X-Git-Url: https://git.whamcloud.com/?p=fs%2Flustre-release.git;a=blobdiff_plain;f=lustre%2Fmdd%2Fmdd_device.c;h=2728b207e70b2459cdfd7b069b807ca688a6bc8d;hp=876a51a3906269eab42009d305715c2d2fda9969;hb=be7efb4ce06996444c08914305a73833a7123eeb;hpb=2a50b6bf048517aa1f63c4e29c3637f09c148fdb diff --git a/lustre/mdd/mdd_device.c b/lustre/mdd/mdd_device.c index 876a51a..2728b20 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. @@ -28,9 +26,8 @@ /* * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved. * Use is subject to license terms. - */ -/* - * Copyright (c) 2011 Whamcloud, Inc. + * + * Copyright (c) 2011, 2014, Intel Corporation. */ /* * This file is part of Lustre, http://www.lustre.org/ @@ -43,296 +40,418 @@ * 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 -#include -#ifdef HAVE_EXT4_LDISKFS -#include -#else -#include -#endif +#include #include -#include -#include /* for changelogs */ +#include +#include #include #include +#include #include "mdd_internal.h" -const struct md_device_operations mdd_ops; +static const struct md_device_operations mdd_ops; static struct lu_device_type mdd_device_type; static const char mdd_root_dir_name[] = "ROOT"; static const char mdd_obf_dir_name[] = "fid"; +static const char mdd_lpf_dir_name[] = "lost+found"; + +/* Slab for MDD object allocation */ +struct kmem_cache *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) +static int mdd_connect_to_next(const struct lu_env *env, struct mdd_device *m, + const char *nextdev) { - struct mdd_device *mdd = lu2mdd_dev(d); - int rc; - ENTRY; + struct obd_connect_data *data = NULL; + struct lu_device *lud = mdd2lu_dev(m); + struct obd_device *obd; + int rc; + ENTRY; + + LASSERT(m->mdd_child_exp == NULL); + + OBD_ALLOC(data, sizeof(*data)); + if (data == NULL) + GOTO(out, rc = -ENOMEM); + + obd = class_name2obd(nextdev); + if (obd == NULL) { + CERROR("can't locate next device: %s\n", nextdev); + GOTO(out, rc = -ENOTCONN); + } + + data->ocd_connect_flags = OBD_CONNECT_VERSION; + data->ocd_version = LUSTRE_VERSION_CODE; + + rc = obd_connect(NULL, &m->mdd_child_exp, obd, &obd->obd_uuid, data, NULL); + if (rc) { + CERROR("cannot connect to next dev %s (%d)\n", nextdev, rc); + GOTO(out, rc); + } + + lud->ld_site = m->mdd_child_exp->exp_obd->obd_lu_dev->ld_site; + LASSERT(lud->ld_site); + m->mdd_child = lu2dt_dev(m->mdd_child_exp->exp_obd->obd_lu_dev); + m->mdd_bottom = lu2dt_dev(lud->ld_site->ls_bottom_dev); + lu_dev_add_linkage(lud->ld_site, lud); + +out: + if (data) + OBD_FREE(data, sizeof(*data)); + RETURN(rc); +} + +static int mdd_init0(const struct lu_env *env, struct mdd_device *mdd, + struct lu_device_type *t, struct lustre_cfg *lcfg) +{ + int rc = -EINVAL; + const char *dev; + ENTRY; + + /* LU-8040 Set defaults here, before values configs */ + mdd->mdd_cl.mc_flags = 0; /* off by default */ + mdd->mdd_cl.mc_mask = CHANGELOG_DEFMASK; + + dev = lustre_cfg_string(lcfg, 0); + if (dev == NULL) + RETURN(rc); - mdd->mdd_child = lu2dt_dev(next); + mdd->mdd_md_dev.md_lu_dev.ld_obd = class_name2obd(dev); + if (mdd->mdd_md_dev.md_lu_dev.ld_obd == NULL) + RETURN(rc); + mdd->mdd_md_dev.md_lu_dev.ld_ops = &mdd_lu_ops; + mdd->mdd_md_dev.md_ops = &mdd_ops; - /* Prepare transactions callbacks. */ - mdd->mdd_txn_cb.dtc_txn_start = mdd_txn_start_cb; - mdd->mdd_txn_cb.dtc_txn_stop = mdd_txn_stop_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; + rc = mdd_connect_to_next(env, mdd, lustre_cfg_string(lcfg, 3)); + if (rc != 0) + RETURN(rc); + + mdd->mdd_atime_diff = MAX_ATIME_DIFF; /* sync permission changes */ mdd->mdd_sync_permission = 1; - rc = mdd_procfs_init(mdd, name); + dt_conf_get(env, mdd->mdd_child, &mdd->mdd_dt_conf); + + /* we are using service name but not mdd obd name + * for compatibility reasons. + * It is passed from MDT in lustre_cfg[2] buffer */ + rc = mdd_procfs_init(mdd, lustre_cfg_string(lcfg, 2)); + if (rc < 0) + obd_disconnect(mdd->mdd_child_exp); + RETURN(rc); } 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; + struct mdd_device *mdd = lu2mdd_dev(d); - rc = mdd_procfs_fini(mdd); - if (rc) { - CERROR("proc fini error %d \n", rc); - return ERR_PTR(rc); - } - return next; -} + if (d->ld_site) + lu_dev_del_linkage(d->ld_site, d); -static void mdd_changelog_fini(const struct lu_env *env, - struct mdd_device *mdd); - -static void mdd_device_shutdown(const struct lu_env *env, - struct mdd_device *m, struct lustre_cfg *cfg) -{ - ENTRY; - mdd_changelog_fini(env, m); - dt_txn_callback_del(m->mdd_child, &m->mdd_txn_cb); - 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); - /* remove upcall device*/ - md_upcall_fini(&m->mdd_md_dev); - EXIT; + mdd_procfs_fini(mdd); + return NULL; } -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; - ENTRY; + struct mdd_device *mdd = (struct mdd_device *)data; + struct llog_changelog_rec *rec = (struct llog_changelog_rec *)hdr; - LASSERT(llh->lgh_hdr->llh_flags & LLOG_F_IS_PLAIN); - LASSERT(rec->cr_hdr.lrh_type == CHANGELOG_REC); + LASSERT(llh->lgh_hdr->llh_flags & LLOG_F_IS_PLAIN); + LASSERT(rec->cr_hdr.lrh_type == CHANGELOG_REC); - 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.cr_index, - rec->cr.cr_type, rec->cr.cr_namelen, rec->cr.cr_name, - llh->lgh_id.lgl_oid); + CDEBUG(D_INFO, + "seeing record at index %d/%d/"LPU64" t=%x %.*s in log" + DOSTID"\n", hdr->lrh_index, rec->cr_hdr.lrh_index, + rec->cr.cr_index, rec->cr.cr_type, rec->cr.cr_namelen, + changelog_rec_name(&rec->cr), POSTID(&llh->lgh_id.lgl_oi)); - mdd->mdd_cl.mc_index = rec->cr.cr_index; - RETURN(LLOG_PROC_BREAK); + mdd->mdd_cl.mc_index = rec->cr.cr_index; + 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 = (struct llog_changelog_user_rec *)hdr; - ENTRY; LASSERT(llh->lgh_hdr->llh_flags & LLOG_F_IS_PLAIN); LASSERT(rec->cur_hdr.lrh_type == CHANGELOG_USER_REC); CDEBUG(D_INFO, "seeing user at index %d/%d id=%d endrec="LPU64 - " in log "LPX64"\n", hdr->lrh_index, rec->cur_hdr.lrh_index, - rec->cur_id, rec->cur_endrec, llh->lgh_id.lgl_oid); + " in log "DOSTID"\n", hdr->lrh_index, rec->cur_hdr.lrh_index, + rec->cur_id, rec->cur_endrec, POSTID(&llh->lgh_id.lgl_oi)); - cfs_spin_lock(&mdd->mdd_cl.mc_user_lock); - mdd->mdd_cl.mc_lastuser = rec->cur_id; - cfs_spin_unlock(&mdd->mdd_cl.mc_user_lock); + spin_lock(&mdd->mdd_cl.mc_user_lock); + mdd->mdd_cl.mc_lastuser = rec->cur_id; + if (rec->cur_endrec > mdd->mdd_cl.mc_index) + mdd->mdd_cl.mc_index = rec->cur_endrec; + spin_unlock(&mdd->mdd_cl.mc_user_lock); - RETURN(LLOG_PROC_BREAK); + return LLOG_PROC_BREAK; } - -static int mdd_changelog_llog_init(struct mdd_device *mdd) +static int llog_changelog_cancel_cb(const struct lu_env *env, + struct llog_handle *llh, + struct llog_rec_hdr *hdr, void *data) { - struct obd_device *obd = mdd2obd_dev(mdd); - struct llog_ctxt *ctxt; - int rc; + struct llog_changelog_rec *rec = (struct llog_changelog_rec *)hdr; + struct llog_cookie cookie; + long long endrec = *(long long *)data; + int rc; - /* Find last changelog entry number */ - ctxt = llog_get_context(obd, LLOG_CHANGELOG_ORIG_CTXT); - if (ctxt == NULL) { - CERROR("no changelog context\n"); - return -EINVAL; - } - if (!ctxt->loc_handle) { - llog_ctxt_put(ctxt); - return -EINVAL; - } + ENTRY; - rc = llog_cat_reverse_process(ctxt->loc_handle, changelog_init_cb, mdd); - llog_ctxt_put(ctxt); + /* This is always a (sub)log, not the catalog */ + LASSERT(llh->lgh_hdr->llh_flags & LLOG_F_IS_PLAIN); - if (rc < 0) { - CERROR("changelog init failed: %d\n", rc); - return rc; - } - CDEBUG(D_IOCTL, "changelog starting index="LPU64"\n", - mdd->mdd_cl.mc_index); - - /* Find last changelog user id */ - ctxt = llog_get_context(obd, LLOG_CHANGELOG_USER_ORIG_CTXT); - if (ctxt == NULL) { - CERROR("no changelog user context\n"); - return -EINVAL; - } - if (!ctxt->loc_handle) { - llog_ctxt_put(ctxt); - return -EINVAL; - } + if (rec->cr.cr_index > endrec) + /* records are in order, so we're done */ + RETURN(LLOG_PROC_BREAK); - rc = llog_cat_reverse_process(ctxt->loc_handle, changelog_user_init_cb, - mdd); - llog_ctxt_put(ctxt); + cookie.lgc_lgl = llh->lgh_id; + cookie.lgc_index = hdr->lrh_index; - if (rc < 0) { - CERROR("changelog user init failed: %d\n", rc); - return rc; - } + /* cancel them one at a time. I suppose we could store up the cookies + * and cancel them all at once; probably more efficient, but this is + * done as a user call, so who cares... */ + rc = llog_cat_cancel_records(env, llh->u.phd.phd_cat_handle, 1, + &cookie); + RETURN(rc < 0 ? rc : 0); +} - /* If we have registered users, assume we want changelogs on */ - if (mdd->mdd_cl.mc_lastuser > 0) - rc = mdd_changelog_on(mdd, 1); +static int llog_changelog_cancel(const struct lu_env *env, + struct llog_ctxt *ctxt, + struct llog_cookie *cookies, int flags) +{ + struct llog_handle *cathandle = ctxt->loc_handle; + int rc; - return rc; + ENTRY; + + /* This should only be called with the catalog handle */ + LASSERT(cathandle->lgh_hdr->llh_flags & LLOG_F_IS_CAT); + + rc = llog_cat_process(env, cathandle, llog_changelog_cancel_cb, + (void *)cookies, 0, 0); + if (rc >= 0) + /* 0 or 1 means we're done */ + rc = 0; + else + CERROR("%s: cancel idx %u of catalog "DOSTID" rc=%d\n", + ctxt->loc_obd->obd_name, cathandle->lgh_last_idx, + POSTID(&cathandle->lgh_id.lgl_oi), rc); + + RETURN(rc); } -static int mdd_changelog_init(const struct lu_env *env, struct mdd_device *mdd) +static struct llog_operations changelog_orig_logops; + +static int +mdd_changelog_write_header(const struct lu_env *env, struct mdd_device *mdd, + int markerflags); + +static int +mdd_changelog_on(const struct lu_env *env, struct mdd_device *mdd) { - int rc; + int rc = 0; - mdd->mdd_cl.mc_index = 0; - 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; - cfs_spin_lock_init(&mdd->mdd_cl.mc_user_lock); - mdd->mdd_cl.mc_lastuser = 0; - - rc = mdd_changelog_llog_init(mdd); - if (rc) { - CERROR("Changelog setup during init failed %d\n", rc); - mdd->mdd_cl.mc_flags |= CLM_ERR; - } + if ((mdd->mdd_cl.mc_flags & CLM_ON) != 0) + return rc; - return rc; + LCONSOLE_INFO("%s: changelog on\n", mdd2obd_dev(mdd)->obd_name); + if (mdd->mdd_cl.mc_flags & CLM_ERR) { + CERROR("Changelogs cannot be enabled due to error " + "condition (see %s log).\n", + mdd2obd_dev(mdd)->obd_name); + rc = -ESRCH; + } else { + spin_lock(&mdd->mdd_cl.mc_lock); + mdd->mdd_cl.mc_flags |= CLM_ON; + spin_unlock(&mdd->mdd_cl.mc_lock); + rc = mdd_changelog_write_header(env, mdd, CLM_START); + } + return rc; } -static void mdd_changelog_fini(const struct lu_env *env, struct mdd_device *mdd) +static int +mdd_changelog_off(const struct lu_env *env, struct mdd_device *mdd) { - mdd->mdd_cl.mc_flags = 0; + int rc = 0; + + if ((mdd->mdd_cl.mc_flags & CLM_ON) != CLM_ON) + return rc; + + LCONSOLE_INFO("%s: changelog off\n", mdd2obd_dev(mdd)->obd_name); + rc = mdd_changelog_write_header(env, mdd, CLM_FINI); + spin_lock(&mdd->mdd_cl.mc_lock); + mdd->mdd_cl.mc_flags &= ~CLM_ON; + spin_unlock(&mdd->mdd_cl.mc_lock); + + return rc; +} + +static int mdd_changelog_llog_init(const struct lu_env *env, + struct mdd_device *mdd) +{ + struct obd_device *obd = mdd2obd_dev(mdd); + struct llog_ctxt *ctxt = NULL, *uctxt = NULL; + int rc; + + ENTRY; + + /* LU-2844 mdd setup failure should not cause umount oops */ + if (OBD_FAIL_CHECK(OBD_FAIL_MDS_CHANGELOG_INIT)) + RETURN(-EIO); + + OBD_SET_CTXT_MAGIC(&obd->obd_lvfs_ctxt); + obd->obd_lvfs_ctxt.dt = mdd->mdd_bottom; + rc = llog_setup(env, obd, &obd->obd_olg, LLOG_CHANGELOG_ORIG_CTXT, + obd, &changelog_orig_logops); + if (rc) { + CERROR("%s: changelog llog setup failed: rc = %d\n", + obd->obd_name, rc); + RETURN(rc); + } + + ctxt = llog_get_context(obd, LLOG_CHANGELOG_ORIG_CTXT); + LASSERT(ctxt); + + rc = llog_open_create(env, ctxt, &ctxt->loc_handle, NULL, + CHANGELOG_CATALOG); + if (rc) + GOTO(out_cleanup, rc); + + rc = llog_init_handle(env, ctxt->loc_handle, LLOG_F_IS_CAT, NULL); + if (rc) + GOTO(out_close, rc); + + rc = llog_cat_reverse_process(env, ctxt->loc_handle, + changelog_init_cb, mdd); + + if (rc < 0) { + CERROR("%s: changelog init failed: rc = %d\n", obd->obd_name, + rc); + GOTO(out_close, rc); + } + + CDEBUG(D_IOCTL, "changelog starting index="LPU64"\n", + mdd->mdd_cl.mc_index); + + /* setup user changelog */ + rc = llog_setup(env, obd, &obd->obd_olg, LLOG_CHANGELOG_USER_ORIG_CTXT, + obd, &changelog_orig_logops); + if (rc) { + CERROR("%s: changelog users llog setup failed: rc = %d\n", + obd->obd_name, rc); + GOTO(out_close, rc); + } + + uctxt = llog_get_context(obd, LLOG_CHANGELOG_USER_ORIG_CTXT); + LASSERT(ctxt); + + rc = llog_open_create(env, uctxt, &uctxt->loc_handle, NULL, + CHANGELOG_USERS); + if (rc) + GOTO(out_ucleanup, rc); + + uctxt->loc_handle->lgh_logops->lop_add = llog_cat_add_rec; + uctxt->loc_handle->lgh_logops->lop_declare_add = llog_cat_declare_add_rec; + + rc = llog_init_handle(env, uctxt->loc_handle, LLOG_F_IS_CAT, NULL); + if (rc) + GOTO(out_uclose, rc); + + rc = llog_cat_reverse_process(env, uctxt->loc_handle, + changelog_user_init_cb, mdd); + if (rc < 0) { + CERROR("%s: changelog user init failed: rc = %d\n", + obd->obd_name, rc); + GOTO(out_uclose, rc); + } + + /* If we have registered users, assume we want changelogs on */ + if (mdd->mdd_cl.mc_lastuser > 0) { + rc = mdd_changelog_on(env, mdd); + if (rc < 0) + GOTO(out_uclose, rc); + } + llog_ctxt_put(ctxt); + llog_ctxt_put(uctxt); + RETURN(0); +out_uclose: + llog_cat_close(env, uctxt->loc_handle); +out_ucleanup: + llog_cleanup(env, uctxt); +out_close: + llog_cat_close(env, ctxt->loc_handle); +out_cleanup: + llog_cleanup(env, ctxt); + return rc; } -/* Start / stop recording */ -int mdd_changelog_on(struct mdd_device *mdd, int on) +static int mdd_changelog_init(const struct lu_env *env, struct mdd_device *mdd) { - int rc = 0; + struct obd_device *obd = mdd2obd_dev(mdd); + int rc; - if ((on == 1) && ((mdd->mdd_cl.mc_flags & CLM_ON) == 0)) { - LCONSOLE_INFO("%s: changelog on\n", mdd2obd_dev(mdd)->obd_name); - if (mdd->mdd_cl.mc_flags & CLM_ERR) { - CERROR("Changelogs cannot be enabled due to error " - "condition (see %s log).\n", - mdd2obd_dev(mdd)->obd_name); - rc = -ESRCH; - } else { - cfs_spin_lock(&mdd->mdd_cl.mc_lock); - mdd->mdd_cl.mc_flags |= CLM_ON; - 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); - cfs_spin_lock(&mdd->mdd_cl.mc_lock); - mdd->mdd_cl.mc_flags &= ~CLM_ON; - cfs_spin_unlock(&mdd->mdd_cl.mc_lock); - } - return rc; -} + mdd->mdd_cl.mc_index = 0; + spin_lock_init(&mdd->mdd_cl.mc_lock); + mdd->mdd_cl.mc_starttime = cfs_time_current_64(); + spin_lock_init(&mdd->mdd_cl.mc_user_lock); + mdd->mdd_cl.mc_lastuser = 0; -static __u64 cl_time(void) { - cfs_fs_time_t time; + rc = mdd_changelog_llog_init(env, mdd); + if (rc) { + CERROR("%s: changelog setup during init failed: rc = %d\n", + obd->obd_name, rc); + mdd->mdd_cl.mc_flags |= CLM_ERR; + } - cfs_fs_time_current(&time); - return (((__u64)time.tv_sec) << 30) + time.tv_nsec; + return rc; } -/** Add a changelog 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_llog_write(struct mdd_device *mdd, - struct llog_changelog_rec *rec, - struct thandle *handle) +static void mdd_changelog_fini(const struct lu_env *env, + struct mdd_device *mdd) { - struct obd_device *obd = mdd2obd_dev(mdd); - struct llog_ctxt *ctxt; - int rc; + struct obd_device *obd = mdd2obd_dev(mdd); + struct llog_ctxt *ctxt; - 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(ctxt, &rec->cr_hdr, NULL, NULL, 0); - llog_ctxt_put(ctxt); + mdd->mdd_cl.mc_flags = 0; - return rc; + ctxt = llog_get_context(obd, LLOG_CHANGELOG_ORIG_CTXT); + if (ctxt) { + llog_cat_close(env, ctxt->loc_handle); + llog_cleanup(env, ctxt); + } + ctxt = llog_get_context(obd, LLOG_CHANGELOG_USER_ORIG_CTXT); + if (ctxt) { + llog_cat_close(env, ctxt->loc_handle); + llog_cleanup(env, ctxt); + } } /** Remove entries with indicies up to and including \a endrec from the @@ -341,7 +460,9 @@ int mdd_changelog_llog_write(struct mdd_device *mdd, * \param endrec * \retval 0 ok */ -int mdd_changelog_llog_cancel(struct mdd_device *mdd, long long endrec) +static 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; @@ -352,9 +473,9 @@ int mdd_changelog_llog_cancel(struct mdd_device *mdd, long long endrec) if (ctxt == NULL) return -ENXIO; - cfs_spin_lock(&mdd->mdd_cl.mc_lock); - cur = (long long)mdd->mdd_cl.mc_index; - cfs_spin_unlock(&mdd->mdd_cl.mc_lock); + spin_lock(&mdd->mdd_cl.mc_lock); + cur = (long long)mdd->mdd_cl.mc_index; + spin_unlock(&mdd->mdd_cl.mc_lock); if (endrec > cur) endrec = cur; @@ -367,7 +488,8 @@ 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) { - rc = mdd_changelog_write_header(mdd, CLM_PURGE); + /* XXX: transaction is started by llog itself */ + rc = mdd_changelog_write_header(env, mdd, CLM_PURGE); if (rc) goto out; } @@ -377,7 +499,7 @@ 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(); - rc = llog_cancel(ctxt, NULL, 1, (struct llog_cookie *)&endrec, 0); + rc = llog_cancel(env, ctxt, (struct llog_cookie *)&endrec, 0); out: llog_ctxt_put(ctxt); return rc; @@ -388,582 +510,389 @@ out: * \param markerflags - CLM_* * \retval 0 ok */ -int mdd_changelog_write_header(struct mdd_device *mdd, int markerflags) -{ - struct obd_device *obd = mdd2obd_dev(mdd); - struct llog_changelog_rec *rec; - int reclen; - int len = strlen(obd->obd_name); - int rc; - ENTRY; - - reclen = llog_data_len(sizeof(*rec) + len); - OBD_ALLOC(rec, reclen); - if (rec == NULL) - RETURN(-ENOMEM); +int mdd_changelog_write_header(const struct lu_env *env, + struct mdd_device *mdd, int markerflags) +{ + struct obd_device *obd = mdd2obd_dev(mdd); + struct llog_changelog_rec *rec; + struct lu_buf *buf; + struct llog_ctxt *ctxt; + int reclen; + int len = strlen(obd->obd_name); + int rc; + + ENTRY; + + if (mdd->mdd_cl.mc_mask & (1 << CL_MARK)) { + mdd->mdd_cl.mc_starttime = cfs_time_current_64(); + RETURN(0); + } + + reclen = llog_data_len(sizeof(*rec) + len); + buf = lu_buf_check_and_alloc(&mdd_env_info(env)->mti_big_buf, reclen); + if (buf->lb_buf == NULL) + RETURN(-ENOMEM); + rec = buf->lb_buf; 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); + memcpy(changelog_rec_name(&rec->cr), obd->obd_name, rec->cr.cr_namelen); /* Status and action flags */ - rec->cr.cr_markerflags = mdd->mdd_cl.mc_flags | markerflags; + rec->cr.cr_markerflags = mdd->mdd_cl.mc_flags | markerflags; + rec->cr_hdr.lrh_len = llog_data_len(changelog_rec_size(&rec->cr) + + rec->cr.cr_namelen); + rec->cr_hdr.lrh_type = CHANGELOG_REC; + rec->cr.cr_time = cl_time(); + spin_lock(&mdd->mdd_cl.mc_lock); + rec->cr.cr_index = ++mdd->mdd_cl.mc_index; + spin_unlock(&mdd->mdd_cl.mc_lock); - rc = (mdd->mdd_cl.mc_mask & (1 << CL_MARK)) ? - mdd_changelog_llog_write(mdd, rec, NULL) : 0; + ctxt = llog_get_context(obd, LLOG_CHANGELOG_ORIG_CTXT); + LASSERT(ctxt); - /* assume on or off event; reset repeat-access time */ - mdd->mdd_cl.mc_starttime = cfs_time_current_64(); + rc = llog_cat_add(env, ctxt->loc_handle, &rec->cr_hdr, NULL); + if (rc > 0) + rc = 0; + llog_ctxt_put(ctxt); - OBD_FREE(rec, reclen); - RETURN(rc); + /* assume on or off event; reset repeat-access time */ + mdd->mdd_cl.mc_starttime = cfs_time_current_64(); + RETURN(rc); } /** - * Create ".lustre" directory. + * Lookup method for "fid" object. Only filenames with correct SEQ:OID format + * are valid. We also check if object with passed fid exists or not. */ -static int create_dot_lustre_dir(const struct lu_env *env, struct mdd_device *m) -{ - struct lu_fid *fid = &mdd_env_info(env)->mti_fid; - struct md_object *mdo; - int rc; - - 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, 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", - dot_lustre_name, PFID(fid), rc); - RETURN(rc); - } - - if (!IS_ERR(mdo)) - lu_object_put(env, &mdo->mo_lu); - - return 0; -} - - -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_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) -{ - 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) +static int obf_lookup(const struct lu_env *env, struct md_object *p, + const struct lu_name *lname, struct lu_fid *f, + struct md_op_spec *spec) { - struct mdd_object *mdd_obj = md2mdd_obj(obj); + char *name = (char *)lname->ln_name; + struct mdd_device *mdd = mdo2mdd(p); + struct mdd_object *child; + int rc = 0; - mdd_write_lock(env, mdd_obj, MOR_TGT_CHILD); - mdd_obj->mod_count++; - mdd_write_unlock(env, mdd_obj); + while (*name == '[') + name++; - return 0; -} + sscanf(name, SFID, RFID(f)); + if (!fid_is_sane(f)) { + CWARN("%s: Trying to lookup invalid FID [%s] in %s/%s, FID " + "format should be "DFID"\n", mdd2obd_dev(mdd)->obd_name, + lname->ln_name, dot_lustre_name, mdd_obf_dir_name, + (__u64)FID_SEQ_NORMAL, 1, 0); + GOTO(out, rc = -EINVAL); + } -static int dot_lustre_mdd_close(const struct lu_env *env, struct md_object *obj, - struct md_attr *ma) -{ - struct mdd_object *mdd_obj = md2mdd_obj(obj); + if (!fid_is_norm(f) && !fid_is_igif(f) && !fid_is_root(f) && + !fid_seq_is_dot(f->f_seq)) { + CWARN("%s: Trying to lookup invalid FID "DFID" in %s/%s, " + "sequence should be >= "LPX64" or within ["LPX64"," + ""LPX64"].\n", mdd2obd_dev(mdd)->obd_name, PFID(f), + dot_lustre_name, mdd_obf_dir_name, (__u64)FID_SEQ_NORMAL, + (__u64)FID_SEQ_IGIF, (__u64)FID_SEQ_IGIF_MAX); + GOTO(out, rc = -EINVAL); + } - mdd_write_lock(env, mdd_obj, MOR_TGT_CHILD); - mdd_obj->mod_count--; - mdd_write_unlock(env, mdd_obj); + /* Check if object with this fid exists */ + child = mdd_object_find(env, mdd, f); + if (IS_ERR(child)) + GOTO(out, rc = PTR_ERR(child)); - return 0; -} + if (mdd_object_exists(child) == 0) + rc = -ENOENT; -static int dot_lustre_mdd_object_sync(const struct lu_env *env, - struct md_object *obj) -{ - return -ENOSYS; -} + mdd_object_put(env, child); -static dt_obj_version_t dot_lustre_mdd_version_get(const struct lu_env *env, - struct md_object *obj) -{ - return 0; +out: + return rc; } -static void dot_lustre_mdd_version_set(const struct lu_env *env, - struct md_object *obj, - dt_obj_version_t version) +static int mdd_dummy_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; + return -EPERM; } -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 mdd_dummy_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 -ENOSYS; + return -EPERM; } -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) +static int mdd_dummy_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 -ENOSYS; + return -EPERM; } -static int dot_file_unlock(const struct lu_env *env, struct md_object *obj, - struct lov_mds_md *lmm, struct lustre_handle *lockh) +static int mdd_dummy_unlink(const struct lu_env *env, + struct md_object *pobj, + struct md_object *cobj, + const struct lu_name *lname, + struct md_attr *ma, + int no_name) { - return -ENOSYS; + return -EPERM; } -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_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_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, +static struct md_dir_operations mdd_obf_dir_ops = { + .mdo_lookup = obf_lookup, + .mdo_create = mdd_dummy_create, + .mdo_rename = mdd_dummy_rename, + .mdo_link = mdd_dummy_link, + .mdo_unlink = mdd_dummy_unlink }; +static struct md_dir_operations mdd_lpf_dir_ops = { + .mdo_lookup = mdd_lookup, + .mdo_create = mdd_dummy_create, + .mdo_rename = mdd_dummy_rename, + .mdo_link = mdd_dummy_link, + .mdo_unlink = mdd_dummy_unlink +}; -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; - else - return -ENOENT; - - return 0; -} - -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) +static struct md_object *mdo_locate(const struct lu_env *env, + struct md_device *md, + const struct lu_fid *fid) { - return -EPERM; -} + struct lu_object *obj; + struct md_object *mdo; -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; + obj = lu_object_find(env, &md->md_lu_dev, fid, NULL); + if (!IS_ERR(obj)) { + obj = lu_object_locate(obj->lo_header, md->md_lu_dev.ld_type); + LASSERT(obj != NULL); + mdo = lu2md(obj); + } else { + mdo = ERR_PTR(PTR_ERR(obj)); + } + return mdo; } -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) +static int mdd_lpf_setup(const struct lu_env *env, struct mdd_device *m) { - return -EPERM; -} + struct md_object *mdo; + struct mdd_object *mdd_lpf; + struct lu_fid fid = LU_LPF_FID; + int rc; + ENTRY; -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; -} + rc = mdd_local_file_create(env, m, mdd_object_fid(m->mdd_dot_lustre), + mdd_lpf_dir_name, S_IFDIR | S_IRUSR | S_IXUSR, + &fid); + if (rc != 0) + RETURN(rc); -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; -} + mdo = mdo_locate(env, &m->mdd_md_dev, &fid); + if (IS_ERR(mdo)) + RETURN(PTR_ERR(mdo)); -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; -} + LASSERT(lu_object_exists(&mdo->mo_lu)); -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; -} + mdd_lpf = md2mdd_obj(mdo); + mdd_lpf->mod_obj.mo_dir_ops = &mdd_lpf_dir_ops; + m->mdd_dot_lustre_objs.mdd_lpf = mdd_lpf; -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; + RETURN(0); } - -static struct md_dir_operations mdd_dot_lustre_dir_ops = { - .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, - struct md_attr *ma) +/** + * Create special in-memory "fid" object for open-by-fid. + */ +static int mdd_obf_setup(const struct lu_env *env, struct mdd_device *m) { - int rc = 0; + struct md_object *mdo; + struct mdd_object *mdd_obf; + struct lu_fid fid = LU_OBF_FID; + int rc; - if (ma->ma_need & MA_INODE) { - struct mdd_device *mdd = mdo2mdd(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); - if (rc) - return rc; - ma->ma_valid |= MA_INODE; - } + rc = mdd_local_file_create(env, m, mdd_object_fid(m->mdd_dot_lustre), + mdd_obf_dir_name, S_IFDIR | S_IXUSR, &fid); + if (rc < 0) + RETURN(rc); - /* "fid" directory does not have any striping information. */ - if (ma->ma_need & MA_LOV) { - struct mdd_object *mdd_obj = md2mdd_obj(obj); - - if (ma->ma_valid & MA_LOV) - return 0; - - if (!(S_ISREG(mdd_object_type(mdd_obj)) || - S_ISDIR(mdd_object_type(mdd_obj)))) - return 0; - - if (ma->ma_need & MA_LOV_DEF) { - 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; - } - } - } + mdo = mdo_locate(env, &m->mdd_md_dev, &fid); + if (IS_ERR(mdo)) + RETURN(PTR_ERR(mdo)); - return rc; -} + LASSERT(lu_object_exists(&mdo->mo_lu)); -static int obf_attr_set(const struct lu_env *env, struct md_object *obj, - const struct md_attr *ma) -{ - return -EPERM; -} + mdd_obf = md2mdd_obj(mdo); + mdd_obf->mod_obj.mo_dir_ops = &mdd_obf_dir_ops; + m->mdd_dot_lustre_objs.mdd_obf = mdd_obf; -static int obf_xattr_get(const struct lu_env *env, - struct md_object *obj, struct lu_buf *buf, - const char *name) -{ - return 0; + return 0; } -static int obf_mdd_open(const struct lu_env *env, struct md_object *obj, - int flags) +static void mdd_dot_lustre_cleanup(const struct lu_env *env, + struct mdd_device *m) { - struct mdd_object *mdd_obj = md2mdd_obj(obj); - - mdd_write_lock(env, mdd_obj, MOR_TGT_CHILD); - mdd_obj->mod_count++; - mdd_write_unlock(env, mdd_obj); - - return 0; + if (m->mdd_dot_lustre_objs.mdd_lpf != NULL) { + mdd_object_put(env, m->mdd_dot_lustre_objs.mdd_lpf); + m->mdd_dot_lustre_objs.mdd_lpf = NULL; + } + if (m->mdd_dot_lustre_objs.mdd_obf != NULL) { + mdd_object_put(env, m->mdd_dot_lustre_objs.mdd_obf); + m->mdd_dot_lustre_objs.mdd_obf = NULL; + } + if (m->mdd_dot_lustre != NULL) { + mdd_object_put(env, m->mdd_dot_lustre); + m->mdd_dot_lustre = NULL; + } } -static int obf_mdd_close(const struct lu_env *env, struct md_object *obj, - struct md_attr *ma) +/** Setup ".lustre" directory object */ +static int mdd_dot_lustre_setup(const struct lu_env *env, struct mdd_device *m) { - struct mdd_object *mdd_obj = md2mdd_obj(obj); + struct md_object *mdo; + struct lu_fid fid; + int rc; + + ENTRY; + /* Create ".lustre" directory in ROOT. */ + fid = LU_DOT_LUSTRE_FID; + rc = mdd_local_file_create(env, m, &m->mdd_root_fid, + dot_lustre_name, + S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO, + &fid); + if (rc < 0) + RETURN(rc); + mdo = mdo_locate(env, &m->mdd_md_dev, &fid); + if (IS_ERR(mdo)) + RETURN(PTR_ERR(mdo)); + LASSERT(lu_object_exists(&mdo->mo_lu)); + + m->mdd_dot_lustre = md2mdd_obj(mdo); + + rc = mdd_obf_setup(env, m); + if (rc) { + CERROR("%s: error initializing \"fid\" object: rc = %d.\n", + mdd2obd_dev(m)->obd_name, rc); + GOTO(out, rc); + } + + rc = mdd_lpf_setup(env, m); + if (rc != 0) { + CERROR("%s: error initializing \"lost+found\": rc = %d.\n", + mdd2obd_dev(m)->obd_name, rc); + GOTO(out, rc); + } + + RETURN(0); - mdd_write_lock(env, mdd_obj, MOR_TGT_CHILD); - mdd_obj->mod_count--; - mdd_write_unlock(env, mdd_obj); - - return 0; -} +out: + mdd_dot_lustre_cleanup(env, m); -/** Nothing to list in "fid" directory */ -static int obf_mdd_readpage(const struct lu_env *env, struct md_object *obj, - const struct lu_rdpg *rdpg) -{ - return -EPERM; + return rc; } -static int obf_path(const struct lu_env *env, struct md_object *obj, - char *path, int pathlen, __u64 *recno, int *linkno) -{ - return -ENOSYS; -} -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 -}; +static struct llog_operations hsm_actions_logops; /** - * Lookup method for "fid" object. Only filenames with correct SEQ:OID format - * are valid. We also check if object with passed fid exists or not. + * set llog methods and create LLOG_AGENT_ORIG_CTXT llog + * object in obd_device */ -static int obf_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 mdd_hsm_actions_llog_init(const struct lu_env *env, + struct mdd_device *m) { - char *name = (char *)lname->ln_name; - struct mdd_device *mdd = mdo2mdd(p); - struct mdd_object *child; - int rc = 0; + struct obd_device *obd = mdd2obd_dev(m); + struct llog_ctxt *ctxt = NULL; + int rc; + ENTRY; - while (*name == '[') - name++; + OBD_SET_CTXT_MAGIC(&obd->obd_lvfs_ctxt); + obd->obd_lvfs_ctxt.dt = m->mdd_bottom; - 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); - GOTO(out, rc = -EINVAL); - } + rc = llog_setup(env, obd, &obd->obd_olg, LLOG_AGENT_ORIG_CTXT, + obd, &hsm_actions_logops); + if (rc) { + CERROR("%s: hsm actions llog setup failed: rc = %d\n", + obd->obd_name, rc); + RETURN(rc); + } - /* Check if object with this fid exists */ - child = mdd_object_find(env, mdd, f); - if (child == NULL) - GOTO(out, rc = 0); - if (IS_ERR(child)) - GOTO(out, rc = PTR_ERR(child)); + ctxt = llog_get_context(obd, LLOG_AGENT_ORIG_CTXT); + LASSERT(ctxt); - if (mdd_object_exists(child) == 0) - rc = -ENOENT; + rc = llog_open_create(env, ctxt, &ctxt->loc_handle, NULL, + HSM_ACTIONS); + if (rc) { + CERROR("%s: hsm actions llog open_create failed: rc = %d\n", + obd->obd_name, rc); + GOTO(out_cleanup, rc); + } - mdd_object_put(env, child); + rc = llog_init_handle(env, ctxt->loc_handle, LLOG_F_IS_CAT, NULL); + if (rc) + GOTO(out_close, rc); -out: - return rc; -} - -static int obf_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; -} + llog_ctxt_put(ctxt); + RETURN(0); -static int obf_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 obf_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; -} +out_close: + llog_cat_close(env, ctxt->loc_handle); + ctxt->loc_handle = NULL; +out_cleanup: + llog_cleanup(env, ctxt); -static int obf_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; + return rc; } -static struct md_dir_operations mdd_obf_dir_ops = { - .mdo_lookup = obf_lookup, - .mdo_create = obf_create, - .mdo_rename = obf_rename, - .mdo_link = obf_link, - .mdo_unlink = obf_unlink -}; - /** - * Create special in-memory "fid" object for open-by-fid. + * cleanup the context created by llog_setup_named() */ -static int mdd_obf_setup(const struct lu_env *env, struct mdd_device *m) +static int mdd_hsm_actions_llog_fini(const struct lu_env *env, + struct mdd_device *m) { - struct mdd_object *mdd_obf; - struct lu_object *obf_lu_obj; - int rc = 0; - - m->mdd_dot_lustre_objs.mdd_obf = mdd_object_find(env, m, - &LU_OBF_FID); - if (m->mdd_dot_lustre_objs.mdd_obf == NULL || - IS_ERR(m->mdd_dot_lustre_objs.mdd_obf)) - GOTO(out, rc = -ENOENT); + struct obd_device *obd = mdd2obd_dev(m); + struct llog_ctxt *lctxt; + ENTRY; - mdd_obf = m->mdd_dot_lustre_objs.mdd_obf; - mdd_obf->mod_obj.mo_dir_ops = &mdd_obf_dir_ops; - mdd_obf->mod_obj.mo_ops = &mdd_obf_obj_ops; - /* Don't allow objects to be created in "fid" dir */ - mdd_obf->mod_flags |= IMMUTE_OBJ; + lctxt = llog_get_context(obd, LLOG_AGENT_ORIG_CTXT); + if (lctxt) { + llog_cat_close(env, lctxt->loc_handle); + lctxt->loc_handle = NULL; + llog_cleanup(env, lctxt); + } - obf_lu_obj = mdd2lu_obj(mdd_obf); - obf_lu_obj->lo_header->loh_attr |= (LOHA_EXISTS | S_IFDIR); - -out: - return rc; + RETURN(0); } -/** Setup ".lustre" directory object */ -static int mdd_dot_lustre_setup(const struct lu_env *env, struct mdd_device *m) +static void mdd_device_shutdown(const struct lu_env *env, struct mdd_device *m, + struct lustre_cfg *cfg) { - struct dt_object *dt_dot_lustre; - struct lu_fid *fid = &mdd_env_info(env)->mti_fid; - int rc; + mdd_generic_thread_stop(&m->mdd_orph_cleanup_thread); + lfsck_degister(env, m->mdd_bottom); + mdd_hsm_actions_llog_fini(env, m); + mdd_changelog_fini(env, m); + orph_index_fini(env, m); + mdd_dot_lustre_cleanup(env, m); + nodemap_fs_fini(env, mdd2obd_dev(m)); + if (m->mdd_los != NULL) { + local_oid_storage_fini(env, m->mdd_los); + m->mdd_los = NULL; + } + lu_site_purge(env, mdd2lu_dev(m)->ld_site, ~0); - rc = create_dot_lustre_dir(env, m); - if (rc) - return rc; - - dt_dot_lustre = dt_store_open(env, m->mdd_child, mdd_root_dir_name, - dot_lustre_name, fid); - if (IS_ERR(dt_dot_lustre)) { - rc = PTR_ERR(dt_dot_lustre); - GOTO(out, rc); - } - - /* references are released in mdd_device_shutdown() */ - m->mdd_dot_lustre = lu2mdd_obj(lu_object_locate(dt_dot_lustre->do_lu.lo_header, - &mdd_device_type)); - - m->mdd_dot_lustre->mod_obj.mo_dir_ops = &mdd_dot_lustre_dir_ops; - m->mdd_dot_lustre->mod_obj.mo_ops = &mdd_dot_lustre_obj_ops; - - rc = mdd_obf_setup(env, m); - if (rc) - CERROR("Error initializing \"fid\" object - %d.\n", rc); - -out: - RETURN(rc); + if (m->mdd_child_exp) + obd_disconnect(m->mdd_child_exp); } static int mdd_process_config(const struct lu_env *env, @@ -976,35 +905,26 @@ static int mdd_process_config(const struct lu_env *env, ENTRY; switch (cfg->lcfg_command) { - case LCFG_PARAM: { - struct lprocfs_static_vars lvars; - - lprocfs_mdd_init_vars(&lvars); - rc = class_process_proc_param(PARAM_MDD, lvars.obd_vars, cfg,m); - if (rc > 0 || rc == -ENOSYS) - /* we don't understand; pass it on */ - rc = next->ld_ops->ldo_process_config(env, next, cfg); - break; - } + case LCFG_PARAM: { + struct obd_device *obd = mdd2obd_dev(m); + + rc = class_process_proc_param(PARAM_MDD, obd->obd_vars, cfg, m); + if (rc > 0 || rc == -ENOSYS) + /* we don't understand; pass it on */ + rc = next->ld_ops->ldo_process_config(env, next, cfg); + break; + } case LCFG_SETUP: rc = next->ld_ops->ldo_process_config(env, next, cfg); if (rc) GOTO(out, rc); - dt->dd_ops->dt_conf_get(env, dt, &m->mdd_dt_conf); - - rc = mdd_init_obd(env, m, cfg); - if (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); + dt_conf_get(env, dt, &m->mdd_dt_conf); break; case LCFG_CLEANUP: - mdd_device_shutdown(env, m, cfg); + rc = next->ld_ops->ldo_process_config(env, next, cfg); + lu_dev_del_linkage(d->ld_site, d); + mdd_device_shutdown(env, m, cfg); + break; default: rc = next->ld_ops->ldo_process_config(env, next, cfg); break; @@ -1013,110 +933,163 @@ out: RETURN(rc); } -#if 0 -static int mdd_lov_set_nextid(const struct lu_env *env, - struct mdd_device *mdd) -{ - struct mds_obd *mds = &mdd->mdd_obd_dev->u.mds; - int rc; - ENTRY; - - LASSERT(mds->mds_lov_objids != NULL); - 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); - - RETURN(rc); -} - -static int mdd_cleanup_unlink_llog(const struct lu_env *env, - struct mdd_device *mdd) -{ - /* XXX: to be implemented! */ - return 0; -} -#endif - static int mdd_recovery_complete(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; - struct obd_device *obd = mdd2obd_dev(mdd); + struct lu_device *next; int rc; ENTRY; LASSERT(mdd != NULL); - LASSERT(obd != NULL); -#if 0 - /* XXX: Do we need this in new stack? */ - rc = mdd_lov_set_nextid(env, mdd); - if (rc) { - CERROR("mdd_lov_set_nextid() failed %d\n", - rc); - RETURN(rc); - } - - /* XXX: cleanup unlink. */ - rc = mdd_cleanup_unlink_llog(env, mdd); - if (rc) { - CERROR("mdd_cleanup_unlink_llog() failed %d\n", - rc); - RETURN(rc); - } -#endif - /* Call that with obd_recovering = 1 just to update objids */ - 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); + next = &mdd->mdd_child->dd_lu_dev; /* XXX: orphans handling. */ - __mdd_orphan_cleanup(env, mdd); + mdd_orphan_cleanup(env, mdd); rc = next->ld_ops->ldo_recovery_complete(env, next); RETURN(rc); } -static int mdd_prepare(const struct lu_env *env, - struct lu_device *pdev, - struct lu_device *cdev) +int mdd_local_file_create(const struct lu_env *env, struct mdd_device *mdd, + const struct lu_fid *pfid, const char *name, + __u32 mode, struct lu_fid *fid) { - struct mdd_device *mdd = lu2mdd_dev(cdev); - struct lu_device *next = &mdd->mdd_child->dd_lu_dev; - struct dt_object *root; - int rc; + struct dt_object *parent, *dto; + int rc; - ENTRY; - rc = next->ld_ops->ldo_prepare(env, cdev, next); - if (rc) - GOTO(out, rc); - - dt_txn_callback_add(mdd->mdd_child, &mdd->mdd_txn_cb); - root = dt_store_open(env, mdd->mdd_child, "", mdd_root_dir_name, - &mdd->mdd_root_fid); - if (!IS_ERR(root)) { - LASSERT(root != NULL); - lu_object_put(env, &root->do_lu); - rc = orph_index_init(env, mdd); - } else { - rc = PTR_ERR(root); - } - if (rc) - GOTO(out, rc); + ENTRY; - rc = mdd_dot_lustre_setup(env, mdd); - if (rc) { - CERROR("Error(%d) initializing .lustre objects\n", rc); - GOTO(out, rc); - } + LASSERT(!fid_is_zero(pfid)); + parent = dt_locate(env, mdd->mdd_bottom, pfid); + if (unlikely(IS_ERR(parent))) + RETURN(PTR_ERR(parent)); -out: - RETURN(rc); + /* create local file/dir, if @fid is passed then try to use it */ + if (fid_is_zero(fid)) + dto = local_file_find_or_create(env, mdd->mdd_los, parent, + name, mode); + else + dto = local_file_find_or_create_with_fid(env, mdd->mdd_bottom, + fid, parent, name, + mode); + if (IS_ERR(dto)) + GOTO(out_put, rc = PTR_ERR(dto)); + *fid = *lu_object_fid(&dto->do_lu); + /* since stack is not fully set up the local_storage uses own stack + * and we should drop its object from cache */ + lu_object_put_nocache(env, &dto->do_lu); + EXIT; +out_put: + lu_object_put(env, &parent->do_lu); + return 0; +} + +static int mdd_lfsck_out_notify(const struct lu_env *env, void *data, + enum lfsck_events event) +{ + return 0; +} + +static int mdd_prepare(const struct lu_env *env, + struct lu_device *pdev, + struct lu_device *cdev) +{ + struct mdd_device *mdd = lu2mdd_dev(cdev); + struct lu_device *next = &mdd->mdd_child->dd_lu_dev; + struct lu_fid fid; + int rc; + + ENTRY; + + rc = next->ld_ops->ldo_prepare(env, cdev, next); + if (rc) + RETURN(rc); + + /* Setup local dirs */ + fid.f_seq = FID_SEQ_LOCAL_NAME; + fid.f_oid = 1; + fid.f_ver = 0; + rc = local_oid_storage_init(env, mdd->mdd_bottom, &fid, + &mdd->mdd_los); + if (rc) + RETURN(rc); + + rc = dt_root_get(env, mdd->mdd_child, &mdd->mdd_local_root_fid); + if (rc < 0) + GOTO(out_los, rc); + + lu_root_fid(&fid); + if (mdd_seq_site(mdd)->ss_node_id == 0) { + rc = mdd_local_file_create(env, mdd, &mdd->mdd_local_root_fid, + mdd_root_dir_name, S_IFDIR | + S_IRUGO | S_IWUSR | S_IXUGO, &fid); + if (rc != 0) { + CERROR("%s: create root fid failed: rc = %d\n", + mdd2obd_dev(mdd)->obd_name, rc); + GOTO(out_los, rc); + } + + mdd->mdd_root_fid = fid; + rc = mdd_dot_lustre_setup(env, mdd); + if (rc != 0) { + CERROR("%s: initializing .lustre failed: rc = %d\n", + mdd2obd_dev(mdd)->obd_name, rc); + GOTO(out_los, rc); + } + } else { + /* Normal client usually send root access to MDT0 directly, + * the root FID on non-MDT0 will only be used by echo client. */ + mdd->mdd_root_fid = fid; + } + + rc = orph_index_init(env, mdd); + if (rc < 0) + GOTO(out_dot, rc); + + rc = mdd_changelog_init(env, mdd); + if (rc != 0) { + CERROR("%s: failed to initialize changelog: rc = %d\n", + mdd2obd_dev(mdd)->obd_name, rc); + GOTO(out_orph, rc); + } + + rc = mdd_hsm_actions_llog_init(env, mdd); + if (rc != 0) + GOTO(out_changelog, rc); + + rc = nodemap_fs_init(env, mdd->mdd_bottom, mdd2obd_dev(mdd), + mdd->mdd_los); + if (rc != 0) + GOTO(out_hsm, rc); + + rc = lfsck_register(env, mdd->mdd_bottom, mdd->mdd_child, + mdd2obd_dev(mdd), mdd_lfsck_out_notify, + mdd, true); + if (rc != 0) { + CERROR("%s: failed to initialize lfsck: rc = %d\n", + mdd2obd_dev(mdd)->obd_name, rc); + GOTO(out_nodemap, rc); + } + + RETURN(0); + +out_nodemap: + nodemap_fs_fini(env, mdd2obd_dev(mdd)); +out_hsm: + mdd_hsm_actions_llog_fini(env, mdd); +out_changelog: + mdd_changelog_fini(env, mdd); +out_orph: + orph_index_fini(env, mdd); +out_dot: + if (mdd_seq_site(mdd)->ss_node_id == 0) + mdd_dot_lustre_cleanup(env, mdd); +out_los: + local_oid_storage_fini(env, mdd->mdd_los); + mdd->mdd_los = NULL; + + return rc; } const struct lu_device_operations mdd_lu_ops = { @@ -1126,82 +1099,43 @@ const struct lu_device_operations mdd_lu_ops = { .ldo_prepare = mdd_prepare, }; -/* - * No permission check is needed. - */ static int mdd_root_get(const struct lu_env *env, - struct md_device *m, struct lu_fid *f) + struct md_device *m, struct lu_fid *f) { - struct mdd_device *mdd = lu2mdd_dev(&m->md_lu_dev); + struct mdd_device *mdd = lu2mdd_dev(&m->md_lu_dev); - ENTRY; - *f = mdd->mdd_root_fid; - RETURN(0); + ENTRY; + *f = mdd->mdd_root_fid; + RETURN(0); } /* * 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; - - ENTRY; - - rc = mdd_child_ops(mdd)->dt_statfs(env, mdd->mdd_child, sfs); + struct mdd_device *mdd = lu2mdd_dev(&m->md_lu_dev); + int rc; - RETURN(rc); -} + ENTRY; -/* - * No permission check is needed. - */ -static int mdd_maxsize_get(const struct lu_env *env, struct md_device *m, - int *md_size, int *cookie_size) -{ - struct mdd_device *mdd = lu2mdd_dev(&m->md_lu_dev); - ENTRY; + rc = mdd_child_ops(mdd)->dt_statfs(env, mdd->mdd_child, sfs); - *md_size = mdd_lov_mdsize(env, mdd); - *cookie_size = mdd_lov_cookiesize(env, mdd); + sfs->os_namelen = min_t(__u32, sfs->os_namelen, NAME_MAX); - RETURN(0); + RETURN(rc); } -static int mdd_init_capa_ctxt(const struct lu_env *env, struct md_device *m, - int mode, unsigned long timeout, __u32 alg, - struct lustre_capa_key *keys) +static int mdd_maxeasize_get(const struct lu_env *env, struct md_device *m, + int *easize) { - struct mdd_device *mdd = lu2mdd_dev(&m->md_lu_dev); - struct mds_obd *mds = &mdd2obd_dev(mdd)->u.mds; - int rc; - ENTRY; + struct mdd_device *mdd = lu2mdd_dev(&m->md_lu_dev); + 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); + *easize = mdd->mdd_dt_conf.ddp_max_ea_size; - rc = mdd_child_ops(mdd)->dt_init_capa_ctxt(env, mdd->mdd_child, mode, - timeout, alg, keys); - RETURN(rc); -} - -static int mdd_update_capa_key(const struct lu_env *env, - struct md_device *m, - struct lustre_capa_key *key) -{ - 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_lov_exp; - int rc; - ENTRY; - - rc = obd_set_info_async(lov_exp, sizeof(KEY_CAPA_KEY), KEY_CAPA_KEY, - sizeof(info), &info, NULL); - RETURN(rc); + RETURN(0); } static int mdd_llog_ctxt_get(const struct lu_env *env, struct md_device *m, @@ -1213,6 +1147,18 @@ static int mdd_llog_ctxt_get(const struct lu_env *env, struct md_device *m, return (*h == NULL ? -ENOENT : 0); } +static struct lu_device *mdd_device_free(const struct lu_env *env, + struct lu_device *lu) +{ + struct mdd_device *m = lu2mdd_dev(lu); + ENTRY; + + LASSERT(atomic_read(&lu->ld_ref) == 0); + md_device_fini(&m->mdd_md_dev); + OBD_FREE_PTR(m); + RETURN(NULL); +} + static struct lu_device *mdd_device_alloc(const struct lu_env *env, struct lu_device_type *t, struct lustre_cfg *lcfg) @@ -1224,97 +1170,129 @@ static struct lu_device *mdd_device_alloc(const struct lu_env *env, if (m == NULL) { l = ERR_PTR(-ENOMEM); } else { - md_device_init(&m->mdd_md_dev, t); + int rc; + l = mdd2lu_dev(m); - l->ld_ops = &mdd_lu_ops; - m->mdd_md_dev.md_ops = &mdd_ops; - md_upcall_init(&m->mdd_md_dev, NULL); + md_device_init(&m->mdd_md_dev, t); + rc = mdd_init0(env, m, t, lcfg); + if (rc != 0) { + mdd_device_free(env, l); + l = ERR_PTR(rc); + } } return l; } -static struct lu_device *mdd_device_free(const struct lu_env *env, - struct lu_device *lu) +/* + * we use exports to track all mdd users + */ +static int mdd_obd_connect(const struct lu_env *env, struct obd_export **exp, + struct obd_device *obd, struct obd_uuid *cluuid, + struct obd_connect_data *data, void *localdata) { - struct mdd_device *m = lu2mdd_dev(lu); - struct lu_device *next = &m->mdd_child->dd_lu_dev; - ENTRY; + struct mdd_device *mdd = lu2mdd_dev(obd->obd_lu_dev); + struct lustre_handle conn; + int rc; + ENTRY; - LASSERT(cfs_atomic_read(&lu->ld_ref) == 0); - md_device_fini(&m->mdd_md_dev); - OBD_FREE_PTR(m); - RETURN(next); -} + CDEBUG(D_CONFIG, "connect #%d\n", mdd->mdd_connects); -static struct obd_ops mdd_obd_device_ops = { - .o_owner = THIS_MODULE -}; + rc = class_connect(&conn, obd, cluuid); + if (rc) + RETURN(rc); -/* context key constructor/destructor: mdd_ucred_key_init, mdd_ucred_key_fini */ -LU_KEY_INIT_FINI(mdd_ucred, struct md_ucred); + *exp = class_conn2export(&conn); -static struct lu_context_key mdd_ucred_key = { - .lct_tags = LCT_SESSION, - .lct_init = mdd_ucred_key_init, - .lct_fini = mdd_ucred_key_fini -}; + /* Why should there ever be more than 1 connect? */ + LASSERT(mdd->mdd_connects == 0); + mdd->mdd_connects++; -struct md_ucred *md_ucred(const struct lu_env *env) -{ - LASSERT(env->le_ses != NULL); - return lu_context_key_get(env->le_ses, &mdd_ucred_key); + RETURN(0); } -EXPORT_SYMBOL(md_ucred); /* - * context key constructor/destructor: - * mdd_capainfo_key_init, mdd_capainfo_key_fini + * once last export (we don't count self-export) disappeared + * mdd can be released */ -LU_KEY_INIT_FINI(mdd_capainfo, struct md_capainfo); +static int mdd_obd_disconnect(struct obd_export *exp) +{ + struct obd_device *obd = exp->exp_obd; + struct mdd_device *mdd = lu2mdd_dev(obd->obd_lu_dev); + int rc, release = 0; + ENTRY; -struct lu_context_key mdd_capainfo_key = { - .lct_tags = LCT_SESSION, - .lct_init = mdd_capainfo_key_init, - .lct_fini = mdd_capainfo_key_fini -}; + mdd->mdd_connects--; + if (mdd->mdd_connects == 0) + release = 1; -struct md_capainfo *md_capainfo(const struct lu_env *env) -{ - /* NB, in mdt_init0 */ - if (env->le_ses == NULL) - return NULL; - return lu_context_key_get(env->le_ses, &mdd_capainfo_key); + rc = class_disconnect(exp); + + if (rc == 0 && release) + class_manual_cleanup(obd); + RETURN(rc); } -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); +static int mdd_obd_get_info(const struct lu_env *env, struct obd_export *exp, + __u32 keylen, void *key, __u32 *vallen, void *val) +{ + int rc = -EINVAL; -struct lu_context_key mdd_quota_key = { - .lct_tags = LCT_SESSION, - .lct_init = mdd_quota_key_init, - .lct_fini = mdd_quota_key_fini -}; + if (KEY_IS(KEY_OSP_CONNECTED)) { + struct obd_device *obd = exp->exp_obd; + struct mdd_device *mdd; + + if (!obd->obd_set_up || obd->obd_stopping) + RETURN(-EAGAIN); + + mdd = lu2mdd_dev(obd->obd_lu_dev); + LASSERT(mdd); + rc = obd_get_info(env, mdd->mdd_child_exp, keylen, key, vallen, + val); + RETURN(rc); + } -struct md_quota *md_quota(const struct lu_env *env) + RETURN(rc); +} + +static int mdd_obd_set_info_async(const struct lu_env *env, + struct obd_export *exp, + __u32 keylen, void *key, + __u32 vallen, void *val, + struct ptlrpc_request_set *set) { - LASSERT(env->le_ses != NULL); - return lu_context_key_get(env->le_ses, &mdd_quota_key); + struct obd_device *obd = exp->exp_obd; + struct mdd_device *mdd; + int rc; + + if (!obd->obd_set_up || obd->obd_stopping) + RETURN(-EAGAIN); + + mdd = lu2mdd_dev(obd->obd_lu_dev); + LASSERT(mdd); + rc = obd_set_info_async(env, mdd->mdd_child_exp, keylen, key, + vallen, val, set); + RETURN(rc); } -EXPORT_SYMBOL(md_quota); -static int mdd_changelog_user_register(struct mdd_device *mdd, int *id) +static struct obd_ops mdd_obd_device_ops = { + .o_owner = THIS_MODULE, + .o_connect = mdd_obd_connect, + .o_disconnect = mdd_obd_disconnect, + .o_get_info = mdd_obd_get_info, + .o_set_info_async = mdd_obd_set_info_async, +}; + +static int mdd_changelog_user_register(const struct lu_env *env, + struct mdd_device *mdd, int *id) { struct llog_ctxt *ctxt; struct llog_changelog_user_rec *rec; int rc; ENTRY; - ctxt = llog_get_context(mdd2obd_dev(mdd),LLOG_CHANGELOG_USER_ORIG_CTXT); + ctxt = llog_get_context(mdd2obd_dev(mdd), + LLOG_CHANGELOG_USER_ORIG_CTXT); if (ctxt == NULL) RETURN(-ENXIO); @@ -1324,24 +1302,24 @@ static int mdd_changelog_user_register(struct mdd_device *mdd, int *id) RETURN(-ENOMEM); } - /* Assume we want it on since somebody registered */ - rc = mdd_changelog_on(mdd, 1); - if (rc) - GOTO(out, rc); + /* Assume we want it on since somebody registered */ + rc = mdd_changelog_on(env, mdd); + if (rc) + GOTO(out, rc); rec->cur_hdr.lrh_len = sizeof(*rec); rec->cur_hdr.lrh_type = CHANGELOG_USER_REC; - cfs_spin_lock(&mdd->mdd_cl.mc_user_lock); - if (mdd->mdd_cl.mc_lastuser == (unsigned int)(-1)) { - 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; - cfs_spin_unlock(&mdd->mdd_cl.mc_user_lock); - - rc = llog_add(ctxt, &rec->cur_hdr, NULL, NULL, 0); + 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); + 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); + + rc = llog_cat_add(env, ctxt->loc_handle, &rec->cur_hdr, NULL); CDEBUG(D_IOCTL, "Registered changelog user %d\n", *id); out: @@ -1356,7 +1334,7 @@ struct mdd_changelog_user_data { __u32 mcud_id; __u32 mcud_minid; /**< user id with lowest rec reference */ __u32 mcud_usercount; - int mcud_found:1; + unsigned int mcud_found:1; }; #define MCUD_UNREGISTER -1LL @@ -1364,13 +1342,14 @@ 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); @@ -1399,12 +1378,15 @@ 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; + 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); + + rc = llog_cat_cancel_records(env, llh->u.phd.phd_cat_handle, + 1, &cookie); if (rc == 0) mcud->mcud_usercount--; + RETURN(rc); } @@ -1412,15 +1394,13 @@ static int mdd_changelog_user_purge_cb(struct llog_handle *llh, CDEBUG(D_IOCTL, "Rewriting changelog user %d endrec to "LPU64"\n", mcud->mcud_id, rec->cur_endrec); - /* 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(env, llh, hdr, 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; @@ -1435,26 +1415,29 @@ static int mdd_changelog_user_purge(struct mdd_device *mdd, int id, data.mcud_minrec = 0; data.mcud_usercount = 0; data.mcud_endrec = endrec; - cfs_spin_lock(&mdd->mdd_cl.mc_lock); - endrec = mdd->mdd_cl.mc_index; - cfs_spin_unlock(&mdd->mdd_cl.mc_lock); + spin_lock(&mdd->mdd_cl.mc_lock); + endrec = mdd->mdd_cl.mc_index; + spin_unlock(&mdd->mdd_cl.mc_lock); if ((data.mcud_endrec == 0) || ((data.mcud_endrec > endrec) && (data.mcud_endrec != MCUD_UNREGISTER))) data.mcud_endrec = endrec; - ctxt = llog_get_context(mdd2obd_dev(mdd),LLOG_CHANGELOG_USER_ORIG_CTXT); + ctxt = llog_get_context(mdd2obd_dev(mdd), + LLOG_CHANGELOG_USER_ORIG_CTXT); if (ctxt == NULL) 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); @@ -1469,11 +1452,11 @@ static int mdd_changelog_user_purge(struct mdd_device *mdd, int id, rc = -ENOENT; } - if (!rc && data.mcud_usercount == 0) - /* No more users; turn changelogs off */ - rc = mdd_changelog_on(mdd, 0); + if (!rc && data.mcud_usercount == 0) + /* No more users; turn changelogs off */ + rc = mdd_changelog_off(env, mdd); - RETURN (rc); + RETURN(rc); } /** mdd_iocontrol @@ -1498,7 +1481,8 @@ static int mdd_iocontrol(const struct lu_env *env, struct md_device *m, 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: { @@ -1506,7 +1490,22 @@ static int mdd_iocontrol(const struct lu_env *env, struct md_device *m, *mntopts = mdd->mdd_dt_conf.ddp_mntopts; RETURN(0); } - } + case OBD_IOC_START_LFSCK: { + rc = lfsck_start(env, mdd->mdd_bottom, + (struct lfsck_start_param *)karg); + RETURN(rc); + } + case OBD_IOC_STOP_LFSCK: { + rc = lfsck_stop(env, mdd->mdd_bottom, + (struct lfsck_stop *)karg); + RETURN(rc); + } + case OBD_IOC_QUERY_LFSCK: { + rc = lfsck_query(env, mdd->mdd_bottom, NULL, NULL, + (struct lfsck_query *)karg); + RETURN(rc); + } + } /* Below ioctls use obd_ioctl_data */ if (len != sizeof(*data)) { @@ -1521,10 +1520,10 @@ static int mdd_iocontrol(const struct lu_env *env, struct md_device *m, switch (cmd) { case OBD_IOC_CHANGELOG_REG: - rc = mdd_changelog_user_register(mdd, &data->ioc_u32_1); + rc = mdd_changelog_user_register(env, 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: @@ -1535,36 +1534,14 @@ static int mdd_iocontrol(const struct lu_env *env, struct md_device *m, } /* type constructor/destructor: mdd_type_init, mdd_type_fini */ -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, - .mdo_root_get = mdd_root_get, - .mdo_maxsize_get = mdd_maxsize_get, - .mdo_init_capa_ctxt = mdd_init_capa_ctxt, - .mdo_update_capa_key= mdd_update_capa_key, - .mdo_llog_ctxt_get = mdd_llog_ctxt_get, - .mdo_iocontrol = mdd_iocontrol, -#ifdef HAVE_QUOTA_SUPPORT - .mdo_quota = { - .mqo_notify = mdd_quota_notify, - .mqo_setup = mdd_quota_setup, - .mqo_cleanup = mdd_quota_cleanup, - .mqo_recovery = mdd_quota_recovery, - .mqo_check = mdd_quota_check, - .mqo_on = mdd_quota_on, - .mqo_off = mdd_quota_off, - .mqo_setinfo = mdd_quota_setinfo, - .mqo_getinfo = mdd_quota_getinfo, - .mqo_setquota = mdd_quota_setquota, - .mqo_getquota = mdd_quota_getquota, - .mqo_getoinfo = mdd_quota_getoinfo, - .mqo_getoquota = mdd_quota_getoquota, - .mqo_invalidate = mdd_quota_invalidate, - .mqo_finvalidate = mdd_quota_finvalidate - } -#endif +LU_TYPE_INIT_FINI(mdd, &mdd_thread_key); + +static const struct md_device_operations mdd_ops = { + .mdo_statfs = mdd_statfs, + .mdo_root_get = mdd_root_get, + .mdo_llog_ctxt_get = mdd_llog_ctxt_get, + .mdo_iocontrol = mdd_iocontrol, + .mdo_maxeasize_get = mdd_maxeasize_get, }; static struct lu_device_type_operations mdd_device_type_ops = { @@ -1577,7 +1554,6 @@ static struct lu_device_type_operations mdd_device_type_ops = { .ldto_device_alloc = mdd_device_alloc, .ldto_device_free = mdd_device_free, - .ldto_device_init = mdd_device_init, .ldto_device_fini = mdd_device_fini }; @@ -1595,11 +1571,9 @@ static void mdd_key_fini(const struct lu_context *ctx, struct lu_context_key *key, void *data) { struct mdd_thread_info *info = data; - if (info->mti_max_lmm != NULL) - OBD_FREE(info->mti_max_lmm, info->mti_max_lmm_size); - if (info->mti_max_cookie != NULL) - OBD_FREE(info->mti_max_cookie, info->mti_max_cookie_size); - mdd_buf_put(&info->mti_big_buf); + + lu_buf_free(&info->mti_big_buf); + lu_buf_free(&info->mti_link_buf); OBD_FREE_PTR(info); } @@ -1607,50 +1581,69 @@ static void mdd_key_fini(const struct lu_context *ctx, /* context key: mdd_thread_key */ LU_CONTEXT_KEY_DEFINE(mdd, LCT_MD_THREAD); -static struct lu_local_obj_desc llod_capa_key = { - .llod_name = CAPA_KEYS, - .llod_oid = MDD_CAPA_KEYS_OID, - .llod_is_index = 0, -}; +int mdd_generic_thread_start(struct mdd_generic_thread *thread, + int (*func)(void *), void *data, char *name) +{ + struct task_struct *task; -static struct lu_local_obj_desc llod_mdd_orphan = { - .llod_name = orph_index_name, - .llod_oid = MDD_ORPHAN_OID, - .llod_is_index = 1, - .llod_feat = &dt_directory_features, -}; + LASSERT(thread->mgt_init == false); + init_completion(&thread->mgt_started); + init_completion(&thread->mgt_finished); + thread->mgt_data = data; + thread->mgt_abort = false; + thread->mgt_init = true; -static struct lu_local_obj_desc llod_mdd_root = { - .llod_name = mdd_root_dir_name, - .llod_oid = MDD_ROOT_INDEX_OID, - .llod_is_index = 1, - .llod_feat = &dt_directory_features, -}; + task = kthread_run(func, thread, name); + if (IS_ERR(task)) { + complete(&thread->mgt_finished); + return PTR_ERR(task); + } + wait_for_completion(&thread->mgt_started); + return 0; +} -static int __init mdd_mod_init(void) +void mdd_generic_thread_stop(struct mdd_generic_thread *thread) { - struct lprocfs_static_vars lvars; - lprocfs_mdd_init_vars(&lvars); + if (thread->mgt_init == true) { + thread->mgt_abort = true; + wait_for_completion(&thread->mgt_finished); + } +} + +static int __init mdd_init(void) +{ + int rc; - llo_local_obj_register(&llod_capa_key); - llo_local_obj_register(&llod_mdd_orphan); - llo_local_obj_register(&llod_mdd_root); + rc = lu_kmem_init(mdd_caches); + if (rc) + return rc; - return class_register_type(&mdd_obd_device_ops, NULL, lvars.module_vars, - LUSTRE_MDD_NAME, &mdd_device_type); + changelog_orig_logops = llog_osd_ops; + changelog_orig_logops.lop_cancel = llog_changelog_cancel; + changelog_orig_logops.lop_add = llog_cat_add_rec; + changelog_orig_logops.lop_declare_add = llog_cat_declare_add_rec; + + hsm_actions_logops = llog_osd_ops; + hsm_actions_logops.lop_add = llog_cat_add_rec; + hsm_actions_logops.lop_declare_add = llog_cat_declare_add_rec; + + rc = class_register_type(&mdd_obd_device_ops, NULL, true, NULL, + LUSTRE_MDD_NAME, &mdd_device_type); + if (rc) + lu_kmem_fini(mdd_caches); + return rc; } -static void __exit mdd_mod_exit(void) +static void __exit mdd_exit(void) { - llo_local_obj_unregister(&llod_capa_key); - llo_local_obj_unregister(&llod_mdd_orphan); - llo_local_obj_unregister(&llod_mdd_root); - - class_unregister_type(LUSTRE_MDD_NAME); + class_unregister_type(LUSTRE_MDD_NAME); + lu_kmem_fini(mdd_caches); } -MODULE_AUTHOR("Sun Microsystems, Inc. "); -MODULE_DESCRIPTION("Lustre Meta-data Device Prototype ("LUSTRE_MDD_NAME")"); +MODULE_AUTHOR("OpenSFS, Inc. "); +MODULE_DESCRIPTION("Lustre Meta-data Device Driver ("LUSTRE_MDD_NAME")"); +MODULE_VERSION(LUSTRE_VERSION_STRING); MODULE_LICENSE("GPL"); -cfs_module(mdd, "0.1.0", mdd_mod_init, mdd_mod_exit); +module_init(mdd_init); +module_exit(mdd_exit);