[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,
		     "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()) {
		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);
	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,
8bb8aefd	70	"orangefs: service_operation: %s %p\n",
1182fca3 MM	71	op_name,
	72	op);
	73	gossip_debug(GOSSIP_WAIT_DEBUG,
8bb8aefd	74	"orangefs: operation posted by process: %s, pid: %i\n",
1182fca3 MM	75	current->comm,
	76	current->pid);
	77
8bb8aefd	78	if (!(flags & ORANGEFS_OP_NO_SEMAPHORE)) {
c72f15b7 AV	79	if (flags & ORANGEFS_OP_INTERRUPTIBLE)
	80	ret = mutex_lock_interruptible(&request_mutex);
	81	else
	82	ret = mutex_lock_killable(&request_mutex);
1182fca3 MM	83	/*
	84	* check to see if we were interrupted while waiting for
	85	* semaphore
	86	*/
	87	if (ret < 0) {
1182fca3 MM	88	op->downcall.status = ret;
1182fca3 MM	89	gossip_debug(GOSSIP_WAIT_DEBUG,
8bb8aefd	90	"orangefs: service_operation interrupted.\n");
1182fca3 MM	91	return ret;
	92	}
	93	}
	94
98815ade AV	95	/* queue up the operation */
	96	spin_lock(&orangefs_request_list_lock);
	97	spin_lock(&op->lock);
	98	set_op_state_waiting(op);
	99	if (flags & ORANGEFS_OP_PRIORITY)
	100	list_add(&op->list, &orangefs_request_list);
	101	else
	102	list_add_tail(&op->list, &orangefs_request_list);
	103	spin_unlock(&op->lock);
	104	wake_up_interruptible(&orangefs_request_list_waitq);
	105	if (!__is_daemon_in_service()) {
1182fca3	106	gossip_debug(GOSSIP_WAIT_DEBUG,
98815ade	107	"%s:client core is NOT in service.\n",
1182fca3	108	__func__);
05b39a8b	109	timeout = op_timeout_secs * HZ;
1182fca3	110	}
98815ade	111	spin_unlock(&orangefs_request_list_lock);
1182fca3	112
8bb8aefd	113	if (!(flags & ORANGEFS_OP_NO_SEMAPHORE))
1182fca3 MM	114	mutex_unlock(&request_mutex);
	115
	116	/*
	117	* If we are asked to service an asynchronous operation from
	118	* VFS perspective, we are done.
	119	*/
8bb8aefd	120	if (flags & ORANGEFS_OP_ASYNC)
1182fca3 MM	121	return 0;
1182fca3 MM	122
05b39a8b AV	123	ret = wait_for_matching_downcall(op, timeout,
	124	flags & ORANGEFS_OP_INTERRUPTIBLE);
	125	if (!ret) {
d2d87a3b	126	spin_unlock(&op->lock);
1182fca3 MM	127	/* got matching downcall; make sure status is in errno format */
1182fca3 MM	128	op->downcall.status =
8bb8aefd	129	orangefs_normalize_to_errno(op->downcall.status);
1182fca3	130	ret = op->downcall.status;
05b39a8b	131	goto out;
1182fca3 MM	132	}
1182fca3 MM	133
05b39a8b AV	134	/* failed to get matching downcall */
	135	if (ret == -ETIMEDOUT) {
	136	gossip_err("orangefs: %s -- wait timed out; aborting attempt.\n",
	137	op_name);
	138	}
	139	orangefs_clean_up_interrupted_operation(op);
	140	op->downcall.status = ret;
1182fca3	141	/* retry if operation has not been serviced and if requested */
05b39a8b AV	142	if (ret == -EAGAIN) {
	143	op->attempts++;
	144	timeout = op_timeout_secs * HZ;
1182fca3	145	gossip_debug(GOSSIP_WAIT_DEBUG,
8bb8aefd	146	"orangefs: tag %llu (%s)"
1182fca3 MM	147	" -- operation to be retried (%d attempt)\n",
	148	llu(op->tag),
	149	op_name,
05b39a8b	150	op->attempts);
1182fca3 MM	151
	152	if (!op->uses_shared_memory)
	153	/*
	154	* this operation doesn't use the shared memory
	155	* system
	156	*/
	157	goto retry_servicing;
1182fca3 MM	158	}
1182fca3 MM	159
05b39a8b	160	out:
1182fca3	161	gossip_debug(GOSSIP_WAIT_DEBUG,
8bb8aefd	162	"orangefs: service_operation %s returning: %d for %p.\n",
1182fca3 MM	163	op_name,
	164	ret,
	165	op);
	166	return ret;
	167	}
	168
78699e29 AV	169	bool orangefs_cancel_op_in_progress(struct orangefs_kernel_op_s *op)
	170	{
	171	u64 tag = op->tag;
	172	if (!op_state_in_progress(op))
	173	return false;
	174
	175	op->slot_to_free = op->upcall.req.io.buf_index;
	176	memset(&op->upcall, 0, sizeof(op->upcall));
	177	memset(&op->downcall, 0, sizeof(op->downcall));
	178	op->upcall.type = ORANGEFS_VFS_OP_CANCEL;
	179	op->upcall.req.cancel.op_tag = tag;
	180	op->downcall.type = ORANGEFS_VFS_OP_INVALID;
	181	op->downcall.status = -1;
	182	orangefs_new_tag(op);
	183
	184	spin_lock(&orangefs_request_list_lock);
	185	/* orangefs_request_list_lock is enough of a barrier here */
	186	if (!__is_daemon_in_service()) {
	187	spin_unlock(&orangefs_request_list_lock);
	188	return false;
	189	}
98815ade AV	190	spin_lock(&op->lock);
	191	set_op_state_waiting(op);
	192	list_add(&op->list, &orangefs_request_list);
	193	spin_unlock(&op->lock);
78699e29 AV	194	spin_unlock(&orangefs_request_list_lock);
	195
	196	gossip_debug(GOSSIP_UTILS_DEBUG,
	197	"Attempting ORANGEFS operation cancellation of tag %llu\n",
	198	llu(tag));
	199	return true;
	200	}
	201
e07db0a2	202	static void orangefs_clean_up_interrupted_operation(struct orangefs_kernel_op_s *op)
1182fca3 MM	203	{
	204	/*
	205	* handle interrupted cases depending on what state we were in when
	206	* the interruption is detected. there is a coarse grained lock
	207	* across the operation.
	208	*
eab9b389	209	* Called with op->lock held.
1182fca3	210	*/
ed42fe05	211	op->op_state \|= OP_VFS_STATE_GIVEN_UP;
1182fca3 MM	212
	213	if (op_state_waiting(op)) {
	214	/*
	215	* upcall hasn't been read; remove op from upcall request
	216	* list.
	217	*/
	218	spin_unlock(&op->lock);
ed42fe05 AV	219	spin_lock(&orangefs_request_list_lock);
	220	list_del(&op->list);
	221	spin_unlock(&orangefs_request_list_lock);
1182fca3 MM	222	gossip_debug(GOSSIP_WAIT_DEBUG,
	223	"Interrupted: Removed op %p from request_list\n",
	224	op);
	225	} else if (op_state_in_progress(op)) {
	226	/* op must be removed from the in progress htable */
	227	spin_unlock(&op->lock);
	228	spin_lock(&htable_ops_in_progress_lock);
	229	list_del(&op->list);
	230	spin_unlock(&htable_ops_in_progress_lock);
	231	gossip_debug(GOSSIP_WAIT_DEBUG,
	232	"Interrupted: Removed op %p"
	233	" from htable_ops_in_progress\n",
	234	op);
	235	} else if (!op_state_serviced(op)) {
	236	spin_unlock(&op->lock);
	237	gossip_err("interrupted operation is in a weird state 0x%x\n",
	238	op->op_state);
84d02150 MM	239	} else {
	240	/*
	241	* It is not intended for execution to flow here,
	242	* but having this unlock here makes sparse happy.
	243	*/
	244	gossip_err("%s: can't get here.\n", __func__);
	245	spin_unlock(&op->lock);
1182fca3	246	}
d2d87a3b	247	reinit_completion(&op->waitq);
1182fca3 MM	248	}
	249
	250	/*
	251	* sleeps on waitqueue waiting for matching downcall.
	252	* if client-core finishes servicing, then we are good to go.
	253	* else if client-core exits, we get woken up here, and retry with a timeout
	254	*
	255	* Post when this call returns to the caller, the specified op will no
	256	* longer be on any list or htable.
	257	*
	258	* Returns 0 on success and -errno on failure
	259	* Errors are:
	260	* EAGAIN in case we want the caller to requeue and try again..
	261	* EINTR/EIO/ETIMEDOUT indicating we are done trying to service this
	262	* operation since client-core seems to be exiting too often
	263	* or if we were interrupted.
d2d87a3b AV	264	*
d2d87a3b AV	265	* Returns with op->lock taken.
1182fca3	266	*/
c72f15b7	267	static int wait_for_matching_downcall(struct orangefs_kernel_op_s *op,
05b39a8b	268	long timeout,
c72f15b7	269	bool interruptible)
1182fca3	270	{
05b39a8b	271	long n;
c72f15b7 AV	272
	273	if (interruptible)
	274	n = wait_for_completion_interruptible_timeout(&op->waitq, timeout);
	275	else
	276	n = wait_for_completion_killable_timeout(&op->waitq, timeout);
	277
d2d87a3b	278	spin_lock(&op->lock);
1182fca3	279
d2d87a3b AV	280	if (op_state_serviced(op))
d2d87a3b AV	281	return 0;
70c6ea26	282
d2d87a3b AV	283	if (unlikely(n < 0)) {
	284	gossip_debug(GOSSIP_WAIT_DEBUG,
	285	"*** %s:"
	286	" operation interrupted by a signal (tag "
	287	"%llu, op %p)\n",
	288	__func__,
	289	llu(op->tag),
	290	op);
	291	return -EINTR;
1182fca3	292	}
d2d87a3b AV	293	if (op_state_purged(op)) {
	294	gossip_debug(GOSSIP_WAIT_DEBUG,
	295	"*** %s:"
	296	" operation purged (tag "
	297	"%llu, %p, att %d)\n",
	298	__func__,
	299	llu(op->tag),
	300	op,
	301	op->attempts);
	302	return (op->attempts < ORANGEFS_PURGE_RETRY_COUNT) ?
	303	-EAGAIN :
	304	-EIO;
	305	}
	306	/* must have timed out, then... */
	307	gossip_debug(GOSSIP_WAIT_DEBUG,
	308	"*** %s:"
	309	" operation timed out (tag"
	310	" %llu, %p, att %d)\n",
	311	__func__,
	312	llu(op->tag),
	313	op,
	314	op->attempts);
	315	return -ETIMEDOUT;
1182fca3	316	}