X-Git-Url: https://git.whamcloud.com/?p=fs%2Flustre-release.git;a=blobdiff_plain;f=lustre%2Fmdd%2Fmdd_device.c;h=35f80f41a50acb69e092eb87a08a8f034ea50e4f;hp=95408d88845e361f129ebfb2cf4665d98e2d5216;hb=b7d25639966c7d8fbfa660e7b135fb6b359a9a81;hpb=8eaf21f9a9a6264598b9fed93b2c5d423b00bd9a;ds=sidebyside diff --git a/lustre/mdd/mdd_device.c b/lustre/mdd/mdd_device.c index 95408d8..35f80f4 100644 --- a/lustre/mdd/mdd_device.c +++ b/lustre/mdd/mdd_device.c @@ -1,6 +1,4 @@ -/* -*- 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. @@ -26,8 +24,10 @@ * GPL HEADER END */ /* - * Copyright 2008 Sun Microsystems, Inc. All rights reserved + * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved. * Use is subject to license terms. + * + * Copyright (c) 2011, 2012, Whamcloud, Inc. */ /* * This file is part of Lustre, http://www.lustre.org/ @@ -40,32 +40,12 @@ * Author: Wang Di */ -#ifndef EXPORT_SYMTAB -# define EXPORT_SYMTAB -#endif #define DEBUG_SUBSYSTEM S_MDS -#include -#ifdef HAVE_EXT4_LDISKFS -#include -#else -#include -#endif -#include #include -#include -#include #include - -#include #include -#ifdef HAVE_EXT4_LDISKFS -#include -#else -#include -#endif #include -#include #include /* for changelogs */ #include #include @@ -78,6 +58,20 @@ static struct lu_device_type mdd_device_type; static const char mdd_root_dir_name[] = "ROOT"; static const char mdd_obf_dir_name[] = "fid"; +/* Slab for MDD object allocation */ +cfs_mem_cache_t *mdd_object_kmem; + +static struct lu_kmem_descr mdd_caches[] = { + { + .ckd_cache = &mdd_object_kmem, + .ckd_name = "mdd_obj", + .ckd_size = sizeof(struct mdd_object) + }, + { + .ckd_cache = NULL + } +}; + static int mdd_device_init(const struct lu_env *env, struct lu_device *d, const char *name, struct lu_device *next) { @@ -88,9 +82,9 @@ static int mdd_device_init(const struct lu_env *env, struct lu_device *d, mdd->mdd_child = lu2dt_dev(next); /* Prepare transactions callbacks. */ - mdd->mdd_txn_cb.dtc_txn_start = mdd_txn_start_cb; + mdd->mdd_txn_cb.dtc_txn_start = NULL; mdd->mdd_txn_cb.dtc_txn_stop = mdd_txn_stop_cb; - mdd->mdd_txn_cb.dtc_txn_commit = mdd_txn_commit_cb; + mdd->mdd_txn_cb.dtc_txn_commit = NULL; mdd->mdd_txn_cb.dtc_cookie = mdd; mdd->mdd_txn_cb.dtc_tag = LCT_MD_THREAD; CFS_INIT_LIST_HEAD(&mdd->mdd_txn_cb.dtc_linkage); @@ -106,7 +100,6 @@ static struct lu_device *mdd_device_fini(const struct lu_env *env, struct lu_device *d) { struct mdd_device *mdd = lu2mdd_dev(d); - struct lu_device *next = &mdd->mdd_child->dd_lu_dev; int rc; rc = mdd_procfs_fini(mdd); @@ -114,7 +107,7 @@ static struct lu_device *mdd_device_fini(const struct lu_env *env, CERROR("proc fini error %d \n", rc); return ERR_PTR(rc); } - return next; + return NULL; } static void mdd_changelog_fini(const struct lu_env *env, @@ -124,6 +117,7 @@ static void mdd_device_shutdown(const struct lu_env *env, struct mdd_device *m, struct lustre_cfg *cfg) { ENTRY; + mdd_lfsck_cleanup(env, m); mdd_changelog_fini(env, m); dt_txn_callback_del(m->mdd_child, &m->mdd_txn_cb); if (m->mdd_dot_lustre_objs.mdd_obf) @@ -133,13 +127,17 @@ static void mdd_device_shutdown(const struct lu_env *env, if (m->mdd_obd_dev) mdd_fini_obd(env, m, cfg); orph_index_fini(env, m); + if (m->mdd_capa != NULL) { + lu_object_put(env, &m->mdd_capa->do_lu); + m->mdd_capa = NULL; + } /* remove upcall device*/ md_upcall_fini(&m->mdd_md_dev); EXIT; } -static int changelog_init_cb(struct llog_handle *llh, struct llog_rec_hdr *hdr, - void *data) +static int changelog_init_cb(const struct lu_env *env, struct llog_handle *llh, + struct llog_rec_hdr *hdr, void *data) { struct mdd_device *mdd = (struct mdd_device *)data; struct llog_changelog_rec *rec = (struct llog_changelog_rec *)hdr; @@ -158,8 +156,9 @@ static int changelog_init_cb(struct llog_handle *llh, struct llog_rec_hdr *hdr, RETURN(LLOG_PROC_BREAK); } -static int changelog_user_init_cb(struct llog_handle *llh, - struct llog_rec_hdr *hdr, void *data) +static int changelog_user_init_cb(const struct lu_env *env, + struct llog_handle *llh, + struct llog_rec_hdr *hdr, void *data) { struct mdd_device *mdd = (struct mdd_device *)data; struct llog_changelog_user_rec *rec = @@ -198,7 +197,8 @@ static int mdd_changelog_llog_init(struct mdd_device *mdd) return -EINVAL; } - rc = llog_cat_reverse_process(ctxt->loc_handle, changelog_init_cb, mdd); + rc = llog_cat_reverse_process(NULL, ctxt->loc_handle, + changelog_init_cb, mdd); llog_ctxt_put(ctxt); if (rc < 0) { @@ -219,8 +219,8 @@ static int mdd_changelog_llog_init(struct mdd_device *mdd) return -EINVAL; } - rc = llog_cat_reverse_process(ctxt->loc_handle, changelog_user_init_cb, - mdd); + rc = llog_cat_reverse_process(NULL, ctxt->loc_handle, + changelog_user_init_cb, mdd); llog_ctxt_put(ctxt); if (rc < 0) { @@ -241,7 +241,6 @@ static int mdd_changelog_init(const struct lu_env *env, struct mdd_device *mdd) mdd->mdd_cl.mc_index = 0; cfs_spin_lock_init(&mdd->mdd_cl.mc_lock); - cfs_waitq_init(&mdd->mdd_cl.mc_waitq); mdd->mdd_cl.mc_starttime = cfs_time_current_64(); mdd->mdd_cl.mc_flags = 0; /* off by default */ mdd->mdd_cl.mc_mask = CHANGELOG_DEFMASK; @@ -312,9 +311,6 @@ int mdd_changelog_llog_write(struct mdd_device *mdd, struct llog_ctxt *ctxt; int rc; - if ((mdd->mdd_cl.mc_mask & (1 << rec->cr.cr_type)) == 0) - return 0; - rec->cr_hdr.lrh_len = llog_data_len(sizeof(*rec) + rec->cr.cr_namelen); /* llog_lvfs_write_rec sets the llog tail len */ rec->cr_hdr.lrh_type = CHANGELOG_REC; @@ -330,21 +326,56 @@ int mdd_changelog_llog_write(struct mdd_device *mdd, return -ENXIO; /* nested journal transaction */ - rc = llog_add(ctxt, &rec->cr_hdr, NULL, NULL, 0); + rc = llog_add(NULL, ctxt, &rec->cr_hdr, NULL, NULL, 0); llog_ctxt_put(ctxt); - cfs_waitq_signal(&mdd->mdd_cl.mc_waitq); - return rc; } +/** Add a changelog_ext entry \a rec to the changelog llog + * \param mdd + * \param rec + * \param handle - currently ignored since llogs start their own transaction; + * this will hopefully be fixed in llog rewrite + * \retval 0 ok + */ +int mdd_changelog_ext_llog_write(struct mdd_device *mdd, + struct llog_changelog_ext_rec *rec, + struct thandle *handle) +{ + struct obd_device *obd = mdd2obd_dev(mdd); + struct llog_ctxt *ctxt; + int rc; + + rec->cr_hdr.lrh_len = llog_data_len(sizeof(*rec) + rec->cr.cr_namelen); + /* llog_lvfs_write_rec sets the llog tail len */ + rec->cr_hdr.lrh_type = CHANGELOG_REC; + rec->cr.cr_time = cl_time(); + cfs_spin_lock(&mdd->mdd_cl.mc_lock); + /* NB: I suppose it's possible llog_add adds out of order wrt cr_index, + * but as long as the MDD transactions are ordered correctly for e.g. + * rename conflicts, I don't think this should matter. */ + rec->cr.cr_index = ++mdd->mdd_cl.mc_index; + cfs_spin_unlock(&mdd->mdd_cl.mc_lock); + ctxt = llog_get_context(obd, LLOG_CHANGELOG_ORIG_CTXT); + if (ctxt == NULL) + return -ENXIO; + + /* nested journal transaction */ + rc = llog_add(NULL, ctxt, &rec->cr_hdr, NULL, NULL, 0); + llog_ctxt_put(ctxt); + + return rc; +} + /** Remove entries with indicies up to and including \a endrec from the * changelog * \param mdd * \param endrec * \retval 0 ok */ -int mdd_changelog_llog_cancel(struct mdd_device *mdd, long long endrec) +int mdd_changelog_llog_cancel(const struct lu_env *env, + struct mdd_device *mdd, long long endrec) { struct obd_device *obd = mdd2obd_dev(mdd); struct llog_ctxt *ctxt; @@ -370,6 +401,7 @@ int mdd_changelog_llog_cancel(struct mdd_device *mdd, long long endrec) time. In case of crash, we just restart with old log so we're allright. */ if (endrec == cur) { + /* XXX: transaction is started by llog itself */ rc = mdd_changelog_write_header(mdd, CLM_PURGE); if (rc) goto out; @@ -380,7 +412,8 @@ int mdd_changelog_llog_cancel(struct mdd_device *mdd, long long endrec) changed since the last purge) */ mdd->mdd_cl.mc_starttime = cfs_time_current_64(); - rc = llog_cancel(ctxt, NULL, 1, (struct llog_cookie *)&endrec, 0); + /* XXX: transaction is started by llog itself */ + rc = llog_cancel(env, ctxt, NULL, 1, (struct llog_cookie *)&endrec, 0); out: llog_ctxt_put(ctxt); return rc; @@ -412,10 +445,12 @@ int mdd_changelog_write_header(struct mdd_device *mdd, int markerflags) /* Status and action flags */ rec->cr.cr_markerflags = mdd->mdd_cl.mc_flags | markerflags; - rc = mdd_changelog_llog_write(mdd, rec, NULL); + /* XXX: transaction is started by llog itself */ + rc = (mdd->mdd_cl.mc_mask & (1 << CL_MARK)) ? + mdd_changelog_llog_write(mdd, rec, NULL) : 0; /* assume on or off event; reset repeat-access time */ - mdd->mdd_cl.mc_starttime = rec->cr.cr_time; + mdd->mdd_cl.mc_starttime = cfs_time_current_64(); OBD_FREE(rec, reclen); RETURN(rc); @@ -439,7 +474,7 @@ static int create_dot_lustre_dir(const struct lu_env *env, struct mdd_device *m) rc = PTR_ERR(mdo); CERROR("creating obj [%s] fid = "DFID" rc = %d\n", dot_lustre_name, PFID(fid), rc); - RETURN(rc); + return rc; } if (!IS_ERR(mdo)) @@ -460,21 +495,6 @@ static int dot_lustre_mdd_permission(const struct lu_env *env, return 0; } -static int dot_lustre_mdd_attr_get(const struct lu_env *env, - struct md_object *obj, struct md_attr *ma) -{ - struct mdd_object *mdd_obj = md2mdd_obj(obj); - - return mdd_attr_get_internal_locked(env, mdd_obj, ma); -} - -static int dot_lustre_mdd_attr_set(const struct lu_env *env, - struct md_object *obj, - const struct md_attr *ma) -{ - return -EPERM; -} - static int dot_lustre_mdd_xattr_get(const struct lu_env *env, struct md_object *obj, struct lu_buf *buf, const char *name) @@ -544,7 +564,7 @@ static int dot_lustre_mdd_open(const struct lu_env *env, struct md_object *obj, } static int dot_lustre_mdd_close(const struct lu_env *env, struct md_object *obj, - struct md_attr *ma) + struct md_attr *ma, int mode) { struct mdd_object *mdd_obj = md2mdd_obj(obj); @@ -561,30 +581,29 @@ static int dot_lustre_mdd_object_sync(const struct lu_env *env, return -ENOSYS; } -static dt_obj_version_t dot_lustre_mdd_version_get(const struct lu_env *env, - struct md_object *obj) +static int dot_lustre_mdd_path(const struct lu_env *env, struct md_object *obj, + char *path, int pathlen, __u64 *recno, int *linkno) { - return 0; + return -ENOSYS; } -static void dot_lustre_mdd_version_set(const struct lu_env *env, - struct md_object *obj, - dt_obj_version_t version) +static int dot_file_lock(const struct lu_env *env, struct md_object *obj, + struct lov_mds_md *lmm, struct ldlm_extent *extent, + struct lustre_handle *lockh) { - return; + return -ENOSYS; } -static int dot_lustre_mdd_path(const struct lu_env *env, struct md_object *obj, - char *path, int pathlen, __u64 *recno, int *linkno) +static int dot_file_unlock(const struct lu_env *env, struct md_object *obj, + struct lov_mds_md *lmm, struct lustre_handle *lockh) { return -ENOSYS; } - static struct md_object_operations mdd_dot_lustre_obj_ops = { .moo_permission = dot_lustre_mdd_permission, - .moo_attr_get = dot_lustre_mdd_attr_get, - .moo_attr_set = dot_lustre_mdd_attr_set, + .moo_attr_get = mdd_attr_get, + .moo_attr_set = mdd_attr_set, .moo_xattr_get = dot_lustre_mdd_xattr_get, .moo_xattr_list = dot_lustre_mdd_xattr_list, .moo_xattr_set = dot_lustre_mdd_xattr_set, @@ -598,9 +617,9 @@ static struct md_object_operations mdd_dot_lustre_obj_ops = { .moo_close = dot_lustre_mdd_close, .moo_capa_get = mdd_capa_get, .moo_object_sync = dot_lustre_mdd_object_sync, - .moo_version_get = dot_lustre_mdd_version_get, - .moo_version_set = dot_lustre_mdd_version_set, .moo_path = dot_lustre_mdd_path, + .moo_file_lock = dot_file_lock, + .moo_file_unlock = dot_file_unlock, }; @@ -725,8 +744,7 @@ static int obf_attr_get(const struct lu_env *env, struct md_object *obj, /* "fid" is a virtual object and hence does not have any "real" * attributes. So we reuse attributes of .lustre for "fid" dir */ ma->ma_need |= MA_INODE; - rc = dot_lustre_mdd_attr_get(env, &mdd->mdd_dot_lustre->mod_obj, - ma); + rc = mdd_attr_get(env, &mdd->mdd_dot_lustre->mod_obj, ma); if (rc) return rc; ma->ma_valid |= MA_INODE; @@ -744,9 +762,9 @@ static int obf_attr_get(const struct lu_env *env, struct md_object *obj, return 0; if (ma->ma_need & MA_LOV_DEF) { - rc = mdd_get_default_md(mdd_obj, ma->ma_lmm, - &ma->ma_lmm_size); + rc = mdd_get_default_md(mdd_obj, ma->ma_lmm); if (rc > 0) { + ma->ma_lmm_size = rc; ma->ma_valid |= MA_LOV; rc = 0; } @@ -762,6 +780,12 @@ static int obf_attr_set(const struct lu_env *env, struct md_object *obj, return -EPERM; } +static int obf_xattr_list(const struct lu_env *env, + struct md_object *obj, struct lu_buf *buf) +{ + return 0; +} + static int obf_xattr_get(const struct lu_env *env, struct md_object *obj, struct lu_buf *buf, const char *name) @@ -769,6 +793,21 @@ static int obf_xattr_get(const struct lu_env *env, return 0; } +static int obf_xattr_set(const struct lu_env *env, + struct md_object *obj, + const struct lu_buf *buf, const char *name, + int fl) +{ + return -EPERM; +} + +static int obf_xattr_del(const struct lu_env *env, + struct md_object *obj, + const char *name) +{ + return -EPERM; +} + static int obf_mdd_open(const struct lu_env *env, struct md_object *obj, int flags) { @@ -782,7 +821,7 @@ static int obf_mdd_open(const struct lu_env *env, struct md_object *obj, } static int obf_mdd_close(const struct lu_env *env, struct md_object *obj, - struct md_attr *ma) + struct md_attr *ma, int mode) { struct mdd_object *mdd_obj = md2mdd_obj(obj); @@ -807,13 +846,16 @@ static int obf_path(const struct lu_env *env, struct md_object *obj, } static struct md_object_operations mdd_obf_obj_ops = { - .moo_attr_get = obf_attr_get, - .moo_attr_set = obf_attr_set, - .moo_xattr_get = obf_xattr_get, - .moo_open = obf_mdd_open, - .moo_close = obf_mdd_close, - .moo_readpage = obf_mdd_readpage, - .moo_path = obf_path + .moo_attr_get = obf_attr_get, + .moo_attr_set = obf_attr_set, + .moo_xattr_list = obf_xattr_list, + .moo_xattr_get = obf_xattr_get, + .moo_xattr_set = obf_xattr_set, + .moo_xattr_del = obf_xattr_del, + .moo_open = obf_mdd_open, + .moo_close = obf_mdd_close, + .moo_readpage = obf_mdd_readpage, + .moo_path = obf_path }; /** @@ -834,11 +876,19 @@ static int obf_lookup(const struct lu_env *env, struct md_object *p, sscanf(name, SFID, RFID(f)); if (!fid_is_sane(f)) { - CWARN("bad FID format [%s], should be "DFID"\n", lname->ln_name, - (__u64)1, 2, 0); + CWARN("%s: bad FID format [%s], should be "DFID"\n", + mdd->mdd_obd_dev->obd_name, lname->ln_name, + (__u64)FID_SEQ_NORMAL, 1, 0); GOTO(out, rc = -EINVAL); } + if (!fid_is_norm(f)) { + CWARN("%s: "DFID" is invalid, sequence should be " + ">= "LPX64"\n", mdd->mdd_obd_dev->obd_name, PFID(f), + (__u64)FID_SEQ_NORMAL); + GOTO(out, rc = -EINVAL); + } + /* Check if object with this fid exists */ child = mdd_object_find(env, mdd, f); if (child == NULL) @@ -982,16 +1032,14 @@ static int mdd_process_config(const struct lu_env *env, rc = mdd_init_obd(env, m, cfg); if (rc) { - CERROR("lov init error %d \n", rc); + CERROR("lov init error %d\n", rc); GOTO(out, rc); } - rc = mdd_txn_init_credits(env, m); - if (rc) - break; mdd_changelog_init(env, m); break; case LCFG_CLEANUP: + lu_dev_del_linkage(d->ld_site, d); mdd_device_shutdown(env, m, cfg); default: rc = next->ld_ops->ldo_process_config(env, next, cfg); @@ -1010,7 +1058,7 @@ static int mdd_lov_set_nextid(const struct lu_env *env, ENTRY; LASSERT(mds->mds_lov_objids != NULL); - rc = obd_set_info_async(mds->mds_osc_exp, strlen(KEY_NEXT_ID), + rc = obd_set_info_async(mds->mds_lov_exp, strlen(KEY_NEXT_ID), KEY_NEXT_ID, mds->mds_lov_desc.ld_tgt_count, mds->mds_lov_objids, NULL); @@ -1054,11 +1102,13 @@ static int mdd_recovery_complete(const struct lu_env *env, } #endif /* Call that with obd_recovering = 1 just to update objids */ - obd_notify(obd->u.mds.mds_osc_obd, NULL, (obd->obd_async_recov ? + obd_notify(obd->u.mds.mds_lov_obd, NULL, (obd->obd_async_recov ? OBD_NOTIFY_SYNC_NONBLOCK : OBD_NOTIFY_SYNC), NULL); /* Drop obd_recovering to 0 and call o_postrecov to recover mds_lov */ + cfs_spin_lock(&obd->obd_dev_lock); obd->obd_recovering = 0; + cfs_spin_unlock(&obd->obd_dev_lock); obd->obd_type->typ_dt_ops->o_postrecov(obd); /* XXX: orphans handling. */ @@ -1075,6 +1125,7 @@ static int mdd_prepare(const struct lu_env *env, struct mdd_device *mdd = lu2mdd_dev(cdev); struct lu_device *next = &mdd->mdd_child->dd_lu_dev; struct dt_object *root; + struct lu_fid fid; int rc; ENTRY; @@ -1101,8 +1152,19 @@ static int mdd_prepare(const struct lu_env *env, GOTO(out, rc); } + /* we use capa file to declare llog changes, + * will be fixed with new llog in 2.3 */ + root = dt_store_open(env, mdd->mdd_child, "", CAPA_KEYS, &fid); + if (IS_ERR(root)) + GOTO(out, rc = PTR_ERR(root)); + + mdd->mdd_capa = root; + rc = mdd_lfsck_setup(env, mdd); + + GOTO(out, rc); + out: - RETURN(rc); + return rc; } const struct lu_device_operations mdd_lu_ops = { @@ -1129,7 +1191,7 @@ static int mdd_root_get(const struct lu_env *env, * No permission check is needed. */ static int mdd_statfs(const struct lu_env *env, struct md_device *m, - cfs_kstatfs_t *sfs) + struct obd_statfs *sfs) { struct mdd_device *mdd = lu2mdd_dev(&m->md_lu_dev); int rc; @@ -1165,7 +1227,11 @@ static int mdd_init_capa_ctxt(const struct lu_env *env, struct md_device *m, int rc; ENTRY; + /* need barrier for mds_capa_keys access. */ + cfs_down_write(&mds->mds_notify_lock); mds->mds_capa_keys = keys; + cfs_up_write(&mds->mds_notify_lock); + rc = mdd_child_ops(mdd)->dt_init_capa_ctxt(env, mdd->mdd_child, mode, timeout, alg, keys); RETURN(rc); @@ -1177,12 +1243,12 @@ static int mdd_update_capa_key(const struct lu_env *env, { struct mds_capa_info info = { .uuid = NULL, .capa = key }; struct mdd_device *mdd = lu2mdd_dev(&m->md_lu_dev); - struct obd_export *lov_exp = mdd2obd_dev(mdd)->u.mds.mds_osc_exp; + struct obd_export *lov_exp = mdd2obd_dev(mdd)->u.mds.mds_lov_exp; int rc; ENTRY; - rc = obd_set_info_async(lov_exp, sizeof(KEY_CAPA_KEY), KEY_CAPA_KEY, - sizeof(info), &info, NULL); + rc = obd_set_info_async(env, lov_exp, sizeof(KEY_CAPA_KEY), + KEY_CAPA_KEY, sizeof(info), &info, NULL); RETURN(rc); } @@ -1220,13 +1286,12 @@ static struct lu_device *mdd_device_free(const struct lu_env *env, struct lu_device *lu) { struct mdd_device *m = lu2mdd_dev(lu); - struct lu_device *next = &m->mdd_child->dd_lu_dev; ENTRY; LASSERT(cfs_atomic_read(&lu->ld_ref) == 0); md_device_fini(&m->mdd_md_dev); OBD_FREE_PTR(m); - RETURN(next); + RETURN(NULL); } static struct obd_ops mdd_obd_device_ops = { @@ -1323,7 +1388,7 @@ static int mdd_changelog_user_register(struct mdd_device *mdd, int *id) rec->cur_endrec = mdd->mdd_cl.mc_index; cfs_spin_unlock(&mdd->mdd_cl.mc_user_lock); - rc = llog_add(ctxt, &rec->cur_hdr, NULL, NULL, 0); + rc = llog_add(NULL, ctxt, &rec->cur_hdr, NULL, NULL, 0); CDEBUG(D_IOCTL, "Registered changelog user %d\n", *id); out: @@ -1332,6 +1397,35 @@ out: RETURN(rc); } +int mdd_declare_llog_cancel(const struct lu_env *env, struct mdd_device *mdd, + struct thandle *handle) +{ + int rc; + + + /* XXX: this is a temporary solution to declare llog changes + * will be fixed in 2.3 with new llog implementation */ + + LASSERT(mdd->mdd_capa); + + /* the llog record could be canceled either by modifying + * the plain llog's header or by destroying the llog itself + * when this record is the last one in it, it can't be known + * here, but the catlog's header will also be modified for + * the second case, then the first case can be covered and + * is no need to declare it */ + + /* destroy empty plain log */ + rc = dt_declare_destroy(env, mdd->mdd_capa, handle); + if (rc) + return rc; + + /* record the catlog's header if an empty plain log was destroyed */ + rc = dt_declare_record_write(env, mdd->mdd_capa, + sizeof(struct llog_logid_rec), 0, handle); + return rc; +} + struct mdd_changelog_user_data { __u64 mcud_endrec; /**< purge record for this user */ __u64 mcud_minrec; /**< lowest changelog recno still referenced */ @@ -1339,6 +1433,7 @@ struct mdd_changelog_user_data { __u32 mcud_minid; /**< user id with lowest rec reference */ __u32 mcud_usercount; int mcud_found:1; + struct mdd_device *mcud_mdd; }; #define MCUD_UNREGISTER -1LL @@ -1346,13 +1441,14 @@ struct mdd_changelog_user_data { * 1. Find the smallest record everyone is willing to purge * 2. Update the last purgeable record for this user */ -static int mdd_changelog_user_purge_cb(struct llog_handle *llh, - struct llog_rec_hdr *hdr, void *data) +static int mdd_changelog_user_purge_cb(const struct lu_env *env, + struct llog_handle *llh, + struct llog_rec_hdr *hdr, void *data) { - struct llog_changelog_user_rec *rec; - struct mdd_changelog_user_data *mcud = - (struct mdd_changelog_user_data *)data; - int rc; + struct llog_changelog_user_rec *rec; + struct mdd_changelog_user_data *mcud = data; + int rc; + ENTRY; LASSERT(llh->lgh_hdr->llh_flags & LLOG_F_IS_PLAIN); @@ -1381,12 +1477,35 @@ static int mdd_changelog_user_purge_cb(struct llog_handle *llh, /* Special case: unregister this user */ if (mcud->mcud_endrec == MCUD_UNREGISTER) { struct llog_cookie cookie; + void *th; + struct mdd_device *mdd = mcud->mcud_mdd; + cookie.lgc_lgl = llh->lgh_id; cookie.lgc_index = hdr->lrh_index; - rc = llog_cat_cancel_records(llh->u.phd.phd_cat_handle, - 1, &cookie); + + /* XXX This is a workaround for the deadlock of changelog + * adding vs. changelog cancelling. LU-81. */ + th = mdd_trans_create(env, mdd); + if (IS_ERR(th)) { + CERROR("Cannot get thandle\n"); + RETURN(-ENOMEM); + } + + rc = mdd_declare_llog_cancel(env, mdd, th); + if (rc) + GOTO(stop, rc); + + rc = mdd_trans_start(env, mdd, th); + if (rc) + GOTO(stop, rc); + + rc = llog_cat_cancel_records(env, llh->u.phd.phd_cat_handle, + 1, &cookie); if (rc == 0) mcud->mcud_usercount--; + +stop: + mdd_trans_stop(env, mdd, 0, th); RETURN(rc); } @@ -1396,13 +1515,14 @@ static int mdd_changelog_user_purge_cb(struct llog_handle *llh, /* hdr+1 is loc of data */ hdr->lrh_len -= sizeof(*hdr) + sizeof(struct llog_rec_tail); - rc = llog_write_rec(llh, hdr, NULL, 0, (void *)(hdr + 1), - hdr->lrh_index); + rc = llog_write_rec(env, llh, hdr, NULL, 0, (void *)(hdr + 1), + hdr->lrh_index); RETURN(rc); } -static int mdd_changelog_user_purge(struct mdd_device *mdd, int id, +static int mdd_changelog_user_purge(const struct lu_env *env, + struct mdd_device *mdd, int id, long long endrec) { struct mdd_changelog_user_data data; @@ -1417,6 +1537,7 @@ static int mdd_changelog_user_purge(struct mdd_device *mdd, int id, data.mcud_minrec = 0; data.mcud_usercount = 0; data.mcud_endrec = endrec; + data.mcud_mdd = mdd; cfs_spin_lock(&mdd->mdd_cl.mc_lock); endrec = mdd->mdd_cl.mc_index; cfs_spin_unlock(&mdd->mdd_cl.mc_lock); @@ -1430,13 +1551,14 @@ static int mdd_changelog_user_purge(struct mdd_device *mdd, int id, return -ENXIO; LASSERT(ctxt->loc_handle->lgh_hdr->llh_flags & LLOG_F_IS_CAT); - rc = llog_cat_process(ctxt->loc_handle, mdd_changelog_user_purge_cb, - (void *)&data, 0, 0); + rc = llog_cat_process(env, ctxt->loc_handle, + mdd_changelog_user_purge_cb, (void *)&data, + 0, 0); if ((rc >= 0) && (data.mcud_minrec > 0)) { CDEBUG(D_IOCTL, "Purging changelog entries up to "LPD64 ", referenced by "CHANGELOG_USER_PREFIX"%d\n", data.mcud_minrec, data.mcud_minid); - rc = mdd_changelog_llog_cancel(mdd, data.mcud_minrec); + rc = mdd_changelog_llog_cancel(env, mdd, data.mcud_minrec); } else { CWARN("Could not determine changelog records to purge; rc=%d\n", rc); @@ -1477,11 +1599,33 @@ static int mdd_iocontrol(const struct lu_env *env, struct md_device *m, mdd = lu2mdd_dev(&m->md_lu_dev); /* Doesn't use obd_ioctl_data */ - if (cmd == OBD_IOC_CHANGELOG_CLEAR) { + switch (cmd) { + case OBD_IOC_CHANGELOG_CLEAR: { struct changelog_setinfo *cs = karg; - rc = mdd_changelog_user_purge(mdd, cs->cs_id, cs->cs_recno); + rc = mdd_changelog_user_purge(env, mdd, cs->cs_id, + cs->cs_recno); RETURN(rc); } + case OBD_IOC_GET_MNTOPT: { + mntopt_t *mntopts = (mntopt_t *)karg; + *mntopts = mdd->mdd_dt_conf.ddp_mntopts; + RETURN(0); + } + case OBD_IOC_START_LFSCK: { + struct lfsck_start *start = karg; + struct md_lfsck *lfsck = &mdd->mdd_lfsck; + + /* Return the kernel service version. */ + /* XXX: version can be used for compatibility in the future. */ + start->ls_version = lfsck->ml_version; + rc = mdd_lfsck_start(env, lfsck, start); + RETURN(rc); + } + case OBD_IOC_STOP_LFSCK: { + rc = mdd_lfsck_stop(env, &mdd->mdd_lfsck); + RETURN(rc); + } + } /* Below ioctls use obd_ioctl_data */ if (len != sizeof(*data)) { @@ -1499,11 +1643,11 @@ static int mdd_iocontrol(const struct lu_env *env, struct md_device *m, rc = mdd_changelog_user_register(mdd, &data->ioc_u32_1); break; case OBD_IOC_CHANGELOG_DEREG: - rc = mdd_changelog_user_purge(mdd, data->ioc_u32_1, + rc = mdd_changelog_user_purge(env, mdd, data->ioc_u32_1, MCUD_UNREGISTER); break; default: - rc = -EOPNOTSUPP; + rc = -ENOTTY; } RETURN (rc); @@ -1602,26 +1746,44 @@ static struct lu_local_obj_desc llod_mdd_root = { .llod_feat = &dt_directory_features, }; +static struct lu_local_obj_desc llod_lfsck_bookmark = { + .llod_name = lfsck_bookmark_name, + .llod_oid = LFSCK_BOOKMARK_OID, + .llod_is_index = 0, +}; + static int __init mdd_mod_init(void) { - struct lprocfs_static_vars lvars; - lprocfs_mdd_init_vars(&lvars); + struct lprocfs_static_vars lvars; + int rc; + + lprocfs_mdd_init_vars(&lvars); + + rc = lu_kmem_init(mdd_caches); + if (rc) + return rc; - llo_local_obj_register(&llod_capa_key); - llo_local_obj_register(&llod_mdd_orphan); - llo_local_obj_register(&llod_mdd_root); + llo_local_obj_register(&llod_capa_key); + llo_local_obj_register(&llod_mdd_orphan); + llo_local_obj_register(&llod_mdd_root); + llo_local_obj_register(&llod_lfsck_bookmark); - return class_register_type(&mdd_obd_device_ops, NULL, lvars.module_vars, - LUSTRE_MDD_NAME, &mdd_device_type); + rc = class_register_type(&mdd_obd_device_ops, NULL, lvars.module_vars, + LUSTRE_MDD_NAME, &mdd_device_type); + if (rc) + lu_kmem_fini(mdd_caches); + return rc; } static void __exit mdd_mod_exit(void) { - llo_local_obj_unregister(&llod_capa_key); - llo_local_obj_unregister(&llod_mdd_orphan); - llo_local_obj_unregister(&llod_mdd_root); + llo_local_obj_unregister(&llod_capa_key); + llo_local_obj_unregister(&llod_mdd_orphan); + llo_local_obj_unregister(&llod_mdd_root); + llo_local_obj_unregister(&llod_lfsck_bookmark); - class_unregister_type(LUSTRE_MDD_NAME); + class_unregister_type(LUSTRE_MDD_NAME); + lu_kmem_fini(mdd_caches); } MODULE_AUTHOR("Sun Microsystems, Inc. ");