[deliverable/linux.git] / drivers / staging / lustre / lustre / llite / llite_close.c

/*
 * GPL HEADER START
 *
 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
 *
 * This program is free software; you can redistribute it and/or modify
 * it under the terms of the GNU General Public License version 2 only,
 * as published by the Free Software Foundation.
 *
 * This program is distributed in the hope that it will be useful, but
 * WITHOUT ANY WARRANTY; without even the implied warranty of
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
 * General Public License version 2 for more details (a copy is included
 * in the LICENSE file that accompanied this code).
 *
 * You should have received a copy of the GNU General Public License
 * version 2 along with this program; If not, see
 * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
 *
 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
 * CA 95054 USA or visit www.sun.com if you need additional information or
 * have any questions.
 *
 * GPL HEADER END
 */
/*
 * Copyright (c) 2003, 2010, Oracle and/or its affiliates. All rights reserved.
 * Use is subject to license terms.
 *
 * Copyright (c) 2011, 2012, Intel Corporation.
 */
/*
 * This file is part of Lustre, http://www.lustre.org/
 * Lustre is a trademark of Sun Microsystems, Inc.
 *
 * lustre/llite/llite_close.c
 *
 * Lustre Lite routines to issue a secondary close after writeback
 */

#include <linux/module.h>

#define DEBUG_SUBSYSTEM S_LLITE

#include <lustre_lite.h>
#include "llite_internal.h"

/** records that a write is in flight */
void vvp_write_pending(struct ccc_object *club, struct ccc_page *page)
{
	struct ll_inode_info *lli = ll_i2info(club->cob_inode);

	spin_lock(&lli->lli_lock);
	lli->lli_flags |= LLIF_SOM_DIRTY;
	if (page != NULL && list_empty(&page->cpg_pending_linkage))
		list_add(&page->cpg_pending_linkage,
			     &club->cob_pending_list);
	spin_unlock(&lli->lli_lock);
}

/** records that a write has completed */
void vvp_write_complete(struct ccc_object *club, struct ccc_page *page)
{
	struct ll_inode_info *lli = ll_i2info(club->cob_inode);
	int rc = 0;

	spin_lock(&lli->lli_lock);
	if (page != NULL && !list_empty(&page->cpg_pending_linkage)) {
		list_del_init(&page->cpg_pending_linkage);
		rc = 1;
	}
	spin_unlock(&lli->lli_lock);
	if (rc)
		ll_queue_done_writing(club->cob_inode, 0);
}

/** Queues DONE_WRITING if
 * - done writing is allowed;
 * - inode has no no dirty pages; */
void ll_queue_done_writing(struct inode *inode, unsigned long flags)
{
	struct ll_inode_info *lli = ll_i2info(inode);
	struct ccc_object *club = cl2ccc(ll_i2info(inode)->lli_clob);

	spin_lock(&lli->lli_lock);
	lli->lli_flags |= flags;

	if ((lli->lli_flags & LLIF_DONE_WRITING) &&
	    list_empty(&club->cob_pending_list)) {
		struct ll_close_queue *lcq = ll_i2sbi(inode)->ll_lcq;

		if (lli->lli_flags & LLIF_MDS_SIZE_LOCK)
			CWARN("ino %lu/%u(flags %u) som valid it just after "
			      "recovery\n",
			      inode->i_ino, inode->i_generation,
			      lli->lli_flags);
		/* DONE_WRITING is allowed and inode has no dirty page. */
		spin_lock(&lcq->lcq_lock);

		LASSERT(list_empty(&lli->lli_close_list));
		CDEBUG(D_INODE, "adding inode %lu/%u to close list\n",
		       inode->i_ino, inode->i_generation);
		list_add_tail(&lli->lli_close_list, &lcq->lcq_head);

		/* Avoid a concurrent insertion into the close thread queue:
		 * an inode is already in the close thread, open(), write(),
		 * close() happen, epoch is closed as the inode is marked as
		 * LLIF_EPOCH_PENDING. When pages are written inode should not
		 * be inserted into the queue again, clear this flag to avoid
		 * it. */
		lli->lli_flags &= ~LLIF_DONE_WRITING;

		wake_up(&lcq->lcq_waitq);
		spin_unlock(&lcq->lcq_lock);
	}
	spin_unlock(&lli->lli_lock);
}

/** Pack SOM attributes info @opdata for CLOSE, DONE_WRITING rpc. */
void ll_done_writing_attr(struct inode *inode, struct md_op_data *op_data)
{
	struct ll_inode_info *lli = ll_i2info(inode);

	op_data->op_flags |= MF_SOM_CHANGE;
	/* Check if Size-on-MDS attributes are valid. */
	if (lli->lli_flags & LLIF_MDS_SIZE_LOCK)
		CERROR("ino %lu/%u(flags %u) som valid it just after "
		       "recovery\n", inode->i_ino, inode->i_generation,
		       lli->lli_flags);

	if (!cl_local_size(inode)) {
		/* Send Size-on-MDS Attributes if valid. */
		op_data->op_attr.ia_valid |= ATTR_MTIME_SET | ATTR_CTIME_SET |
				ATTR_ATIME_SET | ATTR_SIZE | ATTR_BLOCKS;
	}
}

/** Closes ioepoch and packs Size-on-MDS attribute if needed into @op_data. */
void ll_ioepoch_close(struct inode *inode, struct md_op_data *op_data,
		      struct obd_client_handle **och, unsigned long flags)
{
	struct ll_inode_info *lli = ll_i2info(inode);
	struct ccc_object *club = cl2ccc(ll_i2info(inode)->lli_clob);

	spin_lock(&lli->lli_lock);
	if (!(list_empty(&club->cob_pending_list))) {
		if (!(lli->lli_flags & LLIF_EPOCH_PENDING)) {
			LASSERT(*och != NULL);
			LASSERT(lli->lli_pending_och == NULL);
			/* Inode is dirty and there is no pending write done
			 * request yet, DONE_WRITE is to be sent later. */
			lli->lli_flags |= LLIF_EPOCH_PENDING;
			lli->lli_pending_och = *och;
			spin_unlock(&lli->lli_lock);

			inode = igrab(inode);
			LASSERT(inode);
			GOTO(out, 0);
		}
		if (flags & LLIF_DONE_WRITING) {
			/* Some pages are still dirty, it is early to send
			 * DONE_WRITE. Wait untill all pages will be flushed
			 * and try DONE_WRITE again later. */
			LASSERT(!(lli->lli_flags & LLIF_DONE_WRITING));
			lli->lli_flags |= LLIF_DONE_WRITING;
			spin_unlock(&lli->lli_lock);

			inode = igrab(inode);
			LASSERT(inode);
			GOTO(out, 0);
		}
	}
	CDEBUG(D_INODE, "Epoch "LPU64" closed on "DFID"\n",
	       ll_i2info(inode)->lli_ioepoch, PFID(&lli->lli_fid));
	op_data->op_flags |= MF_EPOCH_CLOSE;

	if (flags & LLIF_DONE_WRITING) {
		LASSERT(lli->lli_flags & LLIF_SOM_DIRTY);
		LASSERT(!(lli->lli_flags & LLIF_DONE_WRITING));
		*och = lli->lli_pending_och;
		lli->lli_pending_och = NULL;
		lli->lli_flags &= ~LLIF_EPOCH_PENDING;
	} else {
		/* Pack Size-on-MDS inode attributes only if they has changed */
		if (!(lli->lli_flags & LLIF_SOM_DIRTY)) {
			spin_unlock(&lli->lli_lock);
			GOTO(out, 0);
		}

		/* There is a pending DONE_WRITE -- close epoch with no
		 * attribute change. */
		if (lli->lli_flags & LLIF_EPOCH_PENDING) {
			spin_unlock(&lli->lli_lock);
			GOTO(out, 0);
		}
	}

	LASSERT(list_empty(&club->cob_pending_list));
	lli->lli_flags &= ~LLIF_SOM_DIRTY;
	spin_unlock(&lli->lli_lock);
	ll_done_writing_attr(inode, op_data);

out:
	return;
}

/**
 * Cliens updates SOM attributes on MDS (including llog cookies):
 * obd_getattr with no lock and md_setattr.
 */
int ll_som_update(struct inode *inode, struct md_op_data *op_data)
{
	struct ll_inode_info *lli = ll_i2info(inode);
	struct ptlrpc_request *request = NULL;
	__u32 old_flags;
	struct obdo *oa;
	int rc;

	LASSERT(op_data != NULL);
	if (lli->lli_flags & LLIF_MDS_SIZE_LOCK)
		CERROR("ino %lu/%u(flags %u) som valid it just after "
		       "recovery\n", inode->i_ino, inode->i_generation,
		       lli->lli_flags);

	OBDO_ALLOC(oa);
	if (!oa) {
		CERROR("can't allocate memory for Size-on-MDS update.\n");
		return -ENOMEM;
	}

	old_flags = op_data->op_flags;
	op_data->op_flags = MF_SOM_CHANGE;

	/* If inode is already in another epoch, skip getattr from OSTs. */
	if (lli->lli_ioepoch == op_data->op_ioepoch) {
		rc = ll_inode_getattr(inode, oa, op_data->op_ioepoch,
				      old_flags & MF_GETATTR_LOCK);
		if (rc) {
			oa->o_valid = 0;
			if (rc != -ENOENT)
				CERROR("inode_getattr failed (%d): unable to "
				       "send a Size-on-MDS attribute update "
				       "for inode %lu/%u\n", rc, inode->i_ino,
				       inode->i_generation);
		} else {
			CDEBUG(D_INODE, "Size-on-MDS update on "DFID"\n",
			       PFID(&lli->lli_fid));
		}
		/* Install attributes into op_data. */
		md_from_obdo(op_data, oa, oa->o_valid);
	}

	rc = md_setattr(ll_i2sbi(inode)->ll_md_exp, op_data,
			NULL, 0, NULL, 0, &request, NULL);
	ptlrpc_req_finished(request);

	OBDO_FREE(oa);
	return rc;
}

/**
 * Closes the ioepoch and packs all the attributes into @op_data for
 * DONE_WRITING rpc.
 */
static void ll_prepare_done_writing(struct inode *inode,
				    struct md_op_data *op_data,
				    struct obd_client_handle **och)
{
	ll_ioepoch_close(inode, op_data, och, LLIF_DONE_WRITING);
	/* If there is no @och, we do not do D_W yet. */
	if (*och == NULL)
		return;

	ll_pack_inode2opdata(inode, op_data, &(*och)->och_fh);
	ll_prep_md_op_data(op_data, inode, NULL, NULL,
			   0, 0, LUSTRE_OPC_ANY, NULL);
}

/** Send a DONE_WRITING rpc. */
static void ll_done_writing(struct inode *inode)
{
	struct obd_client_handle *och = NULL;
	struct md_op_data *op_data;
	int rc;

	LASSERT(exp_connect_som(ll_i2mdexp(inode)));

	OBD_ALLOC_PTR(op_data);
	if (op_data == NULL) {
		CERROR("can't allocate op_data\n");
		return;
	}

	ll_prepare_done_writing(inode, op_data, &och);
	/* If there is no @och, we do not do D_W yet. */
	if (och == NULL)
		GOTO(out, 0);

	rc = md_done_writing(ll_i2sbi(inode)->ll_md_exp, op_data, NULL);
	if (rc == -EAGAIN) {
		/* MDS has instructed us to obtain Size-on-MDS attribute from
		 * OSTs and send setattr to back to MDS. */
		rc = ll_som_update(inode, op_data);
	} else if (rc) {
		CERROR("inode %lu mdc done_writing failed: rc = %d\n",
		       inode->i_ino, rc);
	}
out:
	ll_finish_md_op_data(op_data);
	if (och) {
		md_clear_open_replay_data(ll_i2sbi(inode)->ll_md_exp, och);
		OBD_FREE_PTR(och);
	}
}

static struct ll_inode_info *ll_close_next_lli(struct ll_close_queue *lcq)
{
	struct ll_inode_info *lli = NULL;

	spin_lock(&lcq->lcq_lock);

	if (!list_empty(&lcq->lcq_head)) {
		lli = list_entry(lcq->lcq_head.next, struct ll_inode_info,
				     lli_close_list);
		list_del_init(&lli->lli_close_list);
	} else if (atomic_read(&lcq->lcq_stop))
		lli = ERR_PTR(-EALREADY);

	spin_unlock(&lcq->lcq_lock);
	return lli;
}

static int ll_close_thread(void *arg)
{
	struct ll_close_queue *lcq = arg;

	complete(&lcq->lcq_comp);

	while (1) {
		struct l_wait_info lwi = { 0 };
		struct ll_inode_info *lli;
		struct inode *inode;

		l_wait_event_exclusive(lcq->lcq_waitq,
				       (lli = ll_close_next_lli(lcq)) != NULL,
				       &lwi);
		if (IS_ERR(lli))
			break;

		inode = ll_info2i(lli);
		CDEBUG(D_INFO, "done_writting for inode %lu/%u\n",
		       inode->i_ino, inode->i_generation);
		ll_done_writing(inode);
		iput(inode);
	}

	CDEBUG(D_INFO, "ll_close exiting\n");
	complete(&lcq->lcq_comp);
	return 0;
}

int ll_close_thread_start(struct ll_close_queue **lcq_ret)
{
	struct ll_close_queue *lcq;
	task_t *task;

	if (OBD_FAIL_CHECK(OBD_FAIL_LDLM_CLOSE_THREAD))
		return -EINTR;

	OBD_ALLOC(lcq, sizeof(*lcq));
	if (lcq == NULL)
		return -ENOMEM;

	spin_lock_init(&lcq->lcq_lock);
	INIT_LIST_HEAD(&lcq->lcq_head);
	init_waitqueue_head(&lcq->lcq_waitq);
	init_completion(&lcq->lcq_comp);

	task = kthread_run(ll_close_thread, lcq, "ll_close");
	if (IS_ERR(task)) {
		OBD_FREE(lcq, sizeof(*lcq));
		return PTR_ERR(task);
	}

	wait_for_completion(&lcq->lcq_comp);
	*lcq_ret = lcq;
	return 0;
}

void ll_close_thread_shutdown(struct ll_close_queue *lcq)
{
	init_completion(&lcq->lcq_comp);
	atomic_inc(&lcq->lcq_stop);
	wake_up(&lcq->lcq_waitq);
	wait_for_completion(&lcq->lcq_comp);
	OBD_FREE(lcq, sizeof(*lcq));
}
Commit	Line	Data
d7e09d03 PT	1	/*
	2	* GPL HEADER START
	3	*
	4	* DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
	5	*
	6	* This program is free software; you can redistribute it and/or modify
	7	* it under the terms of the GNU General Public License version 2 only,
	8	* as published by the Free Software Foundation.
	9	*
	10	* This program is distributed in the hope that it will be useful, but
	11	* WITHOUT ANY WARRANTY; without even the implied warranty of
	12	* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
	13	* General Public License version 2 for more details (a copy is included
	14	* in the LICENSE file that accompanied this code).
	15	*
	16	* You should have received a copy of the GNU General Public License
	17	* version 2 along with this program; If not, see
	18	* http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
	19	*
	20	* Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
	21	* CA 95054 USA or visit www.sun.com if you need additional information or
	22	* have any questions.
	23	*
	24	* GPL HEADER END
	25	*/
	26	/*
	27	* Copyright (c) 2003, 2010, Oracle and/or its affiliates. All rights reserved.
	28	* Use is subject to license terms.
	29	*
	30	* Copyright (c) 2011, 2012, Intel Corporation.
	31	*/
	32	/*
	33	* This file is part of Lustre, http://www.lustre.org/
	34	* Lustre is a trademark of Sun Microsystems, Inc.
	35	*
	36	* lustre/llite/llite_close.c
	37	*
	38	* Lustre Lite routines to issue a secondary close after writeback
	39	*/
	40
	41	#include <linux/module.h>
	42
	43	#define DEBUG_SUBSYSTEM S_LLITE
	44
	45	#include <lustre_lite.h>
	46	#include "llite_internal.h"
	47
	48	/** records that a write is in flight */
	49	void vvp_write_pending(struct ccc_object club, struct ccc_page page)
	50	{
	51	struct ll_inode_info *lli = ll_i2info(club->cob_inode);
	52
d7e09d03 PT	53	spin_lock(&lli->lli_lock);
	54	lli->lli_flags \|= LLIF_SOM_DIRTY;
	55	if (page != NULL && list_empty(&page->cpg_pending_linkage))
	56	list_add(&page->cpg_pending_linkage,
	57	&club->cob_pending_list);
	58	spin_unlock(&lli->lli_lock);
d7e09d03 PT	59	}
	60
	61	/** records that a write has completed */
	62	void vvp_write_complete(struct ccc_object club, struct ccc_page page)
	63	{
	64	struct ll_inode_info *lli = ll_i2info(club->cob_inode);
	65	int rc = 0;
	66
d7e09d03 PT	67	spin_lock(&lli->lli_lock);
	68	if (page != NULL && !list_empty(&page->cpg_pending_linkage)) {
	69	list_del_init(&page->cpg_pending_linkage);
	70	rc = 1;
	71	}
	72	spin_unlock(&lli->lli_lock);
	73	if (rc)
	74	ll_queue_done_writing(club->cob_inode, 0);
d7e09d03 PT	75	}
	76
	77	/** Queues DONE_WRITING if
	78	* - done writing is allowed;
	79	* - inode has no no dirty pages; */
	80	void ll_queue_done_writing(struct inode *inode, unsigned long flags)
	81	{
	82	struct ll_inode_info *lli = ll_i2info(inode);
	83	struct ccc_object *club = cl2ccc(ll_i2info(inode)->lli_clob);
d7e09d03 PT	84
	85	spin_lock(&lli->lli_lock);
	86	lli->lli_flags \|= flags;
	87
	88	if ((lli->lli_flags & LLIF_DONE_WRITING) &&
	89	list_empty(&club->cob_pending_list)) {
	90	struct ll_close_queue *lcq = ll_i2sbi(inode)->ll_lcq;
	91
	92	if (lli->lli_flags & LLIF_MDS_SIZE_LOCK)
	93	CWARN("ino %lu/%u(flags %u) som valid it just after "
	94	"recovery\n",
	95	inode->i_ino, inode->i_generation,
	96	lli->lli_flags);
	97	/* DONE_WRITING is allowed and inode has no dirty page. */
	98	spin_lock(&lcq->lcq_lock);
	99
	100	LASSERT(list_empty(&lli->lli_close_list));
	101	CDEBUG(D_INODE, "adding inode %lu/%u to close list\n",
	102	inode->i_ino, inode->i_generation);
	103	list_add_tail(&lli->lli_close_list, &lcq->lcq_head);
	104
	105	/* Avoid a concurrent insertion into the close thread queue:
	106	* an inode is already in the close thread, open(), write(),
	107	* close() happen, epoch is closed as the inode is marked as
	108	* LLIF_EPOCH_PENDING. When pages are written inode should not
	109	* be inserted into the queue again, clear this flag to avoid
	110	* it. */
	111	lli->lli_flags &= ~LLIF_DONE_WRITING;
	112
	113	wake_up(&lcq->lcq_waitq);
	114	spin_unlock(&lcq->lcq_lock);
	115	}
	116	spin_unlock(&lli->lli_lock);
d7e09d03 PT	117	}
	118
	119	/** Pack SOM attributes info @opdata for CLOSE, DONE_WRITING rpc. */
	120	void ll_done_writing_attr(struct inode inode, struct md_op_data op_data)
	121	{
	122	struct ll_inode_info *lli = ll_i2info(inode);
d7e09d03 PT	123
	124	op_data->op_flags \|= MF_SOM_CHANGE;
	125	/* Check if Size-on-MDS attributes are valid. */
	126	if (lli->lli_flags & LLIF_MDS_SIZE_LOCK)
	127	CERROR("ino %lu/%u(flags %u) som valid it just after "
	128	"recovery\n", inode->i_ino, inode->i_generation,
	129	lli->lli_flags);
	130
	131	if (!cl_local_size(inode)) {
	132	/* Send Size-on-MDS Attributes if valid. */
	133	op_data->op_attr.ia_valid \|= ATTR_MTIME_SET \| ATTR_CTIME_SET \|
	134	ATTR_ATIME_SET \| ATTR_SIZE \| ATTR_BLOCKS;
	135	}
d7e09d03 PT	136	}
	137
	138	/** Closes ioepoch and packs Size-on-MDS attribute if needed into @op_data. */
	139	void ll_ioepoch_close(struct inode inode, struct md_op_data op_data,
	140	struct obd_client_handle **och, unsigned long flags)
	141	{
	142	struct ll_inode_info *lli = ll_i2info(inode);
	143	struct ccc_object *club = cl2ccc(ll_i2info(inode)->lli_clob);
d7e09d03 PT	144
	145	spin_lock(&lli->lli_lock);
	146	if (!(list_empty(&club->cob_pending_list))) {
	147	if (!(lli->lli_flags & LLIF_EPOCH_PENDING)) {
	148	LASSERT(*och != NULL);
	149	LASSERT(lli->lli_pending_och == NULL);
	150	/* Inode is dirty and there is no pending write done
	151	* request yet, DONE_WRITE is to be sent later. */
	152	lli->lli_flags \|= LLIF_EPOCH_PENDING;
	153	lli->lli_pending_och = *och;
	154	spin_unlock(&lli->lli_lock);
	155
	156	inode = igrab(inode);
	157	LASSERT(inode);
	158	GOTO(out, 0);
	159	}
	160	if (flags & LLIF_DONE_WRITING) {
	161	/* Some pages are still dirty, it is early to send
	162	* DONE_WRITE. Wait untill all pages will be flushed
	163	* and try DONE_WRITE again later. */
	164	LASSERT(!(lli->lli_flags & LLIF_DONE_WRITING));
	165	lli->lli_flags \|= LLIF_DONE_WRITING;
	166	spin_unlock(&lli->lli_lock);
	167
	168	inode = igrab(inode);
	169	LASSERT(inode);
	170	GOTO(out, 0);
	171	}
	172	}
	173	CDEBUG(D_INODE, "Epoch "LPU64" closed on "DFID"\n",
	174	ll_i2info(inode)->lli_ioepoch, PFID(&lli->lli_fid));
	175	op_data->op_flags \|= MF_EPOCH_CLOSE;
	176
	177	if (flags & LLIF_DONE_WRITING) {
	178	LASSERT(lli->lli_flags & LLIF_SOM_DIRTY);
	179	LASSERT(!(lli->lli_flags & LLIF_DONE_WRITING));
	180	*och = lli->lli_pending_och;
	181	lli->lli_pending_och = NULL;
	182	lli->lli_flags &= ~LLIF_EPOCH_PENDING;
	183	} else {
	184	/* Pack Size-on-MDS inode attributes only if they has changed */
	185	if (!(lli->lli_flags & LLIF_SOM_DIRTY)) {
	186	spin_unlock(&lli->lli_lock);
	187	GOTO(out, 0);
	188	}
	189
	190	/* There is a pending DONE_WRITE -- close epoch with no
	191	* attribute change. */
	192	if (lli->lli_flags & LLIF_EPOCH_PENDING) {
	193	spin_unlock(&lli->lli_lock);
	194	GOTO(out, 0);
	195	}
	196	}
	197
	198	LASSERT(list_empty(&club->cob_pending_list));
	199	lli->lli_flags &= ~LLIF_SOM_DIRTY;
	200	spin_unlock(&lli->lli_lock);
	201	ll_done_writing_attr(inode, op_data);
	202
d7e09d03 PT	203	out:
	204	return;
	205	}
	206
	207	/**
	208	* Cliens updates SOM attributes on MDS (including llog cookies):
	209	* obd_getattr with no lock and md_setattr.
	210	*/
	211	int ll_som_update(struct inode inode, struct md_op_data op_data)
	212	{
	213	struct ll_inode_info *lli = ll_i2info(inode);
	214	struct ptlrpc_request *request = NULL;
	215	__u32 old_flags;
	216	struct obdo *oa;
	217	int rc;
d7e09d03 PT	218
	219	LASSERT(op_data != NULL);
	220	if (lli->lli_flags & LLIF_MDS_SIZE_LOCK)
	221	CERROR("ino %lu/%u(flags %u) som valid it just after "
	222	"recovery\n", inode->i_ino, inode->i_generation,
	223	lli->lli_flags);
	224
	225	OBDO_ALLOC(oa);
	226	if (!oa) {
	227	CERROR("can't allocate memory for Size-on-MDS update.\n");
0a3bdb00	228	return -ENOMEM;
d7e09d03 PT	229	}
	230
	231	old_flags = op_data->op_flags;
	232	op_data->op_flags = MF_SOM_CHANGE;
	233
	234	/* If inode is already in another epoch, skip getattr from OSTs. */
	235	if (lli->lli_ioepoch == op_data->op_ioepoch) {
	236	rc = ll_inode_getattr(inode, oa, op_data->op_ioepoch,
	237	old_flags & MF_GETATTR_LOCK);
	238	if (rc) {
	239	oa->o_valid = 0;
	240	if (rc != -ENOENT)
	241	CERROR("inode_getattr failed (%d): unable to "
	242	"send a Size-on-MDS attribute update "
	243	"for inode %lu/%u\n", rc, inode->i_ino,
	244	inode->i_generation);
	245	} else {
	246	CDEBUG(D_INODE, "Size-on-MDS update on "DFID"\n",
	247	PFID(&lli->lli_fid));
	248	}
	249	/* Install attributes into op_data. */
	250	md_from_obdo(op_data, oa, oa->o_valid);
	251	}
	252
	253	rc = md_setattr(ll_i2sbi(inode)->ll_md_exp, op_data,
	254	NULL, 0, NULL, 0, &request, NULL);
	255	ptlrpc_req_finished(request);
	256
	257	OBDO_FREE(oa);
0a3bdb00	258	return rc;
d7e09d03 PT	259	}
	260
	261	/**
	262	* Closes the ioepoch and packs all the attributes into @op_data for
	263	* DONE_WRITING rpc.
	264	*/
	265	static void ll_prepare_done_writing(struct inode *inode,
	266	struct md_op_data *op_data,
	267	struct obd_client_handle **och)
	268	{
	269	ll_ioepoch_close(inode, op_data, och, LLIF_DONE_WRITING);
	270	/* If there is no @och, we do not do D_W yet. */
	271	if (*och == NULL)
	272	return;
	273
	274	ll_pack_inode2opdata(inode, op_data, &(*och)->och_fh);
	275	ll_prep_md_op_data(op_data, inode, NULL, NULL,
	276	0, 0, LUSTRE_OPC_ANY, NULL);
	277	}
	278
	279	/** Send a DONE_WRITING rpc. */
	280	static void ll_done_writing(struct inode *inode)
	281	{
	282	struct obd_client_handle *och = NULL;
	283	struct md_op_data *op_data;
	284	int rc;
d7e09d03 PT	285
	286	LASSERT(exp_connect_som(ll_i2mdexp(inode)));
	287
	288	OBD_ALLOC_PTR(op_data);
	289	if (op_data == NULL) {
	290	CERROR("can't allocate op_data\n");
d7e09d03 PT	291	return;
	292	}
	293
	294	ll_prepare_done_writing(inode, op_data, &och);
	295	/* If there is no @och, we do not do D_W yet. */
	296	if (och == NULL)
	297	GOTO(out, 0);
	298
	299	rc = md_done_writing(ll_i2sbi(inode)->ll_md_exp, op_data, NULL);
	300	if (rc == -EAGAIN) {
	301	/* MDS has instructed us to obtain Size-on-MDS attribute from
	302	* OSTs and send setattr to back to MDS. */
	303	rc = ll_som_update(inode, op_data);
	304	} else if (rc) {
	305	CERROR("inode %lu mdc done_writing failed: rc = %d\n",
	306	inode->i_ino, rc);
	307	}
	308	out:
	309	ll_finish_md_op_data(op_data);
	310	if (och) {
	311	md_clear_open_replay_data(ll_i2sbi(inode)->ll_md_exp, och);
	312	OBD_FREE_PTR(och);
	313	}
d7e09d03 PT	314	}
	315
	316	static struct ll_inode_info ll_close_next_lli(struct ll_close_queue lcq)
	317	{
	318	struct ll_inode_info *lli = NULL;
	319
	320	spin_lock(&lcq->lcq_lock);
	321
	322	if (!list_empty(&lcq->lcq_head)) {
	323	lli = list_entry(lcq->lcq_head.next, struct ll_inode_info,
	324	lli_close_list);
	325	list_del_init(&lli->lli_close_list);
	326	} else if (atomic_read(&lcq->lcq_stop))
	327	lli = ERR_PTR(-EALREADY);
	328
	329	spin_unlock(&lcq->lcq_lock);
	330	return lli;
	331	}
	332
	333	static int ll_close_thread(void *arg)
	334	{
	335	struct ll_close_queue *lcq = arg;
d7e09d03 PT	336
	337	complete(&lcq->lcq_comp);
	338
	339	while (1) {
	340	struct l_wait_info lwi = { 0 };
	341	struct ll_inode_info *lli;
	342	struct inode *inode;
	343
	344	l_wait_event_exclusive(lcq->lcq_waitq,
	345	(lli = ll_close_next_lli(lcq)) != NULL,
	346	&lwi);
	347	if (IS_ERR(lli))
	348	break;
	349
	350	inode = ll_info2i(lli);
	351	CDEBUG(D_INFO, "done_writting for inode %lu/%u\n",
	352	inode->i_ino, inode->i_generation);
	353	ll_done_writing(inode);
	354	iput(inode);
	355	}
	356
	357	CDEBUG(D_INFO, "ll_close exiting\n");
	358	complete(&lcq->lcq_comp);
0a3bdb00	359	return 0;
d7e09d03 PT	360	}
	361
	362	int ll_close_thread_start(struct ll_close_queue **lcq_ret)
	363	{
	364	struct ll_close_queue *lcq;
	365	task_t *task;
	366
	367	if (OBD_FAIL_CHECK(OBD_FAIL_LDLM_CLOSE_THREAD))
	368	return -EINTR;
	369
	370	OBD_ALLOC(lcq, sizeof(*lcq));
	371	if (lcq == NULL)
	372	return -ENOMEM;
	373
	374	spin_lock_init(&lcq->lcq_lock);
	375	INIT_LIST_HEAD(&lcq->lcq_head);
	376	init_waitqueue_head(&lcq->lcq_waitq);
	377	init_completion(&lcq->lcq_comp);
	378
	379	task = kthread_run(ll_close_thread, lcq, "ll_close");
	380	if (IS_ERR(task)) {
	381	OBD_FREE(lcq, sizeof(*lcq));
	382	return PTR_ERR(task);
	383	}
	384
	385	wait_for_completion(&lcq->lcq_comp);
	386	*lcq_ret = lcq;
	387	return 0;
	388	}
	389
	390	void ll_close_thread_shutdown(struct ll_close_queue *lcq)
	391	{
	392	init_completion(&lcq->lcq_comp);
	393	atomic_inc(&lcq->lcq_stop);
	394	wake_up(&lcq->lcq_waitq);
	395	wait_for_completion(&lcq->lcq_comp);
	396	OBD_FREE(lcq, sizeof(*lcq));
	397	}