#define D_MGS D_CONFIG
#ifdef __KERNEL__
-#include <linux/module.h>
#include <linux/pagemap.h>
-#include <linux/fs.h>
#endif
#include <obd.h>
-#include <obd_lov.h>
#include <obd_class.h>
-#include <lustre_log.h>
-#include <obd_ost.h>
-#include <libcfs/list.h>
-#include <linux/lvfs.h>
-#include <lustre_fsfilt.h>
#include <lustre_disk.h>
-#include <lustre_param.h>
+
#include "mgs_internal.h"
static unsigned int ir_timeout;
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;
LASSERT((unit_size & (unit_size - 1)) == 0);
LASSERT(nrpages << CFS_PAGE_SHIFT >= units_total * unit_size);
- cfs_mutex_lock(&tbl->mn_lock);
+ mutex_lock(&tbl->mn_lock);
LASSERT(nidtbl_is_sane(tbl));
/* no more entries ? */
/* check if we need to consume remaining bytes. */
if (last_in_unit != NULL && bytes_in_unit) {
-#
#if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 6, 50, 0)
/* May need to swab back to update the length.*/
if (exp->exp_need_mne_swab)
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",
struct mgs_device *mgs,
struct mgs_nidtbl *tbl)
{
- struct lvfs_run_ctxt saved;
- struct file *file = NULL;
- char filename[sizeof(MGS_NIDTBL_DIR) + 9];
- u64 version;
- loff_t off = 0;
- int rc;
- struct obd_device *obd = tbl->mn_fsdb->fsdb_obd;
+ struct dt_object *fsdb;
+ struct thandle *th;
+ u64 version;
+ struct lu_buf buf = {
+ .lb_buf = &version,
+ .lb_len = sizeof(version)
+ };
+ loff_t off = 0;
+ int rc;
ENTRY;
- LASSERT(cfs_mutex_is_locked(&tbl->mn_lock));
- LASSERT(sizeof(filename) < 32);
+ LASSERT(mutex_is_locked(&tbl->mn_lock));
- sprintf(filename, "%s/%s",
- MGS_NIDTBL_DIR, tbl->mn_fsdb->fsdb_name);
+ fsdb = local_file_find_or_create(env, mgs->mgs_los, mgs->mgs_nidtbl_dir,
+ tbl->mn_fsdb->fsdb_name,
+ S_IFREG | S_IRUGO | S_IWUSR);
+ if (IS_ERR(fsdb))
+ RETURN(PTR_ERR(fsdb));
- push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
+ th = dt_trans_create(env, mgs->mgs_bottom);
+ if (IS_ERR(th))
+ GOTO(out_put, rc = PTR_ERR(th));
- file = l_filp_open(filename, O_RDWR|O_CREAT, 0660);
- if (!IS_ERR(file)) {
- version = cpu_to_le64(tbl->mn_version);
- rc = lustre_fwrite(file, &version, sizeof(version), &off);
- if (rc == sizeof(version))
- rc = 0;
- filp_close(file, 0);
- fsfilt_sync(obd, obd->u.mgs.mgs_sb);
- } else {
- rc = PTR_ERR(file);
- }
+ th->th_sync = 1; /* update table synchronously */
+ rc = dt_declare_record_write(env, fsdb, buf.lb_len, off, th);
+ if (rc)
+ GOTO(out, rc);
- pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
- RETURN(rc);
+ rc = dt_trans_start_local(env, mgs->mgs_bottom, th);
+ if (rc)
+ GOTO(out, rc);
+
+ version = cpu_to_le64(tbl->mn_version);
+ rc = dt_record_write(env, fsdb, &buf, &off, th);
+
+out:
+ dt_trans_stop(env, mgs->mgs_bottom, th);
+out_put:
+ lu_object_put(env, &fsdb->do_lu);
+ RETURN(rc);
}
#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 lvfs_run_ctxt saved;
- struct file *file = NULL;
- char filename[sizeof(MGS_NIDTBL_DIR) + 9];
- u64 version;
+ struct dt_object *fsdb;
+ struct lu_fid fid;
+ u64 tmpver;
+ struct lu_buf buf = {
+ .lb_buf = &tmpver,
+ .lb_len = sizeof(tmpver)
+ };
loff_t off = 0;
int rc;
- struct obd_device *obd = tbl->mn_fsdb->fsdb_obd;
ENTRY;
- LASSERT(cfs_mutex_is_locked(&tbl->mn_lock));
- LASSERT(sizeof(filename) < 32);
-
- sprintf(filename, "%s/%s",
- MGS_NIDTBL_DIR, tbl->mn_fsdb->fsdb_name);
-
- push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
-
- file = l_filp_open(filename, O_RDONLY, 0);
- if (!IS_ERR(file)) {
- rc = lustre_fread(file, &version, sizeof(version), &off);
- if (rc == sizeof(version))
- rc = cpu_to_le64(version);
- else if (rc == 0)
- rc = MGS_NIDTBL_VERSION_INIT;
- else
- CERROR("read version file %s error %d\n", filename, rc);
- filp_close(file, 0);
- } else {
- rc = PTR_ERR(file);
- if (rc == -ENOENT)
- rc = MGS_NIDTBL_VERSION_INIT;
- }
+ LASSERT(mutex_is_locked(&tbl->mn_lock));
- pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
- RETURN(rc);
+ LASSERT(mgs->mgs_nidtbl_dir);
+ rc = dt_lookup_dir(env, mgs->mgs_nidtbl_dir, tbl->mn_fsdb->fsdb_name,
+ &fid);
+ 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);
+ if (IS_ERR(fsdb))
+ RETURN(PTR_ERR(fsdb));
+
+ rc = dt_read(env, fsdb, &buf, &off);
+ 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);
}
static int mgs_nidtbl_write(const struct lu_env *env, struct fs_db *fsdb,
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);
-
- return 0;
+ 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 rc;
}
/* --------- Imperative Recovery relies on nidtbl stuff ------- */
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);
mgs_revoke_lock(fsdb->fsdb_mgs, fsdb, CONFIG_T_RECOVER);
}
- cfs_complete(&fsdb->fsdb_notify_comp);
+ complete(&fsdb->fsdb_notify_comp);
return 0;
}
CFS_INIT_LIST_HEAD(&fsdb->fsdb_clients);
/* start notify thread */
- fsdb->fsdb_obd = mgs->mgs_obd;
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);
+ 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);
+ wait_for_completion(&fsdb->fsdb_notify_comp);
else
CERROR("Start notify thread error %d\n", rc);
void mgs_ir_fini_fs(struct mgs_device *mgs, struct fs_db *fsdb)
{
- if (cfs_test_bit(FSDB_MGS_SELF, &fsdb->fsdb_flags))
+ if (test_bit(FSDB_MGS_SELF, &fsdb->fsdb_flags))
return;
mgs_fsc_cleanup_by_fsdb(fsdb);
fsdb->fsdb_notify_stop = 1;
cfs_waitq_signal(&fsdb->fsdb_notify_waitq);
- cfs_wait_for_completion(&fsdb->fsdb_notify_comp);
+ 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);
+ mutex_unlock(&fsdb->fsdb_mutex);
LASSERT(ergo(mti->mti_flags & LDD_F_IR_CAPABLE, notify));
if (notify) {
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));
+ ptlrpc_prep_bulk_page_pin(desc, pages[i], 0,
+ min_t(int, bytes, CFS_PAGE_SIZE));
bytes -= CFS_PAGE_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;
+ cfs_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;
}
!!(exp->exp_connect_flags & OBD_CONNECT_IMP_RECOV);
rc = -EEXIST;
- cfs_mutex_lock(&fsdb->fsdb_mutex);
+ 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);
}