-/* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
- * vim:expandtab:shiftwidth=8:tabstop=8:
- *
+/*
* GPL HEADER START
*
* DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
* GPL HEADER END
*/
/*
- * Copyright 2008 Sun Microsystems, Inc. All rights reserved
+ * 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/
#define DEBUG_SUBSYSTEM S_LLITE
-//#include <lustre_mdc.h>
#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);
-
- ENTRY;
- 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);
- EXIT;
+ struct ll_inode_info *lli = ll_i2info(club->cob_inode);
+
+ ENTRY;
+ 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);
+ EXIT;
}
/** 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;
-
- ENTRY;
- 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);
- EXIT;
+ struct ll_inode_info *lli = ll_i2info(club->cob_inode);
+ int rc = 0;
+
+ ENTRY;
+ 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);
+ EXIT;
}
/** Queues DONE_WRITING if
* - 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);
- ENTRY;
+ struct ll_inode_info *lli = ll_i2info(inode);
+ struct ccc_object *club = cl2ccc(ll_i2info(inode)->lli_clob);
+ ENTRY;
- spin_lock(&lli->lli_lock);
+ spin_lock(&lli->lli_lock);
lli->lli_flags |= flags;
if ((lli->lli_flags & LLIF_DONE_WRITING) &&
- list_empty(&club->cob_pending_list)) {
+ 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 %lu) 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);
- EXIT;
+ CWARN("%s: file "DFID"(flags %u) Size-on-MDS valid, "
+ "done writing allowed and no diry pages\n",
+ ll_get_fsname(inode->i_sb, NULL, 0),
+ PFID(ll_inode2fid(inode)), 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 "DFID" to close list\n",
+ PFID(ll_inode2fid(inode)));
+ 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);
+ EXIT;
}
/** Pack SOM attributes info @opdata for CLOSE, DONE_WRITING rpc. */
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 %lu) som valid it just after "
- "recovery\n", inode->i_ino, inode->i_generation,
- lli->lli_flags);
+ CERROR("%s: inode "DFID"(flags %u) MDS holds lock on "
+ "Size-on-MDS attributes\n",
+ ll_get_fsname(inode->i_sb, NULL, 0),
+ PFID(ll_inode2fid(inode)), lli->lli_flags);
if (!cl_local_size(inode)) {
/* Send Size-on-MDS Attributes if valid. */
/** 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 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);
- ENTRY;
-
- 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);
+ struct ll_inode_info *lli = ll_i2info(inode);
+ struct ccc_object *club = cl2ccc(ll_i2info(inode)->lli_clob);
+ ENTRY;
+
+ 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);
* and try DONE_WRITE again later. */
LASSERT(!(lli->lli_flags & LLIF_DONE_WRITING));
lli->lli_flags |= LLIF_DONE_WRITING;
- spin_unlock(&lli->lli_lock);
+ spin_unlock(&lli->lli_lock);
inode = igrab(inode);
LASSERT(inode);
} 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);
-
- EXIT;
+ 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);
+
+ EXIT;
out:
- return;
+ return;
}
/**
{
struct ll_inode_info *lli = ll_i2info(inode);
struct ptlrpc_request *request = NULL;
+ __u32 old_flags;
struct obdo *oa;
int rc;
ENTRY;
LASSERT(op_data != NULL);
if (lli->lli_flags & LLIF_MDS_SIZE_LOCK)
- CERROR("ino %lu/%u(flags %lu) som valid it just after "
- "recovery\n", inode->i_ino, inode->i_generation,
- lli->lli_flags);
+ CERROR("%s: inode "DFID"(flags %u) MDS holds lock on "
+ "Size-on-MDS attributes\n",
+ ll_get_fsname(inode->i_sb, NULL, 0),
+ PFID(ll_inode2fid(inode)), lli->lli_flags);
OBDO_ALLOC(oa);
if (!oa) {
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);
+ rc = ll_inode_getattr(inode, oa, op_data->op_ioepoch,
+ old_flags & MF_GETATTR_LOCK);
if (rc) {
oa->o_valid = 0;
- if (rc == -ENOENT)
- CDEBUG(D_INODE, "objid "LPX64" is destroyed\n",
- lli->lli_smd->lsm_object_id);
- else
- 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);
+ if (rc != -ENOENT)
+ CERROR("%s: inode_getattr failed - unable to "
+ "send a Size-on-MDS attribute update "
+ "for inode "DFID": rc = %d\n",
+ ll_get_fsname(inode->i_sb, NULL, 0),
+ PFID(ll_inode2fid(inode)), rc);
} else {
CDEBUG(D_INODE, "Size-on-MDS update on "DFID"\n",
PFID(&lli->lli_fid));
* 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);
+ CERROR("%s: inode "DFID" mdc done_writing failed: rc = %d\n",
+ ll_get_fsname(inode->i_sb, NULL, 0),
+ PFID(ll_inode2fid(inode)), rc);
}
out:
ll_finish_md_op_data(op_data);
static struct ll_inode_info *ll_close_next_lli(struct ll_close_queue *lcq)
{
- struct ll_inode_info *lli = NULL;
+ struct ll_inode_info *lli = NULL;
- spin_lock(&lcq->lcq_lock);
+ 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);
+ 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;
+ spin_unlock(&lcq->lcq_lock);
+ return lli;
}
static int ll_close_thread(void *arg)
struct ll_close_queue *lcq = arg;
ENTRY;
- {
- char name[CFS_CURPROC_COMM_MAX];
- snprintf(name, sizeof(name) - 1, "ll_close");
- cfs_daemonize(name);
- }
-
- complete(&lcq->lcq_comp);
+ complete(&lcq->lcq_comp);
while (1) {
struct l_wait_info lwi = { 0 };
break;
inode = ll_info2i(lli);
- CDEBUG(D_INFO, "done_writting for inode %lu/%u\n",
- inode->i_ino, inode->i_generation);
+ CDEBUG(D_INFO, "done_writting for inode "DFID"\n",
+ PFID(ll_inode2fid(inode)));
ll_done_writing(inode);
iput(inode);
}
CDEBUG(D_INFO, "ll_close exiting\n");
- complete(&lcq->lcq_comp);
- RETURN(0);
+ complete(&lcq->lcq_comp);
+ RETURN(0);
}
int ll_close_thread_start(struct ll_close_queue **lcq_ret)
{
- struct ll_close_queue *lcq;
- pid_t pid;
-
- 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);
-
- pid = kernel_thread(ll_close_thread, lcq, 0);
- if (pid < 0) {
- OBD_FREE(lcq, sizeof(*lcq));
- return pid;
- }
-
- wait_for_completion(&lcq->lcq_comp);
- *lcq_ret = lcq;
- return 0;
+ struct ll_close_queue *lcq;
+ struct task_struct *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));
+ 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));
}
git://git.whamcloud.com / fs / lustre-release.git / blobdiff