[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 *op,
		long timeout,
		int flags)
			__acquires(op->lock);
static void orangefs_clean_up_interrupted_operation(struct orangefs_kernel_op_s *op)
	__releases(op->lock);

/*
 * 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, *tmp;

	spin_lock(&orangefs_request_list_lock);
	list_for_each_entry_safe(op, tmp, &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);
	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)
		__releases(op->lock)
{
	/*
	 * 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,
		int flags)
			__acquires(op->lock)
{
	long n;
	int writeback = flags & ORANGEFS_OP_WRITEBACK,
	    interruptible = flags & ORANGEFS_OP_INTERRUPTIBLE;

	/*
	 * 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 (writeback)
		n = wait_for_completion_io_timeout(&op->waitq, timeout);
	else if (!writeback && interruptible)
		n = wait_for_completion_interruptible_timeout(&op->waitq,
								      timeout);
	else /* !writeback && !interruptible but compiler complains */
		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
b1116bc0 MM	20	static int wait_for_matching_downcall(struct orangefs_kernel_op_s *op,
b1116bc0 MM	21	long timeout,
0dcac0f7	22	int flags)
b1116bc0 MM	23	__acquires(op->lock);
	24	static void orangefs_clean_up_interrupted_operation(struct orangefs_kernel_op_s *op)
	25	__releases(op->lock);
ade3d781	26
1182fca3 MM	27	/*
	28	* What we do in this function is to walk the list of operations that are
	29	* present in the request queue and mark them as purged.
	30	* NOTE: This is called from the device close after client-core has
	31	* guaranteed that no new operations could appear on the list since the
	32	* client-core is anyway going to exit.
	33	*/
	34	void purge_waiting_ops(void)
	35	{
0afc0dec	36	struct orangefs_kernel_op_s op, tmp;
1182fca3	37
8bb8aefd	38	spin_lock(&orangefs_request_list_lock);
0afc0dec	39	list_for_each_entry_safe(op, tmp, &orangefs_request_list, list) {
1182fca3 MM	40	gossip_debug(GOSSIP_WAIT_DEBUG,
	41	"pvfs2-client-core: purging op tag %llu %s\n",
	42	llu(op->tag),
	43	get_opname_string(op));
1182fca3	44	set_op_state_purged(op);
9d9e7ba9 MM	45	gossip_debug(GOSSIP_DEV_DEBUG,
	46	"%s: op:%s: op_state:%d: process:%s:\n",
	47	__func__,
	48	get_opname_string(op),
	49	op->op_state,
	50	current->comm);
1182fca3	51	}
8bb8aefd	52	spin_unlock(&orangefs_request_list_lock);
1182fca3 MM	53	}
	54
	55	/*
8bb8aefd	56	* submits a ORANGEFS operation and waits for it to complete
1182fca3 MM	57	*
	58	* Note op->downcall.status will contain the status of the operation (in
	59	* errno format), whether provided by pvfs2-client or a result of failure to
	60	* service the operation. If the caller wishes to distinguish, then
	61	* op->state can be checked to see if it was serviced or not.
	62	*
	63	* Returns contents of op->downcall.status for convenience
	64	*/
8bb8aefd	65	int service_operation(struct orangefs_kernel_op_s *op,
1182fca3 MM	66	const char *op_name,
	67	int flags)
	68	{
05b39a8b	69	long timeout = MAX_SCHEDULE_TIMEOUT;
1182fca3 MM	70	int ret = 0;
1182fca3 MM	71
ce6c414e	72	DEFINE_WAIT(wait_entry);
1182fca3 MM	73
	74	op->upcall.tgid = current->tgid;
	75	op->upcall.pid = current->pid;
	76
	77	retry_servicing:
	78	op->downcall.status = 0;
	79	gossip_debug(GOSSIP_WAIT_DEBUG,
5253487e MM	80	"%s: %s op:%p: process:%s: pid:%d:\n",
5253487e MM	81	__func__,
1182fca3	82	op_name,
5253487e	83	op,
1182fca3 MM	84	current->comm,
	85	current->pid);
	86
adcf34a2 MM	87	/*
adcf34a2 MM	88	* If ORANGEFS_OP_NO_MUTEX was set in flags, we need to avoid
ca9f518e	89	* acquiring the request_mutex because we're servicing a
adcf34a2 MM	90	* high priority remount operation and the request_mutex is
	91	* already taken.
	92	*/
	93	if (!(flags & ORANGEFS_OP_NO_MUTEX)) {
c72f15b7	94	if (flags & ORANGEFS_OP_INTERRUPTIBLE)
1d503617	95	ret = mutex_lock_interruptible(&orangefs_request_mutex);
c72f15b7	96	else
1d503617	97	ret = mutex_lock_killable(&orangefs_request_mutex);
1182fca3 MM	98	/*
1182fca3 MM	99	* check to see if we were interrupted while waiting for
adcf34a2	100	* mutex
1182fca3 MM	101	*/
1182fca3 MM	102	if (ret < 0) {
1182fca3 MM	103	op->downcall.status = ret;
1182fca3 MM	104	gossip_debug(GOSSIP_WAIT_DEBUG,
ca9f518e MM	105	"%s: service_operation interrupted.\n",
ca9f518e MM	106	__func__);
1182fca3 MM	107	return ret;
	108	}
	109	}
	110
98815ade AV	111	/* queue up the operation */
	112	spin_lock(&orangefs_request_list_lock);
	113	spin_lock(&op->lock);
	114	set_op_state_waiting(op);
9d9e7ba9 MM	115	gossip_debug(GOSSIP_DEV_DEBUG,
	116	"%s: op:%s: op_state:%d: process:%s:\n",
	117	__func__,
	118	get_opname_string(op),
	119	op->op_state,
	120	current->comm);
adcf34a2	121	/* add high priority remount op to the front of the line. */
98815ade AV	122	if (flags & ORANGEFS_OP_PRIORITY)
	123	list_add(&op->list, &orangefs_request_list);
	124	else
	125	list_add_tail(&op->list, &orangefs_request_list);
	126	spin_unlock(&op->lock);
	127	wake_up_interruptible(&orangefs_request_list_waitq);
	128	if (!__is_daemon_in_service()) {
1182fca3	129	gossip_debug(GOSSIP_WAIT_DEBUG,
98815ade	130	"%s:client core is NOT in service.\n",
1182fca3	131	__func__);
b5a9d61e MB	132	/*
	133	* Don't wait for the userspace component to return if
	134	* the filesystem is being umounted anyway.
	135	*/
	136	if (op->upcall.type == ORANGEFS_VFS_OP_FS_UMOUNT)
	137	timeout = 0;
	138	else
	139	timeout = op_timeout_secs * HZ;
1182fca3	140	}
98815ade	141	spin_unlock(&orangefs_request_list_lock);
1182fca3	142
adcf34a2	143	if (!(flags & ORANGEFS_OP_NO_MUTEX))
1d503617	144	mutex_unlock(&orangefs_request_mutex);
1182fca3	145
0dcac0f7	146	ret = wait_for_matching_downcall(op, timeout, flags);
5253487e MM	147	gossip_debug(GOSSIP_WAIT_DEBUG,
	148	"%s: wait_for_matching_downcall returned %d for %p\n",
	149	__func__,
	150	ret,
	151	op);
	152
ca9f518e	153	/* got matching downcall; make sure status is in errno format */
05b39a8b	154	if (!ret) {
d2d87a3b	155	spin_unlock(&op->lock);
1182fca3	156	op->downcall.status =
8bb8aefd	157	orangefs_normalize_to_errno(op->downcall.status);
1182fca3	158	ret = op->downcall.status;
05b39a8b	159	goto out;
1182fca3 MM	160	}
1182fca3 MM	161
05b39a8b AV	162	/* failed to get matching downcall */
05b39a8b AV	163	if (ret == -ETIMEDOUT) {
adcf34a2 MM	164	gossip_err("%s: %s -- wait timed out; aborting attempt.\n",
adcf34a2 MM	165	__func__,
05b39a8b AV	166	op_name);
05b39a8b AV	167	}
adcf34a2 MM	168
adcf34a2 MM	169	/*
ca9f518e MM	170	* remove a waiting op from the request list or
ca9f518e MM	171	* remove an in-progress op from the in-progress list.
adcf34a2	172	*/
05b39a8b	173	orangefs_clean_up_interrupted_operation(op);
adcf34a2	174
05b39a8b	175	op->downcall.status = ret;
1182fca3	176	/* retry if operation has not been serviced and if requested */
05b39a8b AV	177	if (ret == -EAGAIN) {
	178	op->attempts++;
	179	timeout = op_timeout_secs * HZ;
1182fca3	180	gossip_debug(GOSSIP_WAIT_DEBUG,
8bb8aefd	181	"orangefs: tag %llu (%s)"
1182fca3 MM	182	" -- operation to be retried (%d attempt)\n",
	183	llu(op->tag),
	184	op_name,
05b39a8b	185	op->attempts);
1182fca3	186
adcf34a2 MM	187	/*
	188	* io ops (ops that use the shared memory buffer) have
	189	* to be returned to their caller for a retry. Other ops
	190	* can just be recycled here.
	191	*/
1182fca3	192	if (!op->uses_shared_memory)
1182fca3	193	goto retry_servicing;
1182fca3 MM	194	}
1182fca3 MM	195
05b39a8b	196	out:
1182fca3	197	gossip_debug(GOSSIP_WAIT_DEBUG,
9d9e7ba9 MM	198	"%s: %s returning: %d for %p.\n",
9d9e7ba9 MM	199	__func__,
1182fca3 MM	200	op_name,
	201	ret,
	202	op);
	203	return ret;
	204	}
	205
ca9f518e	206	/* This can get called on an I/O op if it had a bad service_operation. */
78699e29 AV	207	bool orangefs_cancel_op_in_progress(struct orangefs_kernel_op_s *op)
	208	{
	209	u64 tag = op->tag;
	210	if (!op_state_in_progress(op))
	211	return false;
	212
	213	op->slot_to_free = op->upcall.req.io.buf_index;
	214	memset(&op->upcall, 0, sizeof(op->upcall));
	215	memset(&op->downcall, 0, sizeof(op->downcall));
	216	op->upcall.type = ORANGEFS_VFS_OP_CANCEL;
	217	op->upcall.req.cancel.op_tag = tag;
	218	op->downcall.type = ORANGEFS_VFS_OP_INVALID;
	219	op->downcall.status = -1;
	220	orangefs_new_tag(op);
	221
	222	spin_lock(&orangefs_request_list_lock);
	223	/* orangefs_request_list_lock is enough of a barrier here */
	224	if (!__is_daemon_in_service()) {
	225	spin_unlock(&orangefs_request_list_lock);
	226	return false;
	227	}
98815ade AV	228	spin_lock(&op->lock);
98815ade AV	229	set_op_state_waiting(op);
9d9e7ba9 MM	230	gossip_debug(GOSSIP_DEV_DEBUG,
	231	"%s: op:%s: op_state:%d: process:%s:\n",
	232	__func__,
	233	get_opname_string(op),
	234	op->op_state,
	235	current->comm);
98815ade AV	236	list_add(&op->list, &orangefs_request_list);
98815ade AV	237	spin_unlock(&op->lock);
78699e29 AV	238	spin_unlock(&orangefs_request_list_lock);
78699e29 AV	239
ca9f518e	240	gossip_debug(GOSSIP_WAIT_DEBUG,
78699e29 AV	241	"Attempting ORANGEFS operation cancellation of tag %llu\n",
	242	llu(tag));
	243	return true;
	244	}
	245
ca9f518e MM	246	/*
	247	* Change an op to the "given up" state and remove it from its list.
	248	*/
	249	static void
	250	orangefs_clean_up_interrupted_operation(struct orangefs_kernel_op_s *op)
b1116bc0	251	__releases(op->lock)
1182fca3 MM	252	{
	253	/*
	254	* handle interrupted cases depending on what state we were in when
ca9f518e	255	* the interruption is detected.
1182fca3	256	*
eab9b389	257	* Called with op->lock held.
1182fca3	258	*/
ca9f518e MM	259
	260	/*
	261	* List manipulation code elsewhere will ignore ops that
	262	* have been given up upon.
	263	*/
ed42fe05	264	op->op_state \|= OP_VFS_STATE_GIVEN_UP;
ca9f518e	265
05a50a5b AV	266	if (list_empty(&op->list)) {
	267	/* caught copying to/from daemon */
	268	BUG_ON(op_state_serviced(op));
	269	spin_unlock(&op->lock);
	270	wait_for_completion(&op->waitq);
	271	} else if (op_state_waiting(op)) {
1182fca3 MM	272	/*
	273	* upcall hasn't been read; remove op from upcall request
	274	* list.
	275	*/
	276	spin_unlock(&op->lock);
ed42fe05	277	spin_lock(&orangefs_request_list_lock);
05a50a5b	278	list_del_init(&op->list);
ed42fe05	279	spin_unlock(&orangefs_request_list_lock);
1182fca3 MM	280	gossip_debug(GOSSIP_WAIT_DEBUG,
	281	"Interrupted: Removed op %p from request_list\n",
	282	op);
	283	} else if (op_state_in_progress(op)) {
	284	/* op must be removed from the in progress htable */
	285	spin_unlock(&op->lock);
1d503617	286	spin_lock(&orangefs_htable_ops_in_progress_lock);
05a50a5b	287	list_del_init(&op->list);
1d503617	288	spin_unlock(&orangefs_htable_ops_in_progress_lock);
1182fca3 MM	289	gossip_debug(GOSSIP_WAIT_DEBUG,
	290	"Interrupted: Removed op %p"
	291	" from htable_ops_in_progress\n",
	292	op);
05a50a5b	293	} else {
1182fca3 MM	294	spin_unlock(&op->lock);
	295	gossip_err("interrupted operation is in a weird state 0x%x\n",
	296	op->op_state);
	297	}
d2d87a3b	298	reinit_completion(&op->waitq);
1182fca3 MM	299	}
	300
	301	/*
ca9f518e MM	302	* Sleeps on waitqueue waiting for matching downcall.
ca9f518e MM	303	* If client-core finishes servicing, then we are good to go.
1182fca3 MM	304	* else if client-core exits, we get woken up here, and retry with a timeout
1182fca3 MM	305	*
ca9f518e MM	306	* When this call returns to the caller, the specified op will no
ca9f518e MM	307	* longer be in either the in_progress hash table or on the request list.
1182fca3 MM	308	*
	309	* Returns 0 on success and -errno on failure
	310	* Errors are:
	311	* EAGAIN in case we want the caller to requeue and try again..
	312	* EINTR/EIO/ETIMEDOUT indicating we are done trying to service this
	313	* operation since client-core seems to be exiting too often
	314	* or if we were interrupted.
d2d87a3b AV	315	*
d2d87a3b AV	316	* Returns with op->lock taken.
1182fca3	317	*/
c72f15b7	318	static int wait_for_matching_downcall(struct orangefs_kernel_op_s *op,
b1116bc0	319	long timeout,
0dcac0f7	320	int flags)
b1116bc0	321	__acquires(op->lock)
1182fca3	322	{
05b39a8b	323	long n;
0dcac0f7 MB	324	int writeback = flags & ORANGEFS_OP_WRITEBACK,
0dcac0f7 MB	325	interruptible = flags & ORANGEFS_OP_INTERRUPTIBLE;
c72f15b7	326
ca9f518e MM	327	/*
	328	* There's a "schedule_timeout" inside of these wait
	329	* primitives, during which the op is out of the hands of the
	330	* user process that needs something done and is being
	331	* manipulated by the client-core process.
	332	*/
0dcac0f7 MB	333	if (writeback)
	334	n = wait_for_completion_io_timeout(&op->waitq, timeout);
	335	else if (!writeback && interruptible)
adcf34a2	336	n = wait_for_completion_interruptible_timeout(&op->waitq,
0dcac0f7 MB	337	timeout);
0dcac0f7 MB	338	else /* !writeback && !interruptible but compiler complains */
c72f15b7 AV	339	n = wait_for_completion_killable_timeout(&op->waitq, timeout);
c72f15b7 AV	340
d2d87a3b	341	spin_lock(&op->lock);
1182fca3	342
d2d87a3b AV	343	if (op_state_serviced(op))
d2d87a3b AV	344	return 0;
70c6ea26	345
d2d87a3b AV	346	if (unlikely(n < 0)) {
d2d87a3b AV	347	gossip_debug(GOSSIP_WAIT_DEBUG,
9d9e7ba9	348	"%s: operation interrupted, tag %llu, %p\n",
d2d87a3b AV	349	__func__,
	350	llu(op->tag),
	351	op);
	352	return -EINTR;
1182fca3	353	}
d2d87a3b AV	354	if (op_state_purged(op)) {
d2d87a3b AV	355	gossip_debug(GOSSIP_WAIT_DEBUG,
9d9e7ba9	356	"%s: operation purged, tag %llu, %p, %d\n",
d2d87a3b AV	357	__func__,
	358	llu(op->tag),
	359	op,
	360	op->attempts);
	361	return (op->attempts < ORANGEFS_PURGE_RETRY_COUNT) ?
	362	-EAGAIN :
	363	-EIO;
	364	}
	365	/* must have timed out, then... */
	366	gossip_debug(GOSSIP_WAIT_DEBUG,
9d9e7ba9	367	"%s: operation timed out, tag %llu, %p, %d)\n",
d2d87a3b AV	368	__func__,
	369	llu(op->tag),
	370	op,
	371	op->attempts);
	372	return -ETIMEDOUT;
1182fca3	373	}