X-Git-Url: https://git.whamcloud.com/?a=blobdiff_plain;f=lustre%2Fmdd%2Fmdd_device.c;h=104d2b50366c0dca11b94f9bebe0a59e870ea6d4;hb=702754812e7221102deb41aaeed4715a734a75f0;hp=8236e3d48a113a05b26fa0fbca89effd0ec719c8;hpb=d6b08e30909b3b02ffaf6a7dffabd48642b6f37f;p=fs%2Flustre-release.git diff --git a/lustre/mdd/mdd_device.c b/lustre/mdd/mdd_device.c index 8236e3d..104d2b5 100644 --- a/lustre/mdd/mdd_device.c +++ b/lustre/mdd/mdd_device.c @@ -1,6 +1,4 @@ -/* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*- - * vim:expandtab:shiftwidth=8:tabstop=8: - * +/* * GPL HEADER START * * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. @@ -26,8 +24,10 @@ * GPL HEADER END */ /* - * Copyright 2008 Sun Microsystems, Inc. All rights reserved + * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved. * Use is subject to license terms. + * + * Copyright (c) 2011, 2012, Intel Corporation. */ /* * This file is part of Lustre, http://www.lustre.org/ @@ -40,25 +40,13 @@ * Author: Wang Di */ -#ifndef EXPORT_SYMTAB -# define EXPORT_SYMTAB -#endif #define DEBUG_SUBSYSTEM S_MDS #include -#include -#include #include -#include -#include -#include - -#include -#include -#include #include -#include -#include /* for changelogs */ +#include +#include #include #include @@ -70,24 +58,88 @@ 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 int mdd_device_init(const struct lu_env *env, struct lu_device *d, - const char *name, struct lu_device *next) +/* Slab for MDD object allocation */ +cfs_mem_cache_t *mdd_object_kmem; + +static struct lu_kmem_descr mdd_caches[] = { + { + .ckd_cache = &mdd_object_kmem, + .ckd_name = "mdd_obj", + .ckd_size = sizeof(struct mdd_object) + }, + { + .ckd_cache = NULL + } +}; + +static int mdd_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; + ENTRY; - mdd->mdd_child = lu2dt_dev(next); + 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 = mdd_txn_commit_cb; - mdd->mdd_txn_cb.dtc_cookie = mdd; - 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) + RETURN(rc); + + mdd->mdd_atime_diff = MAX_ATIME_DIFF; + /* sync permission changes */ + mdd->mdd_sync_permission = 1; + + 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); - rc = mdd_procfs_init(mdd, name); RETURN(rc); } @@ -95,15 +147,17 @@ 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; + if (d->ld_site) + lu_dev_del_linkage(d->ld_site, d); + rc = mdd_procfs_fini(mdd); if (rc) { CERROR("proc fini error %d \n", rc); return ERR_PTR(rc); } - return next; + return NULL; } static void mdd_changelog_fini(const struct lu_env *env, @@ -114,134 +168,295 @@ 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_obd_dev) - mdd_fini_obd(env, m, cfg); + 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); orph_index_fini(env, m); + if (m->mdd_capa != NULL) { + lu_object_put(env, &m->mdd_capa->do_lu); + m->mdd_capa = NULL; + } + lu_site_purge(env, m->mdd_md_dev.md_lu_dev.ld_site, -1); /* remove upcall device*/ md_upcall_fini(&m->mdd_md_dev); + + if (m->mdd_child_exp) + obd_disconnect(m->mdd_child_exp); + EXIT; } -static int changelog_init_cb(struct llog_handle *llh, struct llog_rec_hdr *hdr, - void *data) +static int changelog_init_cb(const struct lu_env *env, struct llog_handle *llh, + struct llog_rec_hdr *hdr, void *data) { struct mdd_device *mdd = (struct mdd_device *)data; struct llog_changelog_rec *rec = (struct llog_changelog_rec *)hdr; - ENTRY; - if (!(llh->lgh_hdr->llh_flags & LLOG_F_IS_PLAIN)) { - CERROR("log is not plain\n"); - RETURN(-EINVAL); - } - if (rec->cr_hdr.lrh_type != CHANGELOG_REC) { - CERROR("Not a changelog rec? %d\n", rec->cr_hdr.lrh_type); - RETURN(-EINVAL); - } + LASSERT(llh->lgh_hdr->llh_flags & LLOG_F_IS_PLAIN); + LASSERT(rec->cr_hdr.lrh_type == CHANGELOG_REC); - CDEBUG(D_INODE, + 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; - RETURN(LLOG_PROC_BREAK); + mdd->mdd_cl.mc_index = rec->cr.cr_index; + return LLOG_PROC_BREAK; } -static int mdd_changelog_llog_init(struct mdd_device *mdd) +static int changelog_user_init_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 mdd_device *mdd = (struct mdd_device *)data; + struct llog_changelog_user_rec *rec = + (struct llog_changelog_user_rec *)hdr; - ctxt = llog_get_context(obd, LLOG_CHANGELOG_ORIG_CTXT); - if (ctxt == NULL) { - CERROR("no context\n"); - return -EINVAL; - } - if (!ctxt->loc_handle) { - CERROR("no handle\n"); - return -EINVAL; - } - rc = llog_cat_reverse_process(ctxt->loc_handle, changelog_init_cb, mdd); - llog_ctxt_put(ctxt); + LASSERT(llh->lgh_hdr->llh_flags & LLOG_F_IS_PLAIN); + LASSERT(rec->cur_hdr.lrh_type == CHANGELOG_USER_REC); - if (rc < 0) - CERROR("changelog init failed: %d\n", rc); - else - rc = 0; /* llog_proc_break is ok */ + 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); - CDEBUG(D_INODE, "changelog_init index="LPU64"\n", mdd->mdd_cl.mc_index); + spin_lock(&mdd->mdd_cl.mc_user_lock); + mdd->mdd_cl.mc_lastuser = rec->cur_id; + spin_unlock(&mdd->mdd_cl.mc_user_lock); - return rc; + return LLOG_PROC_BREAK; } -static int mdd_changelog_init(const struct lu_env *env, 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) { - int rc; + struct llog_changelog_rec *rec = (struct llog_changelog_rec *)hdr; + struct llog_cookie cookie; + long long endrec = *(long long *)data; + int rc; - mdd->mdd_cl.mc_index = 0; - spin_lock_init(&mdd->mdd_cl.mc_lock); - cfs_waitq_init(&mdd->mdd_cl.mc_waitq); + ENTRY; - mdd->mdd_cl.mc_starttime = cfs_time_current_64(); - mdd->mdd_cl.mc_flags = 0; /* off by default */ - mdd->mdd_cl.mc_mask = CL_DEFMASK; - rc = mdd_changelog_llog_init(mdd); - if (rc) { - CERROR("Changelog setup during init failed %d\n", rc); - mdd->mdd_cl.mc_flags |= CLM_ERR; - } - return rc; + /* This is always a (sub)log, not the catalog */ + LASSERT(llh->lgh_hdr->llh_flags & LLOG_F_IS_PLAIN); + + if (rec->cr.cr_index > endrec) + /* records are in order, so we're done */ + RETURN(LLOG_PROC_BREAK); + + cookie.lgc_lgl = llh->lgh_id; + cookie.lgc_index = hdr->lrh_index; + + /* 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); } -static void mdd_changelog_fini(const struct lu_env *env, struct mdd_device *mdd) +static int llog_changelog_cancel(const struct lu_env *env, + struct llog_ctxt *ctxt, + struct lov_stripe_md *lsm, int count, + struct llog_cookie *cookies, int flags) { - mdd->mdd_cl.mc_flags = 0; + struct llog_handle *cathandle = ctxt->loc_handle; + int 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 "LPX64" rc=%d\n", + ctxt->loc_obd->obd_name, cathandle->lgh_last_idx, + cathandle->lgh_id.lgl_oid, rc); + + 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 struct llog_operations changelog_orig_logops; + +int mdd_changelog_on(const struct lu_env *env, struct mdd_device *mdd, int on); + +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; - int rc; + struct obd_device *obd = mdd2obd_dev(mdd); + struct llog_ctxt *ctxt = NULL, *uctxt = NULL; + struct lu_fid rfid; + int rc; + + OBD_SET_CTXT_MAGIC(&obd->obd_lvfs_ctxt); + obd->obd_lvfs_ctxt.dt = mdd->mdd_bottom; + rc = dt_root_get(env, mdd->mdd_bottom, &rfid); + if (rc) + RETURN(-ENODEV); + + 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_cat_init_and_process(env, ctxt->loc_handle); + 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_cat_init_and_process(env, uctxt->loc_handle); + 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, 1); + 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; +} - if ((mdd->mdd_cl.mc_mask & (1 << rec->cr_type)) == 0) - return 0; +static int mdd_changelog_init(const struct lu_env *env, struct mdd_device *mdd) +{ + struct obd_device *obd = mdd2obd_dev(mdd); + int rc; + + mdd->mdd_cl.mc_index = 0; + 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); + mdd->mdd_cl.mc_lastuser = 0; + + 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; + } + + return rc; +} - rec->cr_hdr.lrh_len = llog_data_len(sizeof(*rec) + rec->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); - /* 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); - ctxt = llog_get_context(obd, LLOG_CHANGELOG_ORIG_CTXT); - if (ctxt == NULL) - return -ENXIO; +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; + + mdd->mdd_cl.mc_flags = 0; + + 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); + } +} - /* nested journal transaction */ - rc = llog_add(ctxt, &rec->cr_hdr, NULL, NULL, 0); - llog_ctxt_put(ctxt); +int mdd_changelog_write_header(const struct lu_env *env, + struct mdd_device *mdd, int markerflags); - cfs_waitq_signal(&mdd->mdd_cl.mc_waitq); +/* Start / stop recording */ +int mdd_changelog_on(const struct lu_env *env, struct mdd_device *mdd, int on) +{ + int rc = 0; - return 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 { + 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); + } + } 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(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; } /** Remove entries with indicies up to and including \a endrec from the @@ -250,23 +465,47 @@ 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) +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; + long long unsigned cur; int rc; ctxt = llog_get_context(obd, LLOG_CHANGELOG_ORIG_CTXT); if (ctxt == NULL) return -ENXIO; - /* Some records purged; reset repeat-access time */ - mdd->mdd_cl.mc_starttime = cfs_time_current_64(); + 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; + + /* purge to "0" is shorthand for everything */ + if (endrec == 0) + endrec = cur; + + /* If purging all records, write a header entry so we don't have an + empty catalog and we're sure to have a valid starting index next + 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(env, mdd, CLM_PURGE); + if (rc) + goto out; + } - rc = llog_cancel(ctxt, NULL, 1, (struct llog_cookie *)&endrec, 0); + /* Some records were purged, so reset repeat-access time (so we + record new mtime update records, so users can see a file has been + changed since the last purge) */ + mdd->mdd_cl.mc_starttime = cfs_time_current_64(); + rc = llog_cancel(env, ctxt, NULL, 1, (struct llog_cookie *)&endrec, 0); +out: llog_ctxt_put(ctxt); - return rc; } @@ -275,34 +514,54 @@ int mdd_changelog_llog_cancel(struct mdd_device *mdd, long long endrec) * \param markerflags - CLM_* * \retval 0 ok */ -int mdd_changelog_write_header(struct mdd_device *mdd, int markerflags) +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; - 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); - - rec->cr_flags = CLF_VERSION; - rec->cr_type = CL_MARK; - rec->cr_namelen = len; - memcpy(rec->cr_name, obd->obd_name, rec->cr_namelen); + 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 = mdd_buf_alloc(env, 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); /* Status and action flags */ - rec->cr_markerflags = mdd->mdd_cl.mc_flags | markerflags; - - rc = mdd_changelog_llog_write(mdd, rec, NULL); - - /* assume on or off event; reset repeat-access time */ - mdd->mdd_cl.mc_starttime = rec->cr_time; - - OBD_FREE(rec, reclen); - RETURN(rc); + rec->cr.cr_markerflags = mdd->mdd_cl.mc_flags | markerflags; + rec->cr_hdr.lrh_len = llog_data_len(sizeof(*rec) + 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); + + ctxt = llog_get_context(obd, LLOG_CHANGELOG_ORIG_CTXT); + LASSERT(ctxt); + + rc = llog_cat_add(env, ctxt->loc_handle, &rec->cr_hdr, NULL, NULL); + if (rc > 0) + rc = 0; + llog_ctxt_put(ctxt); + + /* assume on or off event; reset repeat-access time */ + mdd->mdd_cl.mc_starttime = cfs_time_current_64(); + RETURN(rc); } /** @@ -316,40 +575,98 @@ 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); - RETURN(rc); + 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_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) { - struct mdd_object *mdd_obj = md2mdd_obj(obj); + if (mask & ~(MAY_READ | MAY_EXEC)) + return -EPERM; + else + return 0; +} + +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; +} - return mdd_attr_get_internal_locked(env, mdd_obj, ma); +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_attr_set(const struct lu_env *env, struct md_object *obj, - const struct md_attr *ma) +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_xattr_get(const struct lu_env *env, - struct md_object *obj, struct lu_buf *buf, - const char *name) +static int dot_lustre_mdd_swap_layouts(const struct lu_env *env, + struct md_object *obj1, + struct md_object *obj2, + __u64 flags) +{ + 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) { @@ -362,14 +679,8 @@ 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); @@ -380,19 +691,58 @@ static int dot_lustre_close(const struct lu_env *env, struct md_object *obj, return 0; } +static int dot_lustre_mdd_object_sync(const struct lu_env *env, + struct md_object *obj) +{ + return -ENOSYS; +} + +static int dot_lustre_mdd_path(const struct lu_env *env, struct md_object *obj, + char *path, int pathlen, __u64 *recno, int *linkno) +{ + 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_path = dot_lustre_path + .moo_permission = dot_lustre_mdd_permission, + .moo_attr_get = mdd_attr_get, + .moo_attr_set = mdd_attr_set, + .moo_xattr_get = dot_lustre_mdd_xattr_get, + .moo_xattr_list = dot_lustre_mdd_xattr_list, + .moo_xattr_set = dot_lustre_mdd_xattr_set, + .moo_xattr_del = dot_lustre_mdd_xattr_del, + .moo_swap_layouts = dot_lustre_mdd_swap_layouts, + .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; @@ -402,86 +752,112 @@ 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_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_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_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_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_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_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_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_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_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, struct md_attr *ma) { - int rc = 0; - - 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_attr_get(env, &mdd->mdd_dot_lustre->mod_obj, ma); - if (rc) - return rc; - ma->ma_valid |= MA_INODE; - } + struct mdd_device *mdd = mdo2mdd(obj); - /* "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, - &ma->ma_lmm_size); - if (rc > 0) { - ma->ma_valid |= MA_LOV; - rc = 0; - } - } - } - - return rc; + /* "fid" is a virtual object and hence does not have any "real" + * attributes. So we reuse attributes of .lustre for "fid" dir */ + return mdd_attr_get(env, &mdd->mdd_dot_lustre->mod_obj, ma); } static int obf_attr_set(const struct lu_env *env, struct md_object *obj, @@ -490,11 +866,53 @@ static int obf_attr_set(const struct lu_env *env, struct md_object *obj, return -EPERM; } +static int obf_xattr_list(const struct lu_env *env, + struct md_object *obj, struct lu_buf *buf) +{ + return 0; +} + static int obf_xattr_get(const struct lu_env *env, struct md_object *obj, struct lu_buf *buf, const char *name) { - return 0; + struct mdd_device *mdd = mdo2mdd(obj); + struct mdd_object *root; + struct lu_fid rootfid; + int rc = 0; + + /* + * .lustre returns default striping which is 'stored' + * in the root + */ + if (strcmp(name, XATTR_NAME_LOV) == 0) { + rc = dt_root_get(env, mdd->mdd_child, &rootfid); + if (rc) + return rc; + root = mdd_object_find(env, mdd, &rootfid); + if (IS_ERR(root)) + return PTR_ERR(root); + rc = mdo_xattr_get(env, root, buf, name, + mdd_object_capa(env, md2mdd_obj(obj))); + mdd_object_put(env, root); + } + + return rc; +} + +static int obf_xattr_set(const struct lu_env *env, + struct md_object *obj, + const struct lu_buf *buf, const char *name, + int fl) +{ + return -EPERM; +} + +static int obf_xattr_del(const struct lu_env *env, + struct md_object *obj, + const char *name) +{ + return -EPERM; } static int obf_mdd_open(const struct lu_env *env, struct md_object *obj, @@ -510,7 +928,7 @@ 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); @@ -529,19 +947,22 @@ static int obf_mdd_readpage(const struct lu_env *env, struct md_object *obj, } static int obf_path(const struct lu_env *env, struct md_object *obj, - char *path, int pathlen, __u64 recno, int *linkno) + 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 + .moo_attr_get = obf_attr_get, + .moo_attr_set = obf_attr_set, + .moo_xattr_list = obf_xattr_list, + .moo_xattr_get = obf_xattr_get, + .moo_xattr_set = obf_xattr_set, + .moo_xattr_del = obf_xattr_del, + .moo_open = obf_mdd_open, + .moo_close = obf_mdd_close, + .moo_readpage = obf_mdd_readpage, + .moo_path = obf_path }; /** @@ -560,14 +981,21 @@ 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); + CWARN("%s: bad FID format [%s], should be "DFID"\n", + mdd2obd_dev(mdd)->obd_name, lname->ln_name, + (__u64)FID_SEQ_NORMAL, 1, 0); GOTO(out, rc = -EINVAL); } + if (!fid_is_norm(f)) { + CWARN("%s: "DFID" is invalid, sequence should be " + ">= "LPX64"\n", mdd2obd_dev(mdd)->obd_name, PFID(f), + (__u64)FID_SEQ_NORMAL); + GOTO(out, rc = -EINVAL); + } + /* Check if object with this fid exists */ child = mdd_object_find(env, mdd, f); if (child == NULL) @@ -662,7 +1090,7 @@ 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); @@ -672,8 +1100,6 @@ static int mdd_dot_lustre_setup(const struct lu_env *env, struct mdd_device *m) m->mdd_dot_lustre = lu2mdd_obj(lu_object_locate(dt_dot_lustre->do_lu.lo_header, &mdd_device_type)); - lu_object_put(env, &dt_dot_lustre->do_lu); - 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; @@ -710,20 +1136,13 @@ static int mdd_process_config(const struct lu_env *env, 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); break; case LCFG_CLEANUP: + mdd_lfsck_cleanup(env, m); + 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; @@ -732,65 +1151,15 @@ 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_osc_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); 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_osc_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 */ - obd->obd_recovering = 0; - obd->obd_type->typ_dt_ops->o_postrecov(obd); /* XXX: orphans handling. */ __mdd_orphan_cleanup(env, mdd); @@ -806,6 +1175,7 @@ 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; @@ -813,7 +1183,6 @@ static int mdd_prepare(const struct lu_env *env, 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)) { @@ -832,8 +1201,26 @@ static int mdd_prepare(const struct lu_env *env, GOTO(out, rc); } + /* we use capa file to declare llog changes, + * will be fixed with new llog in 2.3 */ + root = dt_store_open(env, mdd->mdd_child, "", CAPA_KEYS, &fid); + if (IS_ERR(root)) + GOTO(out, rc = PTR_ERR(root)); + + mdd->mdd_capa = root; + + rc = mdd_lfsck_setup(env, mdd); + if (rc) { + CERROR("%s: failed to initialize lfsck: rc = %d\n", + mdd2obd_dev(mdd)->obd_name, rc); + GOTO(out, rc); + } + rc = mdd_changelog_init(env, mdd); + + GOTO(out, rc); + out: - RETURN(rc); + return rc; } const struct lu_device_operations mdd_lu_ops = { @@ -860,7 +1247,7 @@ 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) + struct obd_statfs *sfs) { struct mdd_device *mdd = lu2mdd_dev(&m->md_lu_dev); int rc; @@ -875,28 +1262,16 @@ static int mdd_statfs(const struct lu_env *env, struct md_device *m, /* * 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; - - *md_size = mdd_lov_mdsize(env, mdd); - *cookie_size = mdd_lov_cookiesize(env, mdd); - - RETURN(0); -} - 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) { struct mdd_device *mdd = lu2mdd_dev(&m->md_lu_dev); - struct mds_obd *mds = &mdd2obd_dev(mdd)->u.mds; int rc; ENTRY; - mds->mds_capa_keys = keys; + /* need barrier for mds_capa_keys access. */ + rc = mdd_child_ops(mdd)->dt_init_capa_ctxt(env, mdd->mdd_child, mode, timeout, alg, keys); RETURN(rc); @@ -906,14 +1281,29 @@ static int mdd_update_capa_key(const struct lu_env *env, struct md_device *m, struct lustre_capa_key *key) { + /* we do not support capabilities ... */ + return -EINVAL; +} + +static int mdd_llog_ctxt_get(const struct lu_env *env, struct md_device *m, + int idx, void **h) +{ struct mdd_device *mdd = lu2mdd_dev(&m->md_lu_dev); - struct obd_export *lov_exp = mdd2obd_dev(mdd)->u.mds.mds_osc_exp; - int rc; - ENTRY; - rc = obd_set_info_async(lov_exp, sizeof(KEY_CAPA_KEY), KEY_CAPA_KEY, - sizeof(*key), key, NULL); - RETURN(rc); + *h = llog_group_get_ctxt(&mdd2obd_dev(mdd)->obd_olg, idx); + 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(cfs_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, @@ -927,48 +1317,87 @@ 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; + + CDEBUG(D_CONFIG, "connect #%d\n", mdd->mdd_connects); + + rc = class_connect(&conn, obd, cluuid); + if (rc) + RETURN(rc); - LASSERT(atomic_read(&lu->ld_ref) == 0); - md_device_fini(&m->mdd_md_dev); - OBD_FREE_PTR(m); - RETURN(next); + *exp = class_conn2export(&conn); + + /* Why should there ever be more than 1 connect? */ + LASSERT(mdd->mdd_connects == 0); + mdd->mdd_connects++; + + RETURN(0); } -static struct obd_ops mdd_obd_device_ops = { - .o_owner = THIS_MODULE -}; +/* + * once last export (we don't count self-export) disappeared + * mdd can be released + */ +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; -/* context key constructor/destructor: mdd_ucred_key_init, mdd_ucred_key_fini */ -LU_KEY_INIT_FINI(mdd_ucred, struct md_ucred); + mdd->mdd_connects--; + if (mdd->mdd_connects == 0) + release = 1; -static struct lu_context_key mdd_ucred_key = { - .lct_tags = LCT_SESSION, - .lct_init = mdd_ucred_key_init, - .lct_fini = mdd_ucred_key_fini -}; + rc = class_disconnect(exp); + + if (rc == 0 && release) + class_manual_cleanup(obd); + RETURN(rc); +} -struct md_ucred *md_ucred(const struct lu_env *env) +static int mdd_obd_health_check(const struct lu_env *env, + struct obd_device *obd) { - LASSERT(env->le_ses != NULL); - return lu_context_key_get(env->le_ses, &mdd_ucred_key); + struct mdd_device *mdd = lu2mdd_dev(obd->obd_lu_dev); + int rc; + ENTRY; + + LASSERT(mdd); + rc = obd_health_check(env, mdd->mdd_child_exp->exp_obd); + RETURN(rc); } -EXPORT_SYMBOL(md_ucred); + +static struct obd_ops mdd_obd_device_ops = { + .o_owner = THIS_MODULE, + .o_connect = mdd_obd_connect, + .o_disconnect = mdd_obd_disconnect, + .o_health_check = mdd_obd_health_check +}; /* * context key constructor/destructor: @@ -991,34 +1420,268 @@ struct md_capainfo *md_capainfo(const struct lu_env *env) } EXPORT_SYMBOL(md_capainfo); +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); + if (ctxt == NULL) + RETURN(-ENXIO); + + OBD_ALLOC_PTR(rec); + if (rec == NULL) { + llog_ctxt_put(ctxt); + RETURN(-ENOMEM); + } + + /* Assume we want it on since somebody registered */ + rc = mdd_changelog_on(env, mdd, 1); + if (rc) + GOTO(out, rc); + + rec->cur_hdr.lrh_len = sizeof(*rec); + rec->cur_hdr.lrh_type = CHANGELOG_USER_REC; + 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, NULL); + + CDEBUG(D_IOCTL, "Registered changelog user %d\n", *id); +out: + OBD_FREE_PTR(rec); + llog_ctxt_put(ctxt); + RETURN(rc); +} + +struct mdd_changelog_user_data { + __u64 mcud_endrec; /**< purge record for this user */ + __u64 mcud_minrec; /**< lowest changelog recno still referenced */ + __u32 mcud_id; + __u32 mcud_minid; /**< user id with lowest rec reference */ + __u32 mcud_usercount; + int mcud_found:1; +}; +#define MCUD_UNREGISTER -1LL + +/** Two things: + * 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(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 = data; + int rc; + + ENTRY; + + LASSERT(llh->lgh_hdr->llh_flags & LLOG_F_IS_PLAIN); + + rec = (struct llog_changelog_user_rec *)hdr; + + mcud->mcud_usercount++; + + /* If we have a new endrec for this id, use it for the following + min check instead of its old value */ + if (rec->cur_id == mcud->mcud_id) + rec->cur_endrec = max(rec->cur_endrec, mcud->mcud_endrec); + + /* Track the minimum referenced record */ + if (mcud->mcud_minid == 0 || mcud->mcud_minrec > rec->cur_endrec) { + mcud->mcud_minid = rec->cur_id; + mcud->mcud_minrec = rec->cur_endrec; + } + + if (rec->cur_id != mcud->mcud_id) + RETURN(0); + + /* Update this user's record */ + mcud->mcud_found = 1; + + /* 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(env, llh->u.phd.phd_cat_handle, + 1, &cookie); + if (rc == 0) + mcud->mcud_usercount--; + + RETURN(rc); + } + + /* Update the endrec */ + 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(env, llh, hdr, NULL, 0, (void *)(hdr + 1), + hdr->lrh_index); + + RETURN(rc); +} + +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; + struct llog_ctxt *ctxt; + int rc; + ENTRY; + + 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); + 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); + if (ctxt == NULL) + return -ENXIO; + + LASSERT(ctxt->loc_handle->lgh_hdr->llh_flags & LLOG_F_IS_CAT); + + 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(env, mdd, data.mcud_minrec); + } else { + CWARN("Could not determine changelog records to purge; rc=%d\n", + rc); + } + + llog_ctxt_put(ctxt); + + if (!data.mcud_found) { + CWARN("No entry for user %d. Last changelog reference is " + LPD64" by changelog user %d\n", data.mcud_id, + data.mcud_minrec, data.mcud_minid); + rc = -ENOENT; + } + + if (!rc && data.mcud_usercount == 0) + /* No more users; turn changelogs off */ + rc = mdd_changelog_on(env, mdd, 0); + + RETURN (rc); +} + +/** mdd_iocontrol + * May be called remotely from mdt_iocontrol_handle or locally from + * mdt_iocontrol. Data may be freeform - remote handling doesn't enforce + * an obd_ioctl_data format (but local ioctl handler does). + * \param cmd - ioc + * \param len - data len + * \param karg - ioctl data, in kernel space + */ +static int mdd_iocontrol(const struct lu_env *env, struct md_device *m, + unsigned int cmd, int len, void *karg) +{ + struct mdd_device *mdd; + struct obd_ioctl_data *data = karg; + int rc; + ENTRY; + + mdd = lu2mdd_dev(&m->md_lu_dev); + + /* Doesn't use obd_ioctl_data */ + switch (cmd) { + case OBD_IOC_CHANGELOG_CLEAR: { + struct changelog_setinfo *cs = karg; + rc = mdd_changelog_user_purge(env, mdd, cs->cs_id, + cs->cs_recno); + RETURN(rc); + } + case OBD_IOC_GET_MNTOPT: { + mntopt_t *mntopts = (mntopt_t *)karg; + *mntopts = mdd->mdd_dt_conf.ddp_mntopts; + RETURN(0); + } + case OBD_IOC_START_LFSCK: { + struct lfsck_start *start = karg; + struct md_lfsck *lfsck = &mdd->mdd_lfsck; + + /* Return the kernel service version. */ + /* XXX: version can be used for compatibility in the future. */ + start->ls_version = lfsck->ml_version; + rc = mdd_lfsck_start(env, lfsck, start); + RETURN(rc); + } + case OBD_IOC_STOP_LFSCK: { + rc = mdd_lfsck_stop(env, &mdd->mdd_lfsck); + RETURN(rc); + } + } + + /* Below ioctls use obd_ioctl_data */ + if (len != sizeof(*data)) { + CERROR("Bad ioctl size %d\n", len); + RETURN(-EINVAL); + } + if (data->ioc_version != OBD_IOCTL_VERSION) { + CERROR("Bad magic %x != %x\n", data->ioc_version, + OBD_IOCTL_VERSION); + RETURN(-EINVAL); + } + + switch (cmd) { + case OBD_IOC_CHANGELOG_REG: + rc = mdd_changelog_user_register(env, mdd, &data->ioc_u32_1); + break; + case OBD_IOC_CHANGELOG_DEREG: + rc = mdd_changelog_user_purge(env, mdd, data->ioc_u32_1, + MCUD_UNREGISTER); + break; + default: + 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_capainfo_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, -#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 + .mdo_llog_ctxt_get = mdd_llog_ctxt_get, + .mdo_iocontrol = mdd_iocontrol, }; static struct lu_device_type_operations mdd_device_type_ops = { @@ -1031,7 +1694,6 @@ 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 }; @@ -1081,26 +1743,49 @@ static struct lu_local_obj_desc llod_mdd_root = { .llod_feat = &dt_directory_features, }; +static struct lu_local_obj_desc llod_lfsck_bookmark = { + .llod_name = lfsck_bookmark_name, + .llod_oid = LFSCK_BOOKMARK_OID, + .llod_is_index = 0, +}; + static int __init mdd_mod_init(void) { - struct lprocfs_static_vars lvars; - lprocfs_mdd_init_vars(&lvars); - - llo_local_obj_register(&llod_capa_key); - llo_local_obj_register(&llod_mdd_orphan); - llo_local_obj_register(&llod_mdd_root); - - return class_register_type(&mdd_obd_device_ops, NULL, lvars.module_vars, - LUSTRE_MDD_NAME, &mdd_device_type); + struct lprocfs_static_vars lvars; + int rc; + + lprocfs_mdd_init_vars(&lvars); + + rc = lu_kmem_init(mdd_caches); + if (rc) + return rc; + + 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; + + llo_local_obj_register(&llod_capa_key); + llo_local_obj_register(&llod_mdd_orphan); + llo_local_obj_register(&llod_mdd_root); + llo_local_obj_register(&llod_lfsck_bookmark); + + rc = class_register_type(&mdd_obd_device_ops, NULL, lvars.module_vars, + LUSTRE_MDD_NAME, &mdd_device_type); + if (rc) + lu_kmem_fini(mdd_caches); + return rc; } static void __exit mdd_mod_exit(void) { - llo_local_obj_unregister(&llod_capa_key); - llo_local_obj_unregister(&llod_mdd_orphan); - llo_local_obj_unregister(&llod_mdd_root); + llo_local_obj_unregister(&llod_capa_key); + llo_local_obj_unregister(&llod_mdd_orphan); + llo_local_obj_unregister(&llod_mdd_root); + llo_local_obj_unregister(&llod_lfsck_bookmark); - class_unregister_type(LUSTRE_MDD_NAME); + class_unregister_type(LUSTRE_MDD_NAME); + lu_kmem_fini(mdd_caches); } MODULE_AUTHOR("Sun Microsystems, Inc. ");