* Copyright 2009 Sun Microsystems, Inc. All rights reserved
* Use is subject to license terms.
*
- * Copyright (c) 2012, 2016, Intel Corporation.
+ * Copyright (c) 2012, 2017, Intel Corporation.
*/
/*
* This file is part of Lustre, http://www.lustre.org/
#include <obd_class.h>
#include <md_object.h>
#include <lustre_fid.h>
-#include <lustre_param.h>
+#include <uapi/linux/lustre/lustre_param.h>
#include <lustre_update.h>
#include <lustre_log.h>
+#include <lustre_lmv.h>
#include "lod_internal.h"
* \retval negative negated errno on error
**/
int lod_fld_lookup(const struct lu_env *env, struct lod_device *lod,
- const struct lu_fid *fid, __u32 *tgt, int *type)
+ const struct lu_fid *fid, u32 *tgt, int *type)
{
- struct lu_seq_range range = { 0 };
- struct lu_server_fld *server_fld;
+ struct lu_seq_range range = { 0 };
+ struct lu_server_fld *server_fld;
int rc;
+
ENTRY;
if (!fid_is_sane(fid)) {
}
server_fld = lu_site2seq(lod2lu_dev(lod)->ld_site)->ss_server_fld;
- if (server_fld == NULL)
+ if (!server_fld)
RETURN(-EIO);
fld_range_set_type(&range, *type);
const struct lu_object_header *hdr,
struct lu_device *dev)
{
- struct lod_object *lod_obj;
- struct lu_object *lu_obj;
+ struct lod_object *lod_obj;
+ struct lu_object *lu_obj;
+
ENTRY;
OBD_SLAB_ALLOC_PTR_GFP(lod_obj, lod_object_kmem, GFP_NOFS);
- if (lod_obj == NULL)
+ if (!lod_obj)
RETURN(ERR_PTR(-ENOMEM));
+ mutex_init(&lod_obj->ldo_layout_mutex);
lu_obj = lod2lu_obj(lod_obj);
dt_object_init(&lod_obj->ldo_obj, NULL, dev);
lod_obj->ldo_obj.do_ops = &lod_obj_ops;
struct lod_tgt_descs *ltd,
struct lustre_cfg *lcfg)
{
- struct lu_device *next;
+ struct lu_device *next;
+ struct lu_tgt_desc *tgt;
int rc = 0;
- unsigned int i;
lod_getref(ltd);
- if (ltd->ltd_tgts_size <= 0) {
- lod_putref(lod, ltd);
- return 0;
- }
- cfs_foreach_bit(ltd->ltd_tgt_bitmap, i) {
- struct lod_tgt_desc *tgt;
+ ltd_foreach_tgt(ltd, tgt) {
int rc1;
- tgt = LTD_TGT(ltd, i);
LASSERT(tgt && tgt->ltd_tgt);
next = &tgt->ltd_tgt->dd_lu_dev;
rc1 = next->ld_ops->ldo_process_config(env, next, lcfg);
if (rc1) {
- CERROR("%s: error cleaning up LOD index %u: cmd %#x"
- ": rc = %d\n", lod2obd(lod)->obd_name, i,
+ CERROR("%s: error cleaning up LOD index %u: cmd %#x : rc = %d\n",
+ lod2obd(lod)->obd_name, tgt->ltd_index,
lcfg->lcfg_command, rc1);
rc = rc1;
}
struct lod_recovery_data {
struct lod_device *lrd_lod;
struct lod_tgt_desc *lrd_ltd;
- struct ptlrpc_thread *lrd_thread;
- __u32 lrd_idx;
+ struct task_struct **lrd_task;
+ u32 lrd_idx;
+ struct lu_env lrd_env;
+ struct completion *lrd_started;
};
struct llog_rec_hdr *rec,
void *data)
{
- struct lod_recovery_data *lrd = data;
- struct llog_cookie *cookie = &lod_env_info(env)->lti_cookie;
- struct lu_target *lut;
- __u32 index = 0;
+ struct lod_recovery_data *lrd = data;
+ struct llog_cookie *cookie = &lod_env_info(env)->lti_cookie;
+ struct lu_target *lut;
+ u32 index = 0;
+
ENTRY;
- if (lrd->lrd_ltd == NULL) {
+ if (!lrd->lrd_ltd) {
int rc;
rc = lodname2mdt_index(lod2obd(lrd->lrd_lod)->obd_name, &index);
if (rec->lrh_len !=
llog_update_record_size((struct llog_update_record *)rec)) {
- CERROR("%s broken update record! index %u "DFID".%u :"
- " rc = %d\n", lod2obd(lrd->lrd_lod)->obd_name, index,
+ CERROR("%s: broken update record! index %u "DFID".%u: rc = %d\n",
+ lod2obd(lrd->lrd_lod)->obd_name, index,
PFID(&llh->lgh_id.lgl_oi.oi_fid), rec->lrh_index, -EIO);
return -EINVAL;
}
*/
static int lod_sub_recovery_thread(void *arg)
{
- struct lod_recovery_data *lrd = arg;
- struct lod_device *lod = lrd->lrd_lod;
- struct dt_device *dt;
- struct ptlrpc_thread *thread = lrd->lrd_thread;
- struct llog_ctxt *ctxt = NULL;
- struct lu_env env;
+ struct lod_recovery_data *lrd = arg;
+ struct lod_device *lod = lrd->lrd_lod;
+ struct dt_device *dt;
+ struct llog_ctxt *ctxt = NULL;
+ struct lu_env *env = &lrd->lrd_env;
struct lu_target *lut;
+ struct lu_tgt_desc *mdt = NULL;
+ time64_t start;
+ int retries = 0;
+ int rc;
-
- int rc;
ENTRY;
- thread->t_flags = SVC_RUNNING;
- wake_up(&thread->t_ctl_waitq);
-
- rc = lu_env_init(&env, LCT_LOCAL | LCT_MD_THREAD);
- if (rc != 0) {
- OBD_FREE_PTR(lrd);
- CERROR("%s: can't initialize env: rc = %d\n",
- lod2obd(lod)->obd_name, rc);
- RETURN(rc);
- }
-
lut = lod2lu_dev(lod)->ld_site->ls_tgt;
atomic_inc(&lut->lut_tdtd->tdtd_recovery_threads_count);
- if (lrd->lrd_ltd == NULL)
+ if (!lrd->lrd_ltd)
dt = lod->lod_child;
else
dt = lrd->lrd_ltd->ltd_tgt;
+ start = ktime_get_real_seconds();
+ complete(lrd->lrd_started);
+
again:
- rc = lod_sub_prep_llog(&env, lod, dt, lrd->lrd_idx);
+
+ if (unlikely(OBD_FAIL_PRECHECK(OBD_FAIL_TGT_RECOVERY_CONNECT)) &&
+ lrd->lrd_ltd) {
+ OBD_FAIL_TIMEOUT(OBD_FAIL_TGT_RECOVERY_CONNECT, cfs_fail_val);
+ rc = -EIO;
+ } else {
+ rc = lod_sub_prep_llog(env, lod, dt, lrd->lrd_idx);
+ }
if (!rc && !lod->lod_child->dd_rdonly) {
/* Process the recovery record */
ctxt = llog_get_context(dt->dd_lu_dev.ld_obd,
LASSERT(ctxt != NULL);
LASSERT(ctxt->loc_handle != NULL);
- rc = llog_cat_process(&env, ctxt->loc_handle,
+ rc = llog_cat_process(env, ctxt->loc_handle,
lod_process_recovery_updates, lrd, 0, 0);
}
struct lu_device *top_device;
top_device = lod->lod_dt_dev.dd_lu_dev.ld_site->ls_top_dev;
- /* Because the remote target might failover at the same time,
- * let's retry here */
+ /*
+ * Because the remote target might failover at the same time,
+ * let's retry here
+ */
if ((rc == -ETIMEDOUT || rc == -EAGAIN || rc == -EIO) &&
dt != lod->lod_child &&
!top_device->ld_obd->obd_abort_recovery &&
!top_device->ld_obd->obd_stopping) {
- if (ctxt != NULL) {
- if (ctxt->loc_handle != NULL)
- llog_cat_close(&env,
+ if (ctxt) {
+ if (ctxt->loc_handle)
+ llog_cat_close(env,
ctxt->loc_handle);
llog_ctxt_put(ctxt);
}
+ retries++;
+ CDEBUG(D_HA, "%s get update log failed %d, retry\n",
+ dt->dd_lu_dev.ld_obd->obd_name, rc);
goto again;
}
- CERROR("%s getting update log failed: rc = %d\n",
+ CERROR("%s get update log failed: rc = %d\n",
dt->dd_lu_dev.ld_obd->obd_name, rc);
llog_ctxt_put(ctxt);
}
llog_ctxt_put(ctxt);
- CDEBUG(D_HA, "%s retrieve update log: rc = %d\n",
- dt->dd_lu_dev.ld_obd->obd_name, rc);
+ CDEBUG(D_HA, "%s retrieved update log, duration %lld, retries %d\n",
+ dt->dd_lu_dev.ld_obd->obd_name, ktime_get_real_seconds() - start,
+ retries);
- if (lrd->lrd_ltd == NULL)
+ spin_lock(&lod->lod_lock);
+ if (!lrd->lrd_ltd)
lod->lod_child_got_update_log = 1;
else
lrd->lrd_ltd->ltd_got_update_log = 1;
- if (lod->lod_child_got_update_log) {
- struct lod_tgt_descs *ltd = &lod->lod_mdt_descs;
- struct lod_tgt_desc *tgt = NULL;
- bool all_got_log = true;
- int i;
-
- cfs_foreach_bit(ltd->ltd_tgt_bitmap, i) {
- tgt = LTD_TGT(ltd, i);
- if (!tgt->ltd_got_update_log) {
- all_got_log = false;
- break;
- }
- }
+ if (!lod->lod_child_got_update_log) {
+ spin_unlock(&lod->lod_lock);
+ GOTO(out, rc = 0);
+ }
- if (all_got_log) {
- CDEBUG(D_HA, "%s got update logs from all MDTs.\n",
- lut->lut_obd->obd_name);
- lut->lut_tdtd->tdtd_replay_ready = 1;
- wake_up(&lut->lut_obd->obd_next_transno_waitq);
+ lod_foreach_mdt(lod, mdt) {
+ if (!mdt->ltd_got_update_log) {
+ spin_unlock(&lod->lod_lock);
+ GOTO(out, rc = 0);
}
}
+ lut->lut_tdtd->tdtd_replay_ready = 1;
+ spin_unlock(&lod->lod_lock);
+
+ CDEBUG(D_HA, "%s got update logs from all MDTs.\n",
+ lut->lut_obd->obd_name);
+ wake_up(&lut->lut_obd->obd_next_transno_waitq);
+ EXIT;
out:
- OBD_FREE_PTR(lrd);
- thread->t_flags = SVC_STOPPED;
atomic_dec(&lut->lut_tdtd->tdtd_recovery_threads_count);
wake_up(&lut->lut_tdtd->tdtd_recovery_threads_waitq);
- wake_up(&thread->t_ctl_waitq);
- lu_env_fini(&env);
- RETURN(rc);
+ if (xchg(lrd->lrd_task, NULL) == NULL)
+ /* Someone is waiting for us to finish, need
+ * to synchronize cleanly.
+ */
+ wait_var_event(lrd, kthread_should_stop());
+ lu_env_fini(env);
+ OBD_FREE_PTR(lrd);
+ return 0;
}
/**
* \param[in] thread recovery thread on this sub device
*/
void lod_sub_fini_llog(const struct lu_env *env,
- struct dt_device *dt, struct ptlrpc_thread *thread)
+ struct dt_device *dt, struct task_struct **thread)
{
- struct obd_device *obd;
- struct llog_ctxt *ctxt;
+ struct obd_device *obd;
+ struct llog_ctxt *ctxt;
+ struct task_struct *task = NULL;
+
ENTRY;
obd = dt->dd_lu_dev.ld_obd;
CDEBUG(D_INFO, "%s: finish sub llog\n", obd->obd_name);
- /* Stop recovery thread first */
- if (thread != NULL && thread->t_flags & SVC_RUNNING) {
- thread->t_flags = SVC_STOPPING;
- wake_up(&thread->t_ctl_waitq);
- wait_event(thread->t_ctl_waitq, thread->t_flags & SVC_STOPPED);
- }
+ /* Wait for recovery thread to complete */
+ if (thread)
+ task = xchg(thread, NULL);
+ if (task)
+ kthread_stop(task);
ctxt = llog_get_context(obd, LLOG_UPDATELOG_ORIG_CTXT);
- if (ctxt == NULL)
+ if (!ctxt)
RETURN_EXIT;
- if (ctxt->loc_handle != NULL)
+ if (ctxt->loc_handle)
llog_cat_close(env, ctxt->loc_handle);
llog_cleanup(env, ctxt);
* \retval 0 on success
* \retval -EINVAL if the name is invalid
*/
-int lodname2mdt_index(char *lodname, __u32 *mdt_index)
+int lodname2mdt_index(char *lodname, u32 *mdt_index)
{
- unsigned long index;
- char *ptr, *tmp;
+ u32 index;
+ const char *ptr, *tmp;
+ int rc;
/* 1.8 configs don't have "-MDT0000" at the end */
ptr = strstr(lodname, "-MDT");
- if (ptr == NULL) {
+ if (!ptr) {
*mdt_index = 0;
return 0;
}
ptr = strrchr(lodname, '-');
- if (ptr == NULL) {
- CERROR("invalid MDT index in '%s'\n", lodname);
- return -EINVAL;
+ if (!ptr) {
+ rc = -EINVAL;
+ CERROR("invalid MDT index in '%s': rc = %d\n", lodname, rc);
+ return rc;
}
if (strncmp(ptr, "-mdtlov", 7) != 0) {
- CERROR("invalid MDT index in '%s'\n", lodname);
- return -EINVAL;
+ rc = -EINVAL;
+ CERROR("invalid MDT index in '%s': rc = %d\n", lodname, rc);
+ return rc;
}
if ((unsigned long)ptr - (unsigned long)lodname <= 8) {
- CERROR("invalid MDT index in '%s'\n", lodname);
- return -EINVAL;
+ rc = -EINVAL;
+ CERROR("invalid MDT index in '%s': rc = %d\n", lodname, rc);
+ return rc;
}
if (strncmp(ptr - 8, "-MDT", 4) != 0) {
- CERROR("invalid MDT index in '%s'\n", lodname);
- return -EINVAL;
+ rc = -EINVAL;
+ CERROR("invalid MDT index in '%s': rc = %d\n", lodname, rc);
+ return rc;
}
- index = simple_strtol(ptr - 4, &tmp, 16);
- if (*tmp != '-' || index > INT_MAX) {
- CERROR("invalid MDT index in '%s'\n", lodname);
- return -EINVAL;
+ rc = target_name2index(ptr - 7, &index, &tmp);
+ if (rc < 0 || rc & LDD_F_SV_ALL || *tmp != '-') {
+ rc = -EINVAL;
+ CERROR("invalid MDT index in '%s': rc = %d\n", lodname, rc);
+ return rc;
}
*mdt_index = index;
return 0;
int lod_sub_init_llog(const struct lu_env *env, struct lod_device *lod,
struct dt_device *dt)
{
- struct obd_device *obd;
- struct lod_recovery_data *lrd = NULL;
- struct ptlrpc_thread *thread;
- struct task_struct *task;
- struct l_wait_info lwi = { 0 };
- struct lod_tgt_desc *sub_ltd = NULL;
- __u32 index;
- __u32 master_index;
- int rc;
+ struct obd_device *obd;
+ struct lod_recovery_data *lrd = NULL;
+ DECLARE_COMPLETION_ONSTACK(started);
+ struct task_struct **taskp;
+ struct task_struct *task;
+ struct lod_tgt_desc *subtgt = NULL;
+ u32 index;
+ u32 master_index;
+ int rc;
+
ENTRY;
rc = lodname2mdt_index(lod2obd(lod)->obd_name, &master_index);
RETURN(rc);
OBD_ALLOC_PTR(lrd);
- if (lrd == NULL)
+ if (!lrd)
RETURN(-ENOMEM);
if (lod->lod_child == dt) {
- thread = &lod->lod_child_recovery_thread;
+ taskp = &lod->lod_child_recovery_task;
index = master_index;
} else {
- struct lod_tgt_descs *ltd = &lod->lod_mdt_descs;
- struct lod_tgt_desc *tgt = NULL;
- unsigned int i;
-
- cfs_foreach_bit(ltd->ltd_tgt_bitmap, i) {
- tgt = LTD_TGT(ltd, i);
- if (tgt->ltd_tgt == dt) {
- index = tgt->ltd_index;
- sub_ltd = tgt;
+ struct lu_tgt_desc *mdt;
+
+ lod_foreach_mdt(lod, mdt) {
+ if (mdt->ltd_tgt == dt) {
+ index = mdt->ltd_index;
+ subtgt = mdt;
break;
}
}
- LASSERT(sub_ltd != NULL);
- OBD_ALLOC_PTR(sub_ltd->ltd_recovery_thread);
- if (sub_ltd->ltd_recovery_thread == NULL)
- GOTO(free_lrd, rc = -ENOMEM);
-
- thread = sub_ltd->ltd_recovery_thread;
+ LASSERT(subtgt != NULL);
+ taskp = &subtgt->ltd_recovery_task;
}
CDEBUG(D_INFO, "%s init sub log %s\n", lod2obd(lod)->obd_name,
dt->dd_lu_dev.ld_obd->obd_name);
lrd->lrd_lod = lod;
- lrd->lrd_ltd = sub_ltd;
- lrd->lrd_thread = thread;
+ lrd->lrd_ltd = subtgt;
+ lrd->lrd_task = taskp;
lrd->lrd_idx = index;
- init_waitqueue_head(&thread->t_ctl_waitq);
+ lrd->lrd_started = &started;
obd = dt->dd_lu_dev.ld_obd;
obd->obd_lvfs_ctxt.dt = dt;
if (rc < 0) {
CERROR("%s: cannot setup updatelog llog: rc = %d\n",
obd->obd_name, rc);
- GOTO(free_thread, rc);
+ GOTO(free_lrd, rc);
+ }
+
+ rc = lu_env_init(&lrd->lrd_env, LCT_LOCAL | LCT_MD_THREAD);
+ if (rc != 0) {
+ CERROR("%s: can't initialize env: rc = %d\n",
+ lod2obd(lod)->obd_name, rc);
+ GOTO(free_lrd, rc);
}
/* Start the recovery thread */
- task = kthread_run(lod_sub_recovery_thread, lrd, "lod%04x_rec%04x",
- master_index, index);
+ task = kthread_create(lod_sub_recovery_thread, lrd, "lod%04x_rec%04x",
+ master_index, index);
if (IS_ERR(task)) {
rc = PTR_ERR(task);
CERROR("%s: cannot start recovery thread: rc = %d\n",
obd->obd_name, rc);
+ lu_env_fini(&lrd->lrd_env);
GOTO(out_llog, rc);
}
-
- l_wait_event(thread->t_ctl_waitq, thread->t_flags & SVC_RUNNING ||
- thread->t_flags & SVC_STOPPED, &lwi);
+ *taskp = task;
+ wake_up_process(task);
+ wait_for_completion(&started);
RETURN(0);
out_llog:
- lod_sub_fini_llog(env, dt, thread);
-free_thread:
- if (lod->lod_child != dt) {
- OBD_FREE_PTR(sub_ltd->ltd_recovery_thread);
- sub_ltd->ltd_recovery_thread = NULL;
- }
+ lod_sub_fini_llog(env, dt, taskp);
free_lrd:
OBD_FREE_PTR(lrd);
RETURN(rc);
static void lod_sub_stop_recovery_threads(const struct lu_env *env,
struct lod_device *lod)
{
- struct lod_tgt_descs *ltd = &lod->lod_mdt_descs;
- struct ptlrpc_thread *thread;
- unsigned int i;
-
- /* Stop the update log commit cancel threads and finish master
- * llog ctxt */
- thread = &lod->lod_child_recovery_thread;
- /* Stop recovery thread first */
- if (thread != NULL && thread->t_flags & SVC_RUNNING) {
- thread->t_flags = SVC_STOPPING;
- wake_up(&thread->t_ctl_waitq);
- wait_event(thread->t_ctl_waitq, thread->t_flags & SVC_STOPPED);
- }
+ struct task_struct *task;
+ struct lu_tgt_desc *mdt;
- lod_getref(ltd);
- cfs_foreach_bit(ltd->ltd_tgt_bitmap, i) {
- struct lod_tgt_desc *tgt;
-
- tgt = LTD_TGT(ltd, i);
- thread = tgt->ltd_recovery_thread;
- if (thread != NULL && thread->t_flags & SVC_RUNNING) {
- thread->t_flags = SVC_STOPPING;
- wake_up(&thread->t_ctl_waitq);
- wait_event(thread->t_ctl_waitq,
- thread->t_flags & SVC_STOPPED);
- OBD_FREE_PTR(tgt->ltd_recovery_thread);
- tgt->ltd_recovery_thread = NULL;
- }
- }
+ /*
+ * Stop the update log commit cancel threads and finish master
+ * llog ctxt
+ */
+ task = xchg(&lod->lod_child_recovery_task, NULL);
+ if (task)
+ kthread_stop(task);
- lod_putref(lod, ltd);
+ lod_getref(&lod->lod_mdt_descs);
+ lod_foreach_mdt(lod, mdt) {
+ task = xchg(&mdt->ltd_recovery_task, NULL);
+ if (task)
+ kthread_stop(task);
+ }
+ lod_putref(lod, &lod->lod_mdt_descs);
}
/**
static void lod_sub_fini_all_llogs(const struct lu_env *env,
struct lod_device *lod)
{
- struct lod_tgt_descs *ltd = &lod->lod_mdt_descs;
- unsigned int i;
+ struct lu_tgt_desc *mdt;
- /* Stop the update log commit cancel threads and finish master
- * llog ctxt */
+ /*
+ * Stop the update log commit cancel threads and finish master
+ * llog ctxt
+ */
lod_sub_fini_llog(env, lod->lod_child,
- &lod->lod_child_recovery_thread);
- lod_getref(ltd);
- cfs_foreach_bit(ltd->ltd_tgt_bitmap, i) {
- struct lod_tgt_desc *tgt;
-
- tgt = LTD_TGT(ltd, i);
- lod_sub_fini_llog(env, tgt->ltd_tgt,
- tgt->ltd_recovery_thread);
- }
-
- lod_putref(lod, ltd);
+ &lod->lod_child_recovery_task);
+ lod_getref(&lod->lod_mdt_descs);
+ lod_foreach_mdt(lod, mdt)
+ lod_sub_fini_llog(env, mdt->ltd_tgt,
+ &mdt->ltd_recovery_task);
+ lod_putref(lod, &lod->lod_mdt_descs);
}
static char *lod_show_update_logs_retrievers(void *data, int *size, int *count)
{
- struct lod_device *lod = (struct lod_device *)data;
- struct lu_target *lut = lod2lu_dev(lod)->ld_site->ls_tgt;
- struct lod_tgt_descs *ltd = &lod->lod_mdt_descs;
- struct lod_tgt_desc *tgt = NULL;
- char *buf;
- int len = 0;
- int rc;
- int i;
+ struct lod_device *lod = (struct lod_device *)data;
+ struct lu_target *lut = lod2lu_dev(lod)->ld_site->ls_tgt;
+ struct lu_tgt_desc *mdt = NULL;
+ char *buf;
+ int len = 0;
+ int rc;
+ int i;
*count = atomic_read(&lut->lut_tdtd->tdtd_recovery_threads_count);
if (*count == 0) {
*size = 5 * *count + 1;
OBD_ALLOC(buf, *size);
- if (buf == NULL)
+ if (!buf)
return NULL;
*count = 0;
rc = lodname2mdt_index(lod2obd(lod)->obd_name, &i);
LASSERTF(rc == 0, "Fail to parse target index: rc = %d\n", rc);
- rc = snprintf(buf + len, *size - len, " %04x", i);
+ rc = scnprintf(buf + len, *size - len, " %04x", i);
LASSERT(rc > 0);
len += rc;
(*count)++;
}
- cfs_foreach_bit(ltd->ltd_tgt_bitmap, i) {
- tgt = LTD_TGT(ltd, i);
- if (!tgt->ltd_got_update_log) {
- rc = snprintf(buf + len, *size - len, " %04x", i);
+ lod_foreach_mdt(lod, mdt) {
+ if (!mdt->ltd_got_update_log) {
+ rc = scnprintf(buf + len, *size - len, " %04x",
+ mdt->ltd_index);
if (unlikely(rc <= 0))
break;
struct lod_device *lod)
{
struct target_distribute_txn_data *tdtd;
- struct lu_target *lut;
- int rc;
+ struct lu_target *lut;
+ int rc;
+
ENTRY;
/* Init update recovery data */
OBD_ALLOC_PTR(tdtd);
- if (tdtd == NULL)
+ if (!tdtd)
RETURN(-ENOMEM);
lut = lod2lu_dev(lod)->ld_site->ls_tgt;
static void lod_fini_distribute_txn(const struct lu_env *env,
struct lod_device *lod)
{
- struct lu_target *lut;
+ struct lu_target *lut;
lut = lod2lu_dev(lod)->ld_site->ls_tgt;
target_recovery_fini(lut->lut_obd);
- if (lut->lut_tdtd == NULL)
+ if (!lut->lut_tdtd)
return;
distribute_txn_fini(env, lut->lut_tdtd);
struct lustre_cfg *lcfg)
{
struct lod_device *lod = lu2lod_dev(dev);
- struct lu_device *next = &lod->lod_child->dd_lu_dev;
- char *arg1;
- int rc = 0;
+ struct lu_device *next = &lod->lod_child->dd_lu_dev;
+ char *arg1;
+ int rc = 0;
+
ENTRY;
- switch(lcfg->lcfg_command) {
+ switch (lcfg->lcfg_command) {
case LCFG_LOV_DEL_OBD:
case LCFG_LOV_ADD_INA:
case LCFG_LOV_ADD_OBD:
case LCFG_ADD_MDC: {
- __u32 index;
- __u32 mdt_index;
+ u32 index;
+ u32 mdt_index;
int gen;
- /* lov_modify_tgts add 0:lov_mdsA 1:osp 2:0 3:1
+ /*
+ * lov_modify_tgts add 0:lov_mdsA 1:osp 2:0 3:1
* modify_mdc_tgts add 0:lustre-MDT0001
* 1:lustre-MDT0001-mdc0002
- * 2:2 3:1*/
+ * 2:2 3:1
+ */
arg1 = lustre_cfg_string(lcfg, 1);
if (sscanf(lustre_cfg_buf(lcfg, 2), "%d", &index) != 1)
GOTO(out, rc = -EINVAL);
if (lcfg->lcfg_command == LCFG_LOV_ADD_OBD) {
- __u32 mdt_index;
+ u32 mdt_index;
rc = lodname2mdt_index(lustre_cfg_string(lcfg, 0),
&mdt_index);
rc = lod_add_device(env, lod, arg1, index, gen,
mdt_index, LUSTRE_OSC_NAME, 0);
} else {
- rc = lod_del_device(env, lod,
- &lod->lod_ost_descs,
- arg1, index, gen, true);
+ rc = lod_del_device(env, lod, &lod->lod_ost_descs,
+ arg1, index, gen);
}
break;
case LCFG_PARAM: {
struct obd_device *obd;
+ ssize_t count;
char *param;
- /* Check if it is activate/deactivate mdc
- * lustre-MDTXXXX-osp-MDTXXXX.active=1 */
+ /*
+ * Check if it is activate/deactivate mdc
+ * lustre-MDTXXXX-osp-MDTXXXX.active=1
+ */
param = lustre_cfg_buf(lcfg, 1);
- if (strstr(param, "osp") != NULL &&
- strstr(param, ".active=") != NULL) {
- struct lod_tgt_descs *ltd = &lod->lod_mdt_descs;
- struct lod_tgt_desc *sub_tgt = NULL;
+ if (strstr(param, "osp") && strstr(param, ".active=")) {
+ struct lod_tgt_desc *sub_tgt = NULL;
+ struct lu_tgt_desc *mdt;
char *ptr;
char *tmp;
- int i;
ptr = strstr(param, ".");
*ptr = '\0';
obd = class_name2obd(param);
- if (obd == NULL) {
+ if (!obd) {
CERROR("%s: can not find %s: rc = %d\n",
lod2obd(lod)->obd_name, param, -EINVAL);
*ptr = '.';
GOTO(out, rc);
}
- cfs_foreach_bit(ltd->ltd_tgt_bitmap, i) {
- struct lod_tgt_desc *tgt;
-
- tgt = LTD_TGT(ltd, i);
- if (tgt->ltd_tgt->dd_lu_dev.ld_obd == obd) {
- sub_tgt = tgt;
+ lod_foreach_mdt(lod, mdt) {
+ if (mdt->ltd_tgt->dd_lu_dev.ld_obd == obd) {
+ sub_tgt = mdt;
break;
}
}
- if (sub_tgt == NULL) {
+ if (!sub_tgt) {
CERROR("%s: can not find %s: rc = %d\n",
lod2obd(lod)->obd_name, param, -EINVAL);
*ptr = '.';
obd = sub_tgt->ltd_tgt->dd_lu_dev.ld_obd;
ctxt = llog_get_context(obd,
LLOG_UPDATELOG_ORIG_CTXT);
- if (ctxt == NULL) {
+ if (!ctxt) {
rc = llog_setup(env, obd, &obd->obd_olg,
LLOG_UPDATELOG_ORIG_CTXT,
NULL, &llog_common_cat_ops);
GOTO(out, rc);
}
- obd = lod2obd(lod);
- rc = class_process_proc_param(PARAM_LOV, obd->obd_vars,
- lcfg, obd);
- if (rc > 0)
- rc = 0;
+
+ if (strstr(param, PARAM_LOD) != NULL)
+ count = class_modify_config(lcfg, PARAM_LOD,
+ &lod->lod_dt_dev.dd_kobj);
+ else
+ count = class_modify_config(lcfg, PARAM_LOV,
+ &lod->lod_dt_dev.dd_kobj);
+ rc = count > 0 ? 0 : count;
GOTO(out, rc);
}
case LCFG_PRE_CLEANUP: {
- if (lod->lod_md_root != NULL) {
- dt_object_put(env, &lod->lod_md_root->ldo_obj);
- lod->lod_md_root = NULL;
- }
-
lod_sub_process_config(env, lod, &lod->lod_mdt_descs, lcfg);
lod_sub_process_config(env, lod, &lod->lod_ost_descs, lcfg);
+ OBD_FAIL_TIMEOUT(OBD_FAIL_TGT_RECOVERY_CONNECT, cfs_fail_val * 2);
next = &lod->lod_child->dd_lu_dev;
rc = next->ld_ops->ldo_process_config(env, next, lcfg);
if (rc != 0)
break;
}
case LCFG_CLEANUP: {
+ if (lod->lod_md_root) {
+ dt_object_put(env, &lod->lod_md_root->ldo_obj);
+ lod->lod_md_root = NULL;
+ }
+
/*
* do cleanup on underlying storage only when
* all OSPs are cleaned up, as they use that OSD as well
next = &lod->lod_child->dd_lu_dev;
rc = next->ld_ops->ldo_process_config(env, next, lcfg);
if (rc)
- CERROR("%s: can't process %u: %d\n",
+ CERROR("%s: can't process %u: rc = %d\n",
lod2obd(lod)->obd_name, lcfg->lcfg_command, rc);
rc = obd_disconnect(lod->lod_child_exp);
if (rc)
- CERROR("error in disconnect from storage: %d\n", rc);
+ CERROR("error in disconnect from storage: rc = %d\n",
+ rc);
break;
}
default:
- CERROR("%s: unknown command %u\n", lod2obd(lod)->obd_name,
- lcfg->lcfg_command);
- rc = -EINVAL;
- break;
+ CERROR("%s: unknown command %u\n", lod2obd(lod)->obd_name,
+ lcfg->lcfg_command);
+ rc = -EINVAL;
+ break;
}
out:
static int lod_recovery_complete(const struct lu_env *env,
struct lu_device *dev)
{
- struct lod_device *lod = lu2lod_dev(dev);
- struct lu_device *next = &lod->lod_child->dd_lu_dev;
- unsigned int i;
- int rc;
+ struct lod_device *lod = lu2lod_dev(dev);
+ struct lu_device *next = &lod->lod_child->dd_lu_dev;
+ struct lod_tgt_desc *tgt;
+ int rc;
+
ENTRY;
LASSERT(lod->lod_recovery_completed == 0);
rc = next->ld_ops->ldo_recovery_complete(env, next);
lod_getref(&lod->lod_ost_descs);
- if (lod->lod_osts_size > 0) {
- cfs_foreach_bit(lod->lod_ost_bitmap, i) {
- struct lod_tgt_desc *tgt;
- tgt = OST_TGT(lod, i);
+ if (lod->lod_ost_descs.ltd_tgts_size > 0) {
+ lod_foreach_ost(lod, tgt) {
LASSERT(tgt && tgt->ltd_tgt);
- next = &tgt->ltd_ost->dd_lu_dev;
+ next = &tgt->ltd_tgt->dd_lu_dev;
rc = next->ld_ops->ldo_recovery_complete(env, next);
if (rc)
- CERROR("%s: can't complete recovery on #%d:"
- "%d\n", lod2obd(lod)->obd_name, i, rc);
+ CERROR("%s: can't complete recovery on #%d: rc = %d\n",
+ lod2obd(lod)->obd_name, tgt->ltd_index,
+ rc);
}
}
lod_putref(lod, &lod->lod_ost_descs);
*/
static int lod_sub_init_llogs(const struct lu_env *env, struct lod_device *lod)
{
- struct lod_tgt_descs *ltd = &lod->lod_mdt_descs;
- int rc;
- unsigned int i;
+ struct lu_tgt_desc *mdt;
+ int rc;
+
ENTRY;
- /* llog must be setup after LOD is initialized, because llog
- * initialization include FLD lookup */
+ /*
+ * llog must be setup after LOD is initialized, because llog
+ * initialization include FLD lookup
+ */
LASSERT(lod->lod_initialized);
/* Init the llog in its own stack */
if (rc < 0)
RETURN(rc);
- cfs_foreach_bit(ltd->ltd_tgt_bitmap, i) {
- struct lod_tgt_desc *tgt;
-
- tgt = LTD_TGT(ltd, i);
- rc = lod_sub_init_llog(env, lod, tgt->ltd_tgt);
+ lod_foreach_mdt(lod, mdt) {
+ rc = lod_sub_init_llog(env, lod, mdt->ltd_tgt);
if (rc != 0)
break;
}
static int lod_prepare(const struct lu_env *env, struct lu_device *pdev,
struct lu_device *cdev)
{
- struct lod_device *lod = lu2lod_dev(cdev);
- struct lu_device *next = &lod->lod_child->dd_lu_dev;
- struct lu_fid *fid = &lod_env_info(env)->lti_fid;
- int rc;
- struct dt_object *root;
- struct dt_object *dto;
- __u32 index;
+ struct lod_device *lod = lu2lod_dev(cdev);
+ struct lu_device *next = &lod->lod_child->dd_lu_dev;
+ struct lu_fid *fid = &lod_env_info(env)->lti_fid;
+ int rc;
+ struct dt_object *root;
+ struct dt_object *dto;
+ u32 index;
+
ENTRY;
rc = next->ld_ops->ldo_prepare(env, pdev, next);
dto = local_file_find_or_create_with_fid(env, lod->lod_child,
fid, root,
lod_update_log_name,
- S_IFREG | S_IRUGO | S_IWUSR);
+ S_IFREG | 0644);
if (IS_ERR(dto))
GOTO(out_put, rc = PTR_ERR(dto));
dto = local_file_find_or_create_with_fid(env, lod->lod_child,
fid, root,
lod_update_log_dir_name,
- S_IFDIR | S_IRUGO | S_IWUSR);
+ S_IFDIR | 0644);
if (IS_ERR(dto))
GOTO(out_put, rc = PTR_ERR(dto));
RETURN(rc);
}
+/**
+ * Implementation of lu_device_operations::ldo_fid_alloc() for LOD
+ *
+ * Find corresponding device by passed parent and name, and allocate FID from
+ * there.
+ *
+ * see include/lu_object.h for the details.
+ */
+static int lod_fid_alloc(const struct lu_env *env, struct lu_device *d,
+ struct lu_fid *fid, struct lu_object *parent,
+ const struct lu_name *name)
+{
+ struct lod_device *lod = lu2lod_dev(d);
+ struct lod_object *lo = lu2lod_obj(parent);
+ struct dt_device *next;
+ int rc;
+
+ ENTRY;
+
+ /* if @parent is remote, we don't know whether its layout was changed,
+ * always reload layout.
+ */
+ if (lu_object_remote(parent))
+ lod_striping_free(env, lo);
+
+ rc = lod_striping_load(env, lo);
+ if (rc)
+ RETURN(rc);
+
+ if (lo->ldo_dir_stripe_count > 0 && name) {
+ struct dt_object *stripe;
+ int idx;
+
+ idx = __lmv_name_to_stripe_index(lo->ldo_dir_hash_type,
+ lo->ldo_dir_stripe_count,
+ lo->ldo_dir_migrate_hash,
+ lo->ldo_dir_migrate_offset,
+ name->ln_name,
+ name->ln_namelen, true);
+ if (idx < 0)
+ RETURN(idx);
+
+ stripe = lo->ldo_stripe[idx];
+ if (!stripe || !dt_object_exists(stripe))
+ RETURN(-ENODEV);
+
+ next = lu2dt_dev(stripe->do_lu.lo_dev);
+ } else {
+ next = lod->lod_child;
+ }
+
+ rc = dt_fid_alloc(env, next, fid, parent, name);
+
+ RETURN(rc);
+}
+
const struct lu_device_operations lod_lu_ops = {
.ldo_object_alloc = lod_object_alloc,
.ldo_process_config = lod_process_config,
.ldo_recovery_complete = lod_recovery_complete,
.ldo_prepare = lod_prepare,
+ .ldo_fid_alloc = lod_fid_alloc,
};
/**
return dt_root_get(env, dt2lod_dev(dev)->lod_child, f);
}
+static void lod_statfs_sum(struct obd_statfs *sfs,
+ struct obd_statfs *ost_sfs, int *bs)
+{
+ while (ost_sfs->os_bsize < *bs) {
+ *bs >>= 1;
+ sfs->os_bsize >>= 1;
+ sfs->os_bavail <<= 1;
+ sfs->os_blocks <<= 1;
+ sfs->os_bfree <<= 1;
+ sfs->os_granted <<= 1;
+ }
+ while (ost_sfs->os_bsize > *bs) {
+ ost_sfs->os_bsize >>= 1;
+ ost_sfs->os_bavail <<= 1;
+ ost_sfs->os_blocks <<= 1;
+ ost_sfs->os_bfree <<= 1;
+ ost_sfs->os_granted <<= 1;
+ }
+ sfs->os_bavail += ost_sfs->os_bavail;
+ sfs->os_blocks += ost_sfs->os_blocks;
+ sfs->os_bfree += ost_sfs->os_bfree;
+ sfs->os_granted += ost_sfs->os_granted;
+}
+
/**
* Implementation of dt_device_operations::dt_statfs() for LOD
*
* see include/dt_object.h for the details.
*/
-static int lod_statfs(const struct lu_env *env,
- struct dt_device *dev, struct obd_statfs *sfs)
+static int lod_statfs(const struct lu_env *env, struct dt_device *dev,
+ struct obd_statfs *sfs, struct obd_statfs_info *info)
{
- return dt_statfs(env, dt2lod_dev(dev)->lod_child, sfs);
+ struct lod_device *lod = dt2lod_dev(dev);
+ struct lu_tgt_desc *tgt;
+ struct obd_statfs ost_sfs;
+ u64 ost_files = 0;
+ u64 ost_ffree = 0;
+ int rc, bs;
+
+ rc = dt_statfs(env, dt2lod_dev(dev)->lod_child, sfs);
+ if (rc)
+ GOTO(out, rc);
+
+ bs = sfs->os_bsize;
+
+ sfs->os_bavail = 0;
+ sfs->os_blocks = 0;
+ sfs->os_bfree = 0;
+ sfs->os_granted = 0;
+
+ lod_getref(&lod->lod_mdt_descs);
+ lod_foreach_mdt(lod, tgt) {
+ rc = dt_statfs(env, tgt->ltd_tgt, &ost_sfs);
+ /* ignore errors */
+ if (rc)
+ continue;
+ sfs->os_files += ost_sfs.os_files;
+ sfs->os_ffree += ost_sfs.os_ffree;
+ lod_statfs_sum(sfs, &ost_sfs, &bs);
+ }
+ lod_putref(lod, &lod->lod_mdt_descs);
+
+ /*
+ * at some point we can check whether DoM is enabled and
+ * decide how to account MDT space. for simplicity let's
+ * just fallback to pre-DoM policy if any OST is alive
+ */
+ lod_getref(&lod->lod_ost_descs);
+ lod_foreach_ost(lod, tgt) {
+ rc = dt_statfs(env, tgt->ltd_tgt, &ost_sfs);
+ /* ignore errors */
+ if (rc || ost_sfs.os_bsize == 0)
+ continue;
+ if (!ost_files) {
+ /*
+ * if only MDTs with DoM then report only MDT blocks,
+ * otherwise show only OST blocks, and DoM is "free"
+ */
+ sfs->os_bavail = 0;
+ sfs->os_blocks = 0;
+ sfs->os_bfree = 0;
+ sfs->os_granted = 0;
+ }
+ ost_files += ost_sfs.os_files;
+ ost_ffree += ost_sfs.os_ffree;
+ ost_sfs.os_bavail += ost_sfs.os_granted;
+ lod_statfs_sum(sfs, &ost_sfs, &bs);
+ LASSERTF(bs == ost_sfs.os_bsize, "%d != %d\n",
+ (int)sfs->os_bsize, (int)ost_sfs.os_bsize);
+ }
+ lod_putref(lod, &lod->lod_ost_descs);
+ sfs->os_state |= OS_STATFS_SUM;
+
+ /* If we have _some_ OSTs, but don't have as many free objects on the
+ * OSTs as inodes on the MDTs, reduce the reported number of inodes
+ * to compensate, so that the "inodes in use" number is correct.
+ * This should be kept in sync with ll_statfs_internal().
+ */
+ if (ost_files && ost_ffree < sfs->os_ffree) {
+ sfs->os_files = (sfs->os_files - sfs->os_ffree) + ost_ffree;
+ sfs->os_ffree = ost_ffree;
+ }
+
+ /* a single successful statfs should be enough */
+ rc = 0;
+
+out:
+ RETURN(rc);
}
/**
int rc = 0;
top_th = container_of(th, struct top_thandle, tt_super);
- if (top_th->tt_multiple_thandle == NULL)
+ if (!top_th->tt_multiple_thandle)
return 0;
fid_zero(fid);
*/
static int lod_sync(const struct lu_env *env, struct dt_device *dev)
{
- struct lod_device *lod = dt2lod_dev(dev);
- struct lod_ost_desc *ost;
- struct lod_mdt_desc *mdt;
- unsigned int i;
- int rc = 0;
+ struct lod_device *lod = dt2lod_dev(dev);
+ struct lu_tgt_desc *tgt;
+ int rc = 0;
+
ENTRY;
lod_getref(&lod->lod_ost_descs);
- lod_foreach_ost(lod, i) {
- ost = OST_TGT(lod, i);
- LASSERT(ost && ost->ltd_ost);
- rc = dt_sync(env, ost->ltd_ost);
+ lod_foreach_ost(lod, tgt) {
+ if (!tgt->ltd_active)
+ continue;
+ rc = dt_sync(env, tgt->ltd_tgt);
if (rc) {
- CERROR("%s: can't sync ost %u: %d\n",
- lod2obd(lod)->obd_name, i, rc);
- break;
+ if (rc != -ENOTCONN) {
+ CERROR("%s: can't sync ost %u: rc = %d\n",
+ lod2obd(lod)->obd_name, tgt->ltd_index,
+ rc);
+ break;
+ }
+ rc = 0;
}
}
lod_putref(lod, &lod->lod_ost_descs);
RETURN(rc);
lod_getref(&lod->lod_mdt_descs);
- lod_foreach_mdt(lod, i) {
- mdt = MDT_TGT(lod, i);
- LASSERT(mdt && mdt->ltd_mdt);
- rc = dt_sync(env, mdt->ltd_mdt);
+ lod_foreach_mdt(lod, tgt) {
+ if (!tgt->ltd_active)
+ continue;
+ rc = dt_sync(env, tgt->ltd_tgt);
if (rc) {
- CERROR("%s: can't sync mdt %u: %d\n",
- lod2obd(lod)->obd_name, i, rc);
- break;
+ if (rc != -ENOTCONN) {
+ CERROR("%s: can't sync mdt %u: rc = %d\n",
+ lod2obd(lod)->obd_name, tgt->ltd_index,
+ rc);
+ break;
+ }
+ rc = 0;
}
}
lod_putref(lod, &lod->lod_mdt_descs);
struct lustre_cfg *cfg)
{
struct obd_connect_data *data = NULL;
- struct obd_device *obd;
- char *nextdev = NULL, *p, *s;
- int rc, len = 0;
+ struct obd_device *obd;
+ char *nextdev = NULL, *p, *s;
+ int rc, len = 0;
+
ENTRY;
LASSERT(cfg);
LASSERT(lod->lod_child_exp == NULL);
- /* compatibility hack: we still use old config logs
+ /*
+ * compatibility hack: we still use old config logs
* which specify LOV, but we need to learn underlying
* OSD device, which is supposed to be:
* <fsname>-MDTxxxx-osd
* 1.8 MGS generates lines like the following:
* #03 (168)lov_setup 0:lustre-mdtlov 1:(struct lov_desc)
*
- * we use "-MDT" to differentiate 2.x from 1.8 */
-
- if ((p = lustre_cfg_string(cfg, 0)) && strstr(p, "-mdtlov")) {
+ * we use "-MDT" to differentiate 2.x from 1.8
+ */
+ p = lustre_cfg_string(cfg, 0);
+ if (p && strstr(p, "-mdtlov")) {
len = strlen(p) + 6;
OBD_ALLOC(nextdev, len);
- if (nextdev == NULL)
+ if (!nextdev)
GOTO(out, rc = -ENOMEM);
strcpy(nextdev, p);
s = strstr(nextdev, "-mdtlov");
- if (unlikely(s == NULL)) {
- CERROR("unable to parse device name %s\n",
- lustre_cfg_string(cfg, 0));
+ if (unlikely(!s)) {
+ CERROR("%s: unable to parse device name: rc = %d\n",
+ lustre_cfg_string(cfg, 0), -EINVAL);
GOTO(out, rc = -EINVAL);
}
strcpy(s, "-MDT0000-osd");
}
} else {
- CERROR("unable to parse device name %s\n",
- lustre_cfg_string(cfg, 0));
+ CERROR("%s: unable to parse device name: rc = %d\n",
+ lustre_cfg_string(cfg, 0), -EINVAL);
GOTO(out, rc = -EINVAL);
}
OBD_ALLOC_PTR(data);
- if (data == NULL)
+ if (!data)
GOTO(out, rc = -ENOMEM);
obd = class_name2obd(nextdev);
- if (obd == NULL) {
- CERROR("can not locate next device: %s\n", nextdev);
+ if (!obd) {
+ CERROR("%s: can not locate next device: rc = %d\n",
+ nextdev, -ENOTCONN);
GOTO(out, rc = -ENOTCONN);
}
rc = obd_connect(env, &lod->lod_child_exp, obd, &obd->obd_uuid,
data, NULL);
if (rc) {
- CERROR("cannot connect to next dev %s (%d)\n", nextdev, rc);
+ CERROR("%s: cannot connect to next dev: rc = %d\n",
+ nextdev, rc);
GOTO(out, rc);
}
RETURN(rc);
}
-/**
- * Allocate and initialize target table.
- *
- * A helper function to initialize the target table and allocate
- * a bitmap of the available targets.
- *
- * \param[in] ltd target's table to initialize
- *
- * \retval 0 on success
- * \retval negative negated errno on error
- **/
-static int lod_tgt_desc_init(struct lod_tgt_descs *ltd)
+static int lod_lsfs_init(const struct lu_env *env, struct lod_device *d)
{
- mutex_init(<d->ltd_mutex);
- init_rwsem(<d->ltd_rw_sem);
-
- /* the OST array and bitmap are allocated/grown dynamically as OSTs are
- * added to the LOD, see lod_add_device() */
- ltd->ltd_tgt_bitmap = CFS_ALLOCATE_BITMAP(32);
- if (ltd->ltd_tgt_bitmap == NULL)
- RETURN(-ENOMEM);
+ struct obd_statfs sfs;
+ int rc;
- ltd->ltd_tgts_size = 32;
- ltd->ltd_tgtnr = 0;
+ rc = dt_statfs(env, d->lod_child, &sfs);
+ if (rc) {
+ CDEBUG(D_LAYOUT, "%s: failed to get OSD statfs, rc = %d\n",
+ lod2obd(d)->obd_name, rc);
+ return rc;
+ }
- ltd->ltd_death_row = 0;
- ltd->ltd_refcount = 0;
+ /* udpate local OSD cached statfs data */
+ spin_lock_init(&d->lod_lsfs_lock);
+ d->lod_lsfs_age = ktime_get_seconds();
+ d->lod_lsfs_total_mb = (sfs.os_blocks * sfs.os_bsize) >> 20;
+ d->lod_lsfs_free_mb = (sfs.os_bfree * sfs.os_bsize) >> 20;
return 0;
}
struct lu_device_type *ldt, struct lustre_cfg *cfg)
{
struct dt_device_param ddp;
- struct obd_device *obd;
- int rc;
+ struct obd_device *obd;
+ int rc;
+
ENTRY;
obd = class_name2obd(lustre_cfg_string(cfg, 0));
- if (obd == NULL) {
- CERROR("Cannot find obd with name %s\n",
- lustre_cfg_string(cfg, 0));
- RETURN(-ENODEV);
+ if (!obd) {
+ rc = -ENODEV;
+ CERROR("Cannot find obd with name '%s': rc = %d\n",
+ lustre_cfg_string(cfg, 0), rc);
+ RETURN(rc);
}
obd->obd_lu_dev = &lod->lod_dt_dev.dd_lu_dev;
dt_conf_get(env, &lod->lod_dt_dev, &ddp);
lod->lod_osd_max_easize = ddp.ddp_max_ea_size;
+ lod->lod_dom_stripesize_max_kb = (1ULL << 10); /* 1Mb is default */
+
+ /* initialize local statfs cached values */
+ rc = lod_lsfs_init(env, lod);
+ if (rc)
+ GOTO(out_disconnect, rc);
+
+ /* default threshold as half of total space, in MiB */
+ lod->lod_dom_threshold_free_mb = lod->lod_lsfs_total_mb / 2;
+ /* set default DoM stripe size based on free space amount */
+ lod_dom_stripesize_recalc(lod);
/* setup obd to be used with old lov code */
rc = lod_pools_init(lod, cfg);
spin_lock_init(&lod->lod_lock);
spin_lock_init(&lod->lod_connects_lock);
- lod_tgt_desc_init(&lod->lod_mdt_descs);
- lod_tgt_desc_init(&lod->lod_ost_descs);
+ lu_tgt_descs_init(&lod->lod_mdt_descs, true);
+ lu_tgt_descs_init(&lod->lod_ost_descs, false);
RETURN(0);
{
struct lod_device *lod = lu2lod_dev(lu);
struct lu_device *next = &lod->lod_child->dd_lu_dev;
+
ENTRY;
+ if (atomic_read(&lu->ld_site->ls_obj_hash.nelems)) {
+ lu_site_print(env, lu->ld_site, &lu->ld_ref, D_ERROR,
+ lu_cdebug_printer);
+ }
LASSERTF(atomic_read(&lu->ld_ref) == 0, "lu is %p\n", lu);
dt_device_fini(&lod->lod_dt_dev);
OBD_FREE_PTR(lod);
struct lustre_cfg *lcfg)
{
struct lod_device *lod;
- struct lu_device *lu_dev;
+ struct lu_device *lu_dev;
OBD_ALLOC_PTR(lod);
- if (lod == NULL) {
+ if (!lod) {
lu_dev = ERR_PTR(-ENOMEM);
} else {
int rc;
return lu_dev;
}
+static void lod_avoid_guide_fini(struct lod_avoid_guide *lag)
+{
+ if (lag->lag_oss_avoid_array)
+ OBD_FREE_PTR_ARRAY(lag->lag_oss_avoid_array,
+ lag->lag_oaa_size);
+ bitmap_free(lag->lag_ost_avoid_bitmap);
+}
+
/**
* Implementation of lu_device_type_operations::ldto_device_fini() for LOD
*
struct lu_device *d)
{
struct lod_device *lod = lu2lod_dev(d);
- int rc;
+ int rc;
+
ENTRY;
lod_pools_fini(lod);
lod_procfs_fini(lod);
- rc = lod_fini_tgt(env, lod, &lod->lod_ost_descs, true);
+ rc = lod_fini_tgt(env, lod, &lod->lod_ost_descs);
if (rc)
- CERROR("%s:can not fini ost descs %d\n",
+ CERROR("%s: can not fini ost descriptors: rc = %d\n",
lod2obd(lod)->obd_name, rc);
- rc = lod_fini_tgt(env, lod, &lod->lod_mdt_descs, false);
+ rc = lod_fini_tgt(env, lod, &lod->lod_mdt_descs);
if (rc)
- CERROR("%s:can not fini mdt descs %d\n",
+ CERROR("%s: can not fini mdt descriptors: rc = %d\n",
lod2obd(lod)->obd_name, rc);
RETURN(NULL);
struct obd_device *obd, struct obd_uuid *cluuid,
struct obd_connect_data *data, void *localdata)
{
- struct lod_device *lod = lu2lod_dev(obd->obd_lu_dev);
- struct lustre_handle conn;
- int rc;
+ struct lod_device *lod = lu2lod_dev(obd->obd_lu_dev);
+ struct lustre_handle conn;
+ int rc;
+
ENTRY;
CDEBUG(D_CONFIG, "connect #%d\n", lod->lod_connects);
{
struct obd_device *obd = exp->exp_obd;
struct lod_device *lod = lu2lod_dev(obd->obd_lu_dev);
- int rc, release = 0;
+ int rc, release = 0;
+
ENTRY;
/* Only disconnect the underlying layers on the final disconnect. */
struct lod_thread_info *info = data;
struct lod_layout_component *lds =
info->lti_def_striping.lds_def_comp_entries;
- struct ost_pool *inuse = &info->lti_inuse_osts;
- /* allocated in lod_get_lov_ea
+ /*
+ * allocated in lod_get_lov_ea
* XXX: this is overload, a tread may have such store but used only
* once. Probably better would be pool of such stores per LOD.
*/
}
lu_buf_free(&info->lti_linkea_buf);
- if (lds != NULL)
+ if (lds)
lod_free_def_comp_entries(&info->lti_def_striping);
- if (inuse->op_size)
- OBD_FREE(inuse->op_array, inuse->op_size);
+ if (info->lti_comp_size > 0)
+ OBD_FREE_PTR_ARRAY(info->lti_comp_idx,
+ info->lti_comp_size);
+
+ lod_avoid_guide_fini(&info->lti_avoid);
OBD_FREE_PTR(info);
}
* \retval -EINVAL if not supported key is requested
**/
static int lod_obd_get_info(const struct lu_env *env, struct obd_export *exp,
- __u32 keylen, void *key, __u32 *vallen, void *val)
+ u32 keylen, void *key, u32 *vallen, void *val)
{
int rc = -EINVAL;
if (KEY_IS(KEY_OSP_CONNECTED)) {
- struct obd_device *obd = exp->exp_obd;
- struct lod_device *d;
- struct lod_tgt_desc *tgt;
- unsigned int i;
- int rc = 1;
+ struct obd_device *obd = exp->exp_obd;
+ struct lod_device *d;
+ struct lod_tgt_desc *tgt;
+ int rc = 1;
if (!obd->obd_set_up || obd->obd_stopping)
RETURN(-EAGAIN);
d = lu2lod_dev(obd->obd_lu_dev);
lod_getref(&d->lod_ost_descs);
- lod_foreach_ost(d, i) {
- tgt = OST_TGT(d, i);
- LASSERT(tgt && tgt->ltd_tgt);
+ lod_foreach_ost(d, tgt) {
rc = obd_get_info(env, tgt->ltd_exp, keylen, key,
- vallen, val);
+ vallen, val);
/* one healthy device is enough */
if (rc == 0)
break;
lod_putref(d, &d->lod_ost_descs);
lod_getref(&d->lod_mdt_descs);
- lod_foreach_mdt(d, i) {
+ lod_foreach_mdt(d, tgt) {
struct llog_ctxt *ctxt;
- tgt = MDT_TGT(d, i);
- LASSERT(tgt != NULL);
- LASSERT(tgt->ltd_tgt != NULL);
if (!tgt->ltd_active)
continue;
ctxt = llog_get_context(tgt->ltd_tgt->dd_lu_dev.ld_obd,
LLOG_UPDATELOG_ORIG_CTXT);
- if (ctxt == NULL) {
+ if (!ctxt) {
CDEBUG(D_INFO, "%s: %s is not ready.\n",
obd->obd_name,
tgt->ltd_tgt->dd_lu_dev.ld_obd->obd_name);
rc = -EAGAIN;
break;
}
- if (ctxt->loc_handle == NULL) {
+ if (!ctxt->loc_handle) {
CDEBUG(D_INFO, "%s: %s is not ready.\n",
obd->obd_name,
tgt->ltd_tgt->dd_lu_dev.ld_obd->obd_name);
static int lod_obd_set_info_async(const struct lu_env *env,
struct obd_export *exp,
- __u32 keylen, void *key,
- __u32 vallen, void *val,
+ u32 keylen, void *key,
+ u32 vallen, void *val,
struct ptlrpc_request_set *set)
{
struct obd_device *obd = class_exp2obd(exp);
struct lod_device *d;
struct lod_tgt_desc *tgt;
int no_set = 0;
- int i, rc = 0, rc2;
+ int rc = 0, rc2;
+
ENTRY;
- if (set == NULL) {
+ if (!set) {
no_set = 1;
set = ptlrpc_prep_set();
if (!set)
d = lu2lod_dev(obd->obd_lu_dev);
lod_getref(&d->lod_ost_descs);
- lod_foreach_ost(d, i) {
- tgt = OST_TGT(d, i);
- LASSERT(tgt && tgt->ltd_tgt);
+ lod_foreach_ost(d, tgt) {
if (!tgt->ltd_active)
continue;
lod_putref(d, &d->lod_ost_descs);
lod_getref(&d->lod_mdt_descs);
- lod_foreach_mdt(d, i) {
- tgt = MDT_TGT(d, i);
- LASSERT(tgt && tgt->ltd_tgt);
+ lod_foreach_mdt(d, tgt) {
if (!tgt->ltd_active)
continue;
rc2 = obd_set_info_async(env, tgt->ltd_exp, keylen, key,
if (no_set) {
- rc2 = ptlrpc_set_wait(set);
+ rc2 = ptlrpc_set_wait(env, set);
if (rc2 == 0 && rc == 0)
rc = rc2;
ptlrpc_set_destroy(set);
RETURN(rc);
}
-static struct obd_ops lod_obd_device_ops = {
+static const struct obd_ops lod_obd_device_ops = {
.o_owner = THIS_MODULE,
.o_connect = lod_obd_connect,
.o_disconnect = lod_obd_disconnect,
static int __init lod_init(void)
{
- struct obd_type *type;
+ struct obd_type *sym;
int rc;
rc = lu_kmem_init(lod_caches);
return rc;
}
- /* create "lov" entry in procfs for compatibility purposes */
- type = class_search_type(LUSTRE_LOV_NAME);
- if (type != NULL && type->typ_procroot != NULL)
- return rc;
-
- type = class_search_type(LUSTRE_LOD_NAME);
- type->typ_procsym = lprocfs_register("lov", proc_lustre_root,
- NULL, NULL);
- if (IS_ERR(type->typ_procsym)) {
- CERROR("lod: can't create compat entry \"lov\": %d\n",
- (int)PTR_ERR(type->typ_procsym));
- type->typ_procsym = NULL;
+ /* create "lov" entry for compatibility purposes */
+ sym = class_add_symlinks(LUSTRE_LOV_NAME, true);
+ if (IS_ERR(sym)) {
+ rc = PTR_ERR(sym);
+ /* does real "lov" already exist ? */
+ if (rc == -EEXIST)
+ rc = 0;
}
+
return rc;
}
static void __exit lod_exit(void)
{
+ struct obd_type *sym = class_search_type(LUSTRE_LOV_NAME);
+
+ /* if this was never fully initialized by the lov layer
+ * then we are responsible for freeing this obd_type
+ */
+ if (sym) {
+ /* final put if we manage this obd type */
+ if (sym->typ_sym_filter)
+ kobject_put(&sym->typ_kobj);
+ /* put reference taken by class_search_type */
+ kobject_put(&sym->typ_kobj);
+ }
+
class_unregister_type(LUSTRE_LOD_NAME);
lu_kmem_fini(lod_caches);
}