-/* -*- 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) 2002, 2010, Oracle and/or its affiliates. All rights reserved.
* Use is subject to license terms.
+ *
+ * Copyright (c) 2011, 2012, Whamcloud, Inc.
*/
/*
* This file is part of Lustre, http://www.lustre.org/
struct mds_obd *mds = &obd->u.mds;
unsigned int i=0, j;
+ if ((libcfs_debug & D_INFO) == 0)
+ return;
+
CDEBUG(D_INFO, "dump from %s\n", label);
if (mds->mds_lov_page_dirty == NULL) {
CERROR("NULL bitmap!\n");
/* workaround - New target not in objids file; increase mdsize */
/* ld_tgt_count is used as the max index everywhere, despite its name. */
if (data[off] == 0) {
+ __u32 max_easize;
__u32 stripes;
+ max_easize = mds->mds_obt.obt_osd_properties.osd_max_ea_size;
data[off] = 1;
mds->mds_lov_objid_count++;
- stripes = min_t(__u32, LOV_MAX_STRIPE_COUNT,
- mds->mds_lov_objid_count);
+ stripes = min(lov_mds_md_stripecnt(max_easize, LOV_MAGIC_V3),
+ mds->mds_lov_objid_count);
mds->mds_max_mdsize = lov_mds_md_size(stripes, LOV_MAGIC_V3);
mds->mds_max_cookiesize = stripes * sizeof(struct llog_cookie);
return 0;
}
+static int mds_lov_objinit(struct mds_obd *mds, __u32 index)
+{
+ __u32 page = index / OBJID_PER_PAGE();
+ __u32 off = index % OBJID_PER_PAGE();
+ obd_id *data = mds->mds_lov_page_array[page];
+
+ return (data[off] > 0);
+}
+
int mds_lov_prepare_objids(struct obd_device *obd, struct lov_mds_md *lmm)
{
struct lov_ost_data_v1 *data;
- __u32 count;
+ __u16 count;
int rc = 0;
__u32 j;
switch (le32_to_cpu(lmm->lmm_magic)) {
case LOV_MAGIC_V1:
- count = le32_to_cpu(((struct lov_mds_md_v1*)lmm)->lmm_stripe_count);
+ count = le16_to_cpu(((struct lov_mds_md_v1*)lmm)->lmm_stripe_count);
data = &(((struct lov_mds_md_v1*)lmm)->lmm_objects[0]);
break;
case LOV_MAGIC_V3:
- count = le32_to_cpu(((struct lov_mds_md_v3*)lmm)->lmm_stripe_count);
+ count = le16_to_cpu(((struct lov_mds_md_v3*)lmm)->lmm_stripe_count);
data = &(((struct lov_mds_md_v3*)lmm)->lmm_objects[0]);
break;
default:
}
- cfs_mutex_down(&obd->obd_dev_sem);
+ cfs_mutex_lock(&obd->obd_dev_mutex);
for (j = 0; j < count; j++) {
__u32 i = le32_to_cpu(data[j].l_ost_idx);
if (mds_lov_update_max_ost(&obd->u.mds, i)) {
break;
}
}
- cfs_mutex_up(&obd->obd_dev_sem);
+ cfs_mutex_unlock(&obd->obd_dev_mutex);
RETURN(rc);
}
* after use
*/
static int mds_log_lost_precreated(struct obd_device *obd,
- struct lov_stripe_md **lsmp, int *stripes,
+ struct lov_stripe_md **lsmp, __u16 *stripes,
obd_id id, obd_count count, int idx)
{
struct lov_stripe_md *lsm = *lsmp;
}
lsm->lsm_oinfo[0]->loi_id = id;
- lsm->lsm_oinfo[0]->loi_gr = mdt_to_obd_objgrp(obd->u.mds.mds_id);
+ lsm->lsm_oinfo[0]->loi_seq = mdt_to_obd_objseq(obd->u.mds.mds_id);
lsm->lsm_oinfo[0]->loi_ost_idx = idx;
rc = mds_log_op_orphan(obd, lsm, count);
int j;
struct lov_ost_data_v1 *obj;
struct lov_stripe_md *lsm = NULL;
- int stripes = 0;
+ __u16 stripes = 0;
int count;
ENTRY;
switch (le32_to_cpu(lmm->lmm_magic)) {
case LOV_MAGIC_V1:
- count = le32_to_cpu(((struct lov_mds_md_v1*)lmm)->lmm_stripe_count);
+ count = le16_to_cpu(((struct lov_mds_md_v1*)lmm)->lmm_stripe_count);
obj = ((struct lov_mds_md_v1*)lmm)->lmm_objects;
break;
case LOV_MAGIC_V3:
- count = le32_to_cpu(((struct lov_mds_md_v3*)lmm)->lmm_stripe_count);
+ count = le16_to_cpu(((struct lov_mds_md_v3*)lmm)->lmm_stripe_count);
obj = ((struct lov_mds_md_v3*)lmm)->lmm_objects;
break;
default:
static int mds_lov_update_from_read(struct mds_obd *mds, obd_id *data,
__u32 count)
{
- __u32 i;
- __u32 stripes;
+ __u32 max_easize = mds->mds_obt.obt_osd_properties.osd_max_ea_size;
+ __u32 i, stripes;
for (i = 0; i < count; i++) {
if (data[i] == 0)
mds->mds_lov_objid_count++;
}
- stripes = min_t(__u32, LOV_MAX_STRIPE_COUNT,
+ stripes = min(lov_mds_md_stripecnt(max_easize, LOV_MAGIC_V3),
mds->mds_lov_objid_count);
mds->mds_max_mdsize = lov_mds_md_size(stripes, LOV_MAGIC_V3);
if (size == 0)
RETURN(0);
- page = (size / (OBJID_PER_PAGE() * sizeof(obd_id))) + 1;
+ page = (size + MDS_LOV_ALLOC_SIZE - 1) / MDS_LOV_ALLOC_SIZE;
CDEBUG(D_INFO, "file size %lu pages %d\n", size, page);
for (i = 0; i < page; i++) {
obd_id *data;
data = mds->mds_lov_page_array[i];
rc = fsfilt_read_record(obd, mds->mds_lov_objid_filp, data,
- OBJID_PER_PAGE()*sizeof(obd_id), &off);
+ MDS_LOV_ALLOC_SIZE, &off);
if (rc < 0) {
CERROR("Error reading objids %d\n", rc);
GOTO(out, rc);
}
+ if (off == off_old) /* hole is read */
+ off += MDS_LOV_ALLOC_SIZE;
- count += (off - off_old) / sizeof(obd_id);
+ count = (off - off_old) / sizeof(obd_id);
if (mds_lov_update_from_read(mds, data, count)) {
CERROR("Can't update mds data\n");
GOTO(out, rc = -EIO);
}
-
- if (off == off_old)
- break; /* eof */
}
- mds->mds_lov_objid_lastpage = i;
- mds->mds_lov_objid_lastidx = count % OBJID_PER_PAGE();
+ mds->mds_lov_objid_lastpage = page - 1;
+ mds->mds_lov_objid_lastidx = count - 1;
CDEBUG(D_INFO, "Read %u - %u %u objid\n", mds->mds_lov_objid_count,
mds->mds_lov_objid_lastpage, mds->mds_lov_objid_lastidx);
cfs_foreach_bit(mds->mds_lov_page_dirty, i) {
obd_id *data = mds->mds_lov_page_array[i];
- unsigned int size = OBJID_PER_PAGE()*sizeof(obd_id);
+ unsigned int size = MDS_LOV_ALLOC_SIZE;
loff_t off = i * size;
LASSERT(data != NULL);
unsigned int page;
unsigned int off;
obd_id *data;
- __u64 connect_flags;
__u32 size;
int rc = 0;
ENTRY;
off = idx % OBJID_PER_PAGE();
data = mds->mds_lov_page_array[page];
- size = sizeof(__u64);
- connect_flags = idx;
- rc = obd_get_info(lov_exp, sizeof(KEY_CONNECT_FLAG), KEY_CONNECT_FLAG,
- &size, &connect_flags, NULL);
- if (rc)
- GOTO(out, rc);
-
- if (data[off] < 2 || connect_flags & OBD_CONNECT_SKIP_ORPHAN) {
+ if (data[off] < 2) {
/* We never read this lastid; ask the osc */
struct obd_id_info lastid;
size = sizeof(lastid);
lastid.idx = idx;
lastid.data = &data[off];
- rc = obd_get_info(lov_exp, sizeof(KEY_LAST_ID), KEY_LAST_ID,
- &size, &lastid, NULL);
+ rc = obd_get_info(NULL, lov_exp, sizeof(KEY_LAST_ID),
+ KEY_LAST_ID, &size, &lastid, NULL);
if (rc)
GOTO(out, rc);
* objects above this ID, they will be removed. */
memset(&oa, 0, sizeof(oa));
oa.o_flags = OBD_FL_DELORPHAN;
- oa.o_gr = mdt_to_obd_objgrp(mds->mds_id);
+ oa.o_seq = mdt_to_obd_objseq(mds->mds_id);
oa.o_valid = OBD_MD_FLFLAGS | OBD_MD_FLGROUP;
if (ost_uuid != NULL)
oti.oti_ost_uuid = ost_uuid;
- rc = obd_create(mds->mds_lov_exp, &oa, &empty_ea, &oti);
+ rc = obd_create(NULL, mds->mds_lov_exp, &oa, &empty_ea, &oti);
RETURN(rc);
}
info.idx = idx;
info.data = id;
- rc = obd_set_info_async(mds->mds_lov_exp, sizeof(KEY_NEXT_ID),
+ rc = obd_set_info_async(NULL, mds->mds_lov_exp, sizeof(KEY_NEXT_ID),
KEY_NEXT_ID, sizeof(info), &info, NULL);
if (rc)
CERROR ("%s: mds_lov_set_nextid failed (%d)\n",
/* Update the lov desc for a new size lov. */
static int mds_lov_update_desc(struct obd_device *obd, int idx,
- struct obd_uuid *uuid, enum obd_notify_event ev)
+ struct obd_uuid *uuid)
{
struct mds_obd *mds = &obd->u.mds;
struct lov_desc *ld;
if (!ld)
RETURN(-ENOMEM);
- rc = obd_get_info(mds->mds_lov_exp, sizeof(KEY_LOVDESC), KEY_LOVDESC,
- &valsize, ld, NULL);
+ rc = obd_get_info(NULL, mds->mds_lov_exp, sizeof(KEY_LOVDESC),
+ KEY_LOVDESC, &valsize, ld, NULL);
if (rc)
GOTO(out, rc);
CDEBUG(D_CONFIG, "updated lov_desc, tgt_count: %d - idx %d / uuid %s\n",
mds->mds_lov_desc.ld_tgt_count, idx, uuid->uuid);
- cfs_mutex_down(&obd->obd_dev_sem);
+ cfs_mutex_lock(&obd->obd_dev_mutex);
rc = mds_lov_update_max_ost(mds, idx);
- cfs_mutex_up(&obd->obd_dev_sem);
+ cfs_mutex_unlock(&obd->obd_dev_mutex);
if (rc != 0)
GOTO(out, rc );
if (rc)
GOTO(out, rc);
- /*XXX this notifies the MDD until lov handling use old mds code */
- if (obd->obd_upcall.onu_owner) {
- LASSERT(obd->obd_upcall.onu_upcall != NULL);
- rc = obd->obd_upcall.onu_upcall(obd, NULL, ev,
- obd->obd_upcall.onu_owner,
- &mds->mds_mount_count);
- }
out:
OBD_FREE(ld, sizeof(*ld));
RETURN(rc);
/* Inform MDS about new/updated target */
static int mds_lov_update_mds(struct obd_device *obd,
struct obd_device *watched,
- __u32 idx, enum obd_notify_event ev)
+ __u32 idx)
{
struct mds_obd *mds = &obd->u.mds;
int rc = 0;
int page;
int off;
obd_id *data;
-
ENTRY;
- /* Don't let anyone else mess with mds_lov_objids now */
- rc = mds_lov_update_desc(obd, idx, &watched->u.cli.cl_target_uuid, ev);
- if (rc)
- GOTO(out, rc);
+ LASSERT(mds_lov_objinit(mds, idx));
CDEBUG(D_CONFIG, "idx=%d, recov=%d/%d, cnt=%d\n",
idx, obd->obd_recovering, obd->obd_async_recov,
RETURN(-ENOTCONN);
}
- cfs_mutex_down(&obd->obd_dev_sem);
+ cfs_mutex_lock(&obd->obd_dev_mutex);
rc = mds_lov_read_objids(obd);
- cfs_mutex_up(&obd->obd_dev_sem);
+ cfs_mutex_unlock(&obd->obd_dev_mutex);
if (rc) {
CERROR("cannot read %s: rc = %d\n", "lov_objids", rc);
GOTO(err_exit, rc);
GOTO(err_exit, rc);
}
- /* try init too early */
- rc = obd_llog_init(obd, &obd->obd_olg, obd, NULL);
- if (rc)
- GOTO(err_exit, rc);
+ /* ask lov to generate OBD_NOTIFY_CREATE events for already registered
+ * targets */
+ obd_notify(mds->mds_lov_obd, NULL, OBD_NOTIFY_CREATE, NULL);
mds->mds_lov_obd->u.lov.lov_sp_me = LUSTRE_SP_MDT;
data->ocd_connect_flags = OBD_CONNECT_VERSION | OBD_CONNECT_INDEX |
OBD_CONNECT_REQPORTAL | OBD_CONNECT_QUOTA64 |
- OBD_CONNECT_OSS_CAPA | OBD_CONNECT_FID |
- OBD_CONNECT_BRW_SIZE | OBD_CONNECT_CKSUM |
+ OBD_CONNECT_OSS_CAPA | OBD_CONNECT_FULL20 |
OBD_CONNECT_CHANGE_QS | OBD_CONNECT_AT |
OBD_CONNECT_MDS | OBD_CONNECT_SKIP_ORPHAN |
- OBD_CONNECT_SOM;
+ OBD_CONNECT_SOM | OBD_CONNECT_MAX_EASIZE;
#ifdef HAVE_LRU_RESIZE_SUPPORT
data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
#endif
data->ocd_version = LUSTRE_VERSION_CODE;
- data->ocd_group = mdt_to_obd_objgrp(mds->mds_id);
+ data->ocd_group = mdt_to_obd_objseq(mds->mds_id);
+ data->ocd_max_easize = mds->mds_obt.obt_osd_properties.osd_max_ea_size;
+
/* send max bytes per rpc */
data->ocd_brw_size = PTLRPC_MAX_BRW_PAGES << CFS_PAGE_SHIFT;
/* send the list of supported checksum types */
- data->ocd_cksum_types = OBD_CKSUM_ALL;
+ data->ocd_cksum_types = cksum_types_supported_client();
/* NB: lov_connect() needs to fill in .ocd_index for each OST */
rc = obd_connect(NULL, &mds->mds_lov_exp, mds->mds_lov_obd, &obd->obd_uuid, data, NULL);
OBD_FREE(data, sizeof(*data));
struct obd_device *mlsi_obd; /* the lov device to sync */
struct obd_device *mlsi_watched; /* target osc */
__u32 mlsi_index; /* index of target */
- enum obd_notify_event mlsi_ev; /* event type */
};
static int mds_propagate_capa_keys(struct mds_obd *mds, struct obd_uuid *uuid)
DEBUG_CAPA_KEY(D_SEC, key, "propagate");
info.capa = key;
- rc = obd_set_info_async(mds->mds_lov_exp, sizeof(KEY_CAPA_KEY),
- KEY_CAPA_KEY, sizeof(info), &info, NULL);
+ rc = obd_set_info_async(NULL, mds->mds_lov_exp,
+ sizeof(KEY_CAPA_KEY), KEY_CAPA_KEY,
+ sizeof(info), &info, NULL);
if (rc) {
DEBUG_CAPA_KEY(D_ERROR, key,
"propagate failed (rc = %d) for", rc);
struct mds_obd *mds = &obd->u.mds;
struct obd_uuid *uuid;
__u32 idx = mlsi->mlsi_index;
- enum obd_notify_event ev = mlsi->mlsi_ev;
struct mds_group_info mgi;
struct llog_ctxt *ctxt;
int rc = 0;
GOTO(out, rc = -ENODEV);
OBD_RACE(OBD_FAIL_MDS_LOV_SYNC_RACE);
- rc = mds_lov_update_mds(obd, watched, idx, ev);
+ rc = mds_lov_update_mds(obd, watched, idx);
if (rc != 0) {
CERROR("%s failed at update_mds: %d\n", obd_uuid2str(uuid), rc);
GOTO(out, rc);
}
- mgi.group = mdt_to_obd_objgrp(mds->mds_id);
+ mgi.group = mdt_to_obd_objseq(mds->mds_id);
mgi.uuid = uuid;
- rc = obd_set_info_async(mds->mds_lov_exp, sizeof(KEY_MDS_CONN),
+ rc = obd_set_info_async(NULL, mds->mds_lov_exp, sizeof(KEY_MDS_CONN),
KEY_MDS_CONN, sizeof(mgi), &mgi, NULL);
if (rc != 0)
GOTO(out, rc);
LCONSOLE_INFO("MDS %s: %s now active, resetting orphans\n",
obd->obd_name, obd_uuid2str(uuid));
+
rc = mds_lov_clear_orphans(mds, uuid);
if (rc != 0) {
CERROR("%s failed at mds_lov_clear_orphans: %d\n",
}
#ifdef HAVE_QUOTA_SUPPORT
- if (obd->obd_upcall.onu_owner) {
+ if (obd->obd_upcall.onu_owner) {
/*
* This is a hack for mds_notify->mdd_notify. When the mds obd
* in mdd is removed, This hack should be removed.
#endif
EXIT;
out:
- cfs_up_read(&mds->mds_notify_lock);
if (rc) {
/* Deactivate it for safety */
CERROR("%s sync failed %d, deactivating\n", obd_uuid2str(uuid),
obd_notify(mds->mds_lov_obd, watched,
OBD_NOTIFY_INACTIVE, NULL);
}
+ cfs_up_read(&mds->mds_notify_lock);
class_decref(obd, "mds_lov_synchronize", obd);
return rc;
mlsi->mlsi_obd = obd;
mlsi->mlsi_watched = watched;
mlsi->mlsi_index = *(__u32 *)data;
- mlsi->mlsi_ev = ev;
/* Although class_export_get(obd->obd_self_export) would lock
the MDS in place, since it's only a self-export
if (ev != OBD_NOTIFY_SYNC) {
/* Synchronize in the background */
- rc = cfs_kernel_thread(mds_lov_synchronize, mlsi,
- CLONE_VM | CLONE_FILES);
+ rc = cfs_create_thread(mds_lov_synchronize, mlsi,
+ CFS_DAEMON_FLAGS);
if (rc < 0) {
CERROR("%s: error starting mds_lov_synchronize: %d\n",
obd->obd_name, rc);
int mds_notify(struct obd_device *obd, struct obd_device *watched,
enum obd_notify_event ev, void *data)
{
+ struct mds_obd *mds = &obd->u.mds;
int rc = 0;
ENTRY;
CDEBUG(D_CONFIG, "notify %s ev=%d\n", watched->obd_name, ev);
+ if (strcmp(watched->obd_type->typ_name, LUSTRE_OSC_NAME) != 0) {
+ CERROR("unexpected notification of %s %s!\n",
+ watched->obd_type->typ_name, watched->obd_name);
+ RETURN(-EINVAL);
+ }
+
+ /*XXX this notifies the MDD until lov handling use old mds code
+ * must non block!
+ */
+ if (obd->obd_upcall.onu_owner) {
+ LASSERT(obd->obd_upcall.onu_upcall != NULL);
+ rc = obd->obd_upcall.onu_upcall(obd, NULL, ev,
+ obd->obd_upcall.onu_owner,
+ &mds->mds_obt.obt_mount_count);
+ }
+
switch (ev) {
/* We only handle these: */
+ case OBD_NOTIFY_CREATE:
+ CDEBUG(D_CONFIG, "%s: add target %s\n", obd->obd_name,
+ obd_uuid2str(&watched->u.cli.cl_target_uuid));
+ /* We still have to fix the lov descriptor for ost's */
+ LASSERT(data);
+ rc = mds_lov_update_desc(obd, *(__u32 *)data,
+ &watched->u.cli.cl_target_uuid);
+ RETURN(rc);
case OBD_NOTIFY_ACTIVE:
/* lov want one or more _active_ targets for work */
/* activate event should be pass lov idx as argument */
RETURN(0);
}
- if (strcmp(watched->obd_type->typ_name, LUSTRE_OSC_NAME) != 0) {
- CERROR("unexpected notification of %s %s!\n",
- watched->obd_type->typ_name, watched->obd_name);
- RETURN(-EINVAL);
- }
-
if (obd->obd_recovering) {
- CWARN("MDS %s: in recovery, not resetting orphans on %s\n",
- obd->obd_name,
- obd_uuid2str(&watched->u.cli.cl_target_uuid));
+ CDEBUG(D_CONFIG, "%s: Is in recovery, "
+ "not resetting orphans on %s\n",
+ obd->obd_name,
+ obd_uuid2str(&watched->u.cli.cl_target_uuid));
/* We still have to fix the lov descriptor for ost's added
after the mdt in the config log. They didn't make it into
mds_lov_connect. */
rc = mds_lov_update_desc(obd, *(__u32 *)data,
- &watched->u.cli.cl_target_uuid, ev);
+ &watched->u.cli.cl_target_uuid);
} else {
rc = mds_lov_start_synchronize(obd, watched, data, ev);
}
git://git.whamcloud.com / fs / lustre-release.git / blobdiff