Whamcloud - gitweb
LU-4528 llog: don't write llog in 3-steps
[fs/lustre-release.git] / lustre / obdclass / llog.c
index 0c643b2..7146001 100644 (file)
@@ -26,6 +26,8 @@
 /*
  * Copyright (c) 2003, 2010, Oracle and/or its affiliates. All rights reserved.
  * Use is subject to license terms.
+ *
+ * Copyright (c) 2012, 2013, Intel Corporation.
  */
 /*
  * This file is part of Lustre, http://www.lustre.org/
@@ -63,11 +65,12 @@ struct llog_handle *llog_alloc_handle(void)
 
        OBD_ALLOC_PTR(loghandle);
        if (loghandle == NULL)
-               return ERR_PTR(-ENOMEM);
+               return NULL;
 
-       cfs_init_rwsem(&loghandle->lgh_lock);
-       cfs_spin_lock_init(&loghandle->lgh_hdr_lock);
+       init_rwsem(&loghandle->lgh_lock);
+       spin_lock_init(&loghandle->lgh_hdr_lock);
        CFS_INIT_LIST_HEAD(&loghandle->u.phd.phd_entry);
+       atomic_set(&loghandle->lgh_refcount, 1);
 
        return loghandle;
 }
@@ -77,22 +80,34 @@ struct llog_handle *llog_alloc_handle(void)
  */
 void llog_free_handle(struct llog_handle *loghandle)
 {
-       if (!loghandle)
-               return;
+       LASSERT(loghandle != NULL);
 
+       /* failed llog_init_handle */
        if (!loghandle->lgh_hdr)
                goto out;
+
        if (loghandle->lgh_hdr->llh_flags & LLOG_F_IS_PLAIN)
-               cfs_list_del_init(&loghandle->u.phd.phd_entry);
-       if (loghandle->lgh_hdr->llh_flags & LLOG_F_IS_CAT)
+               LASSERT(cfs_list_empty(&loghandle->u.phd.phd_entry));
+       else if (loghandle->lgh_hdr->llh_flags & LLOG_F_IS_CAT)
                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_PTR(loghandle);
 }
 
+void llog_handle_get(struct llog_handle *loghandle)
+{
+       atomic_inc(&loghandle->lgh_refcount);
+}
+
+void llog_handle_put(struct llog_handle *loghandle)
+{
+       LASSERT(atomic_read(&loghandle->lgh_refcount) > 0);
+       if (atomic_dec_and_test(&loghandle->lgh_refcount))
+               llog_free_handle(loghandle);
+}
+
 /* returns negative on error; 0 if success; 1 if success & log destroyed */
 int llog_cancel_rec(const struct lu_env *env, struct llog_handle *loghandle,
                    int index)
@@ -101,57 +116,55 @@ int llog_cancel_rec(const struct lu_env *env, struct llog_handle *loghandle,
         int rc = 0;
         ENTRY;
 
-        CDEBUG(D_RPCTRACE, "Canceling %d in log "LPX64"\n",
-               index, loghandle->lgh_id.lgl_oid);
+        CDEBUG(D_RPCTRACE, "Canceling %d in log "DOSTID"\n",
+               index, POSTID(&loghandle->lgh_id.lgl_oi));
 
         if (index == 0) {
                 CERROR("Can't cancel index 0 which is header\n");
                 RETURN(-EINVAL);
         }
 
-       cfs_spin_lock(&loghandle->lgh_hdr_lock);
+       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);
-        }
+               spin_unlock(&loghandle->lgh_hdr_lock);
+               CDEBUG(D_RPCTRACE, "Catalog index %u already clear?\n", index);
+               RETURN(-ENOENT);
+       }
 
-        llh->llh_count--;
+       llh->llh_count--;
 
-        if ((llh->llh_flags & LLOG_F_ZAP_WHEN_EMPTY) &&
-            (llh->llh_count == 1) &&
-            (loghandle->lgh_last_idx == (LLOG_BITMAP_BYTES * 8) - 1)) {
-               cfs_spin_unlock(&loghandle->lgh_hdr_lock);
+       if ((llh->llh_flags & LLOG_F_ZAP_WHEN_EMPTY) &&
+           (llh->llh_count == 1) &&
+           (loghandle->lgh_last_idx == (LLOG_BITMAP_BYTES * 8) - 1)) {
+               spin_unlock(&loghandle->lgh_hdr_lock);
                rc = llog_destroy(env, loghandle);
                if (rc < 0) {
-                       CERROR("%s: can't destroy empty llog #"LPX64"#"LPX64
+                       CERROR("%s: can't destroy empty llog #"DOSTID
                               "#%08x: rc = %d\n",
                               loghandle->lgh_ctxt->loc_obd->obd_name,
-                              loghandle->lgh_id.lgl_oid,
-                              loghandle->lgh_id.lgl_oseq,
+                              POSTID(&loghandle->lgh_id.lgl_oi),
                               loghandle->lgh_id.lgl_ogen, rc);
                        GOTO(out_err, rc);
                }
-               RETURN(1);
+               RETURN(LLOG_DEL_PLAIN);
        }
-       cfs_spin_unlock(&loghandle->lgh_hdr_lock);
+       spin_unlock(&loghandle->lgh_hdr_lock);
 
-       rc = llog_write(env, loghandle, &llh->llh_hdr, NULL, 0, NULL, 0);
+       rc = llog_write(env, loghandle, &llh->llh_hdr, LLOG_HEADER_IDX);
        if (rc < 0) {
-               CERROR("%s: fail to write header for llog #"LPX64"#"LPX64
+               CERROR("%s: fail to write header for llog #"DOSTID
                       "#%08x: rc = %d\n",
                       loghandle->lgh_ctxt->loc_obd->obd_name,
-                      loghandle->lgh_id.lgl_oid,
-                      loghandle->lgh_id.lgl_oseq,
+                      POSTID(&loghandle->lgh_id.lgl_oi),
                       loghandle->lgh_id.lgl_ogen, rc);
                GOTO(out_err, rc);
        }
        RETURN(0);
 out_err:
-       cfs_spin_lock(&loghandle->lgh_hdr_lock);
+       spin_lock(&loghandle->lgh_hdr_lock);
        ext2_set_bit(index, llh->llh_bitmap);
        llh->llh_count++;
-       cfs_spin_unlock(&loghandle->lgh_hdr_lock);
+       spin_unlock(&loghandle->lgh_hdr_lock);
        return rc;
 }
 EXPORT_SYMBOL(llog_cancel_rec);
@@ -365,10 +378,9 @@ repeat:
                                if (rc == LLOG_PROC_BREAK) {
                                        GOTO(out, rc);
                                } else if (rc == LLOG_DEL_RECORD) {
-                                       llog_cancel_rec(lpi->lpi_env,
-                                                       loghandle,
-                                                       rec->lrh_index);
-                                        rc = 0;
+                                       rc = llog_cancel_rec(lpi->lpi_env,
+                                                            loghandle,
+                                                            rec->lrh_index);
                                 }
                                 if (rc)
                                         GOTO(out, rc);
@@ -387,6 +399,20 @@ out:
         if (cd != NULL)
                 cd->lpcd_last_idx = last_called_index;
 
+       if (unlikely(rc == -EIO)) {
+               /* something bad happened to the processing, probably I/O
+                * error or the log got corrupted..
+                * to be able to finally release the log we discard any
+                * remaining bits in the header */
+               CERROR("llog found corrupted\n");
+               while (index <= last_index) {
+                       if (ext2_test_bit(index, llh->llh_bitmap) != 0)
+                               llog_cancel_rec(lpi->lpi_env, loghandle, index);
+                       index++;
+               }
+               rc = 0;
+       }
+
        OBD_FREE(buf, LLOG_CHUNK_SIZE);
         lpi->lpi_rc = rc;
         return 0;
@@ -399,10 +425,10 @@ static int llog_process_thread_daemonize(void *arg)
        struct lu_env                    env;
        int                              rc;
 
-       cfs_daemonize_ctxt("llog_process_thread");
+       unshare_fs_struct();
 
        /* client env has no keys, tags is just 0 */
-       rc = lu_env_init(&env, LCT_LOCAL);
+       rc = lu_env_init(&env, LCT_LOCAL | LCT_MG_THREAD);
        if (rc)
                goto out;
        lpi->lpi_env = &env;
@@ -411,7 +437,7 @@ static int llog_process_thread_daemonize(void *arg)
 
        lu_env_fini(&env);
 out:
-       cfs_complete(&lpi->lpi_completion);
+       complete(&lpi->lpi_completion);
        return rc;
 }
 #endif
@@ -437,19 +463,21 @@ int llog_process_or_fork(const struct lu_env *env,
 
 #ifdef __KERNEL__
        if (fork) {
+               struct task_struct *task;
+
                /* The new thread can't use parent env,
                 * init the new one in llog_process_thread_daemonize. */
                lpi->lpi_env = NULL;
-               cfs_init_completion(&lpi->lpi_completion);
-               rc = cfs_create_thread(llog_process_thread_daemonize, lpi,
-                                      CFS_DAEMON_FLAGS);
-               if (rc < 0) {
+               init_completion(&lpi->lpi_completion);
+               task = kthread_run(llog_process_thread_daemonize, lpi,
+                                  "llog_process_thread");
+               if (IS_ERR(task)) {
+                       rc = PTR_ERR(task);
                        CERROR("%s: cannot start thread: rc = %d\n",
                               loghandle->lgh_ctxt->loc_obd->obd_name, rc);
-                       OBD_FREE_PTR(lpi);
-                       RETURN(rc);
+                       GOTO(out_lpi, rc);
                }
-               cfs_wait_for_completion(&lpi->lpi_completion);
+               wait_for_completion(&lpi->lpi_completion);
        } else {
                lpi->lpi_env = env;
                llog_process_thread(lpi);
@@ -458,27 +486,25 @@ int llog_process_or_fork(const struct lu_env *env,
        lpi->lpi_env = env;
        llog_process_thread(lpi);
 #endif
-        rc = lpi->lpi_rc;
-        OBD_FREE_PTR(lpi);
-        RETURN(rc);
+       rc = lpi->lpi_rc;
+
+#ifdef __KERNEL__
+out_lpi:
+#endif
+       OBD_FREE_PTR(lpi);
+       RETURN(rc);
 }
 EXPORT_SYMBOL(llog_process_or_fork);
 
 int llog_process(const struct lu_env *env, struct llog_handle *loghandle,
                 llog_cb_t cb, void *data, void *catdata)
 {
-       return llog_process_or_fork(env, loghandle, cb, data, catdata, true);
+       int rc;
+       rc = llog_process_or_fork(env, loghandle, cb, data, catdata, true);
+       return rc == LLOG_DEL_PLAIN ? 0 : rc;
 }
 EXPORT_SYMBOL(llog_process);
 
-inline int llog_get_size(struct llog_handle *loghandle)
-{
-        if (loghandle && loghandle->lgh_hdr)
-                return loghandle->lgh_hdr->llh_count;
-        return 0;
-}
-EXPORT_SYMBOL(llog_get_size);
-
 int llog_reverse_process(const struct lu_env *env,
                         struct llog_handle *loghandle, llog_cb_t cb,
                         void *data, void *catdata)
@@ -546,9 +572,8 @@ int llog_reverse_process(const struct lu_env *env,
                                if (rc == LLOG_PROC_BREAK) {
                                        GOTO(out, rc);
                                } else if (rc == LLOG_DEL_RECORD) {
-                                       llog_cancel_rec(env, loghandle,
-                                                       tail->lrt_index);
-                                       rc = 0;
+                                       rc = llog_cancel_rec(env, loghandle,
+                                                            tail->lrt_index);
                                }
                                 if (rc)
                                         GOTO(out, rc);
@@ -678,7 +703,7 @@ EXPORT_SYMBOL(llog_declare_write_rec);
 
 int llog_write_rec(const struct lu_env *env, struct llog_handle *handle,
                   struct llog_rec_hdr *rec, struct llog_cookie *logcookies,
-                  int numcookies, void *buf, int idx, struct thandle *th)
+                  int idx, struct thandle *th)
 {
        struct llog_operations  *lop;
        int                      raised, rc, buflen;
@@ -693,18 +718,13 @@ int llog_write_rec(const struct lu_env *env, struct llog_handle *handle,
        if (lop->lop_write_rec == NULL)
                RETURN(-EOPNOTSUPP);
 
-       if (buf)
-               buflen = rec->lrh_len + sizeof(struct llog_rec_hdr) +
-                        sizeof(struct llog_rec_tail);
-       else
-               buflen = rec->lrh_len;
+       buflen = rec->lrh_len;
        LASSERT(cfs_size_round(buflen) == buflen);
 
        raised = cfs_cap_raised(CFS_CAP_SYS_RESOURCE);
        if (!raised)
                cfs_cap_raise(CFS_CAP_SYS_RESOURCE);
-       rc = lop->lop_write_rec(env, handle, rec, logcookies, numcookies,
-                               buf, idx, th);
+       rc = lop->lop_write_rec(env, handle, rec, logcookies, idx, th);
        if (!raised)
                cfs_cap_lower(CFS_CAP_SYS_RESOURCE);
        RETURN(rc);
@@ -713,7 +733,7 @@ EXPORT_SYMBOL(llog_write_rec);
 
 int llog_add(const struct lu_env *env, struct llog_handle *lgh,
             struct llog_rec_hdr *rec, struct llog_cookie *logcookies,
-            void *buf, struct thandle *th)
+            struct thandle *th)
 {
        int raised, rc;
 
@@ -725,7 +745,7 @@ int llog_add(const struct lu_env *env, struct llog_handle *lgh,
        raised = cfs_cap_raised(CFS_CAP_SYS_RESOURCE);
        if (!raised)
                cfs_cap_raise(CFS_CAP_SYS_RESOURCE);
-       rc = lgh->lgh_logops->lop_add(env, lgh, rec, logcookies, buf, th);
+       rc = lgh->lgh_logops->lop_add(env, lgh, rec, logcookies, th);
        if (!raised)
                cfs_cap_lower(CFS_CAP_SYS_RESOURCE);
        RETURN(rc);
@@ -760,8 +780,9 @@ int llog_open_create(const struct lu_env *env, struct llog_ctxt *ctxt,
                     struct llog_handle **res, struct llog_logid *logid,
                     char *name)
 {
-       struct thandle  *th;
-       int              rc;
+       struct dt_device        *d;
+       struct thandle          *th;
+       int                      rc;
 
        ENTRY;
 
@@ -772,27 +793,21 @@ int llog_open_create(const struct lu_env *env, struct llog_ctxt *ctxt,
        if (llog_exist(*res))
                RETURN(0);
 
-       if ((*res)->lgh_obj != NULL) {
-               struct dt_device *d;
+       LASSERT((*res)->lgh_obj != NULL);
 
-               d = lu2dt_dev((*res)->lgh_obj->do_lu.lo_dev);
+       d = lu2dt_dev((*res)->lgh_obj->do_lu.lo_dev);
 
-               th = dt_trans_create(env, d);
-               if (IS_ERR(th))
-                       GOTO(out, rc = PTR_ERR(th));
+       th = dt_trans_create(env, d);
+       if (IS_ERR(th))
+               GOTO(out, rc = PTR_ERR(th));
 
-               rc = llog_declare_create(env, *res, th);
-               if (rc == 0) {
-                       rc = dt_trans_start_local(env, d, th);
-                       if (rc == 0)
-                               rc = llog_create(env, *res, th);
-               }
-               dt_trans_stop(env, d, th);
-       } else {
-               /* lvfs compat code */
-               LASSERT((*res)->lgh_file == NULL);
-               rc = llog_create(env, *res, NULL);
+       rc = llog_declare_create(env, *res, th);
+       if (rc == 0) {
+               rc = dt_trans_start_local(env, d, th);
+               if (rc == 0)
+                       rc = llog_create(env, *res, th);
        }
+       dt_trans_stop(env, d, th);
 out:
        if (rc)
                llog_close(env, *res);
@@ -836,46 +851,37 @@ EXPORT_SYMBOL(llog_erase);
  * Valid only with local llog.
  */
 int llog_write(const struct lu_env *env, struct llog_handle *loghandle,
-              struct llog_rec_hdr *rec, struct llog_cookie *reccookie,
-              int cookiecount, void *buf, int idx)
+              struct llog_rec_hdr *rec, int idx)
 {
-       int rc;
+       struct dt_device        *dt;
+       struct thandle          *th;
+       int                      rc;
 
        ENTRY;
 
        LASSERT(loghandle);
        LASSERT(loghandle->lgh_ctxt);
+       LASSERT(loghandle->lgh_obj != NULL);
 
-       if (loghandle->lgh_obj != NULL) {
-               struct dt_device        *dt;
-               struct thandle          *th;
+       dt = lu2dt_dev(loghandle->lgh_obj->do_lu.lo_dev);
 
-               dt = lu2dt_dev(loghandle->lgh_obj->do_lu.lo_dev);
+       th = dt_trans_create(env, dt);
+       if (IS_ERR(th))
+               RETURN(PTR_ERR(th));
 
-               th = dt_trans_create(env, dt);
-               if (IS_ERR(th))
-                       RETURN(PTR_ERR(th));
-
-               rc = llog_declare_write_rec(env, loghandle, rec, idx, th);
-               if (rc)
-                       GOTO(out_trans, rc);
+       rc = llog_declare_write_rec(env, loghandle, rec, idx, th);
+       if (rc)
+               GOTO(out_trans, rc);
 
-               rc = dt_trans_start_local(env, dt, th);
-               if (rc)
-                       GOTO(out_trans, rc);
+       rc = dt_trans_start_local(env, dt, th);
+       if (rc)
+               GOTO(out_trans, rc);
 
-               cfs_down_write(&loghandle->lgh_lock);
-               rc = llog_write_rec(env, loghandle, rec, reccookie,
-                                   cookiecount, buf, idx, th);
-               cfs_up_write(&loghandle->lgh_lock);
+       down_write(&loghandle->lgh_lock);
+       rc = llog_write_rec(env, loghandle, rec, NULL, idx, th);
+       up_write(&loghandle->lgh_lock);
 out_trans:
-               dt_trans_stop(env, dt, th);
-       } else { /* lvfs compatibility */
-               cfs_down_write(&loghandle->lgh_lock);
-               rc = llog_write_rec(env, loghandle, rec, reccookie,
-                                   cookiecount, buf, idx, NULL);
-               cfs_up_write(&loghandle->lgh_lock);
-       }
+       dt_trans_stop(env, dt, th);
        RETURN(rc);
 }
 EXPORT_SYMBOL(llog_write);
@@ -928,10 +934,111 @@ int llog_close(const struct lu_env *env, struct llog_handle *loghandle)
        if (rc)
                GOTO(out, rc);
        if (lop->lop_close == NULL)
-               GOTO(out, -EOPNOTSUPP);
+               GOTO(out, rc = -EOPNOTSUPP);
        rc = lop->lop_close(env, loghandle);
 out:
-       llog_free_handle(loghandle);
+       llog_handle_put(loghandle);
        RETURN(rc);
 }
 EXPORT_SYMBOL(llog_close);
+
+int llog_is_empty(const struct lu_env *env, struct llog_ctxt *ctxt,
+                 char *name)
+{
+       struct llog_handle      *llh;
+       int                      rc = 0;
+
+       rc = llog_open(env, ctxt, &llh, NULL, name, LLOG_OPEN_EXISTS);
+       if (rc < 0) {
+               if (likely(rc == -ENOENT))
+                       rc = 0;
+               GOTO(out, rc);
+       }
+
+       rc = llog_init_handle(env, llh, LLOG_F_IS_PLAIN, NULL);
+       if (rc)
+               GOTO(out_close, rc);
+       rc = llog_get_size(llh);
+
+out_close:
+       llog_close(env, llh);
+out:
+       /* header is record 1 */
+       return (rc <= 1);
+}
+EXPORT_SYMBOL(llog_is_empty);
+
+int llog_copy_handler(const struct lu_env *env, struct llog_handle *llh,
+                     struct llog_rec_hdr *rec, void *data)
+{
+       struct llog_handle      *copy_llh = data;
+
+       /* Append all records */
+       return llog_write(env, copy_llh, rec, LLOG_NEXT_IDX);
+}
+EXPORT_SYMBOL(llog_copy_handler);
+
+/* backup plain llog */
+int llog_backup(const struct lu_env *env, struct obd_device *obd,
+               struct llog_ctxt *ctxt, struct llog_ctxt *bctxt,
+               char *name, char *backup)
+{
+       struct llog_handle      *llh, *bllh;
+       int                      rc;
+
+       ENTRY;
+
+       /* open original log */
+       rc = llog_open(env, ctxt, &llh, NULL, name, LLOG_OPEN_EXISTS);
+       if (rc < 0) {
+               /* the -ENOENT case is also reported to the caller
+                * but silently so it should handle that if needed.
+                */
+               if (rc != -ENOENT)
+                       CERROR("%s: failed to open log %s: rc = %d\n",
+                              obd->obd_name, name, rc);
+               RETURN(rc);
+       }
+
+       rc = llog_init_handle(env, llh, LLOG_F_IS_PLAIN, NULL);
+       if (rc)
+               GOTO(out_close, rc);
+
+       /* Make sure there's no old backup log */
+       rc = llog_erase(env, bctxt, NULL, backup);
+       if (rc < 0 && rc != -ENOENT)
+               GOTO(out_close, rc);
+
+       /* open backup log */
+       rc = llog_open_create(env, bctxt, &bllh, NULL, backup);
+       if (rc) {
+               CERROR("%s: failed to open backup logfile %s: rc = %d\n",
+                      obd->obd_name, backup, rc);
+               GOTO(out_close, rc);
+       }
+
+       /* check that backup llog is not the same object as original one */
+       if (llh->lgh_obj == bllh->lgh_obj) {
+               CERROR("%s: backup llog %s to itself (%s), objects %p/%p\n",
+                      obd->obd_name, name, backup, llh->lgh_obj,
+                      bllh->lgh_obj);
+               GOTO(out_backup, rc = -EEXIST);
+       }
+
+       rc = llog_init_handle(env, bllh, LLOG_F_IS_PLAIN, NULL);
+       if (rc)
+               GOTO(out_backup, rc);
+
+       /* Copy log record by record */
+       rc = llog_process_or_fork(env, llh, llog_copy_handler, (void *)bllh,
+                                 NULL, false);
+       if (rc)
+               CERROR("%s: failed to backup log %s: rc = %d\n",
+                      obd->obd_name, name, rc);
+out_backup:
+       llog_close(env, bllh);
+out_close:
+       llog_close(env, llh);
+       RETURN(rc);
+}
+EXPORT_SYMBOL(llog_backup);