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

// SPDX-License-Identifier: GPL-2.0
/*
 * (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
b2441318	1	// SPDX-License-Identifier: GPL-2.0
1182fca3 MM	2	/*
	3	* (C) 2001 Clemson University and The University of Chicago
	4	* (C) 2011 Omnibond Systems
	5	*
	6	* Changes by Acxiom Corporation to implement generic service_operation()
	7	* function, Copyright Acxiom Corporation, 2005.
	8	*
	9	* See COPYING in top-level directory.
	10	*/
	11
	12	/*
	13	* In-kernel waitqueue operations.
	14	*/
	15
	16	#include "protocol.h"
575e9461 MM	17	#include "orangefs-kernel.h"
575e9461 MM	18	#include "orangefs-bufmap.h"
1182fca3	19
05b39a8b	20	static int wait_for_matching_downcall(struct orangefs_kernel_op_s *, long, bool);
d2d87a3b	21	static void orangefs_clean_up_interrupted_operation(struct orangefs_kernel_op_s *);
ade3d781	22
1182fca3 MM	23	/*
	24	* What we do in this function is to walk the list of operations that are
	25	* present in the request queue and mark them as purged.
	26	* NOTE: This is called from the device close after client-core has
	27	* guaranteed that no new operations could appear on the list since the
	28	* client-core is anyway going to exit.
	29	*/
	30	void purge_waiting_ops(void)
	31	{
8bb8aefd	32	struct orangefs_kernel_op_s *op;
1182fca3	33
8bb8aefd YL	34	spin_lock(&orangefs_request_list_lock);
8bb8aefd YL	35	list_for_each_entry(op, &orangefs_request_list, list) {
1182fca3 MM	36	gossip_debug(GOSSIP_WAIT_DEBUG,
	37	"pvfs2-client-core: purging op tag %llu %s\n",
	38	llu(op->tag),
	39	get_opname_string(op));
1182fca3	40	set_op_state_purged(op);
9d9e7ba9 MM	41	gossip_debug(GOSSIP_DEV_DEBUG,
	42	"%s: op:%s: op_state:%d: process:%s:\n",
	43	__func__,
	44	get_opname_string(op),
	45	op->op_state,
	46	current->comm);
1182fca3	47	}
8bb8aefd	48	spin_unlock(&orangefs_request_list_lock);
1182fca3 MM	49	}
	50
	51	/*
8bb8aefd	52	* submits a ORANGEFS operation and waits for it to complete
1182fca3 MM	53	*
	54	* Note op->downcall.status will contain the status of the operation (in
	55	* errno format), whether provided by pvfs2-client or a result of failure to
	56	* service the operation. If the caller wishes to distinguish, then
	57	* op->state can be checked to see if it was serviced or not.
	58	*
	59	* Returns contents of op->downcall.status for convenience
	60	*/
8bb8aefd	61	int service_operation(struct orangefs_kernel_op_s *op,
1182fca3 MM	62	const char *op_name,
	63	int flags)
	64	{
05b39a8b	65	long timeout = MAX_SCHEDULE_TIMEOUT;
1182fca3 MM	66	int ret = 0;
1182fca3 MM	67
ce6c414e	68	DEFINE_WAIT(wait_entry);
1182fca3 MM	69
	70	op->upcall.tgid = current->tgid;
	71	op->upcall.pid = current->pid;
	72
	73	retry_servicing:
	74	op->downcall.status = 0;
	75	gossip_debug(GOSSIP_WAIT_DEBUG,
5253487e MM	76	"%s: %s op:%p: process:%s: pid:%d:\n",
5253487e MM	77	__func__,
1182fca3	78	op_name,
5253487e	79	op,
1182fca3 MM	80	current->comm,
	81	current->pid);
	82
adcf34a2 MM	83	/*
adcf34a2 MM	84	* If ORANGEFS_OP_NO_MUTEX was set in flags, we need to avoid
ca9f518e	85	* acquiring the request_mutex because we're servicing a
adcf34a2 MM	86	* high priority remount operation and the request_mutex is
	87	* already taken.
	88	*/
	89	if (!(flags & ORANGEFS_OP_NO_MUTEX)) {
c72f15b7	90	if (flags & ORANGEFS_OP_INTERRUPTIBLE)
1d503617	91	ret = mutex_lock_interruptible(&orangefs_request_mutex);
c72f15b7	92	else
1d503617	93	ret = mutex_lock_killable(&orangefs_request_mutex);
1182fca3 MM	94	/*
1182fca3 MM	95	* check to see if we were interrupted while waiting for
adcf34a2	96	* mutex
1182fca3 MM	97	*/
1182fca3 MM	98	if (ret < 0) {
1182fca3 MM	99	op->downcall.status = ret;
1182fca3 MM	100	gossip_debug(GOSSIP_WAIT_DEBUG,
ca9f518e MM	101	"%s: service_operation interrupted.\n",
ca9f518e MM	102	__func__);
1182fca3 MM	103	return ret;
	104	}
	105	}
	106
98815ade AV	107	/* queue up the operation */
	108	spin_lock(&orangefs_request_list_lock);
	109	spin_lock(&op->lock);
	110	set_op_state_waiting(op);
9d9e7ba9 MM	111	gossip_debug(GOSSIP_DEV_DEBUG,
	112	"%s: op:%s: op_state:%d: process:%s:\n",
	113	__func__,
	114	get_opname_string(op),
	115	op->op_state,
	116	current->comm);
adcf34a2	117	/* add high priority remount op to the front of the line. */
98815ade AV	118	if (flags & ORANGEFS_OP_PRIORITY)
	119	list_add(&op->list, &orangefs_request_list);
	120	else
	121	list_add_tail(&op->list, &orangefs_request_list);
	122	spin_unlock(&op->lock);
	123	wake_up_interruptible(&orangefs_request_list_waitq);
	124	if (!__is_daemon_in_service()) {
1182fca3	125	gossip_debug(GOSSIP_WAIT_DEBUG,
98815ade	126	"%s:client core is NOT in service.\n",
1182fca3	127	__func__);
b5a9d61e MB	128	/*
	129	* Don't wait for the userspace component to return if
	130	* the filesystem is being umounted anyway.
	131	*/
	132	if (op->upcall.type == ORANGEFS_VFS_OP_FS_UMOUNT)
	133	timeout = 0;
	134	else
	135	timeout = op_timeout_secs * HZ;
1182fca3	136	}
98815ade	137	spin_unlock(&orangefs_request_list_lock);
1182fca3	138
adcf34a2	139	if (!(flags & ORANGEFS_OP_NO_MUTEX))
1d503617	140	mutex_unlock(&orangefs_request_mutex);
1182fca3	141
05b39a8b AV	142	ret = wait_for_matching_downcall(op, timeout,
05b39a8b AV	143	flags & ORANGEFS_OP_INTERRUPTIBLE);
5253487e MM	144
	145	gossip_debug(GOSSIP_WAIT_DEBUG,
	146	"%s: wait_for_matching_downcall returned %d for %p\n",
	147	__func__,
	148	ret,
	149	op);
	150
ca9f518e	151	/* got matching downcall; make sure status is in errno format */
05b39a8b	152	if (!ret) {
d2d87a3b	153	spin_unlock(&op->lock);
1182fca3	154	op->downcall.status =
8bb8aefd	155	orangefs_normalize_to_errno(op->downcall.status);
1182fca3	156	ret = op->downcall.status;
05b39a8b	157	goto out;
1182fca3 MM	158	}
1182fca3 MM	159
05b39a8b AV	160	/* failed to get matching downcall */
05b39a8b AV	161	if (ret == -ETIMEDOUT) {
adcf34a2 MM	162	gossip_err("%s: %s -- wait timed out; aborting attempt.\n",
adcf34a2 MM	163	__func__,
05b39a8b AV	164	op_name);
05b39a8b AV	165	}
adcf34a2 MM	166
adcf34a2 MM	167	/*
ca9f518e MM	168	* remove a waiting op from the request list or
ca9f518e MM	169	* remove an in-progress op from the in-progress list.
adcf34a2	170	*/
05b39a8b	171	orangefs_clean_up_interrupted_operation(op);
adcf34a2	172
05b39a8b	173	op->downcall.status = ret;
1182fca3	174	/* retry if operation has not been serviced and if requested */
05b39a8b AV	175	if (ret == -EAGAIN) {
	176	op->attempts++;
	177	timeout = op_timeout_secs * HZ;
1182fca3	178	gossip_debug(GOSSIP_WAIT_DEBUG,
8bb8aefd	179	"orangefs: tag %llu (%s)"
1182fca3 MM	180	" -- operation to be retried (%d attempt)\n",
	181	llu(op->tag),
	182	op_name,
05b39a8b	183	op->attempts);
1182fca3	184
adcf34a2 MM	185	/*
	186	* io ops (ops that use the shared memory buffer) have
	187	* to be returned to their caller for a retry. Other ops
	188	* can just be recycled here.
	189	*/
1182fca3	190	if (!op->uses_shared_memory)
1182fca3	191	goto retry_servicing;
1182fca3 MM	192	}
1182fca3 MM	193
05b39a8b	194	out:
1182fca3	195	gossip_debug(GOSSIP_WAIT_DEBUG,
9d9e7ba9 MM	196	"%s: %s returning: %d for %p.\n",
9d9e7ba9 MM	197	__func__,
1182fca3 MM	198	op_name,
	199	ret,
	200	op);
	201	return ret;
	202	}
	203
ca9f518e	204	/* This can get called on an I/O op if it had a bad service_operation. */
78699e29 AV	205	bool orangefs_cancel_op_in_progress(struct orangefs_kernel_op_s *op)
	206	{
	207	u64 tag = op->tag;
	208	if (!op_state_in_progress(op))
	209	return false;
	210
	211	op->slot_to_free = op->upcall.req.io.buf_index;
	212	memset(&op->upcall, 0, sizeof(op->upcall));
	213	memset(&op->downcall, 0, sizeof(op->downcall));
	214	op->upcall.type = ORANGEFS_VFS_OP_CANCEL;
	215	op->upcall.req.cancel.op_tag = tag;
	216	op->downcall.type = ORANGEFS_VFS_OP_INVALID;
	217	op->downcall.status = -1;
	218	orangefs_new_tag(op);
	219
	220	spin_lock(&orangefs_request_list_lock);
	221	/* orangefs_request_list_lock is enough of a barrier here */
	222	if (!__is_daemon_in_service()) {
	223	spin_unlock(&orangefs_request_list_lock);
	224	return false;
	225	}
98815ade AV	226	spin_lock(&op->lock);
98815ade AV	227	set_op_state_waiting(op);
9d9e7ba9 MM	228	gossip_debug(GOSSIP_DEV_DEBUG,
	229	"%s: op:%s: op_state:%d: process:%s:\n",
	230	__func__,
	231	get_opname_string(op),
	232	op->op_state,
	233	current->comm);
98815ade AV	234	list_add(&op->list, &orangefs_request_list);
98815ade AV	235	spin_unlock(&op->lock);
78699e29 AV	236	spin_unlock(&orangefs_request_list_lock);
78699e29 AV	237
ca9f518e	238	gossip_debug(GOSSIP_WAIT_DEBUG,
78699e29 AV	239	"Attempting ORANGEFS operation cancellation of tag %llu\n",
	240	llu(tag));
	241	return true;
	242	}
	243
ca9f518e MM	244	/*
	245	* Change an op to the "given up" state and remove it from its list.
	246	*/
	247	static void
	248	orangefs_clean_up_interrupted_operation(struct orangefs_kernel_op_s *op)
1182fca3 MM	249	{
	250	/*
	251	* handle interrupted cases depending on what state we were in when
ca9f518e	252	* the interruption is detected.
1182fca3	253	*
eab9b389	254	* Called with op->lock held.
1182fca3	255	*/
ca9f518e MM	256
	257	/*
	258	* List manipulation code elsewhere will ignore ops that
	259	* have been given up upon.
	260	*/
ed42fe05	261	op->op_state \|= OP_VFS_STATE_GIVEN_UP;
ca9f518e	262
05a50a5b AV	263	if (list_empty(&op->list)) {
	264	/* caught copying to/from daemon */
	265	BUG_ON(op_state_serviced(op));
	266	spin_unlock(&op->lock);
	267	wait_for_completion(&op->waitq);
	268	} else if (op_state_waiting(op)) {
1182fca3 MM	269	/*
	270	* upcall hasn't been read; remove op from upcall request
	271	* list.
	272	*/
	273	spin_unlock(&op->lock);
ed42fe05	274	spin_lock(&orangefs_request_list_lock);
05a50a5b	275	list_del_init(&op->list);
ed42fe05	276	spin_unlock(&orangefs_request_list_lock);
1182fca3 MM	277	gossip_debug(GOSSIP_WAIT_DEBUG,
	278	"Interrupted: Removed op %p from request_list\n",
	279	op);
	280	} else if (op_state_in_progress(op)) {
	281	/* op must be removed from the in progress htable */
	282	spin_unlock(&op->lock);
1d503617	283	spin_lock(&orangefs_htable_ops_in_progress_lock);
05a50a5b	284	list_del_init(&op->list);
1d503617	285	spin_unlock(&orangefs_htable_ops_in_progress_lock);
1182fca3 MM	286	gossip_debug(GOSSIP_WAIT_DEBUG,
	287	"Interrupted: Removed op %p"
	288	" from htable_ops_in_progress\n",
	289	op);
05a50a5b	290	} else {
1182fca3 MM	291	spin_unlock(&op->lock);
	292	gossip_err("interrupted operation is in a weird state 0x%x\n",
	293	op->op_state);
	294	}
d2d87a3b	295	reinit_completion(&op->waitq);
1182fca3 MM	296	}
	297
	298	/*
ca9f518e MM	299	* Sleeps on waitqueue waiting for matching downcall.
ca9f518e MM	300	* If client-core finishes servicing, then we are good to go.
1182fca3 MM	301	* else if client-core exits, we get woken up here, and retry with a timeout
1182fca3 MM	302	*
ca9f518e MM	303	* When this call returns to the caller, the specified op will no
ca9f518e MM	304	* longer be in either the in_progress hash table or on the request list.
1182fca3 MM	305	*
	306	* Returns 0 on success and -errno on failure
	307	* Errors are:
	308	* EAGAIN in case we want the caller to requeue and try again..
	309	* EINTR/EIO/ETIMEDOUT indicating we are done trying to service this
	310	* operation since client-core seems to be exiting too often
	311	* or if we were interrupted.
d2d87a3b AV	312	*
d2d87a3b AV	313	* Returns with op->lock taken.
1182fca3	314	*/
c72f15b7	315	static int wait_for_matching_downcall(struct orangefs_kernel_op_s *op,
05b39a8b	316	long timeout,
c72f15b7	317	bool interruptible)
1182fca3	318	{
05b39a8b	319	long n;
c72f15b7	320
ca9f518e MM	321	/*
	322	* There's a "schedule_timeout" inside of these wait
	323	* primitives, during which the op is out of the hands of the
	324	* user process that needs something done and is being
	325	* manipulated by the client-core process.
	326	*/
c72f15b7	327	if (interruptible)
adcf34a2 MM	328	n = wait_for_completion_interruptible_timeout(&op->waitq,
adcf34a2 MM	329	timeout);
c72f15b7 AV	330	else
	331	n = wait_for_completion_killable_timeout(&op->waitq, timeout);
	332
d2d87a3b	333	spin_lock(&op->lock);
1182fca3	334
d2d87a3b AV	335	if (op_state_serviced(op))
d2d87a3b AV	336	return 0;
70c6ea26	337
d2d87a3b AV	338	if (unlikely(n < 0)) {
d2d87a3b AV	339	gossip_debug(GOSSIP_WAIT_DEBUG,
9d9e7ba9	340	"%s: operation interrupted, tag %llu, %p\n",
d2d87a3b AV	341	__func__,
	342	llu(op->tag),
	343	op);
	344	return -EINTR;
1182fca3	345	}
d2d87a3b AV	346	if (op_state_purged(op)) {
d2d87a3b AV	347	gossip_debug(GOSSIP_WAIT_DEBUG,
9d9e7ba9	348	"%s: operation purged, tag %llu, %p, %d\n",
d2d87a3b AV	349	__func__,
	350	llu(op->tag),
	351	op,
	352	op->attempts);
	353	return (op->attempts < ORANGEFS_PURGE_RETRY_COUNT) ?
	354	-EAGAIN :
	355	-EIO;
	356	}
	357	/* must have timed out, then... */
	358	gossip_debug(GOSSIP_WAIT_DEBUG,
9d9e7ba9	359	"%s: operation timed out, tag %llu, %p, %d)\n",
d2d87a3b AV	360	__func__,
	361	llu(op->tag),
	362	op,
	363	op->attempts);
	364	return -ETIMEDOUT;
1182fca3	365	}