* GPL HEADER END
*/
/*
- * Copyright (c) 2012 Whamcloud, Inc.
+ * Copyright (c) 2012, Intel Corporation.
*/
/*
* lustre/osd-ldiskfs/osd_scrub.c
sf->sf_items_updated = 0;
sf->sf_items_failed = 0;
sf->sf_items_updated_prior = 0;
+ sf->sf_items_noscrub = 0;
+ sf->sf_items_igif = 0;
}
static int osd_scrub_file_load(struct osd_scrub *scrub)
osd_scrub_file_to_le(&scrub->os_file_disk, &scrub->os_file);
rc = osd_ldiskfs_write_record(scrub->os_inode, &scrub->os_file_disk,
- len, &pos, jh);
+ len, 0, &pos, jh);
ldiskfs_journal_stop(jh);
if (rc != 0)
CERROR("%.16s: fail to store scrub file, expected = %d, "
int rc;
ENTRY;
- cfs_down_write(&scrub->os_rwsem);
+ down_write(&scrub->os_rwsem);
if (flags & SS_SET_FAILOUT)
sf->sf_param |= SP_FAILOUT;
sf->sf_time_last_checkpoint = sf->sf_time_latest_start;
rc = osd_scrub_file_store(scrub);
if (rc == 0) {
- cfs_spin_lock(&scrub->os_lock);
+ spin_lock(&scrub->os_lock);
thread_set_flags(thread, SVC_RUNNING);
- cfs_spin_unlock(&scrub->os_lock);
+ spin_unlock(&scrub->os_lock);
cfs_waitq_broadcast(&thread->t_ctl_waitq);
}
- cfs_up_write(&scrub->os_rwsem);
+ up_write(&scrub->os_rwsem);
RETURN(rc);
}
struct osd_scrub *scrub = &dev->od_scrub;
struct scrub_file *sf = &scrub->os_file;
- cfs_down_write(&scrub->os_rwsem);
+ down_write(&scrub->os_rwsem);
scrub->os_new_checked++;
sf->sf_items_failed++;
if (sf->sf_pos_first_inconsistent == 0 ||
sf->sf_pos_first_inconsistent > lid->oii_ino)
sf->sf_pos_first_inconsistent = lid->oii_ino;
- cfs_up_write(&scrub->os_rwsem);
+ up_write(&scrub->os_rwsem);
return sf->sf_param & SP_FAILOUT ? rc : 0;
}
oii = cfs_list_entry(oic, struct osd_inconsistent_item,
oii_cache);
- cfs_down_write(&scrub->os_rwsem);
+ down_write(&scrub->os_rwsem);
scrub->os_new_checked++;
if (lid->oii_ino < sf->sf_pos_latest_start && oii == NULL)
GOTO(out, rc = 0);
}
/* Prevent the inode to be unlinked during OI scrub. */
- cfs_mutex_lock(&inode->i_mutex);
+ mutex_lock(&inode->i_mutex);
if (unlikely(inode->i_nlink == 0)) {
- cfs_mutex_unlock(&inode->i_mutex);
+ mutex_unlock(&inode->i_mutex);
iput(inode);
GOTO(out, rc = 0);
}
GOTO(out, rc = -ENOMEM);
}
- if (ops == DTO_INDEX_UPDATE)
+ if (ops == DTO_INDEX_UPDATE) {
rc = iam_update(jh, bag, (const struct iam_key *)oi_fid,
(struct iam_rec *)oi_id, ipd);
- else
+ } else {
rc = iam_insert(jh, bag, (const struct iam_key *)oi_fid,
(struct iam_rec *)oi_id, ipd);
+ if (rc == -EEXIST)
+ rc = 1;
+ }
osd_ipd_put(info->oti_env, bag, ipd);
ldiskfs_journal_stop(jh);
if (rc == 0) {
GOTO(out, rc);
out:
- if (rc != 0) {
+ if (rc < 0) {
sf->sf_items_failed++;
if (sf->sf_pos_first_inconsistent == 0 ||
sf->sf_pos_first_inconsistent > lid->oii_ino)
sf->sf_pos_first_inconsistent = lid->oii_ino;
+ } else {
+ rc = 0;
}
if (ops == DTO_INDEX_INSERT) {
- cfs_mutex_unlock(&inode->i_mutex);
+ mutex_unlock(&inode->i_mutex);
iput(inode);
}
- cfs_up_write(&scrub->os_rwsem);
+ up_write(&scrub->os_rwsem);
if (oii != NULL) {
LASSERT(!cfs_list_empty(&oii->oii_list));
- cfs_spin_lock(&scrub->os_lock);
+ spin_lock(&scrub->os_lock);
cfs_list_del_init(&oii->oii_list);
- cfs_spin_unlock(&scrub->os_lock);
+ spin_unlock(&scrub->os_lock);
OBD_FREE_PTR(oii);
}
RETURN(sf->sf_param & SP_FAILOUT ? rc : 0);
int rc;
ENTRY;
- cfs_down_write(&scrub->os_rwsem);
+ down_write(&scrub->os_rwsem);
sf->sf_items_checked += scrub->os_new_checked;
scrub->os_new_checked = 0;
sf->sf_pos_last_checkpoint = scrub->os_pos_current;
sf->sf_run_time += cfs_duration_sec(cfs_time_current() + HALF_SEC -
scrub->os_time_last_checkpoint);
rc = osd_scrub_file_store(scrub);
- cfs_up_write(&scrub->os_rwsem);
+ up_write(&scrub->os_rwsem);
RETURN(rc);
}
struct scrub_file *sf = &scrub->os_file;
ENTRY;
- cfs_down_write(&scrub->os_rwsem);
- cfs_spin_lock(&scrub->os_lock);
+ down_write(&scrub->os_rwsem);
+ spin_lock(&scrub->os_lock);
thread_set_flags(&scrub->os_thread, SVC_STOPPING);
- cfs_spin_unlock(&scrub->os_lock);
+ spin_unlock(&scrub->os_lock);
if (scrub->os_new_checked > 0) {
sf->sf_items_checked += scrub->os_new_checked;
scrub->os_new_checked = 0;
sf->sf_time_last_complete = sf->sf_time_last_checkpoint;
sf->sf_success_count++;
} else if (result == 0) {
- sf->sf_status = SS_PAUSED;
+ if (scrub->os_paused)
+ sf->sf_status = SS_PAUSED;
+ else
+ sf->sf_status = SS_STOPPED;
} else {
sf->sf_status = SS_FAILED;
}
CERROR("%.16s: fail to osd_scrub_post, rc = %d\n",
LDISKFS_SB(osd_scrub2sb(scrub))->s_es->s_volume_name,
result);
- cfs_up_write(&scrub->os_rwsem);
+ up_write(&scrub->os_rwsem);
EXIT;
}
#define SCRUB_NEXT_WAIT 4 /* wait for free cache slot */
#define SCRUB_NEXT_CRASH 5 /* simulate system crash during OI scrub */
#define SCRUB_NEXT_FATAL 6 /* simulate failure during OI scrub */
+#define SCRUB_NEXT_NOSCRUB 7 /* new created object, no scrub on it */
+#define SCRUB_NEXT_IGIF 8 /* IGIF object */
struct osd_iit_param {
struct super_block *sb;
}
if (OBD_FAIL_CHECK(OBD_FAIL_OSD_SCRUB_CRASH)) {
- cfs_spin_lock(&scrub->os_lock);
+ spin_lock(&scrub->os_lock);
thread_set_flags(thread, SVC_STOPPING);
- cfs_spin_unlock(&scrub->os_lock);
+ spin_unlock(&scrub->os_lock);
return SCRUB_NEXT_CRASH;
}
if (rc != 0)
return rc;
- if (!fid_is_norm(fid) || inode->i_state & I_LUSTRE_NOSCRUB)
- rc = SCRUB_NEXT_CONTINUE;
+ if (inode->i_state & I_LUSTRE_NOSCRUB) {
+ /* Only skip it for the first OI scrub accessing. */
+ inode->i_state &= ~I_LUSTRE_NOSCRUB;
+ rc = SCRUB_NEXT_NOSCRUB;
+ } else if (!fid_is_norm(fid)) {
+ rc = SCRUB_NEXT_IGIF;
+ }
+
iput(inode);
return rc;
}
{
struct l_wait_info lwi = { 0 };
struct osd_scrub *scrub = &dev->od_scrub;
+ struct scrub_file *sf = &scrub->os_file;
+ __u64 *items = NULL;
struct ptlrpc_thread *thread = &scrub->os_thread;
struct osd_otable_it *it = dev->od_otable_it;
struct osd_otable_cache *ooc = it ? &it->ooi_cache : NULL;
goto next;
case SCRUB_NEXT_WAIT:
goto wait;
+ case SCRUB_NEXT_NOSCRUB:
+ items = &sf->sf_items_noscrub;
+ break;
+ case SCRUB_NEXT_IGIF:
+ items = &sf->sf_items_igif;
+ break;
+ }
+
+ if (items != NULL) {
+ down_write(&scrub->os_rwsem);
+ scrub->os_new_checked++;
+ (*items)++;
+ up_write(&scrub->os_rwsem);
+ goto next;
}
LASSERTF(rc <= 0, "unexpected rc = %d\n", rc);
lu_env_fini(&env);
noenv:
- cfs_spin_lock(&scrub->os_lock);
+ spin_lock(&scrub->os_lock);
thread_set_flags(thread, SVC_STOPPED);
cfs_waitq_broadcast(&thread->t_ctl_waitq);
- cfs_spin_unlock(&scrub->os_lock);
+ spin_unlock(&scrub->os_lock);
return rc;
}
again:
/* os_lock: sync status between stop and scrub thread */
- cfs_spin_lock(&scrub->os_lock);
+ spin_lock(&scrub->os_lock);
if (thread_is_running(thread)) {
- cfs_spin_unlock(&scrub->os_lock);
+ spin_unlock(&scrub->os_lock);
RETURN(-EALREADY);
} else if (unlikely(thread_is_stopping(thread))) {
- cfs_spin_unlock(&scrub->os_lock);
+ spin_unlock(&scrub->os_lock);
l_wait_event(thread->t_ctl_waitq,
thread_is_stopped(thread),
&lwi);
goto again;
}
- cfs_spin_unlock(&scrub->os_lock);
+ spin_unlock(&scrub->os_lock);
+
+ if (scrub->os_file.sf_status == SS_COMPLETED)
+ flags |= SS_RESET;
scrub->os_start_flags = flags;
thread_set_flags(thread, 0);
int osd_scrub_start(struct osd_device *dev)
{
- __u32 flags = SS_AUTO;
- int rc;
+ int rc;
ENTRY;
- if (dev->od_scrub.os_file.sf_status == SS_COMPLETED)
- flags |= SS_RESET;
-
/* od_otable_mutex: prevent curcurrent start/stop */
- cfs_mutex_lock(&dev->od_otable_mutex);
- rc = do_osd_scrub_start(dev, flags);
- cfs_mutex_unlock(&dev->od_otable_mutex);
+ mutex_lock(&dev->od_otable_mutex);
+ rc = do_osd_scrub_start(dev, SS_AUTO);
+ mutex_unlock(&dev->od_otable_mutex);
RETURN(rc == -EALREADY ? 0 : rc);
}
struct l_wait_info lwi = { 0 };
/* os_lock: sync status between stop and scrub thread */
- cfs_spin_lock(&scrub->os_lock);
+ spin_lock(&scrub->os_lock);
if (!thread_is_init(thread) && !thread_is_stopped(thread)) {
thread_set_flags(thread, SVC_STOPPING);
- cfs_spin_unlock(&scrub->os_lock);
+ spin_unlock(&scrub->os_lock);
cfs_waitq_broadcast(&thread->t_ctl_waitq);
l_wait_event(thread->t_ctl_waitq,
thread_is_stopped(thread),
&lwi);
/* Do not skip the last lock/unlock, which can guarantee that
* the caller cannot return until the OI scrub thread exit. */
- cfs_spin_lock(&scrub->os_lock);
+ spin_lock(&scrub->os_lock);
}
- cfs_spin_unlock(&scrub->os_lock);
+ spin_unlock(&scrub->os_lock);
}
static void osd_scrub_stop(struct osd_device *dev)
{
/* od_otable_mutex: prevent curcurrent start/stop */
- cfs_mutex_lock(&dev->od_otable_mutex);
+ mutex_lock(&dev->od_otable_mutex);
+ dev->od_scrub.os_paused = 1;
do_osd_scrub_stop(&dev->od_scrub);
- cfs_mutex_unlock(&dev->od_otable_mutex);
+ mutex_unlock(&dev->od_otable_mutex);
}
static const char osd_scrub_name[] = "OI_scrub";
int rc = 0;
ENTRY;
+ memset(scrub, 0, sizeof(*scrub));
OBD_SET_CTXT_MAGIC(ctxt);
ctxt->pwdmnt = dev->od_mnt;
ctxt->pwd = dev->od_mnt->mnt_root;
ctxt->fs = get_ds();
cfs_waitq_init(&scrub->os_thread.t_ctl_waitq);
- cfs_init_rwsem(&scrub->os_rwsem);
- cfs_spin_lock_init(&scrub->os_lock);
+ init_rwsem(&scrub->os_rwsem);
+ spin_lock_init(&scrub->os_lock);
CFS_INIT_LIST_HEAD(&scrub->os_inconsistent_items);
- if (get_mount_flags(dev->od_mount->lmi_sb) & LMD_FLG_NOSCRUB)
- scrub->os_no_scrub = 1;
push_ctxt(&saved, ctxt, NULL);
filp = filp_open(osd_scrub_name, O_RDWR | O_CREAT, 0644);
}
}
- if (rc == 0 && !scrub->os_no_scrub &&
- ((sf->sf_status == SS_CRASHED &&
+ if (rc == 0 && !dev->od_noscrub &&
+ ((sf->sf_status == SS_PAUSED) ||
+ (sf->sf_status == SS_CRASHED &&
sf->sf_flags & (SF_RECREATED | SF_INCONSISTENT | SF_AUTO)) ||
(sf->sf_status == SS_INIT &&
sf->sf_flags & (SF_RECREATED | SF_INCONSISTENT))))
ENTRY;
/* od_otable_mutex: prevent curcurrent init/fini */
- cfs_mutex_lock(&dev->od_otable_mutex);
+ mutex_lock(&dev->od_otable_mutex);
if (dev->od_otable_it != NULL)
GOTO(out, it = ERR_PTR(-EALREADY));
GOTO(out, it);
out:
- cfs_mutex_unlock(&dev->od_otable_mutex);
+ mutex_unlock(&dev->od_otable_mutex);
return (struct dt_it *)it;
}
struct osd_device *dev = it->ooi_dev;
/* od_otable_mutex: prevent curcurrent init/fini */
- cfs_mutex_lock(&dev->od_otable_mutex);
+ mutex_lock(&dev->od_otable_mutex);
do_osd_scrub_stop(&dev->od_scrub);
LASSERT(dev->od_otable_it == it);
dev->od_otable_it = NULL;
- cfs_mutex_unlock(&dev->od_otable_mutex);
+ mutex_unlock(&dev->od_otable_mutex);
OBD_FREE_PTR(it);
}
/**
+ * XXX: Temporary used to notify otable iteration to be paused.
+ */
+static void osd_otable_it_put(const struct lu_env *env, struct dt_it *di)
+{
+ struct osd_device *dev = ((struct osd_otable_it *)di)->ooi_dev;
+
+ /* od_otable_mutex: prevent curcurrent init/fini */
+ mutex_lock(&dev->od_otable_mutex);
+ dev->od_scrub.os_paused = 1;
+ mutex_unlock(&dev->od_otable_mutex);
+}
+
+/**
* Set the OSD layer iteration start position as the specified key.
*
* The LFSCK out of OSD layer does not know the detail of the key, so if there
.dio_it = {
.init = osd_otable_it_init,
.fini = osd_otable_it_fini,
+ .put = osd_otable_it_put,
.get = osd_otable_it_get,
.next = osd_otable_it_next,
.key = osd_otable_it_key,
oii->oii_cache = *oic;
oii->oii_insert = insert;
- cfs_spin_lock(&scrub->os_lock);
+ spin_lock(&scrub->os_lock);
if (unlikely(!thread_is_running(thread))) {
- cfs_spin_unlock(&scrub->os_lock);
+ spin_unlock(&scrub->os_lock);
OBD_FREE_PTR(oii);
RETURN(-EAGAIN);
}
if (cfs_list_empty(&scrub->os_inconsistent_items))
wakeup = 1;
cfs_list_add_tail(&oii->oii_list, &scrub->os_inconsistent_items);
- cfs_spin_unlock(&scrub->os_lock);
+ spin_unlock(&scrub->os_lock);
if (wakeup != 0)
cfs_waitq_broadcast(&thread->t_ctl_waitq);
struct osd_inconsistent_item *oii;
ENTRY;
- cfs_spin_lock(&scrub->os_lock);
+ spin_lock(&scrub->os_lock);
cfs_list_for_each_entry(oii, &scrub->os_inconsistent_items, oii_list) {
if (lu_fid_eq(fid, &oii->oii_cache.oic_fid)) {
*id = oii->oii_cache.oic_lid;
- cfs_spin_unlock(&scrub->os_lock);
+ spin_unlock(&scrub->os_lock);
RETURN(0);
}
}
- cfs_spin_unlock(&scrub->os_lock);
+ spin_unlock(&scrub->os_lock);
RETURN(-ENOENT);
}
"scanning",
"completed",
"failed",
+ "stopped",
"paused",
"crashed",
NULL
int ret = -ENOSPC;
int rc;
- cfs_down_read(&scrub->os_rwsem);
+ down_read(&scrub->os_rwsem);
rc = snprintf(buf, len,
"name: OI scrub\n"
"magic: 0x%x\n"
"updated: "LPU64"\n"
"failed: "LPU64"\n"
"prior_updated: "LPU64"\n"
+ "noscrub: "LPU64"\n"
+ "igif: "LPU64"\n"
"success_count: %u\n",
checked, sf->sf_items_updated, sf->sf_items_failed,
- sf->sf_items_updated_prior, sf->sf_success_count);
+ sf->sf_items_updated_prior, sf->sf_items_noscrub,
+ sf->sf_items_igif, sf->sf_success_count);
if (rc <= 0)
goto out;
ret = save - len;
out:
- cfs_up_read(&scrub->os_rwsem);
+ up_read(&scrub->os_rwsem);
return ret;
}