X-Git-Url: https://git.whamcloud.com/?p=fs%2Flustre-release.git;a=blobdiff_plain;f=lustre%2Fmdd%2Fmdd_device.c;h=0c891fa1be2eb3197c36aab4c8c7246cd311c228;hp=05bf856d15dc858aa359c0a098cd7c47fad3f4ed;hb=22464d1230ed58461f51d881f512d5e16644a735;hpb=905a547356dab45f06dd535776bde3fab76c4688 diff --git a/lustre/mdd/mdd_device.c b/lustre/mdd/mdd_device.c index 05bf856..0c891fa 100644 --- a/lustre/mdd/mdd_device.c +++ b/lustre/mdd/mdd_device.c @@ -26,10 +26,13 @@ * 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 Whamcloud, Inc. + */ +/* * This file is part of Lustre, http://www.lustre.org/ * Lustre is a trademark of Sun Microsystems, Inc. * @@ -46,7 +49,11 @@ #define DEBUG_SUBSYSTEM S_MDS #include +#ifdef HAVE_EXT4_LDISKFS +#include +#else #include +#endif #include #include #include @@ -55,7 +62,11 @@ #include #include +#ifdef HAVE_EXT4_LDISKFS +#include +#else #include +#endif #include #include #include /* for changelogs */ @@ -80,13 +91,15 @@ 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); mdd->mdd_atime_diff = MAX_ATIME_DIFF; + /* sync permission changes */ + mdd->mdd_sync_permission = 1; rc = mdd_procfs_init(mdd, name); RETURN(rc); @@ -116,11 +129,17 @@ static void mdd_device_shutdown(const struct lu_env *env, ENTRY; mdd_changelog_fini(env, m); dt_txn_callback_del(m->mdd_child, &m->mdd_txn_cb); - mdd_object_put(env, m->mdd_dot_lustre_objs.mdd_obf); - mdd_object_put(env, m->mdd_dot_lustre); + if (m->mdd_dot_lustre_objs.mdd_obf) + mdd_object_put(env, m->mdd_dot_lustre_objs.mdd_obf); + if (m->mdd_dot_lustre) + mdd_object_put(env, m->mdd_dot_lustre); 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; @@ -138,11 +157,11 @@ static int changelog_init_cb(struct llog_handle *llh, struct llog_rec_hdr *hdr, CDEBUG(D_INFO, "seeing record at index %d/%d/"LPU64" t=%x %.*s in log "LPX64"\n", - hdr->lrh_index, rec->cr_hdr.lrh_index, rec->cr_index, - rec->cr_type, rec->cr_namelen, rec->cr_name, + hdr->lrh_index, rec->cr_hdr.lrh_index, rec->cr.cr_index, + rec->cr.cr_type, rec->cr.cr_namelen, rec->cr.cr_name, llh->lgh_id.lgl_oid); - mdd->mdd_cl.mc_index = rec->cr_index; + mdd->mdd_cl.mc_index = rec->cr.cr_index; RETURN(LLOG_PROC_BREAK); } @@ -161,9 +180,9 @@ static int changelog_user_init_cb(struct llog_handle *llh, " in log "LPX64"\n", hdr->lrh_index, rec->cur_hdr.lrh_index, rec->cur_id, rec->cur_endrec, llh->lgh_id.lgl_oid); - spin_lock(&mdd->mdd_cl.mc_user_lock); + cfs_spin_lock(&mdd->mdd_cl.mc_user_lock); mdd->mdd_cl.mc_lastuser = rec->cur_id; - spin_unlock(&mdd->mdd_cl.mc_user_lock); + cfs_spin_unlock(&mdd->mdd_cl.mc_user_lock); RETURN(LLOG_PROC_BREAK); } @@ -228,12 +247,11 @@ static int mdd_changelog_init(const struct lu_env *env, struct mdd_device *mdd) int rc; mdd->mdd_cl.mc_index = 0; - spin_lock_init(&mdd->mdd_cl.mc_lock); - cfs_waitq_init(&mdd->mdd_cl.mc_waitq); + cfs_spin_lock_init(&mdd->mdd_cl.mc_lock); 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; - spin_lock_init(&mdd->mdd_cl.mc_user_lock); + cfs_spin_lock_init(&mdd->mdd_cl.mc_user_lock); mdd->mdd_cl.mc_lastuser = 0; rc = mdd_changelog_llog_init(mdd); @@ -263,21 +281,28 @@ int mdd_changelog_on(struct mdd_device *mdd, int on) mdd2obd_dev(mdd)->obd_name); rc = -ESRCH; } else { - spin_lock(&mdd->mdd_cl.mc_lock); + cfs_spin_lock(&mdd->mdd_cl.mc_lock); mdd->mdd_cl.mc_flags |= CLM_ON; - spin_unlock(&mdd->mdd_cl.mc_lock); + cfs_spin_unlock(&mdd->mdd_cl.mc_lock); rc = mdd_changelog_write_header(mdd, CLM_START); } } else if ((on == 0) && ((mdd->mdd_cl.mc_flags & CLM_ON) == CLM_ON)) { LCONSOLE_INFO("%s: changelog off\n",mdd2obd_dev(mdd)->obd_name); rc = mdd_changelog_write_header(mdd, CLM_FINI); - spin_lock(&mdd->mdd_cl.mc_lock); + cfs_spin_lock(&mdd->mdd_cl.mc_lock); mdd->mdd_cl.mc_flags &= ~CLM_ON; - spin_unlock(&mdd->mdd_cl.mc_lock); + cfs_spin_unlock(&mdd->mdd_cl.mc_lock); } return rc; } +static __u64 cl_time(void) { + cfs_fs_time_t time; + + cfs_fs_time_current(&time); + return (((__u64)time.tv_sec) << 30) + time.tv_nsec; +} + /** Add a changelog entry \a rec to the changelog llog * \param mdd * \param rec @@ -293,19 +318,16 @@ int mdd_changelog_llog_write(struct mdd_device *mdd, struct llog_ctxt *ctxt; int rc; - if ((mdd->mdd_cl.mc_mask & (1 << rec->cr_type)) == 0) - return 0; - - rec->cr_hdr.lrh_len = llog_data_len(sizeof(*rec) + rec->cr_namelen); + 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_time = cfs_time_current_64(); - spin_lock(&mdd->mdd_cl.mc_lock); + 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_index = ++mdd->mdd_cl.mc_index; - spin_unlock(&mdd->mdd_cl.mc_lock); + 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; @@ -314,8 +336,6 @@ int mdd_changelog_llog_write(struct mdd_device *mdd, rc = llog_add(ctxt, &rec->cr_hdr, NULL, NULL, 0); llog_ctxt_put(ctxt); - cfs_waitq_signal(&mdd->mdd_cl.mc_waitq); - return rc; } @@ -336,9 +356,9 @@ int mdd_changelog_llog_cancel(struct mdd_device *mdd, long long endrec) if (ctxt == NULL) return -ENXIO; - spin_lock(&mdd->mdd_cl.mc_lock); + cfs_spin_lock(&mdd->mdd_cl.mc_lock); cur = (long long)mdd->mdd_cl.mc_index; - spin_unlock(&mdd->mdd_cl.mc_lock); + cfs_spin_unlock(&mdd->mdd_cl.mc_lock); if (endrec > cur) endrec = cur; @@ -351,6 +371,7 @@ 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; @@ -361,6 +382,7 @@ 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(); + /* XXX: transaction is started by llog itself */ rc = llog_cancel(ctxt, NULL, 1, (struct llog_cookie *)&endrec, 0); out: llog_ctxt_put(ctxt); @@ -386,17 +408,19 @@ int mdd_changelog_write_header(struct mdd_device *mdd, int markerflags) if (rec == NULL) RETURN(-ENOMEM); - rec->cr_flags = CLF_VERSION; - rec->cr_type = CL_MARK; - rec->cr_namelen = len; - memcpy(rec->cr_name, obd->obd_name, rec->cr_namelen); + rec->cr.cr_flags = CLF_VERSION; + rec->cr.cr_type = CL_MARK; + rec->cr.cr_namelen = len; + memcpy(rec->cr.cr_name, obd->obd_name, rec->cr.cr_namelen); /* Status and action flags */ - rec->cr_markerflags = mdd->mdd_cl.mc_flags | markerflags; + 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_time; + mdd->mdd_cl.mc_starttime = cfs_time_current_64(); OBD_FREE(rec, reclen); RETURN(rc); @@ -413,13 +437,13 @@ static int create_dot_lustre_dir(const struct lu_env *env, struct mdd_device *m) memcpy(fid, &LU_DOT_LUSTRE_FID, sizeof(struct lu_fid)); mdo = llo_store_create_index(env, &m->mdd_md_dev, m->mdd_child, - mdd_root_dir_name, mdd_dot_lustre_name, + mdd_root_dir_name, dot_lustre_name, fid, &dt_directory_features); /* .lustre dir may be already present */ if (IS_ERR(mdo) && PTR_ERR(mdo) != -EEXIST) { rc = PTR_ERR(mdo); CERROR("creating obj [%s] fid = "DFID" rc = %d\n", - mdd_dot_lustre_name, PFID(fid), rc); + dot_lustre_name, PFID(fid), rc); RETURN(rc); } @@ -429,27 +453,89 @@ static int create_dot_lustre_dir(const struct lu_env *env, struct mdd_device *m) return 0; } -static int dot_lustre_attr_get(const struct lu_env *env, struct md_object *obj, - struct md_attr *ma) + +static int dot_lustre_mdd_permission(const struct lu_env *env, + struct md_object *pobj, + struct md_object *cobj, + struct md_attr *attr, int mask) +{ + if (mask & ~(MAY_READ | MAY_EXEC)) + return -EPERM; + else + 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_attr_set(const struct lu_env *env, struct md_object *obj, - const struct md_attr *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_xattr_get(const struct lu_env *env, - struct md_object *obj, struct lu_buf *buf, - const char *name) +static int dot_lustre_mdd_xattr_get(const struct lu_env *env, + struct md_object *obj, struct lu_buf *buf, + const char *name) +{ + return 0; +} + +static int dot_lustre_mdd_xattr_list(const struct lu_env *env, + struct md_object *obj, struct lu_buf *buf) { return 0; } +static int dot_lustre_mdd_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 dot_lustre_mdd_xattr_del(const struct lu_env *env, + struct md_object *obj, + const char *name) +{ + return -EPERM; +} + +static int dot_lustre_mdd_readlink(const struct lu_env *env, + struct md_object *obj, struct lu_buf *buf) +{ + return 0; +} + +static int dot_lustre_mdd_object_create(const struct lu_env *env, + struct md_object *obj, + const struct md_op_spec *spec, + struct md_attr *ma) +{ + return -EPERM; +} + +static int dot_lustre_mdd_ref_add(const struct lu_env *env, + struct md_object *obj, + const struct md_attr *ma) +{ + return -EPERM; +} + +static int dot_lustre_mdd_ref_del(const struct lu_env *env, + struct md_object *obj, + struct md_attr *ma) +{ + return -EPERM; +} + static int dot_lustre_mdd_open(const struct lu_env *env, struct md_object *obj, int flags) { @@ -462,14 +548,8 @@ static int dot_lustre_mdd_open(const struct lu_env *env, struct md_object *obj, return 0; } -static int dot_lustre_path(const struct lu_env *env, struct md_object *obj, - char *path, int pathlen, __u64 *recno, int *linkno) -{ - return -ENOSYS; -} - -static int dot_lustre_close(const struct lu_env *env, struct md_object *obj, - struct md_attr *ma) +static int dot_lustre_mdd_close(const struct lu_env *env, struct md_object *obj, + struct md_attr *ma, int mode) { struct mdd_object *mdd_obj = md2mdd_obj(obj); @@ -480,35 +560,57 @@ static int dot_lustre_close(const struct lu_env *env, struct md_object *obj, return 0; } -static dt_obj_version_t dot_lustre_version_get(const struct lu_env *env, - struct md_object *obj) +static int dot_lustre_mdd_object_sync(const struct lu_env *env, + struct md_object *obj) { - return 0; + return -ENOSYS; } -static void dot_lustre_version_set(const struct lu_env *env, - struct md_object *obj, - dt_obj_version_t version) +static int dot_lustre_mdd_path(const struct lu_env *env, struct md_object *obj, + char *path, int pathlen, __u64 *recno, int *linkno) { - return; + return -ENOSYS; +} + +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 -ENOSYS; } +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_attr_get = dot_lustre_attr_get, - .moo_attr_set = dot_lustre_attr_set, - .moo_xattr_get = dot_lustre_xattr_get, - .moo_open = dot_lustre_mdd_open, - .moo_close = dot_lustre_close, - .moo_readpage = mdd_readpage, - .moo_version_get = dot_lustre_version_get, - .moo_version_set = dot_lustre_version_set, - .moo_path = dot_lustre_path + .moo_permission = dot_lustre_mdd_permission, + .moo_attr_get = dot_lustre_mdd_attr_get, + .moo_attr_set = dot_lustre_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, + .moo_xattr_del = dot_lustre_mdd_xattr_del, + .moo_readpage = mdd_readpage, + .moo_readlink = dot_lustre_mdd_readlink, + .moo_object_create = dot_lustre_mdd_object_create, + .moo_ref_add = dot_lustre_mdd_ref_add, + .moo_ref_del = dot_lustre_mdd_ref_del, + .moo_open = dot_lustre_mdd_open, + .moo_close = dot_lustre_mdd_close, + .moo_capa_get = mdd_capa_get, + .moo_object_sync = dot_lustre_mdd_object_sync, + .moo_path = dot_lustre_mdd_path, + .moo_file_lock = dot_file_lock, + .moo_file_unlock = dot_file_unlock, }; -static int dot_lustre_lookup(const struct lu_env *env, struct md_object *p, - const struct lu_name *lname, struct lu_fid *f, - struct md_op_spec *spec) + +static int dot_lustre_mdd_lookup(const struct lu_env *env, struct md_object *p, + const struct lu_name *lname, struct lu_fid *f, + struct md_op_spec *spec) { if (strcmp(lname->ln_name, mdd_obf_dir_name) == 0) *f = LU_OBF_FID; @@ -518,45 +620,102 @@ static int dot_lustre_lookup(const struct lu_env *env, struct md_object *p, return 0; } -static int dot_lustre_create(const struct lu_env *env, struct md_object *pobj, - const struct lu_name *lname, - struct md_object *child, struct md_op_spec *spec, - struct md_attr* ma) +static mdl_mode_t dot_lustre_mdd_lock_mode(const struct lu_env *env, + struct md_object *obj, + mdl_mode_t mode) +{ + return MDL_MINMODE; +} + +static int dot_lustre_mdd_create(const struct lu_env *env, + struct md_object *pobj, + const struct lu_name *lname, + struct md_object *child, + struct md_op_spec *spec, + struct md_attr* ma) +{ + return -EPERM; +} + +static int dot_lustre_mdd_create_data(const struct lu_env *env, + struct md_object *p, + struct md_object *o, + const struct md_op_spec *spec, + struct md_attr *ma) +{ + return -EPERM; +} + +static int dot_lustre_mdd_rename(const struct lu_env *env, + struct md_object *src_pobj, + struct md_object *tgt_pobj, + const struct lu_fid *lf, + const struct lu_name *lsname, + struct md_object *tobj, + const struct lu_name *ltname, + struct md_attr *ma) +{ + return -EPERM; +} + +static int dot_lustre_mdd_link(const struct lu_env *env, + struct md_object *tgt_obj, + struct md_object *src_obj, + const struct lu_name *lname, + struct md_attr *ma) +{ + return -EPERM; +} + +static int dot_lustre_mdd_unlink(const struct lu_env *env, + struct md_object *pobj, + struct md_object *cobj, + const struct lu_name *lname, + struct md_attr *ma) { return -EPERM; } -static int dot_lustre_rename(const struct lu_env *env, - struct md_object *src_pobj, - struct md_object *tgt_pobj, - const struct lu_fid *lf, - const struct lu_name *lsname, - struct md_object *tobj, - const struct lu_name *ltname, struct md_attr *ma) +static int dot_lustre_mdd_name_insert(const struct lu_env *env, + struct md_object *obj, + const struct lu_name *lname, + const struct lu_fid *fid, + const struct md_attr *ma) { return -EPERM; } -static int dot_lustre_link(const struct lu_env *env, struct md_object *tgt_obj, - struct md_object *src_obj, - const struct lu_name *lname, struct md_attr *ma) +static int dot_lustre_mdd_name_remove(const struct lu_env *env, + struct md_object *obj, + const struct lu_name *lname, + const struct md_attr *ma) { return -EPERM; } -static int dot_lustre_unlink(const struct lu_env *env, struct md_object *pobj, - struct md_object *cobj, const struct lu_name *lname, - struct md_attr *ma) +static int dot_lustre_mdd_rename_tgt(const struct lu_env *env, + struct md_object *pobj, + struct md_object *tobj, + const struct lu_fid *fid, + const struct lu_name *lname, + struct md_attr *ma) { return -EPERM; } + static struct md_dir_operations mdd_dot_lustre_dir_ops = { - .mdo_lookup = dot_lustre_lookup, - .mdo_create = dot_lustre_create, - .mdo_rename = dot_lustre_rename, - .mdo_link = dot_lustre_link, - .mdo_unlink = dot_lustre_unlink, + .mdo_is_subdir = mdd_is_subdir, + .mdo_lookup = dot_lustre_mdd_lookup, + .mdo_lock_mode = dot_lustre_mdd_lock_mode, + .mdo_create = dot_lustre_mdd_create, + .mdo_create_data = dot_lustre_mdd_create_data, + .mdo_rename = dot_lustre_mdd_rename, + .mdo_link = dot_lustre_mdd_link, + .mdo_unlink = dot_lustre_mdd_unlink, + .mdo_name_insert = dot_lustre_mdd_name_insert, + .mdo_name_remove = dot_lustre_mdd_name_remove, + .mdo_rename_tgt = dot_lustre_mdd_rename_tgt, }; static int obf_attr_get(const struct lu_env *env, struct md_object *obj, @@ -570,7 +729,8 @@ 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_attr_get(env, &mdd->mdd_dot_lustre->mod_obj, ma); + rc = dot_lustre_mdd_attr_get(env, &mdd->mdd_dot_lustre->mod_obj, + ma); if (rc) return rc; ma->ma_valid |= MA_INODE; @@ -588,9 +748,9 @@ 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; } @@ -626,7 +786,7 @@ 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); @@ -676,8 +836,7 @@ static int obf_lookup(const struct lu_env *env, struct md_object *p, while (*name == '[') name++; - sscanf(name, SFID, &(f->f_seq), &(f->f_oid), - &(f->f_ver)); + 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); @@ -778,7 +937,7 @@ static int mdd_dot_lustre_setup(const struct lu_env *env, struct mdd_device *m) return rc; dt_dot_lustre = dt_store_open(env, m->mdd_child, mdd_root_dir_name, - mdd_dot_lustre_name, fid); + dot_lustre_name, fid); if (IS_ERR(dt_dot_lustre)) { rc = PTR_ERR(dt_dot_lustre); GOTO(out, rc); @@ -827,12 +986,9 @@ 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; @@ -855,7 +1011,7 @@ 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); @@ -899,11 +1055,13 @@ 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. */ @@ -920,6 +1078,7 @@ 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; @@ -946,6 +1105,14 @@ 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)) + mdd->mdd_capa = root; + else + rc = PTR_ERR(root); + out: RETURN(rc); } @@ -974,7 +1141,7 @@ 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, - struct kstatfs *sfs) + cfs_kstatfs_t *sfs) { struct mdd_device *mdd = lu2mdd_dev(&m->md_lu_dev); int rc; @@ -1010,7 +1177,11 @@ 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); @@ -1022,7 +1193,7 @@ 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; @@ -1068,7 +1239,7 @@ static struct lu_device *mdd_device_free(const struct lu_env *env, struct lu_device *next = &m->mdd_child->dd_lu_dev; ENTRY; - LASSERT(atomic_read(&lu->ld_ref) == 0); + LASSERT(cfs_atomic_read(&lu->ld_ref) == 0); md_device_fini(&m->mdd_md_dev); OBD_FREE_PTR(m); RETURN(next); @@ -1115,6 +1286,25 @@ struct md_capainfo *md_capainfo(const struct lu_env *env) } EXPORT_SYMBOL(md_capainfo); +/* + * context key constructor/destructor: + * mdd_quota_key_init, mdd_quota_key_fini + */ +LU_KEY_INIT_FINI(mdd_quota, struct md_quota); + +struct lu_context_key mdd_quota_key = { + .lct_tags = LCT_SESSION, + .lct_init = mdd_quota_key_init, + .lct_fini = mdd_quota_key_fini +}; + +struct md_quota *md_quota(const struct lu_env *env) +{ + LASSERT(env->le_ses != NULL); + return lu_context_key_get(env->le_ses, &mdd_quota_key); +} +EXPORT_SYMBOL(md_quota); + static int mdd_changelog_user_register(struct mdd_device *mdd, int *id) { struct llog_ctxt *ctxt; @@ -1139,15 +1329,15 @@ static int mdd_changelog_user_register(struct mdd_device *mdd, int *id) rec->cur_hdr.lrh_len = sizeof(*rec); rec->cur_hdr.lrh_type = CHANGELOG_USER_REC; - spin_lock(&mdd->mdd_cl.mc_user_lock); + cfs_spin_lock(&mdd->mdd_cl.mc_user_lock); if (mdd->mdd_cl.mc_lastuser == (unsigned int)(-1)) { - spin_unlock(&mdd->mdd_cl.mc_user_lock); + cfs_spin_unlock(&mdd->mdd_cl.mc_user_lock); CERROR("Maximum number of changelog users exceeded!\n"); GOTO(out, rc = -EOVERFLOW); } *id = rec->cur_id = ++mdd->mdd_cl.mc_lastuser; rec->cur_endrec = mdd->mdd_cl.mc_index; - spin_unlock(&mdd->mdd_cl.mc_user_lock); + cfs_spin_unlock(&mdd->mdd_cl.mc_user_lock); rc = llog_add(ctxt, &rec->cur_hdr, NULL, NULL, 0); @@ -1158,6 +1348,35 @@ 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 */ @@ -1165,6 +1384,8 @@ 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; + const struct lu_env *mcud_env; }; #define MCUD_UNREGISTER -1LL @@ -1207,12 +1428,35 @@ 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; + + /* XXX This is a workaround for the deadlock of changelog + * adding vs. changelog cancelling. LU-81. */ + th = mdd_trans_create(mcud->mcud_env, mdd); + if (IS_ERR(th)) { + CERROR("Cannot get thandle\n"); + RETURN(-ENOMEM); + } + + rc = mdd_declare_llog_cancel(mcud->mcud_env, mdd, th); + if (rc) + GOTO(stop, rc); + + rc = mdd_trans_start(mcud->mcud_env, mdd, th); + if (rc) + GOTO(stop, rc); + rc = llog_cat_cancel_records(llh->u.phd.phd_cat_handle, 1, &cookie); if (rc == 0) mcud->mcud_usercount--; + +stop: + mdd_trans_stop(mcud->mcud_env, mdd, 0, th); RETURN(rc); } @@ -1228,7 +1472,8 @@ static int mdd_changelog_user_purge_cb(struct llog_handle *llh, 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; @@ -1236,16 +1481,18 @@ static int mdd_changelog_user_purge(struct mdd_device *mdd, int id, int rc; ENTRY; - CDEBUG(D_IOCTL, "Purge request: id=%d, endrec="LPD64"\n", id, endrec); + CDEBUG(D_IOCTL, "Purge request: id=%d, endrec=%lld\n", id, endrec); data.mcud_id = id; data.mcud_minid = 0; data.mcud_minrec = 0; data.mcud_usercount = 0; data.mcud_endrec = endrec; - spin_lock(&mdd->mdd_cl.mc_lock); + data.mcud_mdd = mdd; + data.mcud_env = env; + cfs_spin_lock(&mdd->mdd_cl.mc_lock); endrec = mdd->mdd_cl.mc_index; - spin_unlock(&mdd->mdd_cl.mc_lock); + cfs_spin_unlock(&mdd->mdd_cl.mc_lock); if ((data.mcud_endrec == 0) || ((data.mcud_endrec > endrec) && (data.mcud_endrec != MCUD_UNREGISTER))) @@ -1303,11 +1550,19 @@ 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); + } + } /* Below ioctls use obd_ioctl_data */ if (len != sizeof(*data)) { @@ -1325,18 +1580,19 @@ 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); } /* type constructor/destructor: mdd_type_init, mdd_type_fini */ -LU_TYPE_INIT_FINI(mdd, &mdd_thread_key, &mdd_ucred_key, &mdd_capainfo_key); +LU_TYPE_INIT_FINI(mdd, &mdd_thread_key, &mdd_ucred_key, &mdd_capainfo_key, + &mdd_quota_key); const struct md_device_operations mdd_ops = { .mdo_statfs = mdd_statfs,