Whamcloud - gitweb
Land b_head_quota onto HEAD (20081116_0105)
[fs/lustre-release.git] / lustre / obdclass / llog_lvfs.c
index 18a1afe..60ee61c 100644 (file)
@@ -1,29 +1,46 @@
 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
  * vim:expandtab:shiftwidth=8:tabstop=8:
  *
- *  Copyright (C) 2001-2003 Cluster File Systems, Inc.
- *   Author: Andreas Dilger <adilger@clusterfs.com>
+ * GPL HEADER START
  *
- *   This file is part of Lustre, http://www.lustre.org.
+ * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
  *
- *   Lustre is free 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.
+ * 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.
  *
- *   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
- *   GNU General Public License for more details.
+ * 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
- *   along with Lustre; if not, write to the Free Software
- *   Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
+ * 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
  *
- * OST<->MDS recovery logging infrastructure.
+ * 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  2008 Sun Microsystems, Inc. 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_lvfs.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>
  */
 
 #define DEBUG_SUBSYSTEM S_LOG
 #define EXPORT_SYMTAB
 #endif
 
-#ifdef __KERNEL__
-#include <linux/fs.h>
-#else
+#ifndef __KERNEL__
 #include <liblustre.h>
 #endif
 
-#include <linux/obd.h>
-#include <linux/obd_class.h>
-#include <linux/lustre_log.h>
-#include <linux/obd_ost.h>
-#include <portals/list.h>
-#include <linux/lvfs.h>
-#include <linux/lustre_fsfilt.h>
+#include <obd.h>
+#include <obd_class.h>
+#include <lustre_log.h>
+#include <obd_ost.h>
+#include <libcfs/list.h>
+#include <lvfs.h>
+#include <lustre_fsfilt.h>
+#include <lustre_disk.h>
 #include "llog_internal.h"
 
-#ifdef __KERNEL__
+#if defined(__KERNEL__) && defined(LLOG_LVFS)
 
 static int llog_lvfs_pad(struct obd_device *obd, struct l_file *file,
                                 int len, int index)
 {
-        struct llog_rec_hdr rec;
+        struct llog_rec_hdr rec = { 0 };
         struct llog_rec_tail tail;
         int rc;
         ENTRY;
 
         LASSERT(len >= LLOG_MIN_REC_SIZE && (len & 0x7) == 0);
 
-        tail.lrt_len = rec.lrh_len = cpu_to_le32(len);
-        tail.lrt_index = rec.lrh_index = cpu_to_le32(index);
-        rec.lrh_type = 0;
+        tail.lrt_len = rec.lrh_len = len;
+        tail.lrt_index = rec.lrh_index = index;
+        rec.lrh_type = LLOG_PAD_MAGIC;
 
         rc = fsfilt_write_record(obd, file, &rec, sizeof(rec), &file->f_pos, 0);
         if (rc) {
@@ -86,11 +102,15 @@ static int llog_lvfs_write_blob(struct obd_device *obd, struct l_file *file,
         int rc;
         struct llog_rec_tail end;
         loff_t saved_off = file->f_pos;
-        int buflen = le32_to_cpu(rec->lrh_len);
+        int buflen = rec->lrh_len;
 
         ENTRY;
+
         file->f_pos = off;
 
+        if (buflen == 0)
+                CWARN("0-length record\n");
+
         if (!buf) {
                 rc = fsfilt_write_record(obd, file, rec, buflen,&file->f_pos,0);
                 if (rc) {
@@ -101,7 +121,7 @@ static int llog_lvfs_write_blob(struct obd_device *obd, struct l_file *file,
         }
 
         /* the buf case */
-        rec->lrh_len = cpu_to_le32(sizeof(*rec) + buflen + sizeof(end));
+        rec->lrh_len = sizeof(*rec) + buflen + sizeof(end);
         rc = fsfilt_write_record(obd, file, rec, sizeof(*rec), &file->f_pos, 0);
         if (rc) {
                 CERROR("error writing log hdr: rc %d\n", rc);
@@ -155,18 +175,42 @@ static int llog_lvfs_read_header(struct llog_handle *handle)
 
         obd = handle->lgh_ctxt->loc_exp->exp_obd;
 
-        if (handle->lgh_file->f_dentry->d_inode->i_size == 0) {
+        if (i_size_read(handle->lgh_file->f_dentry->d_inode) == 0) {
                 CDEBUG(D_HA, "not reading header from 0-byte log\n");
                 RETURN(LLOG_EEMPTY);
         }
 
         rc = llog_lvfs_read_blob(obd, handle->lgh_file, handle->lgh_hdr,
                                  LLOG_CHUNK_SIZE, 0);
-        if (rc)
-                CERROR("error reading log header\n");
+        if (rc) {
+                CERROR("error reading log header from %.*s\n",
+                       handle->lgh_file->f_dentry->d_name.len,
+                       handle->lgh_file->f_dentry->d_name.name);
+        } else {
+                struct llog_rec_hdr *llh_hdr = &handle->lgh_hdr->llh_hdr;
+
+                if (LLOG_REC_HDR_NEEDS_SWABBING(llh_hdr))
+                        lustre_swab_llog_hdr(handle->lgh_hdr);
+
+                if (llh_hdr->lrh_type != LLOG_HDR_MAGIC) {
+                        CERROR("bad log %.*s header magic: %#x (expected %#x)\n",
+                               handle->lgh_file->f_dentry->d_name.len,
+                               handle->lgh_file->f_dentry->d_name.name,
+                               llh_hdr->lrh_type, LLOG_HDR_MAGIC);
+                        rc = -EIO;
+                } else if (llh_hdr->lrh_len != LLOG_CHUNK_SIZE) {
+                        CERROR("incorrectly sized log %.*s header: %#x "
+                               "(expected %#x)\n",
+                               handle->lgh_file->f_dentry->d_name.len,
+                               handle->lgh_file->f_dentry->d_name.name,
+                               llh_hdr->lrh_len, LLOG_CHUNK_SIZE);
+                        CERROR("you may need to re-run lconf --write_conf.\n");
+                        rc = -EIO;
+                }
+        }
 
-        handle->lgh_last_idx = le32_to_cpu(handle->lgh_hdr->llh_tail.lrt_index);
-        handle->lgh_file->f_pos = handle->lgh_file->f_dentry->d_inode->i_size;
+        handle->lgh_last_idx = handle->lgh_hdr->llh_tail.lrt_index;
+        handle->lgh_file->f_pos = i_size_read(handle->lgh_file->f_dentry->d_inode);
 
         RETURN(rc);
 }
@@ -179,11 +223,10 @@ static int llog_lvfs_write_rec(struct llog_handle *loghandle,
                                void *buf, int idx)
 {
         struct llog_log_hdr *llh;
-        int reclen = le32_to_cpu(rec->lrh_len), index, rc;
+        int reclen = rec->lrh_len, index, rc;
         struct llog_rec_tail *lrt;
         struct obd_device *obd;
         struct file *file;
-        loff_t offset;
         size_t left;
         ENTRY;
 
@@ -193,31 +236,73 @@ static int llog_lvfs_write_rec(struct llog_handle *loghandle,
 
         /* record length should not bigger than LLOG_CHUNK_SIZE */
         if (buf)
-                rc = (reclen > LLOG_CHUNK_SIZE - sizeof(struct llog_rec_hdr)
-                      sizeof(struct llog_rec_tail)) ? -E2BIG : 0;
+                rc = (reclen > LLOG_CHUNK_SIZE - sizeof(struct llog_rec_hdr) -
+                      sizeof(struct llog_rec_tail)) ? -E2BIG : 0;
         else
                 rc = (reclen > LLOG_CHUNK_SIZE) ? -E2BIG : 0;
         if (rc)
                 RETURN(rc);
 
+        if (buf)
+                /* write_blob adds header and tail to lrh_len. */
+                reclen = sizeof(*rec) + rec->lrh_len +
+                         sizeof(struct llog_rec_tail);
+
         if (idx != -1) {
                 loff_t saved_offset;
 
                 /* no header: only allowed to insert record 1 */
-                if (idx != 1 && !file->f_dentry->d_inode->i_size) {
+                if (idx != 1 && !i_size_read(file->f_dentry->d_inode)) {
                         CERROR("idx != -1 in empty log\n");
                         LBUG();
                 }
 
-                if (idx && llh->llh_size && llh->llh_size != reclen)
+                if (idx && llh->llh_size && llh->llh_size != rec->lrh_len)
                         RETURN(-EINVAL);
 
+                if (!ext2_test_bit(idx, llh->llh_bitmap))
+                        CERROR("Modify unset record %u\n", idx);
+                if (idx != rec->lrh_index)
+                        CERROR("Index mismatch %d %u\n", idx, rec->lrh_index);
+
                 rc = llog_lvfs_write_blob(obd, file, &llh->llh_hdr, NULL, 0);
                 /* we are done if we only write the header or on error */
                 if (rc || idx == 0)
                         RETURN(rc);
 
-                saved_offset = sizeof(*llh) + (idx-1)*le32_to_cpu(rec->lrh_len);
+                /* Assumes constant lrh_len */
+                saved_offset = sizeof(*llh) + (idx - 1) * reclen;
+
+                if (buf) {
+                        struct llog_rec_hdr check;
+
+                        /* We assume that caller has set lgh_cur_* */
+                        saved_offset = loghandle->lgh_cur_offset;
+                        CDEBUG(D_OTHER,
+                               "modify record "LPX64": idx:%d/%u/%d, len:%u "
+                               "offset %llu\n",
+                               loghandle->lgh_id.lgl_oid, idx, rec->lrh_index,
+                               loghandle->lgh_cur_idx, rec->lrh_len,
+                               (long long)(saved_offset - sizeof(*llh)));
+                        if (rec->lrh_index != loghandle->lgh_cur_idx) {
+                                CERROR("modify idx mismatch %u/%d\n",
+                                       idx, loghandle->lgh_cur_idx);
+                                RETURN(-EFAULT);
+                        }
+#if 1  /* FIXME remove this safety check at some point */
+                        /* Verify that the record we're modifying is the
+                           right one. */
+                        rc = llog_lvfs_read_blob(obd, file, &check,
+                                                 sizeof(check), saved_offset);
+                        if (check.lrh_index != idx || check.lrh_len != reclen) {
+                                CERROR("Bad modify idx %u/%u size %u/%u (%d)\n",
+                                       idx, check.lrh_index, reclen,
+                                       check.lrh_len, rc);
+                                RETURN(-EFAULT);
+                        }
+#endif
+                }
+
                 rc = llog_lvfs_write_blob(obd, file, rec, buf, saved_offset);
                 if (rc == 0 && reccookie) {
                         reccookie->lgc_lgl = loghandle->lgh_id;
@@ -235,35 +320,39 @@ static int llog_lvfs_write_rec(struct llog_handle *loghandle,
          * big enough to hold reclen, so all we care about is padding here.
          */
         left = LLOG_CHUNK_SIZE - (file->f_pos & (LLOG_CHUNK_SIZE - 1));
-        if (buf)
-                reclen = sizeof(*rec) + le32_to_cpu(rec->lrh_len) +
-                        sizeof(struct llog_rec_tail);
 
         /* NOTE: padding is a record, but no bit is set */
         if (left != 0 && left != reclen &&
             left < (reclen + LLOG_MIN_REC_SIZE)) {
-                loghandle->lgh_last_idx++;
-                rc = llog_lvfs_pad(obd, file, left, loghandle->lgh_last_idx);
-                if (rc)
-                        RETURN(rc);
-        }
-
+                 index = loghandle->lgh_last_idx + 1;
+                 rc = llog_lvfs_pad(obd, file, left, index);
+                 if (rc)
+                         RETURN(rc);
+                 loghandle->lgh_last_idx++; /*for pad rec*/
+         }
+         /* if it's the last idx in log file, then return -ENOSPC */
+         if (loghandle->lgh_last_idx >= LLOG_BITMAP_SIZE(llh) - 1)
+                 RETURN(-ENOSPC);
         loghandle->lgh_last_idx++;
         index = loghandle->lgh_last_idx;
-        rec->lrh_index = cpu_to_le32(index);
+        LASSERT(index < LLOG_BITMAP_SIZE(llh));
+        rec->lrh_index = index;
         if (buf == NULL) {
-                lrt = (void *)rec + le32_to_cpu(rec->lrh_len) - sizeof(*lrt);
+                lrt = (struct llog_rec_tail *)
+                        ((char *)rec + rec->lrh_len - sizeof(*lrt));
                 lrt->lrt_len = rec->lrh_len;
                 lrt->lrt_index = rec->lrh_index;
         }
+        /*The caller should make sure only 1 process access the lgh_last_idx,
+         *Otherwise it might hit the assert.*/
+        LASSERT(index < LLOG_BITMAP_SIZE(llh));
         if (ext2_set_bit(index, llh->llh_bitmap)) {
                 CERROR("argh, index %u already set in log bitmap?\n", index);
                 LBUG(); /* should never happen */
         }
-        llh->llh_count = cpu_to_le32(le32_to_cpu(llh->llh_count) + 1);
-        llh->llh_tail.lrt_index = cpu_to_le32(index);
+        llh->llh_count++;
+        llh->llh_tail.lrt_index = index;
 
-        offset = 0;
         rc = llog_lvfs_write_blob(obd, file, &llh->llh_hdr, NULL, 0);
         if (rc)
                 RETURN(rc);
@@ -272,22 +361,23 @@ static int llog_lvfs_write_rec(struct llog_handle *loghandle,
         if (rc)
                 RETURN(rc);
 
-        CDEBUG(D_HA, "added record "LPX64": idx: %u, %u bytes\n",
-               loghandle->lgh_id.lgl_oid, index, le32_to_cpu(rec->lrh_len));
+        CDEBUG(D_RPCTRACE, "added record "LPX64": idx: %u, %u \n",
+               loghandle->lgh_id.lgl_oid, index, rec->lrh_len);
         if (rc == 0 && reccookie) {
                 reccookie->lgc_lgl = loghandle->lgh_id;
                 reccookie->lgc_index = index;
-                if (le32_to_cpu(rec->lrh_type) == MDS_UNLINK_REC)
-                        reccookie->lgc_subsys = LLOG_UNLINK_ORIG_CTXT;
-                else if (le32_to_cpu(rec->lrh_type) == OST_SZ_REC)
+                if ((rec->lrh_type == MDS_UNLINK_REC) ||
+                                (rec->lrh_type == MDS_SETATTR_REC))
+                        reccookie->lgc_subsys = LLOG_MDS_OST_ORIG_CTXT;
+                else if (rec->lrh_type == OST_SZ_REC)
                         reccookie->lgc_subsys = LLOG_SIZE_ORIG_CTXT;
-                else if (le32_to_cpu(rec->lrh_type) == OST_RAID1_REC)
+                else if (rec->lrh_type == OST_RAID1_REC)
                         reccookie->lgc_subsys = LLOG_RD1_ORIG_CTXT;
                 else
                         reccookie->lgc_subsys = -1;
                 rc = 1;
         }
-        if (rc == 0 && le32_to_cpu(rec->lrh_type) == LLOG_GEN_REC)
+        if (rc == 0 && rec->lrh_type == LLOG_GEN_REC)
                 rc = 1;
 
         RETURN(rc);
@@ -326,7 +416,7 @@ static int llog_lvfs_next_block(struct llog_handle *loghandle, int *cur_idx,
         CDEBUG(D_OTHER, "looking for log index %u (cur idx %u off "LPU64")\n",
                next_idx, *cur_idx, *cur_offset);
 
-        while (*cur_offset < loghandle->lgh_file->f_dentry->d_inode->i_size) {
+        while (*cur_offset < i_size_read(loghandle->lgh_file->f_dentry->d_inode)) {
                 struct llog_rec_hdr *rec;
                 struct llog_rec_tail *tail;
                 loff_t ppos;
@@ -336,19 +426,24 @@ static int llog_lvfs_next_block(struct llog_handle *loghandle, int *cur_idx,
                 ppos = *cur_offset;
                 rc = fsfilt_read_record(loghandle->lgh_ctxt->loc_exp->exp_obd,
                                         loghandle->lgh_file, buf, len,
-                                        cur_offset);
-
+                                        &ppos);
                 if (rc) {
                         CERROR("Cant read llog block at log id "LPU64
                                "/%u offset "LPU64"\n",
                                loghandle->lgh_id.lgl_oid,
                                loghandle->lgh_id.lgl_ogen,
                                *cur_offset);
-                         RETURN(rc);
+                        RETURN(rc);
                 }
 
-                /* put number of readed bytes in rc to make code simpler */
-                rc = *cur_offset - ppos;
+                /* put number of bytes read into rc to make code simpler */
+                rc = ppos - *cur_offset;
+                *cur_offset = ppos;
+
+                if (rc < len) {
+                        /* signal the end of the valid buffer to llog_process */
+                        memset(buf + rc, 0, len - rc);
+                }
 
                 if (rc == 0) /* end of file, nothing to do */
                         RETURN(0);
@@ -357,11 +452,18 @@ static int llog_lvfs_next_block(struct llog_handle *loghandle, int *cur_idx,
                         CERROR("Invalid llog block at log id "LPU64"/%u offset "
                                LPU64"\n", loghandle->lgh_id.lgl_oid,
                                loghandle->lgh_id.lgl_ogen, *cur_offset);
-                         RETURN(-EINVAL);
+                        RETURN(-EINVAL);
                 }
 
-                tail = buf + rc - sizeof(struct llog_rec_tail);
-                *cur_idx = le32_to_cpu(tail->lrt_index);
+                rec = buf;
+                tail = (struct llog_rec_tail *)((char *)buf + rc -
+                                                sizeof(struct llog_rec_tail));
+
+                if (LLOG_REC_HDR_NEEDS_SWABBING(rec)) {
+                        lustre_swab_llog_rec(rec, tail);
+                }
+
+                *cur_idx = tail->lrt_index;
 
                 /* this shouldn't happen */
                 if (tail->lrt_index == 0) {
@@ -370,15 +472,87 @@ static int llog_lvfs_next_block(struct llog_handle *loghandle, int *cur_idx,
                                loghandle->lgh_id.lgl_ogen, *cur_offset);
                         RETURN(-EINVAL);
                 }
-                if (le32_to_cpu(tail->lrt_index) < next_idx)
+                if (tail->lrt_index < next_idx)
+                        continue;
+
+                /* sanity check that the start of the new buffer is no farther
+                 * than the record that we wanted.  This shouldn't happen. */
+                if (rec->lrh_index > next_idx) {
+                        CERROR("missed desired record? %u > %u\n",
+                               rec->lrh_index, next_idx);
+                        RETURN(-ENOENT);
+                }
+                RETURN(0);
+        }
+        RETURN(-EIO);
+}
+
+static int llog_lvfs_prev_block(struct llog_handle *loghandle,
+                                int prev_idx, void *buf, int len)
+{
+        __u64 cur_offset;
+        int rc;
+        ENTRY;
+
+        if (len == 0 || len & (LLOG_CHUNK_SIZE - 1))
+                RETURN(-EINVAL);
+
+        CDEBUG(D_OTHER, "looking for log index %u\n", prev_idx);
+
+        cur_offset = LLOG_CHUNK_SIZE;
+        llog_skip_over(&cur_offset, 0, prev_idx);
+
+        while (cur_offset < i_size_read(loghandle->lgh_file->f_dentry->d_inode)) {
+                struct llog_rec_hdr *rec;
+                struct llog_rec_tail *tail;
+                loff_t ppos;
+
+                ppos = cur_offset;
+
+                rc = fsfilt_read_record(loghandle->lgh_ctxt->loc_exp->exp_obd,
+                                        loghandle->lgh_file, buf, len,
+                                        &ppos);
+                if (rc) {
+                        CERROR("Cant read llog block at log id "LPU64
+                               "/%u offset "LPU64"\n",
+                               loghandle->lgh_id.lgl_oid,
+                               loghandle->lgh_id.lgl_ogen,
+                               cur_offset);
+                        RETURN(rc);
+                }
+
+                /* put number of bytes read into rc to make code simpler */
+                rc = ppos - cur_offset;
+                cur_offset = ppos;
+
+                if (rc == 0) /* end of file, nothing to do */
+                        RETURN(0);
+
+                if (rc < sizeof(*tail)) {
+                        CERROR("Invalid llog block at log id "LPU64"/%u offset "
+                               LPU64"\n", loghandle->lgh_id.lgl_oid,
+                               loghandle->lgh_id.lgl_ogen, cur_offset);
+                        RETURN(-EINVAL);
+                }
+
+                tail = buf + rc - sizeof(struct llog_rec_tail);
+
+                /* this shouldn't happen */
+                if (tail->lrt_index == 0) {
+                        CERROR("Invalid llog tail at log id "LPU64"/%u offset "
+                               LPU64"\n", loghandle->lgh_id.lgl_oid,
+                               loghandle->lgh_id.lgl_ogen, cur_offset);
+                        RETURN(-EINVAL);
+                }
+                if (le32_to_cpu(tail->lrt_index) < prev_idx)
                         continue;
 
                 /* sanity check that the start of the new buffer is no farther
                  * than the record that we wanted.  This shouldn't happen. */
                 rec = buf;
-                if (le32_to_cpu(rec->lrh_index) > next_idx) {
+                if (le32_to_cpu(rec->lrh_index) > prev_idx) {
                         CERROR("missed desired record? %u > %u\n",
-                               le32_to_cpu(rec->lrh_index), next_idx);
+                               le32_to_cpu(rec->lrh_index), prev_idx);
                         RETURN(-ENOENT);
                 }
                 RETURN(0);
@@ -386,7 +560,7 @@ static int llog_lvfs_next_block(struct llog_handle *loghandle, int *cur_idx,
         RETURN(-EIO);
 }
 
-static struct file *llog_filp_open(char *name, int flags, int mode)
+static struct file *llog_filp_open(char *dir, char *name, int flags, int mode)
 {
         char *logname;
         struct file *filp;
@@ -396,7 +570,7 @@ static struct file *llog_filp_open(char *name, int flags, int mode)
         if (logname == NULL)
                 return ERR_PTR(-ENOMEM);
 
-        len = snprintf(logname, PATH_MAX, "LOGS/%s", name);
+        len = snprintf(logname, PATH_MAX, "%s/%s", dir, name);
         if (len >= PATH_MAX - 1) {
                 filp = ERR_PTR(-ENAMETOOLONG);
         } else {
@@ -405,7 +579,6 @@ static struct file *llog_filp_open(char *name, int flags, int mode)
                         CERROR("logfile creation %s: %ld\n", logname,
                                PTR_ERR(filp));
         }
-
         OBD_FREE(logname, PATH_MAX);
         return filp;
 }
@@ -451,7 +624,7 @@ static int llog_lvfs_create(struct llog_ctxt *ctxt, struct llog_handle **res,
                         GOTO(cleanup, rc);
                 }
 
-                handle->lgh_file = l_dentry_open(&obd->obd_ctxt, dchild,
+                handle->lgh_file = l_dentry_open(&obd->obd_lvfs_ctxt, dchild,
                                                     O_RDWR | O_LARGEFILE);
                 if (IS_ERR(handle->lgh_file)) {
                         rc = PTR_ERR(handle->lgh_file);
@@ -464,7 +637,16 @@ static int llog_lvfs_create(struct llog_ctxt *ctxt, struct llog_handle **res,
                 handle->lgh_id = *logid;
 
         } else if (name) {
-                handle->lgh_file = llog_filp_open(name, open_flags, 0644);
+                /* COMPAT_146 */
+                if (strcmp(obd->obd_type->typ_name, LUSTRE_MDS_NAME) == 0) {
+                        handle->lgh_file = llog_filp_open(MDT_LOGS_DIR, name,
+                                                          open_flags, 0644);
+                } else {
+                        /* end COMPAT_146 */
+                        handle->lgh_file = llog_filp_open(MOUNT_CONFIGS_DIR,
+                                                          name, open_flags,
+                                                          0644);
+                }
                 if (IS_ERR(handle->lgh_file))
                         GOTO(cleanup, rc = PTR_ERR(handle->lgh_file));
 
@@ -474,12 +656,13 @@ static int llog_lvfs_create(struct llog_ctxt *ctxt, struct llog_handle **res,
                 handle->lgh_id.lgl_ogen =
                         handle->lgh_file->f_dentry->d_inode->i_generation;
         } else {
-                oa = obdo_alloc();
+                OBDO_ALLOC(oa);
                 if (oa == NULL)
                         GOTO(cleanup, rc = -ENOMEM);
-                /* XXX get some filter group constants */
-                oa->o_gr = 1;
+
+                oa->o_gr = FILTER_GROUP_LLOG;
                 oa->o_valid = OBD_MD_FLGENER | OBD_MD_FLGROUP;
+
                 rc = obd_create(ctxt->loc_exp, oa, NULL, NULL);
                 if (rc)
                         GOTO(cleanup, rc);
@@ -490,7 +673,7 @@ static int llog_lvfs_create(struct llog_ctxt *ctxt, struct llog_handle **res,
                 if (IS_ERR(dchild))
                         GOTO(cleanup, rc = PTR_ERR(dchild));
                 cleanup_phase = 2;
-                handle->lgh_file = l_dentry_open(&obd->obd_ctxt, dchild,
+                handle->lgh_file = l_dentry_open(&obd->obd_lvfs_ctxt, dchild,
                                                  open_flags);
                 if (IS_ERR(handle->lgh_file))
                         GOTO(cleanup, rc = PTR_ERR(handle->lgh_file));
@@ -503,7 +686,7 @@ static int llog_lvfs_create(struct llog_ctxt *ctxt, struct llog_handle **res,
         handle->lgh_ctxt = ctxt;
  finish:
         if (oa)
-                obdo_free(oa);
+                OBDO_FREE(oa);
         RETURN(rc);
 cleanup:
         switch (cleanup_phase) {
@@ -528,11 +711,43 @@ static int llog_lvfs_close(struct llog_handle *handle)
 
 static int llog_lvfs_destroy(struct llog_handle *handle)
 {
+        struct dentry *fdentry;
         struct obdo *oa;
+        struct obd_device *obd = handle->lgh_ctxt->loc_exp->exp_obd;
+        char *dir;
         int rc;
         ENTRY;
 
-        oa = obdo_alloc();
+        /* COMPAT_146 */
+        if (strcmp(obd->obd_type->typ_name, LUSTRE_MDS_NAME) == 0)
+                dir = MDT_LOGS_DIR;
+        else
+                /* end COMPAT_146 */
+                dir = MOUNT_CONFIGS_DIR;
+
+        fdentry = handle->lgh_file->f_dentry;
+        if (strcmp(fdentry->d_parent->d_name.name, dir) == 0) {
+                struct inode *inode = fdentry->d_parent->d_inode;
+                struct lvfs_run_ctxt saved;
+                struct vfsmount *mnt = mntget(handle->lgh_file->f_vfsmnt);
+
+                push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
+                dget(fdentry);
+                rc = llog_lvfs_close(handle);
+
+                if (rc == 0) {
+                        LOCK_INODE_MUTEX_PARENT(inode);
+                        rc = ll_vfs_unlink(inode, fdentry, mnt);
+                        UNLOCK_INODE_MUTEX(inode);
+                }
+                mntput(mnt);
+
+                dput(fdentry);
+                pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
+                RETURN(rc);
+        }
+
+        OBDO_ALLOC(oa);
         if (oa == NULL)
                 RETURN(-ENOMEM);
 
@@ -545,25 +760,27 @@ static int llog_lvfs_destroy(struct llog_handle *handle)
         if (rc)
                 GOTO(out, rc);
 
-        rc = obd_destroy(handle->lgh_ctxt->loc_exp, oa, NULL, NULL);
+        rc = obd_destroy(handle->lgh_ctxt->loc_exp, oa, NULL, NULL, NULL, NULL);
  out:
-        obdo_free(oa);
+        OBDO_FREE(oa);
         RETURN(rc);
 }
 
 /* reads the catalog list */
 int llog_get_cat_list(struct obd_device *obd, struct obd_device *disk_obd,
-                      char *name, int count, struct llog_logid *idarray)
+                      char *name, int idx, int count, struct llog_catid *idarray)
 {
-        struct obd_run_ctxt saved;
+        struct lvfs_run_ctxt saved;
         struct l_file *file;
-        int rc;
+        int rc, rc1 = 0;
         int size = sizeof(*idarray) * count;
-        loff_t off = 0;
+        loff_t off = idx *  sizeof(*idarray);
+        ENTRY;
 
-        LASSERT(count);
+        if (!count)
+                RETURN(0);
 
-        push_ctxt(&saved, &obd->obd_ctxt, NULL);
+        push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
         file = filp_open(name, O_RDWR | O_CREAT | O_LARGEFILE, 0700);
         if (!file || IS_ERR(file)) {
                 rc = PTR_ERR(file);
@@ -578,34 +795,45 @@ int llog_get_cat_list(struct obd_device *obd, struct obd_device *disk_obd,
                 GOTO(out, rc = -ENOENT);
         }
 
+        CDEBUG(D_CONFIG, "cat list: disk size=%d, read=%d\n",
+               (int)i_size_read(file->f_dentry->d_inode), size);
+
+        /* read for new ost index or for empty file */
+        memset(idarray, 0, size);
+        if (i_size_read(file->f_dentry->d_inode) < off)
+                GOTO(out, rc = 0);
+
         rc = fsfilt_read_record(disk_obd, file, idarray, size, &off);
         if (rc) {
-                CDEBUG(D_INODE,"OBD filter: error reading %s: rc %d\n",
-                       name, rc);
+                CERROR("OBD filter: error reading %s: rc %d\n", name, rc);
                 GOTO(out, rc);
         }
 
+        EXIT;
  out:
-        pop_ctxt(&saved, &obd->obd_ctxt, NULL);
+        pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
         if (file && !IS_ERR(file))
-                rc = filp_close(file, 0);
-        RETURN(rc);
+                rc1 = filp_close(file, 0);
+        if (rc == 0)
+                rc = rc1;
+        return rc;
 }
 EXPORT_SYMBOL(llog_get_cat_list);
 
 /* writes the cat list */
 int llog_put_cat_list(struct obd_device *obd, struct obd_device *disk_obd,
-                      char *name, int count, struct llog_logid *idarray)
+                      char *name, int idx, int count, struct llog_catid *idarray)
 {
-        struct obd_run_ctxt saved;
+        struct lvfs_run_ctxt saved;
         struct l_file *file;
-        int rc;
+        int rc, rc1 = 0;
         int size = sizeof(*idarray) * count;
-        loff_t off = 0;
+        loff_t off = idx * sizeof(*idarray);
 
-        LASSERT(count);
+        if (!count)
+                GOTO(out1, rc = 0);
 
-        push_ctxt(&saved, &obd->obd_ctxt, NULL);
+        push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
         file = filp_open(name, O_RDWR | O_CREAT | O_LARGEFILE, 0700);
         if (!file || IS_ERR(file)) {
                 rc = PTR_ERR(file);
@@ -622,21 +850,27 @@ int llog_put_cat_list(struct obd_device *obd, struct obd_device *disk_obd,
 
         rc = fsfilt_write_record(disk_obd, file, idarray, size, &off, 1);
         if (rc) {
-                CDEBUG(D_INODE,"OBD filter: error reading %s: rc %d\n",
+                CDEBUG(D_INODE,"OBD filter: error writeing %s: rc %d\n",
                        name, rc);
                 GOTO(out, rc);
         }
 
- out:
-        pop_ctxt(&saved, &obd->obd_ctxt, NULL);
+out:
+        pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
         if (file && !IS_ERR(file))
-                rc = filp_close(file, 0);
+                rc1 = filp_close(file, 0);
+
+        if (rc == 0)
+                rc = rc1;
+out1:
         RETURN(rc);
 }
+EXPORT_SYMBOL(llog_put_cat_list);
 
 struct llog_operations llog_lvfs_ops = {
         lop_write_rec:   llog_lvfs_write_rec,
         lop_next_block:  llog_lvfs_next_block,
+        lop_prev_block:  llog_lvfs_prev_block,
         lop_read_header: llog_lvfs_read_header,
         lop_create:      llog_lvfs_create,
         lop_destroy:     llog_lvfs_destroy,
@@ -671,7 +905,14 @@ static int llog_lvfs_next_block(struct llog_handle *loghandle, int *cur_idx,
         return 0;
 }
 
-static int llog_lvfs_create(struct llog_obd_ctxt *ctxt,struct llog_handle **res,
+static int llog_lvfs_prev_block(struct llog_handle *loghandle,
+                                int prev_idx, void *buf, int len)
+{
+        LBUG();
+        return 0;
+}
+
+static int llog_lvfs_create(struct llog_ctxt *ctxt, struct llog_handle **res,
                             struct llog_logid *logid, char *name)
 {
         LBUG();
@@ -691,14 +932,14 @@ static int llog_lvfs_destroy(struct llog_handle *handle)
 }
 
 int llog_get_cat_list(struct obd_device *obd, struct obd_device *disk_obd,
-                      char *name, int count, struct llog_logid *idarray)
+                      char *name, int idx, int count, struct llog_catid *idarray)
 {
         LBUG();
         return 0;
 }
 
 int llog_put_cat_list(struct obd_device *obd, struct obd_device *disk_obd,
-                      char *name, int count, struct llog_logid *idarray)
+                      char *name, int idx, int count, struct llog_catid *idarray)
 {
         LBUG();
         return 0;
@@ -707,6 +948,7 @@ int llog_put_cat_list(struct obd_device *obd, struct obd_device *disk_obd,
 struct llog_operations llog_lvfs_ops = {
         lop_write_rec:   llog_lvfs_write_rec,
         lop_next_block:  llog_lvfs_next_block,
+        lop_prev_block:  llog_lvfs_prev_block,
         lop_read_header: llog_lvfs_read_header,
         lop_create:      llog_lvfs_create,
         lop_destroy:     llog_lvfs_destroy,