGEFS_OP_INTERRUPTIBLE)
1d503617	90	ret = mutex_lock_interruptible(&orangefs_request_mutex);
c72f15b7	91	else
1d503617	92	ret = mutex_lock_killable(&orangefs_request_mutex);
1182fca3 MM	93	/*
1182fca3 MM	94	* check to see if we were interrupted while waiting for
adcf34a2	95	* mutex
1182fca3 MM	96	*/
1182fca3 MM	97	if (ret < 0) {
1182fca3 MM	98	op->downcall.status = ret;
1182fca3 MM	99	gossip_debug(GOSSIP_WAIT_DEBUG,
ca9f518e MM	100	"%s: service_operation interrupted.\n",
ca9f518e MM	101	__func__);
1182fca3 MM	102	return ret;
	103	}
	104	}
	105
98815ade AV	106	/* queue up the operation */
	107	spin_lock(&orangefs_request_list_lock);
	108	spin_lock(&op->lock);
	109	set_op_state_waiting(op);
9d9e7ba9 MM	110	gossip_debug(GOSSIP_DEV_DEBUG,
	111	"%s: op:%s: op_state:%d: process:%s:\n",
	112	__func__,
	113	get_opname_string(op),
	114	op->op_state,
	115	current->comm);
adcf34a2	116	/* add high priority remount op to the front of the line. */
98815ade AV	117	if (flags & ORANGEFS_OP_PRIORITY)
	118	list_add(&op->list, &orangefs_request_list);
	119	else
	120	list_add_tail(&op->list, &orangefs_request_list);
	121	spin_unlock(&op->lock);
	122	wake_up_interruptible(&orangefs_request_list_waitq);
	123	if (!__is_daemon_in_service()) {
1182fca3	124	gossip_debug(GOSSIP_WAIT_DEBUG,
98815ade	125	"%s:client core is NOT in service.\n",
1182fca3	126	__func__);
b5a9d61e MB	127	/*
	128	* Don't wait for the userspace component to return if
	129	* the filesystem is being umounted anyway.
	130	*/
	131	if (op->upcall.type == ORANGEFS_VFS_OP_FS_UMOUNT)
	132	timeout = 0;
	133	else
	134	timeout = op_timeout_secs * HZ;
1182fca3	135	}
98815ade	136	spin_unlock(&orangefs_request_list_lock);
1182fca3	137
adcf34a2	138	if (!(flags & ORANGEFS_OP_NO_MUTEX))
1d503617	139	mutex_unlock(&orangefs_request_mutex);
1182fca3	140
05b39a8b AV	141	ret = wait_for_matching_downcall(op, timeout,
05b39a8b AV	142	flags & ORANGEFS_OP_INTERRUPTIBLE);
5253487e MM	143
	144	gossip_debug(GOSSIP_WAIT_DEBUG,
	145	"%s: wait_for_matching_downcall returned %d for %p\n",
	146	__func__,
	147	ret,
	148	op);
	149
ca9f518e	150	/* got matching downcall; make sure status is in errno format */
05b39a8b	151	if (!ret) {
d2d87a3b	152	spin_unlock(&op->lock);
1182fca3	153	op->downcall.status =
8bb8aefd	154	orangefs_normalize_to_errno(op->downcall.status);
1182fca3	155	ret = op->downcall.status;
05b39a8b	156	goto out;
1182fca3 MM	157	}
1182fca3 MM	158
05b39a8b AV	159	/* failed to get matching downcall */
05b39a8b AV	160	if (ret == -ETIMEDOUT) {
adcf34a2 MM	161	gossip_err("%s: %s -- wait timed out; aborting attempt.\n",
adcf34a2 MM	162	__func__,
05b39a8b AV	163	op_name);
05b39a8b AV	164	}
adcf34a2 MM	165
adcf34a2 MM	166	/*
ca9f518e MM	167	* remove a waiting op from the request list or
ca9f518e MM	168	* remove an in-progress op from the in-progress list.
adcf34a2	169	*/
05b39a8b	170	orangefs_clean_up_interrupted_operation(op);
adcf34a2	171
05b39a8b	172	op->downcall.status = ret;
1182fca3	173	/* retry if operation has not been serviced and if requested */
05b39a8b AV	174	if (ret == -EAGAIN) {
	175	op->attempts++;
	176	timeout = op_timeout_secs * HZ;
1182fca3	177	gossip_debug(GOSSIP_WAIT_DEBUG,
8bb8aefd	178	"orangefs: tag %llu (%s)"
1182fca3 MM	179	" -- operation to be retried (%d attempt)\n",
	180	llu(op->tag),
	181	op_name,
05b39a8b	182	op->attempts);
1182fca3	183
adcf34a2 MM	184	/*
	185	* io ops (ops that use the shared memory buffer) have
	186	* to be returned to their caller for a retry. Other ops
	187	* can just be recycled here.
	188	*/
1182fca3	189	if (!op->uses_shared_memory)
1182fca3	190	goto retry_servicing;
1182fca3 MM	191	}
1182fca3 MM	192
05b39a8b	193	out:
1182fca3	194	gossip_debug(GOSSIP_WAIT_DEBUG,
9d9e7ba9 MM	195	"%s: %s returning: %d for %p.\n",
9d9e7ba9 MM	196	__func__,
1182fca3 MM	197	op_name,
	198	ret,
	199	op);
	200	return ret;
	201	}
	202
ca9f518e	203	/* This can get called on an I/O op if it had a bad service_operation. */
78699e29 AV	204	bool orangefs_cancel_op_in_progress(struct orangefs_kernel_op_s *op)
	205	{
	206	u64 tag = op->tag;
	207	if (!op_state_in_progress(op))
	208	return false;
	209
	210	op->slot_to_free = op->upcall.req.io.buf_index;
	211	memset(&op->upcall, 0, sizeof(op->upcall));
	212	memset(&op->downcall, 0, sizeof(op->downcall));
	213	op->upcall.type = ORANGEFS_VFS_OP_CANCEL;
	214	op->upcall.req.cancel.op_tag = tag;
	215	op->downcall.type = ORANGEFS_VFS_OP_INVALID;
	216	op->downcall.status = -1;
	217	orangefs_new_tag(op);
	218
	219	spin_lock(&orangefs_request_list_lock);
	220	/* orangefs_request_list_lock is enough of a barrier here */
	221	if (!__is_daemon_in_service()) {
	222	spin_unlock(&orangefs_request_list_lock);
	223	return false;
	224	}
98815ade AV	225	spin_lock(&op->lock);
98815ade AV	226	set_op_state_waiting(op);
9d9e7ba9 MM	227	gossip_debug(GOSSIP_DEV_DEBUG,
	228	"%s: op:%s: op_state:%d: process:%s:\n",
	229	__func__,
	230	get_opname_string(op),
	231	op->op_state,
	232	current->comm);
98815ade AV	233	list_add(&op->list, &orangefs_request_list);
98815ade AV	234	spin_unlock(&op->lock);
78699e29 AV	235	spin_unlock(&orangefs_request_list_lock);
78699e29 AV	236
ca9f518e	237	gossip_debug(GOSSIP_WAIT_DEBUG,
78699e29 AV	238	"Attempting ORANGEFS operation cancellation of tag %llu\n",
	239	llu(tag));
	240	return true;
	241	}
	242
ca9f518e MM	243	/*
	244	* Change an op to the "given up" state and remove it from its list.
	245	*/
	246	static void
	247	orangefs_clean_up_interrupted_operation(struct orangefs_kernel_op_s *op)
1182fca3 MM	248	{
	249	/*
	250	* handle interrupted cases depending on what state we were in when
ca9f518e	251	* the interruption is detected.
1182fca3	252	*
eab9b389	253	* Called with op->lock held.
1182fca3	254	*/
ca9f518e MM	255
	256	/*
	257	* List manipulation code elsewhere will ignore ops that
	258	* have been given up upon.
	259	*/
ed42fe05	260	op->op_state \|= OP_VFS_STATE_GIVEN_UP;
ca9f518e	261
05a50a5b AV	262	if (list_empty(&op->list)) {
	263	/* caught copying to/from daemon */
	264	BUG_ON(op_state_serviced(op));
	265	spin_unlock(&op->lock);
	266	wait_for_completion(&op->waitq);
	267	} else if (op_state_waiting(op)) {
1182fca3 MM	268	/*
	269	* upcall hasn't been read; remove op from upcall request
	270	* list.
	271	*/
	272	spin_unlock(&op->lock);
ed42fe05	273	spin_lock(&orangefs_request_list_lock);
05a50a5b	274	list_del_init(&op->list);
ed42fe05	275	spin_unlock(&orangefs_request_list_lock);
1182fca3 MM	276	gossip_debug(GOSSIP_WAIT_DEBUG,
	277	"Interrupted: Removed op %p from request_list\n",
	278	op);
	279	} else if (op_state_in_progress(op)) {
	280	/* op must be removed from the in progress htable */
	281	spin_unlock(&op->lock);
1d503617	282	spin_lock(&orangefs_htable_ops_in_progress_lock);
05a50a5b	283	list_del_init(&op->list);
1d503617	284	spin_unlock(&orangefs_htable_ops_in_progress_lock);
1182fca3 MM	285	gossip_debug(GOSSIP_WAIT_DEBUG,
	286	"Interrupted: Removed op %p"
	287	" from htable_ops_in_progress\n",
	288	op);
05a50a5b	289	} else {
1182fca3 MM	290	spin_unlock(&op->lock);
	291	gossip_err("interrupted operation is in a weird state 0x%x\n",
	292	op->op_state);
	293	}
d2d87a3b	294	reinit_completion(&op->waitq);
1182fca3 MM	295	}
	296
	297	/*
ca9f518e MM	298	* Sleeps on waitqueue waiting for matching downcall.
ca9f518e MM	299	* If client-core finishes servicing, then we are good to go.
1182fca3 MM	300	* else if client-core exits, we get woken up here, and retry with a timeout
1182fca3 MM	301	*
ca9f518e MM	302	* When this call returns to the caller, the specified op will no
ca9f518e MM	303	* longer be in either the in_progress hash table or on the request list.
1182fca3 MM	304	*
	305	* Returns 0 on success and -errno on failure
	306	* Errors are:
	307	* EAGAIN in case we want the caller to requeue and try again..
	308	* EINTR/EIO/ETIMEDOUT indicating we are done trying to service this
	309	* operation since client-core seems to be exiting too often
	310	* or if we were interrupted.
d2d87a3b AV	311	*
d2d87a3b AV	312	* Returns with op->lock taken.
1182fca3	313	*/
c72f15b7	314	static int wait_for_matching_downcall(struct orangefs_kernel_op_s *op,
05b39a8b	315	long timeout,
c72f15b7	316	bool interruptible)
1182fca3	317	{
05b39a8b	318	long n;
c72f15b7	319
ca9f518e MM	320	/*
	321	* There's a "schedule_timeout" inside of these wait
	322	* primitives, during which the op is out of the hands of the
	323	* user process that needs something done and is being
	324	* manipulated by the client-core process.
	325	*/
c72f15b7	326	if (interruptible)
adcf34a2 MM	327	n = wait_for_completion_interruptible_timeout(&op->waitq,
adcf34a2 MM	328	timeout);
c72f15b7 AV	329	else
	330	n = wait_for_completion_killable_timeout(&op->waitq, timeout);
	331
d2d87a3b	332	spin_lock(&op->lock);
1182fca3	333
d2d87a3b AV	334	if (op_state_serviced(op))
d2d87a3b AV	335	return 0;
70c6ea26	336
d2d87a3b AV	337	if (unlikely(n < 0)) {
d2d87a3b AV	338	gossip_debug(GOSSIP_WAIT_DEBUG,
9d9e7ba9	339	"%s: operation interrupted, tag %llu, %p\n",
d2d87a3b AV	340	__func__,
	341	llu(op->tag),
	342	op);
	343	return -EINTR;
1182fca3	344	}
d2d87a3b AV	345	if (op_state_purged(op)) {
d2d87a3b AV	346	gossip_debug(GOSSIP_WAIT_DEBUG,
9d9e7ba9	347	"%s: operation purged, tag %llu, %p, %d\n",
d2d87a3b AV	348	__func__,
	349	llu(op->tag),
	350	op,
	351	op->attempts);
	352	return (op->attempts < ORANGEFS_PURGE_RETRY_COUNT) ?
	353	-EAGAIN :
	354	-EIO;
	355	}
	356	/* must have timed out, then... */
	357	gossip_debug(GOSSIP_WAIT_DEBUG,
9d9e7ba9	358	"%s: operation timed out, tag %llu, %p, %d)\n",
d2d87a3b AV	359	__func__,
	360	llu(op->tag),
	361	op,
	362	op->attempts);
	363	return -ETIMEDOUT;
1182fca3	364	}