[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 *, 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);
	}
	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;
	/* 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,
		     "%s: %s op:%p: process:%s: pid:%d:\n",
		     __func__,
		     op_name,
		     op,
		     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()) {
		gossip_debug(GOSSIP_WAIT_DEBUG,
			     "%s:client core is NOT in service.\n",
			     __func__);
		timeout = op_timeout_secs * HZ;
	}
	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, timeout,
					 flags & ORANGEFS_OP_INTERRUPTIBLE);

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

	if (!ret) {
		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;
		goto out;
	}

	/* 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;
	/* 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);

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

out:
	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,
				      long timeout,
				      bool interruptible)
{
	long n;

	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;
	}
	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
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);
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	{
05b39a8b	58	long timeout = MAX_SCHEDULE_TIMEOUT;
1182fca3	59	/* flags to modify behavior */
1182fca3 MM	60	int ret = 0;
1182fca3 MM	61
ce6c414e	62	DEFINE_WAIT(wait_entry);
1182fca3 MM	63
	64	op->upcall.tgid = current->tgid;
	65	op->upcall.pid = current->pid;
	66
	67	retry_servicing:
	68	op->downcall.status = 0;
	69	gossip_debug(GOSSIP_WAIT_DEBUG,
5253487e MM	70	"%s: %s op:%p: process:%s: pid:%d:\n",
5253487e MM	71	__func__,
1182fca3	72	op_name,
5253487e	73	op,
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	105	gossip_debug(GOSSIP_WAIT_DEBUG,
98815ade	106	"%s:client core is NOT in service.\n",
1182fca3	107	__func__);
05b39a8b	108	timeout = op_timeout_secs * HZ;
1182fca3	109	}
98815ade	110	spin_unlock(&orangefs_request_list_lock);
1182fca3	111
8bb8aefd	112	if (!(flags & ORANGEFS_OP_NO_SEMAPHORE))
1182fca3 MM	113	mutex_unlock(&request_mutex);
	114
	115	/*
	116	* If we are asked to service an asynchronous operation from
	117	* VFS perspective, we are done.
	118	*/
8bb8aefd	119	if (flags & ORANGEFS_OP_ASYNC)
1182fca3 MM	120	return 0;
1182fca3 MM	121
05b39a8b AV	122	ret = wait_for_matching_downcall(op, timeout,
05b39a8b AV	123	flags & ORANGEFS_OP_INTERRUPTIBLE);
5253487e MM	124
	125	gossip_debug(GOSSIP_WAIT_DEBUG,
	126	"%s: wait_for_matching_downcall returned %d for %p\n",
	127	__func__,
	128	ret,
	129	op);
	130
05b39a8b	131	if (!ret) {
d2d87a3b	132	spin_unlock(&op->lock);
1182fca3 MM	133	/* got matching downcall; make sure status is in errno format */
1182fca3 MM	134	op->downcall.status =
8bb8aefd	135	orangefs_normalize_to_errno(op->downcall.status);
1182fca3	136	ret = op->downcall.status;
05b39a8b	137	goto out;
1182fca3 MM	138	}
1182fca3 MM	139
05b39a8b AV	140	/* failed to get matching downcall */
	141	if (ret == -ETIMEDOUT) {
	142	gossip_err("orangefs: %s -- wait timed out; aborting attempt.\n",
	143	op_name);
	144	}
	145	orangefs_clean_up_interrupted_operation(op);
	146	op->downcall.status = ret;
1182fca3	147	/* retry if operation has not been serviced and if requested */
05b39a8b AV	148	if (ret == -EAGAIN) {
	149	op->attempts++;
	150	timeout = op_timeout_secs * HZ;
1182fca3	151	gossip_debug(GOSSIP_WAIT_DEBUG,
8bb8aefd	152	"orangefs: tag %llu (%s)"
1182fca3 MM	153	" -- operation to be retried (%d attempt)\n",
	154	llu(op->tag),
	155	op_name,
05b39a8b	156	op->attempts);
1182fca3 MM	157
	158	if (!op->uses_shared_memory)
	159	/*
	160	* this operation doesn't use the shared memory
	161	* system
	162	*/
	163	goto retry_servicing;
1182fca3 MM	164	}
1182fca3 MM	165
05b39a8b	166	out:
1182fca3	167	gossip_debug(GOSSIP_WAIT_DEBUG,
8bb8aefd	168	"orangefs: service_operation %s returning: %d for %p.\n",
1182fca3 MM	169	op_name,
	170	ret,
	171	op);
	172	return ret;
	173	}
	174
78699e29 AV	175	bool orangefs_cancel_op_in_progress(struct orangefs_kernel_op_s *op)
	176	{
	177	u64 tag = op->tag;
	178	if (!op_state_in_progress(op))
	179	return false;
	180
	181	op->slot_to_free = op->upcall.req.io.buf_index;
	182	memset(&op->upcall, 0, sizeof(op->upcall));
	183	memset(&op->downcall, 0, sizeof(op->downcall));
	184	op->upcall.type = ORANGEFS_VFS_OP_CANCEL;
	185	op->upcall.req.cancel.op_tag = tag;
	186	op->downcall.type = ORANGEFS_VFS_OP_INVALID;
	187	op->downcall.status = -1;
	188	orangefs_new_tag(op);
	189
	190	spin_lock(&orangefs_request_list_lock);
	191	/* orangefs_request_list_lock is enough of a barrier here */
	192	if (!__is_daemon_in_service()) {
	193	spin_unlock(&orangefs_request_list_lock);
	194	return false;
	195	}
98815ade AV	196	spin_lock(&op->lock);
	197	set_op_state_waiting(op);
	198	list_add(&op->list, &orangefs_request_list);
	199	spin_unlock(&op->lock);
78699e29 AV	200	spin_unlock(&orangefs_request_list_lock);
	201
	202	gossip_debug(GOSSIP_UTILS_DEBUG,
	203	"Attempting ORANGEFS operation cancellation of tag %llu\n",
	204	llu(tag));
	205	return true;
	206	}
	207
e07db0a2	208	static void orangefs_clean_up_interrupted_operation(struct orangefs_kernel_op_s *op)
1182fca3 MM	209	{
	210	/*
	211	* handle interrupted cases depending on what state we were in when
	212	* the interruption is detected. there is a coarse grained lock
	213	* across the operation.
	214	*
eab9b389	215	* Called with op->lock held.
1182fca3	216	*/
ed42fe05	217	op->op_state \|= OP_VFS_STATE_GIVEN_UP;
1182fca3 MM	218
	219	if (op_state_waiting(op)) {
	220	/*
	221	* upcall hasn't been read; remove op from upcall request
	222	* list.
	223	*/
	224	spin_unlock(&op->lock);
ed42fe05 AV	225	spin_lock(&orangefs_request_list_lock);
	226	list_del(&op->list);
	227	spin_unlock(&orangefs_request_list_lock);
1182fca3 MM	228	gossip_debug(GOSSIP_WAIT_DEBUG,
	229	"Interrupted: Removed op %p from request_list\n",
	230	op);
	231	} else if (op_state_in_progress(op)) {
	232	/* op must be removed from the in progress htable */
	233	spin_unlock(&op->lock);
	234	spin_lock(&htable_ops_in_progress_lock);
	235	list_del(&op->list);
	236	spin_unlock(&htable_ops_in_progress_lock);
	237	gossip_debug(GOSSIP_WAIT_DEBUG,
	238	"Interrupted: Removed op %p"
	239	" from htable_ops_in_progress\n",
	240	op);
	241	} else if (!op_state_serviced(op)) {
	242	spin_unlock(&op->lock);
	243	gossip_err("interrupted operation is in a weird state 0x%x\n",
	244	op->op_state);
84d02150 MM	245	} else {
	246	/*
	247	* It is not intended for execution to flow here,
	248	* but having this unlock here makes sparse happy.
	249	*/
	250	gossip_err("%s: can't get here.\n", __func__);
	251	spin_unlock(&op->lock);
1182fca3	252	}
d2d87a3b	253	reinit_completion(&op->waitq);
1182fca3 MM	254	}
	255
	256	/*
	257	* sleeps on waitqueue waiting for matching downcall.
	258	* if client-core finishes servicing, then we are good to go.
	259	* else if client-core exits, we get woken up here, and retry with a timeout
	260	*
	261	* Post when this call returns to the caller, the specified op will no
	262	* longer be on any list or htable.
	263	*
	264	* Returns 0 on success and -errno on failure
	265	* Errors are:
	266	* EAGAIN in case we want the caller to requeue and try again..
	267	* EINTR/EIO/ETIMEDOUT indicating we are done trying to service this
	268	* operation since client-core seems to be exiting too often
	269	* or if we were interrupted.
d2d87a3b AV	270	*
d2d87a3b AV	271	* Returns with op->lock taken.
1182fca3	272	*/
c72f15b7	273	static int wait_for_matching_downcall(struct orangefs_kernel_op_s *op,
05b39a8b	274	long timeout,
c72f15b7	275	bool interruptible)
1182fca3	276	{
05b39a8b	277	long n;
c72f15b7 AV	278
	279	if (interruptible)
	280	n = wait_for_completion_interruptible_timeout(&op->waitq, timeout);
	281	else
	282	n = wait_for_completion_killable_timeout(&op->waitq, timeout);
	283
d2d87a3b	284	spin_lock(&op->lock);
1182fca3	285
d2d87a3b AV	286	if (op_state_serviced(op))
d2d87a3b AV	287	return 0;
70c6ea26	288
d2d87a3b AV	289	if (unlikely(n < 0)) {
	290	gossip_debug(GOSSIP_WAIT_DEBUG,
	291	"*** %s:"
	292	" operation interrupted by a signal (tag "
	293	"%llu, op %p)\n",
	294	__func__,
	295	llu(op->tag),
	296	op);
	297	return -EINTR;
1182fca3	298	}
d2d87a3b AV	299	if (op_state_purged(op)) {
	300	gossip_debug(GOSSIP_WAIT_DEBUG,
	301	"*** %s:"
	302	" operation purged (tag "
	303	"%llu, %p, att %d)\n",
	304	__func__,
	305	llu(op->tag),
	306	op,
	307	op->attempts);
	308	return (op->attempts < ORANGEFS_PURGE_RETRY_COUNT) ?
	309	-EAGAIN :
	310	-EIO;
	311	}
	312	/* must have timed out, then... */
	313	gossip_debug(GOSSIP_WAIT_DEBUG,
	314	"*** %s:"
	315	" operation timed out (tag"
	316	" %llu, %p, att %d)\n",
	317	__func__,
	318	llu(op->tag),
	319	op,
	320	op->attempts);
	321	return -ETIMEDOUT;
1182fca3	322	}