Whamcloud - gitweb
LU-1302 llog: structures changes, llog_thread_info
[fs/lustre-release.git] / lustre / obdclass / llog.c
index 217b3f8..658dbee 100644 (file)
@@ -1,47 +1,57 @@
-/* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
- * vim:expandtab:shiftwidth=8:tabstop=8:
+/*
+ * GPL HEADER START
  *
- *  Copyright (C) 2001-2003 Cluster File Systems, Inc.
- *   Author: Andreas Dilger <adilger@clusterfs.com>
+ * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
  *
- *   This file is part of the Lustre file system, http://www.lustre.org
- *   Lustre is a trademark of Cluster File Systems, Inc.
+ * 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.
  *
- *   You may have signed or agreed to another license before downloading
- *   this software.  If so, you are bound by the terms and conditions
- *   of that agreement, and the following does not apply to you.  See the
- *   LICENSE file included with this distribution for more information.
+ * 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).
  *
- *   If you did not agree to a different license, then this copy of Lustre
- *   is open source software; you can redistribute it and/or modify it
- *   under the terms of version 2 of the GNU General Public License as
- *   published by the Free Software Foundation.
+ * 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
  *
- *   In either case, Lustre 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
- *   license text for more details.
+ * 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.
  *
- * OST<->MDS recovery logging infrastructure.
+ * GPL HEADER END
+ */
+/*
+ * Copyright (c) 2003, 2010, Oracle and/or its affiliates. All rights reserved.
+ * Use is subject to license terms.
+ */
+/*
+ * This file is part of Lustre, http://www.lustre.org/
+ * Lustre is a trademark of Sun Microsystems, Inc.
+ *
+ * lustre/obdclass/llog.c
  *
+ * OST<->MDS recovery logging infrastructure.
  * Invariants in implementation:
  * - we do not share logs among different OST<->MDS connections, so that
  *   if an OST or MDS fails it need only look at log(s) relevant to itself
+ *
+ * Author: Andreas Dilger <adilger@clusterfs.com>
+ * Author: Alex Zhuravlev <bzzz@whamcloud.com>
+ * Author: Mikhail Pershin <tappro@whamcloud.com>
  */
 
 #define DEBUG_SUBSYSTEM S_LOG
 
-#ifndef EXPORT_SYMTAB
-#define EXPORT_SYMTAB
-#endif
-
 #ifndef __KERNEL__
 #include <liblustre.h>
 #endif
 
 #include <obd_class.h>
 #include <lustre_log.h>
-#include <libcfs/list.h>
+#include "llog_internal.h"
 
 /* Allocate a new log or catalog handle */
 struct llog_handle *llog_alloc_handle(void)
@@ -49,11 +59,13 @@ struct llog_handle *llog_alloc_handle(void)
         struct llog_handle *loghandle;
         ENTRY;
 
-        OBD_ALLOC(loghandle, sizeof(*loghandle));
+       OBD_ALLOC_PTR(loghandle);
         if (loghandle == NULL)
                 RETURN(ERR_PTR(-ENOMEM));
 
-        init_rwsem(&loghandle->lgh_lock);
+        cfs_init_rwsem(&loghandle->lgh_lock);
+       cfs_spin_lock_init(&loghandle->lgh_hdr_lock);
+       CFS_INIT_LIST_HEAD(&loghandle->u.phd.phd_entry);
 
         RETURN(loghandle);
 }
@@ -68,13 +80,14 @@ void llog_free_handle(struct llog_handle *loghandle)
         if (!loghandle->lgh_hdr)
                 goto out;
         if (loghandle->lgh_hdr->llh_flags & LLOG_F_IS_PLAIN)
-                list_del_init(&loghandle->u.phd.phd_entry);
+                cfs_list_del_init(&loghandle->u.phd.phd_entry);
         if (loghandle->lgh_hdr->llh_flags & LLOG_F_IS_CAT)
-                LASSERT(list_empty(&loghandle->u.chd.chd_head));
-        OBD_FREE(loghandle->lgh_hdr, LLOG_CHUNK_SIZE);
+                LASSERT(cfs_list_empty(&loghandle->u.chd.chd_head));
+       LASSERT(sizeof(*(loghandle->lgh_hdr)) == LLOG_CHUNK_SIZE);
+       OBD_FREE(loghandle->lgh_hdr, LLOG_CHUNK_SIZE);
 
- out:
-        OBD_FREE(loghandle, sizeof(*loghandle));
+out:
+       OBD_FREE_PTR(loghandle);
 }
 EXPORT_SYMBOL(llog_free_handle);
 
@@ -85,17 +98,19 @@ int llog_cancel_rec(struct llog_handle *loghandle, int index)
         int rc = 0;
         ENTRY;
 
-        CDEBUG(D_HA, "canceling %d in log "LPX64"\n",
+        CDEBUG(D_RPCTRACE, "Canceling %d in log "LPX64"\n",
                index, loghandle->lgh_id.lgl_oid);
 
         if (index == 0) {
-                CERROR("cannot cancel index 0 (which is header)\n");
+                CERROR("Can't cancel index 0 which is header\n");
                 RETURN(-EINVAL);
         }
 
-        if (!ext2_clear_bit(index, llh->llh_bitmap)) {
-                CDEBUG(D_HA, "catalog index %u already clear?\n", index);
-                RETURN(-EINVAL);
+       cfs_spin_lock(&loghandle->lgh_hdr_lock);
+       if (!ext2_clear_bit(index, llh->llh_bitmap)) {
+               cfs_spin_unlock(&loghandle->lgh_hdr_lock);
+                CDEBUG(D_RPCTRACE, "Catalog index %u already clear?\n", index);
+                RETURN(-ENOENT);
         }
 
         llh->llh_count--;
@@ -103,25 +118,38 @@ int llog_cancel_rec(struct llog_handle *loghandle, int index)
         if ((llh->llh_flags & LLOG_F_ZAP_WHEN_EMPTY) &&
             (llh->llh_count == 1) &&
             (loghandle->lgh_last_idx == (LLOG_BITMAP_BYTES * 8) - 1)) {
-                rc = llog_destroy(loghandle);
-                if (rc) {
-                        CERROR("failure destroying log after last cancel: %d\n",
-                               rc);
-                        ext2_set_bit(index, llh->llh_bitmap);
-                        llh->llh_count++;
-                } else {
-                        rc = 1;
-                }
-                RETURN(rc);
-        }
-
-        rc = llog_write_rec(loghandle, &llh->llh_hdr, NULL, 0, NULL, 0);
-        if (rc) {
-                CERROR("failure re-writing header %d\n", rc);
-                ext2_set_bit(index, llh->llh_bitmap);
-                llh->llh_count++;
-        }
-        RETURN(rc);
+               cfs_spin_unlock(&loghandle->lgh_hdr_lock);
+               rc = llog_destroy(loghandle);
+               if (rc < 0) {
+                       CERROR("%s: can't destroy empty llog #"LPX64"#"LPX64
+                              "#%08x: rc = %d\n",
+                              loghandle->lgh_ctxt->loc_obd->obd_name,
+                              loghandle->lgh_id.lgl_oid,
+                              loghandle->lgh_id.lgl_oseq,
+                              loghandle->lgh_id.lgl_ogen, rc);
+                       GOTO(out_err, rc);
+               }
+               RETURN(1);
+       }
+       cfs_spin_unlock(&loghandle->lgh_hdr_lock);
+
+       rc = llog_write_rec(loghandle, &llh->llh_hdr, NULL, 0, NULL, 0);
+       if (rc < 0) {
+               CERROR("%s: fail to write header for llog #"LPX64"#"LPX64
+                      "#%08x: rc = %d\n",
+                      loghandle->lgh_ctxt->loc_obd->obd_name,
+                      loghandle->lgh_id.lgl_oid,
+                      loghandle->lgh_id.lgl_oseq,
+                      loghandle->lgh_id.lgl_ogen, rc);
+               GOTO(out_err, rc);
+       }
+       RETURN(0);
+out_err:
+       cfs_spin_lock(&loghandle->lgh_hdr_lock);
+       ext2_set_bit(index, llh->llh_bitmap);
+       llh->llh_count++;
+       cfs_spin_unlock(&loghandle->lgh_hdr_lock);
+       return rc;
 }
 EXPORT_SYMBOL(llog_cancel_rec);
 
@@ -133,7 +161,7 @@ int llog_init_handle(struct llog_handle *handle, int flags,
         ENTRY;
         LASSERT(handle->lgh_hdr == NULL);
 
-        OBD_ALLOC(llh, sizeof(*llh));
+       OBD_ALLOC_PTR(llh);
         if (llh == NULL)
                 RETURN(-ENOMEM);
         handle->lgh_hdr = llh;
@@ -160,7 +188,7 @@ int llog_init_handle(struct llog_handle *handle, int flags,
         llh->llh_hdr.lrh_type = LLOG_HDR_MAGIC;
         llh->llh_hdr.lrh_len = llh->llh_tail.lrt_len = LLOG_CHUNK_SIZE;
         llh->llh_hdr.lrh_index = llh->llh_tail.lrt_index = 0;
-        llh->llh_timestamp = CURRENT_SECONDS;
+        llh->llh_timestamp = cfs_time_current_sec();
         if (uuid)
                 memcpy(&llh->llh_tgtuuid, uuid, sizeof(llh->llh_tgtuuid));
         llh->llh_bitmap_offset = offsetof(typeof(*llh),llh_bitmap);
@@ -168,21 +196,20 @@ int llog_init_handle(struct llog_handle *handle, int flags,
 
 out:
         if (flags & LLOG_F_IS_CAT) {
-                CFS_INIT_LIST_HEAD(&handle->u.chd.chd_head);
-                llh->llh_size = sizeof(struct llog_logid_rec);
-        } else if (flags & LLOG_F_IS_PLAIN) {
-                CFS_INIT_LIST_HEAD(&handle->u.phd.phd_entry);
-        } else {
-                CERROR("Unknown flags: %#x (Expected %#x or %#x\n",
-                       flags, LLOG_F_IS_CAT, LLOG_F_IS_PLAIN);
-                LBUG();
-        }
-
-        if (rc) {
-                OBD_FREE(llh, sizeof(*llh));
-                handle->lgh_hdr = NULL;
-        }
-        RETURN(rc);
+               LASSERT(cfs_list_empty(&handle->u.chd.chd_head));
+               CFS_INIT_LIST_HEAD(&handle->u.chd.chd_head);
+               llh->llh_size = sizeof(struct llog_logid_rec);
+       } else if (!(flags & LLOG_F_IS_PLAIN)) {
+               CERROR("Unknown flags: %#x (Expected %#x or %#x\n",
+                      flags, LLOG_F_IS_CAT, LLOG_F_IS_PLAIN);
+               rc = -EINVAL;
+       }
+
+       if (rc) {
+               OBD_FREE_PTR(llh);
+               handle->lgh_hdr = NULL;
+       }
+       RETURN(rc);
 }
 EXPORT_SYMBOL(llog_init_handle);
 
@@ -204,29 +231,38 @@ int llog_close(struct llog_handle *loghandle)
 }
 EXPORT_SYMBOL(llog_close);
 
-int llog_process(struct llog_handle *loghandle, llog_cb_t cb,
-                 void *data, void *catdata)
+static int llog_process_thread(void *arg)
 {
-        struct llog_log_hdr *llh = loghandle->lgh_hdr;
-        struct llog_process_cat_data *cd = catdata;
-        char *buf;
-        __u64 cur_offset = LLOG_CHUNK_SIZE, last_offset;
-        int rc = 0, index = 1, last_index;
-        int saved_index = 0, last_called_index = 0;
-        ENTRY;
+        struct llog_process_info     *lpi = (struct llog_process_info *)arg;
+        struct llog_handle           *loghandle = lpi->lpi_loghandle;
+        struct llog_log_hdr          *llh = loghandle->lgh_hdr;
+        struct llog_process_cat_data *cd  = lpi->lpi_catdata;
+        char                         *buf;
+        __u64                         cur_offset = LLOG_CHUNK_SIZE;
+        __u64                         last_offset;
+        int                           rc = 0, index = 1, last_index;
+        int                           saved_index = 0, last_called_index = 0;
 
         LASSERT(llh);
 
         OBD_ALLOC(buf, LLOG_CHUNK_SIZE);
-        if (!buf)
-                RETURN(-ENOMEM);
+        if (!buf) {
+                lpi->lpi_rc = -ENOMEM;
+#ifdef __KERNEL__
+                cfs_complete(&lpi->lpi_completion);
+#endif
+                return 0;
+        }
+
+        if (!(lpi->lpi_flags & LLOG_FLAG_NODEAMON))
+                cfs_daemonize_ctxt("llog_process_thread");
 
         if (cd != NULL) {
-                last_called_index = cd->first_idx;
-                index = cd->first_idx + 1;
+                last_called_index = cd->lpcd_first_idx;
+                index = cd->lpcd_first_idx + 1;
         }
-        if (cd != NULL && cd->last_idx)
-                last_index = cd->last_idx;
+        if (cd != NULL && cd->lpcd_last_idx)
+                last_index = cd->lpcd_last_idx;
         else
                 last_index = LLOG_BITMAP_BYTES * 8 - 1;
 
@@ -246,6 +282,7 @@ int llog_process(struct llog_handle *loghandle, llog_cb_t cb,
                        index, last_index);
 
                 /* get the buf with our target record; avoid old garbage */
+                memset(buf, 0, LLOG_CHUNK_SIZE);
                 last_offset = cur_offset;
                 rc = llog_next_block(loghandle, &saved_index, index,
                                      &cur_offset, buf, LLOG_CHUNK_SIZE);
@@ -267,7 +304,7 @@ int llog_process(struct llog_handle *loghandle, llog_cb_t cb,
 
                         CDEBUG(D_OTHER, "after swabbing, type=%#x idx=%d\n",
                                rec->lrh_type, rec->lrh_index);
-                        
+
                         if (rec->lrh_index == 0)
                                 GOTO(out, 0); /* no more records */
 
@@ -284,27 +321,25 @@ int llog_process(struct llog_handle *loghandle, llog_cb_t cb,
                                 continue;
                         }
 
-                        CDEBUG(D_OTHER, 
+                        CDEBUG(D_OTHER,
                                "lrh_index: %d lrh_len: %d (%d remains)\n",
                                rec->lrh_index, rec->lrh_len,
                                (int)(buf + LLOG_CHUNK_SIZE - (char *)rec));
 
                         loghandle->lgh_cur_idx = rec->lrh_index;
                         loghandle->lgh_cur_offset = (char *)rec - (char *)buf +
-                                last_offset;
+                                                    last_offset;
 
                         /* if set, process the callback on this record */
                         if (ext2_test_bit(index, llh->llh_bitmap)) {
-                                rc = cb(loghandle, rec, data);
+                                rc = lpi->lpi_cb(loghandle, rec,
+                                                 lpi->lpi_cbdata);
                                 last_called_index = index;
                                 if (rc == LLOG_PROC_BREAK) {
-                                        CDEBUG(D_HA, "recovery from log: "LPX64
-                                               ":%x stopped\n",
-                                               loghandle->lgh_id.lgl_oid,
-                                               loghandle->lgh_id.lgl_ogen);
                                         GOTO(out, rc);
                                 } else if (rc == LLOG_DEL_RECORD) {
-                                        llog_cancel_rec(loghandle, rec->lrh_index);
+                                        llog_cancel_rec(loghandle,
+                                                        rec->lrh_index);
                                         rc = 0;
                                 }
                                 if (rc)
@@ -322,11 +357,57 @@ int llog_process(struct llog_handle *loghandle, llog_cb_t cb,
 
  out:
         if (cd != NULL)
-                cd->last_idx = last_called_index;
+                cd->lpcd_last_idx = last_called_index;
         if (buf)
                 OBD_FREE(buf, LLOG_CHUNK_SIZE);
+        lpi->lpi_rc = rc;
+#ifdef __KERNEL__
+        cfs_complete(&lpi->lpi_completion);
+#endif
+        return 0;
+}
+
+int llog_process_flags(struct llog_handle *loghandle, llog_cb_t cb,
+                       void *data, void *catdata, int flags)
+{
+        struct llog_process_info *lpi;
+        int                      rc;
+        ENTRY;
+
+        OBD_ALLOC_PTR(lpi);
+        if (lpi == NULL) {
+                CERROR("cannot alloc pointer\n");
+                RETURN(-ENOMEM);
+        }
+        lpi->lpi_loghandle = loghandle;
+        lpi->lpi_cb        = cb;
+        lpi->lpi_cbdata    = data;
+        lpi->lpi_catdata   = catdata;
+        lpi->lpi_flags     = flags;
+
+#ifdef __KERNEL__
+        cfs_init_completion(&lpi->lpi_completion);
+        rc = cfs_create_thread(llog_process_thread, lpi, CFS_DAEMON_FLAGS);
+        if (rc < 0) {
+                CERROR("cannot start thread: %d\n", rc);
+                OBD_FREE_PTR(lpi);
+                RETURN(rc);
+        }
+        cfs_wait_for_completion(&lpi->lpi_completion);
+#else
+        llog_process_thread(lpi);
+#endif
+        rc = lpi->lpi_rc;
+        OBD_FREE_PTR(lpi);
         RETURN(rc);
 }
+EXPORT_SYMBOL(llog_process_flags);
+
+int llog_process(struct llog_handle *loghandle, llog_cb_t cb,
+                 void *data, void *catdata)
+{
+        return llog_process_flags(loghandle, cb, data, catdata, 0);
+}
 EXPORT_SYMBOL(llog_process);
 
 inline int llog_get_size(struct llog_handle *loghandle)
@@ -351,9 +432,9 @@ int llog_reverse_process(struct llog_handle *loghandle, llog_cb_t cb,
                 RETURN(-ENOMEM);
 
         if (cd != NULL)
-                first_index = cd->first_idx + 1;
-        if (cd != NULL && cd->last_idx)
-                index = cd->last_idx;
+                first_index = cd->lpcd_first_idx + 1;
+        if (cd != NULL && cd->lpcd_last_idx)
+                index = cd->lpcd_last_idx;
         else
                 index = LLOG_BITMAP_BYTES * 8 - 1;
 
@@ -379,7 +460,7 @@ int llog_reverse_process(struct llog_handle *loghandle, llog_cb_t cb,
                 rec = buf;
                 idx = le32_to_cpu(rec->lrh_index);
                 if (idx < index)
-                        CDEBUG(D_HA, "index %u : idx %u\n", index, idx);
+                        CDEBUG(D_RPCTRACE, "index %u : idx %u\n", index, idx);
                 while (idx < index) {
                         rec = ((void *)rec + le32_to_cpu(rec->lrh_len));
                         idx ++;
@@ -398,10 +479,6 @@ int llog_reverse_process(struct llog_handle *loghandle, llog_cb_t cb,
                         if (ext2_test_bit(index, llh->llh_bitmap)) {
                                 rc = cb(loghandle, rec, data);
                                 if (rc == LLOG_PROC_BREAK) {
-                                        CWARN("recovery from log: "LPX64":%x"
-                                              " stopped\n",
-                                              loghandle->lgh_id.lgl_oid,
-                                              loghandle->lgh_id.lgl_ogen);
                                         GOTO(out, rc);
                                 }
                                 if (rc)