Whamcloud - gitweb
LU-1302 llog: structures changes, llog_thread_info
[fs/lustre-release.git] / lustre / obdclass / llog_obd.c
index bcecaae..aa9c3c0 100644 (file)
-
-/* -*- 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.
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 only,
+ * as published by the Free Software Foundation.
+ *
+ * This program is distributed in the hope that it will be useful, but
+ * WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * General Public License version 2 for more details (a copy is included
+ * in the LICENSE file that accompanied this code).
+ *
+ * You should have received a copy of the GNU General Public License
+ * version 2 along with this program; If not, see
+ * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
+ *
+ * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
+ * CA 95054 USA or visit www.sun.com if you need additional information or
+ * have any questions.
+ *
+ * GPL HEADER END
+ */
+/*
+ * Copyright (c) 2003, 2010, Oracle and/or its affiliates. All rights reserved.
+ * Use is subject to license terms.
  *
- *   You should have received a copy of the GNU General Public License
- *   along with Lustre; if not, write to the Free Software
- *   Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
+ * Copyright (c) 2012, Whamcloud, Inc.
+ */
+/*
+ * This file is part of Lustre, http://www.lustre.org/
+ * Lustre is a trademark of Sun Microsystems, Inc.
  */
 
 #define DEBUG_SUBSYSTEM S_LOG
 
-#ifndef EXPORT_SYMTAB
-#define EXPORT_SYMTAB
-#endif
-
-#ifdef __KERNEL__
-#include <linux/fs.h>
-#else
+#ifndef __KERNEL__
 #include <liblustre.h>
 #endif
 
-#include <linux/obd_class.h>
-#include <linux/lustre_log.h>
-#include <linux/lustre_mds.h>
+#include <obd_class.h>
+#include <lustre_log.h>
 #include <libcfs/list.h>
+#include "llog_internal.h"
 
 /* helper functions for calling the llog obd methods */
+static struct llog_ctxt* llog_new_ctxt(struct obd_device *obd)
+{
+        struct llog_ctxt *ctxt;
+
+        OBD_ALLOC_PTR(ctxt);
+        if (!ctxt)
+                return NULL;
 
-int obd_llog_setup(struct obd_device *obd, struct obd_llogs *llogs, 
-                   int index, struct obd_device *disk_obd, int count, 
-                   struct llog_logid *logid, struct llog_operations *op)
+        ctxt->loc_obd = obd;
+        cfs_atomic_set(&ctxt->loc_refcount, 1);
+
+        return ctxt;
+}
+
+static void llog_ctxt_destroy(struct llog_ctxt *ctxt)
 {
+        if (ctxt->loc_exp) {
+                class_export_put(ctxt->loc_exp);
+                ctxt->loc_exp = NULL;
+        }
+        if (ctxt->loc_imp) {
+                class_import_put(ctxt->loc_imp);
+                ctxt->loc_imp = NULL;
+        }
+        LASSERT(ctxt->loc_llcd == NULL);
+        OBD_FREE_PTR(ctxt);
+}
+
+int __llog_ctxt_put(struct llog_ctxt *ctxt)
+{
+        struct obd_llog_group *olg = ctxt->loc_olg;
+        struct obd_device *obd;
         int rc = 0;
-        struct llog_ctxt *ctxt;
+
+        cfs_spin_lock(&olg->olg_lock);
+        if (!cfs_atomic_dec_and_test(&ctxt->loc_refcount)) {
+                cfs_spin_unlock(&olg->olg_lock);
+                return rc;
+        }
+        olg->olg_ctxts[ctxt->loc_idx] = NULL;
+        cfs_spin_unlock(&olg->olg_lock);
+
+        if (ctxt->loc_lcm)
+                lcm_put(ctxt->loc_lcm);
+
+        obd = ctxt->loc_obd;
+        cfs_spin_lock(&obd->obd_dev_lock);
+        /* sync with llog ctxt user thread */
+        cfs_spin_unlock(&obd->obd_dev_lock);
+
+        /* obd->obd_starting is needed for the case of cleanup
+         * in error case while obd is starting up. */
+        LASSERTF(obd->obd_starting == 1 ||
+                 obd->obd_stopping == 1 || obd->obd_set_up == 0,
+                 "wrong obd state: %d/%d/%d\n", !!obd->obd_starting,
+                 !!obd->obd_stopping, !!obd->obd_set_up);
+
+        /* cleanup the llog ctxt here */
+        if (CTXTP(ctxt, cleanup))
+                rc = CTXTP(ctxt, cleanup)(ctxt);
+
+        llog_ctxt_destroy(ctxt);
+        cfs_waitq_signal(&olg->olg_waitq);
+        return rc;
+}
+EXPORT_SYMBOL(__llog_ctxt_put);
+
+int llog_cleanup(struct llog_ctxt *ctxt)
+{
+        struct l_wait_info lwi = LWI_INTR(LWI_ON_SIGNAL_NOOP, NULL);
+        struct obd_llog_group *olg;
+        int rc, idx;
         ENTRY;
 
-        LASSERT(llogs);
+        LASSERT(ctxt != NULL);
+        LASSERT(ctxt != LP_POISON);
+
+        olg = ctxt->loc_olg;
+        LASSERT(olg != NULL);
+        LASSERT(olg != LP_POISON);
+
+        idx = ctxt->loc_idx;
+
+        /* 
+         * Banlance the ctxt get when calling llog_cleanup()
+         */
+        LASSERT(cfs_atomic_read(&ctxt->loc_refcount) < LI_POISON);
+        LASSERT(cfs_atomic_read(&ctxt->loc_refcount) > 1);
+        llog_ctxt_put(ctxt);
+
+        /* 
+         * Try to free the ctxt. 
+         */
+        rc = __llog_ctxt_put(ctxt);
+        if (rc)
+                CERROR("Error %d while cleaning up ctxt %p\n",
+                       rc, ctxt);
+
+        l_wait_event(olg->olg_waitq,
+                     llog_group_ctxt_null(olg, idx), &lwi);
+
+        RETURN(rc);
+}
+EXPORT_SYMBOL(llog_cleanup);
+
+int llog_setup_named(struct obd_device *obd,  struct obd_llog_group *olg,
+                     int index, struct obd_device *disk_obd, int count,
+                     struct llog_logid *logid, const char *logname,
+                     struct llog_operations *op)
+{
+        struct llog_ctxt *ctxt;
+        int rc = 0;
+        ENTRY;
 
         if (index < 0 || index >= LLOG_MAX_CTXTS)
-                RETURN(-EFAULT);
+                RETURN(-EINVAL);
 
-        OBD_ALLOC(ctxt, sizeof(*ctxt));
+        LASSERT(olg != NULL);
+
+        ctxt = llog_new_ctxt(obd);
         if (!ctxt)
                 RETURN(-ENOMEM);
 
-        llogs->llog_ctxt[index] = ctxt;
-        obd->obd_llog_ctxt[index] = ctxt;
-        ctxt->loc_logops = op;
         ctxt->loc_obd = obd;
-        ctxt->loc_llogs = llogs;
+        ctxt->loc_olg = olg;
         ctxt->loc_idx = index;
-        ctxt->loc_alone = 0;
+        ctxt->loc_logops = op;
+        cfs_mutex_init(&ctxt->loc_mutex);
         ctxt->loc_exp = class_export_get(disk_obd->obd_self_export);
-        sema_init(&ctxt->loc_sem, 1);
-
-        if (op->lop_close == llog_lvfs_ops.lop_close) {
-                ctxt->loc_fsops = disk_obd->obd_fsops;
-                ctxt->loc_lvfs_ctxt = &disk_obd->obd_lvfs_ctxt;
-                if (!strcmp(disk_obd->obd_type->typ_name, LUSTRE_MDS_NAME)) {
-                        struct mds_obd *mds = &disk_obd->u.mds;
-                        ctxt->loc_objects_dir = mds->mds_objects_dir;
-                        ctxt->loc_logs_dir = mds->mds_logs_dir;
+        ctxt->loc_flags = LLOG_CTXT_FLAG_UNINITIALIZED;
+
+        rc = llog_group_set_ctxt(olg, ctxt, index);
+        if (rc) {
+                llog_ctxt_destroy(ctxt);
+                if (rc == -EEXIST) {
+                        ctxt = llog_group_get_ctxt(olg, index);
+                        if (ctxt) {
+                                /*
+                                 * mds_lov_update_desc() might call here multiple
+                                 * times. So if the llog is already set up then
+                                 * don't to do it again. 
+                                 */
+                                CDEBUG(D_CONFIG, "obd %s ctxt %d already set up\n",
+                                       obd->obd_name, index);
+                                LASSERT(ctxt->loc_olg == olg);
+                                LASSERT(ctxt->loc_obd == obd);
+                                LASSERT(ctxt->loc_exp == disk_obd->obd_self_export);
+                                LASSERT(ctxt->loc_logops == op);
+                                llog_ctxt_put(ctxt);
+                        }
+                        rc = 0;
                 }
+                RETURN(rc);
         }
 
-        if (op->lop_setup)
-                rc = op->lop_setup(obd, llogs, index, disk_obd, count, logid);
-        if (ctxt && rc) {
-                obd->obd_llog_ctxt[index] = NULL;
-                OBD_FREE(ctxt, sizeof(*ctxt));
+        if (OBD_FAIL_CHECK(OBD_FAIL_OBD_LLOG_SETUP)) {
+                rc = -ENOTSUPP;
+        } else {
+                if (op->lop_setup)
+                        rc = op->lop_setup(obd, olg, index, disk_obd, count,
+                                           logid, logname);
+        }
+
+        if (rc) {
+                CERROR("obd %s ctxt %d lop_setup=%p failed %d\n",
+                       obd->obd_name, index, op->lop_setup, rc);
+                llog_ctxt_put(ctxt);
+        } else {
+                CDEBUG(D_CONFIG, "obd %s ctxt %d is initialized\n",
+                       obd->obd_name, index);
+                ctxt->loc_flags &= ~LLOG_CTXT_FLAG_UNINITIALIZED;
         }
 
         RETURN(rc);
 }
-EXPORT_SYMBOL(obd_llog_setup);
+EXPORT_SYMBOL(llog_setup_named);
 
-int obd_llog_cleanup(struct llog_ctxt *ctxt)
+int llog_setup(struct obd_device *obd,  struct obd_llog_group *olg,
+               int index, struct obd_device *disk_obd, int count,
+               struct llog_logid *logid, struct llog_operations *op)
+{
+        return llog_setup_named(obd,olg,index,disk_obd,count,logid,NULL,op);
+}
+EXPORT_SYMBOL(llog_setup);
+
+int llog_sync(struct llog_ctxt *ctxt, struct obd_export *exp, int flags)
 {
         int rc = 0;
         ENTRY;
 
-        if (ctxt == NULL)
+        if (!ctxt)
                 RETURN(0);
 
-        if (CTXTP(ctxt, cleanup))  
-                rc = CTXTP(ctxt, cleanup)(ctxt);
-                
-        ctxt->loc_obd->obd_llog_ctxt[ctxt->loc_idx] = NULL;
-        ctxt->loc_llogs->llog_ctxt[ctxt->loc_idx] = NULL;
-        class_export_put(ctxt->loc_exp);
-        ctxt->loc_exp = NULL;
-        OBD_FREE(ctxt, sizeof(*ctxt));
+        if (CTXTP(ctxt, sync))
+               rc = CTXTP(ctxt, sync)(ctxt, exp, flags);
+
+        RETURN(rc);
+}
+EXPORT_SYMBOL(llog_sync);
+
+int llog_add(struct llog_ctxt *ctxt, struct llog_rec_hdr *rec,
+             struct lov_stripe_md *lsm, struct llog_cookie *logcookies,
+             int numcookies)
+{
+        int raised, rc;
+        ENTRY;
+
+        if (!ctxt) {
+                CERROR("No ctxt\n");
+                RETURN(-ENODEV);
+        }
+
+        if (ctxt->loc_flags & LLOG_CTXT_FLAG_UNINITIALIZED)
+                RETURN(-ENXIO);
+
+
+        CTXT_CHECK_OP(ctxt, add, -EOPNOTSUPP);
+        raised = cfs_cap_raised(CFS_CAP_SYS_RESOURCE);
+        if (!raised)
+                cfs_cap_raise(CFS_CAP_SYS_RESOURCE);
+        rc = CTXTP(ctxt, add)(ctxt, rec, lsm, logcookies, numcookies);
+        if (!raised)
+                cfs_cap_lower(CFS_CAP_SYS_RESOURCE);
+        RETURN(rc);
+}
+EXPORT_SYMBOL(llog_add);
+
+int llog_cancel(struct llog_ctxt *ctxt, struct lov_stripe_md *lsm,
+                int count, struct llog_cookie *cookies, int flags)
+{
+        int rc;
+        ENTRY;
+
+        if (!ctxt) {
+                CERROR("No ctxt\n");
+                RETURN(-ENODEV);
+        }
+
+        CTXT_CHECK_OP(ctxt, cancel, -EOPNOTSUPP);
+        rc = CTXTP(ctxt, cancel)(ctxt, lsm, count, cookies, flags);
         RETURN(rc);
 }
-EXPORT_SYMBOL(obd_llog_cleanup);
+EXPORT_SYMBOL(llog_cancel);
 
 /* callback func for llog_process in llog_obd_origin_setup */
 static int cat_cancel_cb(struct llog_handle *cathandle,
@@ -100,35 +296,58 @@ static int cat_cancel_cb(struct llog_handle *cathandle,
 {
         struct llog_logid_rec *lir = (struct llog_logid_rec *)rec;
         struct llog_handle *loghandle;
-        int rc;
+        struct llog_log_hdr *llh;
+        int rc, index;
         ENTRY;
 
         if (rec->lrh_type != LLOG_LOGID_MAGIC) {
                 CERROR("invalid record in catalog\n");
                 RETURN(-EINVAL);
         }
-        CDEBUG(D_HA, "processing log "LPX64":%x at index %u of catalog "LPX64"\n",
-               lir->lid_id.lgl_oid, lir->lid_id.lgl_ogen,
+        CDEBUG(D_HA, "processing log "LPX64":%x at index %u of catalog "
+               LPX64"\n", lir->lid_id.lgl_oid, lir->lid_id.lgl_ogen,
                rec->lrh_index, cathandle->lgh_id.lgl_oid);
 
         rc = llog_cat_id2handle(cathandle, &loghandle, &lir->lid_id);
         if (rc) {
                 CERROR("Cannot find handle for log "LPX64"\n",
                        lir->lid_id.lgl_oid);
+                if (rc == -ENOENT) {
+                        index = rec->lrh_index;
+                        goto cat_cleanup;
+                }
                 RETURN(rc);
         }
 
-        if (cathandle->lgh_last_idx == loghandle->u.phd.phd_cookie.lgc_index)
-                cathandle->u.chd.chd_current_log = loghandle;
+        llh = loghandle->lgh_hdr;
+        if ((llh->llh_flags & LLOG_F_ZAP_WHEN_EMPTY) &&
+            (llh->llh_count == 1)) {
+                rc = llog_destroy(loghandle);
+                if (rc)
+                        CERROR("failure destroying log in postsetup: %d\n", rc);
+
+                index = loghandle->u.phd.phd_cookie.lgc_index;
+                llog_free_handle(loghandle);
+
+cat_cleanup:
+                LASSERT(index);
+                llog_cat_set_first_idx(cathandle, index);
+                rc = llog_cancel_rec(cathandle, index);
+                if (rc == 0)
+                        CDEBUG(D_HA, "cancel log "LPX64":%x at index %u of catalog "
+                              LPX64"\n", lir->lid_id.lgl_oid,
+                              lir->lid_id.lgl_ogen, rec->lrh_index,
+                              cathandle->lgh_id.lgl_oid);
+        }
 
         RETURN(rc);
 }
 
 /* lop_setup method for filter/osc */
 // XXX how to set exports
-int llog_obd_origin_setup(struct obd_device *obd, struct obd_llogs *llogs,
+int llog_obd_origin_setup(struct obd_device *obd, struct obd_llog_group *olg,
                           int index, struct obd_device *disk_obd, int count,
-                          struct llog_logid *logid)
+                          struct llog_logid *logid, const char *name)
 {
         struct llog_ctxt *ctxt;
         struct llog_handle *handle;
@@ -141,99 +360,138 @@ int llog_obd_origin_setup(struct obd_device *obd, struct obd_llogs *llogs,
 
         LASSERT(count == 1);
 
-        ctxt = llog_get_context(llogs, index);
-        LASSERT(ctxt);
-        llog_gen_init(ctxt);
+        LASSERT(olg != NULL);
+        ctxt = llog_group_get_ctxt(olg, index);
+        if (!ctxt)
+                RETURN(-ENODEV);
 
-        if (logid->lgl_oid) {
-                rc = llog_open(ctxt, &handle, logid, NULL, 0);
+        if (logid && logid->lgl_oid) {
+                rc = llog_create(ctxt, &handle, logid, NULL);
         } else {
-                rc = llog_open(ctxt, &handle, NULL, NULL, 0);
-                if (!rc)
+                rc = llog_create(ctxt, &handle, NULL, (char *)name);
+                if (!rc && logid)
                         *logid = handle->lgh_id;
         }
         if (rc)
-                RETURN(rc);
+                GOTO(out, rc);
 
         ctxt->loc_handle = handle;
         push_ctxt(&saved, &disk_obd->obd_lvfs_ctxt, NULL);
         rc = llog_init_handle(handle, LLOG_F_IS_CAT, NULL);
         pop_ctxt(&saved, &disk_obd->obd_lvfs_ctxt, NULL);
         if (rc)
-                RETURN(rc);
+                GOTO(out, rc);
 
         rc = llog_process(handle, (llog_cb_t)cat_cancel_cb, NULL, NULL);
         if (rc)
-                CERROR("llog_process with cat_cancel_cb failed: %d\n", rc);
-
-        if (ctxt && rc) {
-                llogs->llog_ctxt[index] = NULL;
-                OBD_FREE(ctxt, sizeof(*ctxt));
-        }
-        RETURN(rc);
+                CERROR("llog_process() with cat_cancel_cb failed: %d\n", rc);
+        GOTO(out, rc);
+out:
+        llog_ctxt_put(ctxt);
+        return rc;
 }
 EXPORT_SYMBOL(llog_obd_origin_setup);
 
-int obd_llog_cat_initialize(struct obd_device *obd, struct obd_llogs *llogs, 
-                            int count, const char *name)
+int llog_obd_origin_cleanup(struct llog_ctxt *ctxt)
 {
-        struct llog_catid *idarray;
-        int size = sizeof(*idarray) * count;
-        int rc;
+        struct llog_handle *cathandle, *n, *loghandle;
+        struct llog_log_hdr *llh;
+        int rc, index;
         ENTRY;
 
-        OBD_ALLOC(idarray, size);
-        if (!idarray)
-                RETURN(-ENOMEM);
-
-        rc = llog_get_cat_list(&obd->obd_lvfs_ctxt, obd->obd_fsops,
-                               name, count, idarray);
-        if (rc) {
-                CERROR("rc: %d\n", rc);
-                GOTO(out, rc);
-        }
+        if (!ctxt)
+                RETURN(0);
 
-        rc = obd_llog_init(obd, llogs, obd, count, idarray);
-        if (rc) {
-                CERROR("rc: %d\n", rc);
-                GOTO(out, rc);
+        cathandle = ctxt->loc_handle;
+        if (cathandle) {
+                cfs_list_for_each_entry_safe(loghandle, n,
+                                             &cathandle->u.chd.chd_head,
+                                             u.phd.phd_entry) {
+                        llh = loghandle->lgh_hdr;
+                        if ((llh->llh_flags &
+                                LLOG_F_ZAP_WHEN_EMPTY) &&
+                            (llh->llh_count == 1)) {
+                                rc = llog_destroy(loghandle);
+                                if (rc)
+                                        CERROR("failure destroying log during "
+                                               "cleanup: %d\n", rc);
+
+                                index = loghandle->u.phd.phd_cookie.lgc_index;
+                                llog_free_handle(loghandle);
+
+                                LASSERT(index);
+                                llog_cat_set_first_idx(cathandle, index);
+                                rc = llog_cancel_rec(cathandle, index);
+                                if (rc == 0)
+                                        CDEBUG(D_RPCTRACE, "cancel plain log at"
+                                               "index %u of catalog "LPX64"\n",
+                                               index,cathandle->lgh_id.lgl_oid);
+                        }
+                }
+                llog_cat_put(ctxt->loc_handle);
         }
+        RETURN(0);
+}
+EXPORT_SYMBOL(llog_obd_origin_cleanup);
 
-        rc = llog_put_cat_list(&obd->obd_lvfs_ctxt, obd->obd_fsops,
-                               name, count, idarray);
-        if (rc) {
-                CERROR("rc: %d\n", rc);
-                GOTO(out, rc);
-        }
+/* add for obdfilter/sz and mds/unlink */
+int llog_obd_origin_add(struct llog_ctxt *ctxt,
+                        struct llog_rec_hdr *rec, struct lov_stripe_md *lsm,
+                        struct llog_cookie *logcookies, int numcookies)
+{
+        struct llog_handle *cathandle;
+        int rc;
+        ENTRY;
 
- out:
-        OBD_FREE(idarray, size);
+        cathandle = ctxt->loc_handle;
+        LASSERT(cathandle != NULL);
+        rc = llog_cat_add_rec(cathandle, rec, logcookies, NULL);
+        if (rc != 0 && rc != 1)
+                CERROR("write one catalog record failed: %d\n", rc);
         RETURN(rc);
 }
-EXPORT_SYMBOL(obd_llog_cat_initialize);
+EXPORT_SYMBOL(llog_obd_origin_add);
 
-int obd_llog_init(struct obd_device *obd, struct obd_llogs *llogs,
-                  struct obd_device *disk_obd, int count,
-                  struct llog_catid *logid)
+int obd_llog_init(struct obd_device *obd, struct obd_llog_group *olg,
+                  struct obd_device *disk_obd, int *index)
 {
         int rc;
         ENTRY;
-        OBD_CHECK_OP(obd, llog_init, 0);
+        OBD_CHECK_DT_OP(obd, llog_init, 0);
         OBD_COUNTER_INCREMENT(obd, llog_init);
 
-        rc = OBP(obd, llog_init)(obd, llogs, disk_obd, count, logid);
+        rc = OBP(obd, llog_init)(obd, olg, disk_obd, index);
         RETURN(rc);
 }
 EXPORT_SYMBOL(obd_llog_init);
 
-int obd_llog_finish(struct obd_device *obd, struct obd_llogs *llogs, int count)
+int obd_llog_finish(struct obd_device *obd, int count)
 {
         int rc;
         ENTRY;
-        OBD_CHECK_OP(obd, llog_finish, 0);
+        OBD_CHECK_DT_OP(obd, llog_finish, 0);
         OBD_COUNTER_INCREMENT(obd, llog_finish);
 
-        rc = OBP(obd, llog_finish)(obd, llogs, count);
+        rc = OBP(obd, llog_finish)(obd, count);
         RETURN(rc);
 }
 EXPORT_SYMBOL(obd_llog_finish);
+
+/* context key constructor/destructor: llog_key_init, llog_key_fini */
+LU_KEY_INIT_FINI(llog, struct llog_thread_info);
+/* context key: llog_thread_key */
+LU_CONTEXT_KEY_DEFINE(llog, LCT_MD_THREAD | LCT_MG_THREAD | LCT_LOCAL);
+LU_KEY_INIT_GENERIC(llog);
+EXPORT_SYMBOL(llog_thread_key);
+
+int llog_info_init(void)
+{
+       llog_key_init_generic(&llog_thread_key, NULL);
+       lu_context_key_register(&llog_thread_key);
+       return 0;
+}
+
+void llog_info_fini(void)
+{
+       lu_context_key_degister(&llog_thread_key);
+}