-/* -*- 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.
/*
* Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
* Use is subject to license terms.
+ *
+ * Copyright (c) 2011, Whamcloud, Inc.
*/
/*
* This file is part of Lustre, http://www.lustre.org/
* Author: Jinshan Xiong <jinshan.xiong@whamcloud.com>
*/
-#ifndef EXPORT_SYMTAB
-#define EXPORT_SYMTAB
-#endif
#define DEBUG_SUBSYSTEM S_MGS
#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;
* nidtbl entries will be packed in @pages by @unit_size units - entries
* shouldn't cross unit boundaries.
*/
-static int mgs_nidtbl_read(struct obd_device *unused, struct mgs_nidtbl *tbl,
+static int mgs_nidtbl_read(struct obd_export *exp, struct mgs_nidtbl *tbl,
struct mgs_config_res *res, cfs_page_t **pages,
int nrpages, int units_total, int unit_size)
{
/* check if we need to consume remaining bytes. */
if (last_in_unit != NULL && bytes_in_unit) {
- /* entry has been swapped. */
- __swab32s(&last_in_unit->mne_length);
- last_in_unit->mne_length += bytes_in_unit;
- __swab32s(&last_in_unit->mne_length);
+#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)
+ lustre_swab_mgs_nidtbl_entry(last_in_unit);
+#endif
+ last_in_unit->mne_length += bytes_in_unit;
+#if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 6, 50, 0)
+ if (exp->exp_need_mne_swab)
+ lustre_swab_mgs_nidtbl_entry(last_in_unit);
+#endif
rc += bytes_in_unit;
buf += bytes_in_unit;
last_in_unit = NULL;
entry->mne_nid_count = mti->mti_nid_count;
memcpy(entry->u.nids, mti->mti_nids,
mti->mti_nid_count * sizeof(lnet_nid_t));
- lustre_swab_mgs_nidtbl_entry(entry);
+
+#if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 6, 50, 0)
+ /* For LU-1644, swab entry for 2.2 clients. */
+ if (exp->exp_need_mne_swab)
+ lustre_swab_mgs_nidtbl_entry(entry);
+#endif
version = tgt->mnt_version;
rc += entry_len;
RETURN(rc);
}
-static int nidtbl_update_version(struct obd_device *obd, struct mgs_nidtbl *tbl)
+static int nidtbl_update_version(const struct lu_env *env,
+ 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 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);
-
- 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_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);
- }
+ LASSERT(cfs_mutex_is_locked(&tbl->mn_lock));
- pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
- RETURN(rc);
+ 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));
+
+ th = dt_trans_create(env, mgs->mgs_bottom);
+ if (IS_ERR(th))
+ GOTO(out_put, rc = PTR_ERR(th));
+
+ th->th_sync = 1; /* update table synchronously */
+ rc = dt_declare_record_write(env, fsdb, buf.lb_len, off, th);
+ if (rc)
+ GOTO(out, 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(struct obd_device *obd, struct mgs_nidtbl *tbl)
+static int nidtbl_read_version(const struct lu_env *env,
+ 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;
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;
- }
- 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(struct fs_db *fsdb, struct mgs_target_info *mti)
+static int mgs_nidtbl_write(const struct lu_env *env, struct fs_db *fsdb,
+ struct mgs_target_info *mti)
{
struct mgs_nidtbl *tbl;
struct mgs_nidtbl_target *tgt;
cfs_list_move_tail(&tgt->mnt_list, &tbl->mn_targets);
- rc = nidtbl_update_version(fsdb->fsdb_obd, tbl);
+ rc = nidtbl_update_version(env, fsdb->fsdb_mgs, tbl);
EXIT;
out:
}
}
-static int mgs_nidtbl_init_fs(struct fs_db *fsdb)
+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);
tbl->mn_nr_targets = 0;
tbl->mn_fsdb = fsdb;
cfs_mutex_lock(&tbl->mn_lock);
- tbl->mn_version = nidtbl_read_version(fsdb->fsdb_obd, tbl);
+ rc = nidtbl_read_version(env, fsdb->fsdb_mgs, tbl, &tbl->mn_version);
cfs_mutex_unlock(&tbl->mn_lock);
- CDEBUG(D_MGS, "IR: current version is %llu\n", tbl->mn_version);
-
- return 0;
+ 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 ------- */
+void mgs_ir_notify_complete(struct fs_db *fsdb)
+{
+ struct timeval tv;
+ cfs_duration_t delta;
+
+ cfs_atomic_set(&fsdb->fsdb_notify_phase, 0);
+
+ /* do statistic */
+ fsdb->fsdb_notify_count++;
+ delta = cfs_time_sub(cfs_time_current(), fsdb->fsdb_notify_start);
+ fsdb->fsdb_notify_total += delta;
+ if (delta > fsdb->fsdb_notify_max)
+ fsdb->fsdb_notify_max = delta;
+
+ cfs_duration_usec(delta, &tv);
+ CDEBUG(D_MGS, "Revoke recover lock of %s completed after %ld.%06lds\n",
+ fsdb->fsdb_name, tv.tv_sec, tv.tv_usec);
+}
+
static int mgs_ir_notify(void *arg)
{
struct fs_db *fsdb = arg;
cfs_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 };
- struct lustre_handle lockh;
- cfs_time_t curtime;
- int lockrc;
- int delta;
l_wait_event(fsdb->fsdb_notify_waitq,
fsdb->fsdb_notify_stop ||
CDEBUG(D_MGS, "%s woken up, phase is %d\n",
name, cfs_atomic_read(&fsdb->fsdb_notify_phase));
- curtime = cfs_time_current();
- lockrc = mgs_get_lock(fsdb->fsdb_obd, &resid, &lockh);
- if (lockrc == ELDLM_OK) {
- cfs_atomic_set(&fsdb->fsdb_notify_phase, 0);
- mgs_put_lock(&lockh);
-
- /* do statistic */
- fsdb->fsdb_notify_count++;
- delta = (cfs_time_current() - curtime) / NSEC_PER_USEC;
- fsdb->fsdb_notify_total += delta;
- if (delta > fsdb->fsdb_notify_max)
- fsdb->fsdb_notify_max = delta;
- CDEBUG(D_MGS, "Revoke recover lock of %s spent %dus\n",
- fsdb->fsdb_name, delta);
- } else {
- CERROR("Fatal error %d for fs %s\n",
- lockrc, fsdb->fsdb_name);
- }
+ 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;
}
-int mgs_ir_init_fs(struct obd_device *obd, struct fs_db *fsdb)
+int mgs_ir_init_fs(const struct lu_env *env, struct mgs_device *mgs,
+ struct fs_db *fsdb)
{
- struct mgs_obd *mgs = &obd->u.mgs;
int rc;
if (!ir_timeout)
CFS_INIT_LIST_HEAD(&fsdb->fsdb_clients);
/* start notify thread */
- fsdb->fsdb_obd = 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);
else
CERROR("Start notify thread error %d\n", rc);
- mgs_nidtbl_init_fs(fsdb);
+ mgs_nidtbl_init_fs(env, fsdb);
return 0;
}
-void mgs_ir_fini_fs(struct obd_device *obd, struct fs_db *fsdb)
+void mgs_ir_fini_fs(struct mgs_device *mgs, struct fs_db *fsdb)
{
if (cfs_test_bit(FSDB_MGS_SELF, &fsdb->fsdb_flags))
return;
cfs_wait_for_completion(&fsdb->fsdb_notify_comp);
}
-/* caller must have held fsdb_sem */
+/* caller must have held fsdb_mutex */
static inline void ir_state_graduate(struct fs_db *fsdb)
{
- struct mgs_obd *mgs = &fsdb->fsdb_obd->u.mgs;
-
if (fsdb->fsdb_ir_state == IR_STARTUP) {
- if (cfs_time_before(mgs->mgs_start_time + ir_timeout,
+ if (cfs_time_before(fsdb->fsdb_mgs->mgs_start_time + ir_timeout,
cfs_time_current_sec())) {
fsdb->fsdb_ir_state = IR_FULL;
if (fsdb->fsdb_nonir_clients)
}
}
-int mgs_ir_update(struct obd_device *obd, struct mgs_target_info *mti)
+int mgs_ir_update(const struct lu_env *env, struct mgs_device *mgs,
+ struct mgs_target_info *mti)
{
struct fs_db *fsdb;
bool notify = true;
if (mti->mti_instance == 0)
return -EINVAL;
- rc = mgs_find_or_make_fsdb(obd, mti->mti_fsname, &fsdb);
+ rc = mgs_find_or_make_fsdb(env, mgs, mti->mti_fsname, &fsdb);
if (rc)
return rc;
- rc = mgs_nidtbl_write(fsdb, mti);
+ rc = mgs_nidtbl_write(env, fsdb, mti);
if (rc)
return rc;
/* check ir state */
- cfs_down(&fsdb->fsdb_sem);
+ cfs_mutex_lock(&fsdb->fsdb_mutex);
ir_state_graduate(fsdb);
switch (fsdb->fsdb_ir_state) {
case IR_FULL:
default:
LBUG();
}
- cfs_up(&fsdb->fsdb_sem);
+ cfs_mutex_unlock(&fsdb->fsdb_mutex);
LASSERT(ergo(mti->mti_flags & LDD_F_IR_CAPABLE, notify));
if (notify) {
int mgs_get_ir_logs(struct ptlrpc_request *req)
{
- struct obd_device *obd = req->rq_export->exp_obd;
+ struct lu_env *env = req->rq_svc_thread->t_env;
+ struct mgs_device *mgs = exp2mgs_dev(req->rq_export);
struct fs_db *fsdb;
struct mgs_config_body *body;
struct mgs_config_res *res;
char fsname[16];
long bufsize;
int unit_size;
-
int type;
int rc = 0;
int i;
if (rc)
RETURN(rc);
- rc = mgs_find_or_make_fsdb(obd, fsname, &fsdb);
+ rc = mgs_find_or_make_fsdb(env, mgs, fsname, &fsdb);
if (rc)
- GOTO(out, rc);
+ RETURN(rc);
bufsize = body->mcb_units << body->mcb_bits;
nrpages = (bufsize + CFS_PAGE_SIZE - 1) >> CFS_PAGE_SHIFT;
res->mcr_offset = body->mcb_offset;
unit_size = min_t(int, 1 << body->mcb_bits, CFS_PAGE_SIZE);
- bytes = mgs_nidtbl_read(obd, &fsdb->fsdb_nidtbl, res, pages, nrpages,
- bufsize / unit_size, unit_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);
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_down(&fsdb->fsdb_sem);
+ cfs_mutex_lock(&fsdb->fsdb_mutex);
if (state == IR_FULL && fsdb->fsdb_nonir_clients)
state = IR_PARTIAL;
fsdb->fsdb_ir_state = state;
- cfs_up(&fsdb->fsdb_sem);
+ cfs_mutex_unlock(&fsdb->fsdb_mutex);
return 0;
}
struct fs_db *fsdb = data;
struct mgs_nidtbl *tbl = &fsdb->fsdb_nidtbl;
const char *ir_strings[] = IR_STRINGS;
+ struct timeval tv_max;
+ struct timeval tv;
- /* mgs_live_seq_show() already holds fsdb_sem. */
+ /* mgs_live_seq_show() already holds fsdb_mutex. */
ir_state_graduate(fsdb);
+ seq_printf(seq, "\nimperative_recovery_state:\n");
seq_printf(seq,
- "\tstate: %s, nonir clients: %d\n"
- "\tnidtbl version: %lld\n",
+ " state: %s\n"
+ " nonir_clients: %d\n"
+ " nidtbl_version: %lld\n",
ir_strings[fsdb->fsdb_ir_state], fsdb->fsdb_nonir_clients,
tbl->mn_version);
- seq_printf(seq, "\tnotify total/max/count: %u/%u/%u\n",
- fsdb->fsdb_notify_total, fsdb->fsdb_notify_max,
+
+ cfs_duration_usec(fsdb->fsdb_notify_total, &tv);
+ cfs_duration_usec(fsdb->fsdb_notify_max, &tv_max);
+
+ seq_printf(seq, " notify_duration_total: %lu.%06lu\n"
+ " notify_duation_max: %lu.%06lu\n"
+ " notify_count: %u\n",
+ tv.tv_sec, tv.tv_usec,
+ tv_max.tv_sec, tv_max.tv_usec,
fsdb->fsdb_notify_count);
+
return 0;
}
/* --------------- Handle non IR support clients --------------- */
/* attach a lustre file system to an export */
-int mgs_fsc_attach(struct obd_export *exp, char *fsname)
+int mgs_fsc_attach(const struct lu_env *env, struct obd_export *exp,
+ char *fsname)
{
struct mgs_export_data *data = &exp->u.eu_mgs_data;
- struct obd_device *obd = exp->exp_obd;
+ struct mgs_device *mgs = exp2mgs_dev(exp);
struct fs_db *fsdb;
struct mgs_fsc *fsc = NULL;
struct mgs_fsc *new_fsc = NULL;
int rc;
ENTRY;
- rc = mgs_find_or_make_fsdb(obd, fsname, &fsdb);
+ rc = mgs_find_or_make_fsdb(env, mgs, fsname, &fsdb);
if (rc)
RETURN(rc);
!!(exp->exp_connect_flags & OBD_CONNECT_IMP_RECOV);
rc = -EEXIST;
- cfs_down(&fsdb->fsdb_sem);
+ cfs_mutex_lock(&fsdb->fsdb_mutex);
/* tend to find it in export list because this list is shorter. */
cfs_spin_lock(&data->med_lock);
rc = 0;
}
cfs_spin_unlock(&data->med_lock);
- cfs_up(&fsdb->fsdb_sem);
+ cfs_mutex_unlock(&fsdb->fsdb_mutex);
if (new_fsc) {
class_export_put(new_fsc->mfc_export);
LASSERT(fsc->mfc_export == exp);
- cfs_down(&fsdb->fsdb_sem);
+ cfs_mutex_lock(&fsdb->fsdb_mutex);
cfs_list_del_init(&fsc->mfc_fsdb_list);
if (fsc->mfc_ir_capable == 0) {
--fsdb->fsdb_nonir_clients;
fsdb->fsdb_ir_state == IR_PARTIAL)
fsdb->fsdb_ir_state = IR_FULL;
}
- cfs_up(&fsdb->fsdb_sem);
+ cfs_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_sem held */
+/* must be called with fsdb->fsdb_mutex held */
void mgs_fsc_cleanup_by_fsdb(struct fs_db *fsdb)
{
struct mgs_fsc *fsc, *tmp;
git://git.whamcloud.com / fs / lustre-release.git / blobdiff