* Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
* Use is subject to license terms.
*
- * Copyright (c) 2011, Whamcloud, Inc.
+ * Copyright (c) 2011, 2012, Intel Corporation.
*/
/*
* This file is part of Lustre, http://www.lustre.org/
struct mgs_nidtbl_target *tgt;
int version = 0;
- LASSERT(cfs_mutex_is_locked(&tbl->mn_lock));
+ LASSERT(mutex_is_locked(&tbl->mn_lock));
cfs_list_for_each_entry(tgt, &tbl->mn_targets, mnt_list) {
if (!tgt->mnt_version)
continue;
* shouldn't cross unit boundaries.
*/
static int mgs_nidtbl_read(struct obd_export *exp, struct mgs_nidtbl *tbl,
- struct mgs_config_res *res, cfs_page_t **pages,
+ struct mgs_config_res *res, struct page **pages,
int nrpages, int units_total, int unit_size)
{
struct mgs_nidtbl_target *tgt;
/* make sure unit_size is power 2 */
LASSERT((unit_size & (unit_size - 1)) == 0);
- LASSERT(nrpages << CFS_PAGE_SHIFT >= units_total * unit_size);
+ LASSERT(nrpages << PAGE_CACHE_SHIFT >= units_total * unit_size);
- cfs_mutex_lock(&tbl->mn_lock);
+ mutex_lock(&tbl->mn_lock);
LASSERT(nidtbl_is_sane(tbl));
/* no more entries ? */
}
LASSERT((rc & (unit_size - 1)) == 0);
- if (units_in_page == 0) {
- /* allocate a new page */
- pages[index] = cfs_alloc_page(CFS_ALLOC_STD);
- if (pages[index] == NULL) {
- rc = -ENOMEM;
- break;
- }
+ if (units_in_page == 0) {
+ /* allocate a new page */
+ pages[index] = alloc_page(GFP_IOFS);
+ if (pages[index] == NULL) {
+ rc = -ENOMEM;
+ break;
+ }
- /* destroy previous map */
- if (index > 0)
- cfs_kunmap(pages[index - 1]);
+ /* destroy previous map */
+ if (index > 0)
+ kunmap(pages[index - 1]);
- /* reassign buffer */
- buf = cfs_kmap(pages[index]);
- ++index;
+ /* reassign buffer */
+ buf = kmap(pages[index]);
+ ++index;
- units_in_page = CFS_PAGE_SIZE / unit_size;
- LASSERT(units_in_page > 0);
- }
+ units_in_page = PAGE_CACHE_SIZE / unit_size;
+ LASSERT(units_in_page > 0);
+ }
/* allocate an unit */
LASSERT(((long)buf & (unit_size - 1)) == 0);
bytes_in_unit, index, nrpages, units_total);
}
if (index > 0)
- cfs_kunmap(pages[index - 1]);
+ kunmap(pages[index - 1]);
out:
LASSERT(version <= tbl->mn_version);
res->mcr_size = tbl->mn_version;
res->mcr_offset = nobuf ? version : tbl->mn_version;
- cfs_mutex_unlock(&tbl->mn_lock);
+ mutex_unlock(&tbl->mn_lock);
LASSERT(ergo(version == 1, rc == 0)); /* get the log first time */
CDEBUG(D_MGS, "Read IR logs %s return with %d, version %llu\n",
int rc;
ENTRY;
- LASSERT(cfs_mutex_is_locked(&tbl->mn_lock));
+ LASSERT(mutex_is_locked(&tbl->mn_lock));
fsdb = local_file_find_or_create(env, mgs->mgs_los, mgs->mgs_nidtbl_dir,
tbl->mn_fsdb->fsdb_name,
#define MGS_NIDTBL_VERSION_INIT 2
static int nidtbl_read_version(const struct lu_env *env,
- struct mgs_device *mgs, struct mgs_nidtbl *tbl)
+ struct mgs_device *mgs, struct mgs_nidtbl *tbl,
+ u64 *version)
{
struct dt_object *fsdb;
struct lu_fid fid;
int rc;
ENTRY;
- LASSERT(cfs_mutex_is_locked(&tbl->mn_lock));
+ LASSERT(mutex_is_locked(&tbl->mn_lock));
LASSERT(mgs->mgs_nidtbl_dir);
rc = dt_lookup_dir(env, mgs->mgs_nidtbl_dir, tbl->mn_fsdb->fsdb_name,
&fid);
- if (rc == -ENOENT)
- RETURN(MGS_NIDTBL_VERSION_INIT);
- else if (rc < 0)
+ if (rc == -ENOENT) {
+ *version = MGS_NIDTBL_VERSION_INIT;
+ RETURN(0);
+ } else if (rc < 0) {
RETURN(rc);
+ }
fsdb = dt_locate_at(env, mgs->mgs_bottom, &fid,
&mgs->mgs_dt_dev.dd_lu_dev);
RETURN(PTR_ERR(fsdb));
rc = dt_read(env, fsdb, &buf, &off);
- if (rc == buf.lb_len)
- rc = le64_to_cpu(tmpver);
- else if (rc == 0)
- rc = MGS_NIDTBL_VERSION_INIT;
- else
+ if (rc == buf.lb_len) {
+ *version = le64_to_cpu(tmpver);
+ rc = 0;
+ } else if (rc == 0) {
+ *version = MGS_NIDTBL_VERSION_INIT;
+ } else {
CERROR("%s: read version file %s error %d\n",
mgs->mgs_obd->obd_name, tbl->mn_fsdb->fsdb_name, rc);
+ }
lu_object_put(env, &fsdb->do_lu);
RETURN(rc);
}
LASSERT(type != 0);
tbl = &fsdb->fsdb_nidtbl;
- cfs_mutex_lock(&tbl->mn_lock);
+ mutex_lock(&tbl->mn_lock);
cfs_list_for_each_entry(tgt, &tbl->mn_targets, mnt_list) {
struct mgs_target_info *info = &tgt->mnt_mti;
if (type == tgt->mnt_type &&
EXIT;
out:
- cfs_mutex_unlock(&tbl->mn_lock);
+ mutex_unlock(&tbl->mn_lock);
if (rc)
CERROR("Write NID table version for file system %s error %d\n",
fsdb->fsdb_name, rc);
struct mgs_nidtbl *tbl = &fsdb->fsdb_nidtbl;
CFS_LIST_HEAD(head);
- cfs_mutex_lock(&tbl->mn_lock);
+ mutex_lock(&tbl->mn_lock);
tbl->mn_nr_targets = 0;
cfs_list_splice_init(&tbl->mn_targets, &head);
- cfs_mutex_unlock(&tbl->mn_lock);
+ mutex_unlock(&tbl->mn_lock);
while (!cfs_list_empty(&head)) {
struct mgs_nidtbl_target *tgt;
static int mgs_nidtbl_init_fs(const struct lu_env *env, struct fs_db *fsdb)
{
struct mgs_nidtbl *tbl = &fsdb->fsdb_nidtbl;
+ int rc;
CFS_INIT_LIST_HEAD(&tbl->mn_targets);
- cfs_mutex_init(&tbl->mn_lock);
+ mutex_init(&tbl->mn_lock);
tbl->mn_nr_targets = 0;
tbl->mn_fsdb = fsdb;
- cfs_mutex_lock(&tbl->mn_lock);
- tbl->mn_version = nidtbl_read_version(env, fsdb->fsdb_mgs, tbl);
- cfs_mutex_unlock(&tbl->mn_lock);
- CDEBUG(D_MGS, "IR: current version is %llu\n", tbl->mn_version);
+ mutex_lock(&tbl->mn_lock);
+ rc = nidtbl_read_version(env, fsdb->fsdb_mgs, tbl, &tbl->mn_version);
+ mutex_unlock(&tbl->mn_lock);
+ if (rc < 0)
+ CERROR("%s: IR: failed to read current version, rc = %d\n",
+ fsdb->fsdb_mgs->mgs_obd->obd_name, rc);
+ else
+ CDEBUG(D_MGS, "IR: current version is %llu\n",
+ tbl->mn_version);
- return 0;
+ return rc;
}
/* --------- Imperative Recovery relies on nidtbl stuff ------- */
void mgs_ir_notify_complete(struct fs_db *fsdb)
{
- struct timeval tv;
- cfs_duration_t delta;
+ struct timeval tv;
+ cfs_duration_t delta;
- cfs_atomic_set(&fsdb->fsdb_notify_phase, 0);
+ atomic_set(&fsdb->fsdb_notify_phase, 0);
/* do statistic */
fsdb->fsdb_notify_count++;
LASSERTF(sizeof(name) < 32, "name is too large to be in stack.\n");
sprintf(name, "mgs_%s_notify", fsdb->fsdb_name);
- cfs_daemonize(name);
- cfs_complete(&fsdb->fsdb_notify_comp);
+ complete(&fsdb->fsdb_notify_comp);
set_user_nice(current, -2);
- mgc_fsname2resid(fsdb->fsdb_name, &resid, CONFIG_T_RECOVER);
- while (1) {
- struct l_wait_info lwi = { 0 };
+ mgc_fsname2resid(fsdb->fsdb_name, &resid, CONFIG_T_RECOVER);
+ while (1) {
+ struct l_wait_info lwi = { 0 };
- l_wait_event(fsdb->fsdb_notify_waitq,
- fsdb->fsdb_notify_stop ||
- cfs_atomic_read(&fsdb->fsdb_notify_phase),
- &lwi);
- if (fsdb->fsdb_notify_stop)
- break;
+ l_wait_event(fsdb->fsdb_notify_waitq,
+ fsdb->fsdb_notify_stop ||
+ atomic_read(&fsdb->fsdb_notify_phase),
+ &lwi);
+ if (fsdb->fsdb_notify_stop)
+ break;
- CDEBUG(D_MGS, "%s woken up, phase is %d\n",
- name, cfs_atomic_read(&fsdb->fsdb_notify_phase));
+ CDEBUG(D_MGS, "%s woken up, phase is %d\n",
+ name, atomic_read(&fsdb->fsdb_notify_phase));
- fsdb->fsdb_notify_start = cfs_time_current();
+ fsdb->fsdb_notify_start = cfs_time_current();
mgs_revoke_lock(fsdb->fsdb_mgs, fsdb, CONFIG_T_RECOVER);
- }
+ }
- cfs_complete(&fsdb->fsdb_notify_comp);
- return 0;
+ complete(&fsdb->fsdb_notify_comp);
+ return 0;
}
int mgs_ir_init_fs(const struct lu_env *env, struct mgs_device *mgs,
struct fs_db *fsdb)
{
- int rc;
+ struct task_struct *task;
- if (!ir_timeout)
- ir_timeout = OBD_IR_MGS_TIMEOUT;
+ if (!ir_timeout)
+ ir_timeout = OBD_IR_MGS_TIMEOUT;
- fsdb->fsdb_ir_state = IR_FULL;
- if (cfs_time_before(cfs_time_current_sec(),
- mgs->mgs_start_time + ir_timeout))
- fsdb->fsdb_ir_state = IR_STARTUP;
- fsdb->fsdb_nonir_clients = 0;
- CFS_INIT_LIST_HEAD(&fsdb->fsdb_clients);
+ fsdb->fsdb_ir_state = IR_FULL;
+ if (cfs_time_before(cfs_time_current_sec(),
+ mgs->mgs_start_time + ir_timeout))
+ fsdb->fsdb_ir_state = IR_STARTUP;
+ fsdb->fsdb_nonir_clients = 0;
+ CFS_INIT_LIST_HEAD(&fsdb->fsdb_clients);
- /* start notify thread */
- fsdb->fsdb_obd = mgs->mgs_obd;
+ /* start notify thread */
fsdb->fsdb_mgs = mgs;
- cfs_atomic_set(&fsdb->fsdb_notify_phase, 0);
- cfs_waitq_init(&fsdb->fsdb_notify_waitq);
- cfs_init_completion(&fsdb->fsdb_notify_comp);
- rc = cfs_create_thread(mgs_ir_notify, fsdb, CFS_DAEMON_FLAGS);
- if (rc > 0)
- cfs_wait_for_completion(&fsdb->fsdb_notify_comp);
- else
- CERROR("Start notify thread error %d\n", rc);
+ atomic_set(&fsdb->fsdb_notify_phase, 0);
+ init_waitqueue_head(&fsdb->fsdb_notify_waitq);
+ init_completion(&fsdb->fsdb_notify_comp);
+
+ task = kthread_run(mgs_ir_notify, fsdb,
+ "mgs_%s_notify", fsdb->fsdb_name);
+ if (!IS_ERR(task))
+ wait_for_completion(&fsdb->fsdb_notify_comp);
+ else
+ CERROR("Start notify thread error %ld\n", PTR_ERR(task));
mgs_nidtbl_init_fs(env, fsdb);
- return 0;
+ return 0;
}
void mgs_ir_fini_fs(struct mgs_device *mgs, struct fs_db *fsdb)
{
- if (cfs_test_bit(FSDB_MGS_SELF, &fsdb->fsdb_flags))
- return;
+ if (test_bit(FSDB_MGS_SELF, &fsdb->fsdb_flags))
+ return;
- mgs_fsc_cleanup_by_fsdb(fsdb);
+ mgs_fsc_cleanup_by_fsdb(fsdb);
- mgs_nidtbl_fini_fs(fsdb);
+ mgs_nidtbl_fini_fs(fsdb);
- LASSERT(cfs_list_empty(&fsdb->fsdb_clients));
+ LASSERT(cfs_list_empty(&fsdb->fsdb_clients));
- fsdb->fsdb_notify_stop = 1;
- cfs_waitq_signal(&fsdb->fsdb_notify_waitq);
- cfs_wait_for_completion(&fsdb->fsdb_notify_comp);
+ fsdb->fsdb_notify_stop = 1;
+ wake_up(&fsdb->fsdb_notify_waitq);
+ wait_for_completion(&fsdb->fsdb_notify_comp);
}
/* caller must have held fsdb_mutex */
return rc;
/* check ir state */
- cfs_mutex_lock(&fsdb->fsdb_mutex);
+ mutex_lock(&fsdb->fsdb_mutex);
ir_state_graduate(fsdb);
switch (fsdb->fsdb_ir_state) {
case IR_FULL:
default:
LBUG();
}
- cfs_mutex_unlock(&fsdb->fsdb_mutex);
-
- LASSERT(ergo(mti->mti_flags & LDD_F_IR_CAPABLE, notify));
- if (notify) {
- CDEBUG(D_MGS, "Try to revoke recover lock of %s\n",
- fsdb->fsdb_name);
- cfs_atomic_inc(&fsdb->fsdb_notify_phase);
- cfs_waitq_signal(&fsdb->fsdb_notify_waitq);
- }
- return 0;
+ mutex_unlock(&fsdb->fsdb_mutex);
+
+ LASSERT(ergo(mti->mti_flags & LDD_F_IR_CAPABLE, notify));
+ if (notify) {
+ CDEBUG(D_MGS, "Try to revoke recover lock of %s\n",
+ fsdb->fsdb_name);
+ atomic_inc(&fsdb->fsdb_notify_phase);
+ wake_up(&fsdb->fsdb_notify_waitq);
+ }
+ return 0;
}
/* NID table can be cached by two entities: Clients and MDTs */
int bytes;
int page_count;
int nrpages;
- cfs_page_t **pages = NULL;
+ struct page **pages = NULL;
ENTRY;
body = req_capsule_client_get(&req->rq_pill, &RMF_MGS_CONFIG_BODY);
RETURN(rc);
bufsize = body->mcb_units << body->mcb_bits;
- nrpages = (bufsize + CFS_PAGE_SIZE - 1) >> CFS_PAGE_SHIFT;
+ nrpages = (bufsize + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
if (nrpages > PTLRPC_MAX_BRW_PAGES)
RETURN(-EINVAL);
if (pages == NULL)
RETURN(-ENOMEM);
- rc = req_capsule_server_pack(&req->rq_pill);
- if (rc)
- GOTO(out, rc);
-
res = req_capsule_server_get(&req->rq_pill, &RMF_MGS_CONFIG_RES);
if (res == NULL)
GOTO(out, rc = -EINVAL);
res->mcr_offset = body->mcb_offset;
- unit_size = min_t(int, 1 << body->mcb_bits, CFS_PAGE_SIZE);
+ unit_size = min_t(int, 1 << body->mcb_bits, PAGE_CACHE_SIZE);
bytes = mgs_nidtbl_read(req->rq_export, &fsdb->fsdb_nidtbl, res,
pages, nrpages, bufsize / unit_size, unit_size);
- if (bytes < 0)
- GOTO(out, rc = bytes);
-
- /* start bulk transfer */
- page_count = (bytes + CFS_PAGE_SIZE - 1) >> CFS_PAGE_SHIFT;
- LASSERT(page_count <= nrpages);
- desc = ptlrpc_prep_bulk_exp(req, page_count,
- BULK_PUT_SOURCE, MGS_BULK_PORTAL);
- if (desc == NULL)
- GOTO(out, rc = -ENOMEM);
-
- for (i = 0; i < page_count && bytes > 0; i++) {
- ptlrpc_prep_bulk_page(desc, pages[i], 0,
- min_t(int, bytes, CFS_PAGE_SIZE));
- bytes -= CFS_PAGE_SIZE;
+ if (bytes < 0)
+ GOTO(out, rc = bytes);
+
+ /* start bulk transfer */
+ page_count = (bytes + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
+ LASSERT(page_count <= nrpages);
+ desc = ptlrpc_prep_bulk_exp(req, page_count, 1,
+ BULK_PUT_SOURCE, MGS_BULK_PORTAL);
+ if (desc == NULL)
+ GOTO(out, rc = -ENOMEM);
+
+ for (i = 0; i < page_count && bytes > 0; i++) {
+ ptlrpc_prep_bulk_page_pin(desc, pages[i], 0,
+ min_t(int, bytes, PAGE_CACHE_SIZE));
+ bytes -= PAGE_CACHE_SIZE;
}
rc = target_bulk_io(req->rq_export, desc, &lwi);
- ptlrpc_free_bulk(desc);
+ ptlrpc_free_bulk_pin(desc);
out:
- if (pages) {
- for (i = 0; i < nrpages; i++) {
- if (pages[i] == NULL)
- break;
- cfs_free_page(pages[i]);
- }
- OBD_FREE(pages, sizeof(*pages) * nrpages);
- }
- return rc;
+ for (i = 0; i < nrpages; i++) {
+ if (pages[i] == NULL)
+ break;
+ __free_page(pages[i]);
+ }
+ OBD_FREE(pages, sizeof(*pages) * nrpages);
+ return rc;
}
static int lprocfs_ir_set_state(struct fs_db *fsdb, const char *buf)
CDEBUG(D_MGS, "change fsr state of %s from %s to %s\n",
fsdb->fsdb_name, strings[fsdb->fsdb_ir_state], strings[state]);
- cfs_mutex_lock(&fsdb->fsdb_mutex);
+ mutex_lock(&fsdb->fsdb_mutex);
if (state == IR_FULL && fsdb->fsdb_nonir_clients)
state = IR_PARTIAL;
fsdb->fsdb_ir_state = state;
- cfs_mutex_unlock(&fsdb->fsdb_mutex);
+ mutex_unlock(&fsdb->fsdb_mutex);
return 0;
}
char *ptr;
int rc = 0;
- if (count > CFS_PAGE_SIZE)
+ if (count > PAGE_CACHE_SIZE)
return -EINVAL;
OBD_ALLOC(kbuf, count + 1);
if (new_fsc == NULL)
RETURN(-ENOMEM);
- CFS_INIT_LIST_HEAD(&new_fsc->mfc_export_list);
- CFS_INIT_LIST_HEAD(&new_fsc->mfc_fsdb_list);
- new_fsc->mfc_fsdb = fsdb;
- new_fsc->mfc_export = class_export_get(exp);
- new_fsc->mfc_ir_capable =
- !!(exp->exp_connect_flags & OBD_CONNECT_IMP_RECOV);
+ CFS_INIT_LIST_HEAD(&new_fsc->mfc_export_list);
+ CFS_INIT_LIST_HEAD(&new_fsc->mfc_fsdb_list);
+ new_fsc->mfc_fsdb = fsdb;
+ new_fsc->mfc_export = class_export_get(exp);
+ new_fsc->mfc_ir_capable = !!(exp_connect_flags(exp) &
+ OBD_CONNECT_IMP_RECOV);
- rc = -EEXIST;
- cfs_mutex_lock(&fsdb->fsdb_mutex);
+ rc = -EEXIST;
+ mutex_lock(&fsdb->fsdb_mutex);
- /* tend to find it in export list because this list is shorter. */
- cfs_spin_lock(&data->med_lock);
+ /* tend to find it in export list because this list is shorter. */
+ spin_lock(&data->med_lock);
cfs_list_for_each_entry(fsc, &data->med_clients, mfc_export_list) {
if (strcmp(fsname, fsc->mfc_fsdb->fsdb_name) == 0) {
found = true;
}
rc = 0;
}
- cfs_spin_unlock(&data->med_lock);
- cfs_mutex_unlock(&fsdb->fsdb_mutex);
+ spin_unlock(&data->med_lock);
+ mutex_unlock(&fsdb->fsdb_mutex);
if (new_fsc) {
class_export_put(new_fsc->mfc_export);
void mgs_fsc_cleanup(struct obd_export *exp)
{
- struct mgs_export_data *data = &exp->u.eu_mgs_data;
- struct mgs_fsc *fsc, *tmp;
- CFS_LIST_HEAD(head);
-
- cfs_spin_lock(&data->med_lock);
- cfs_list_splice_init(&data->med_clients, &head);
- cfs_spin_unlock(&data->med_lock);
-
- cfs_list_for_each_entry_safe(fsc, tmp, &head, mfc_export_list) {
- struct fs_db *fsdb = fsc->mfc_fsdb;
-
- LASSERT(fsc->mfc_export == exp);
-
- cfs_mutex_lock(&fsdb->fsdb_mutex);
- cfs_list_del_init(&fsc->mfc_fsdb_list);
- if (fsc->mfc_ir_capable == 0) {
- --fsdb->fsdb_nonir_clients;
- LASSERT(fsdb->fsdb_ir_state != IR_FULL);
- if (fsdb->fsdb_nonir_clients == 0 &&
- fsdb->fsdb_ir_state == IR_PARTIAL)
- fsdb->fsdb_ir_state = IR_FULL;
- }
- cfs_mutex_unlock(&fsdb->fsdb_mutex);
- cfs_list_del_init(&fsc->mfc_export_list);
- class_export_put(fsc->mfc_export);
- OBD_FREE_PTR(fsc);
- }
+ struct mgs_export_data *data = &exp->u.eu_mgs_data;
+ struct mgs_fsc *fsc, *tmp;
+ CFS_LIST_HEAD(head);
+
+ spin_lock(&data->med_lock);
+ cfs_list_splice_init(&data->med_clients, &head);
+ spin_unlock(&data->med_lock);
+
+ cfs_list_for_each_entry_safe(fsc, tmp, &head, mfc_export_list) {
+ struct fs_db *fsdb = fsc->mfc_fsdb;
+
+ LASSERT(fsc->mfc_export == exp);
+
+ mutex_lock(&fsdb->fsdb_mutex);
+ cfs_list_del_init(&fsc->mfc_fsdb_list);
+ if (fsc->mfc_ir_capable == 0) {
+ --fsdb->fsdb_nonir_clients;
+ LASSERT(fsdb->fsdb_ir_state != IR_FULL);
+ if (fsdb->fsdb_nonir_clients == 0 &&
+ fsdb->fsdb_ir_state == IR_PARTIAL)
+ fsdb->fsdb_ir_state = IR_FULL;
+ }
+ mutex_unlock(&fsdb->fsdb_mutex);
+ cfs_list_del_init(&fsc->mfc_export_list);
+ class_export_put(fsc->mfc_export);
+ OBD_FREE_PTR(fsc);
+ }
}
/* must be called with fsdb->fsdb_mutex held */
LASSERT(fsdb == fsc->mfc_fsdb);
cfs_list_del_init(&fsc->mfc_fsdb_list);
- cfs_spin_lock(&data->med_lock);
- cfs_list_del_init(&fsc->mfc_export_list);
- cfs_spin_unlock(&data->med_lock);
+ spin_lock(&data->med_lock);
+ cfs_list_del_init(&fsc->mfc_export_list);
+ spin_unlock(&data->med_lock);
class_export_put(fsc->mfc_export);
OBD_FREE_PTR(fsc);
}
git://git.whamcloud.com / fs / lustre-release.git / blobdiff