X-Git-Url: https://git.whamcloud.com/?p=fs%2Flustre-release.git;a=blobdiff_plain;f=lustre%2Fmdd%2Fmdd_device.c;h=52be69acb6fd0a0a4ba32e77d85aef2fd1815ed1;hp=4e863c1a098a57587623fd59729546cb342b6bd8;hb=3442db6faf685fbdbd092bdfdc8d273e4990a141;hpb=8e357d93ccc26e85e6ddf53360df72b47a254852 diff --git a/lustre/mdd/mdd_device.c b/lustre/mdd/mdd_device.c index 4e863c1..52be69a 100644 --- a/lustre/mdd/mdd_device.c +++ b/lustre/mdd/mdd_device.c @@ -15,11 +15,7 @@ * * You should have received a copy of the GNU General Public License * version 2 along with this program; If not, see - * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf - * - * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara, - * CA 95054 USA or visit www.sun.com if you need additional information or - * have any questions. + * http://www.gnu.org/licenses/gpl-2.0.html * * GPL HEADER END */ @@ -27,7 +23,7 @@ * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved. * Use is subject to license terms. * - * Copyright (c) 2011, 2013, Intel Corporation. + * Copyright (c) 2011, 2016, Intel Corporation. */ /* * This file is part of Lustre, http://www.lustre.org/ @@ -43,16 +39,20 @@ #define DEBUG_SUBSYSTEM S_MDS #include +#include #include +#include #include #include #include -#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"; @@ -118,19 +118,41 @@ out: static int mdd_init0(const struct lu_env *env, struct mdd_device *mdd, struct lu_device_type *t, struct lustre_cfg *lcfg) { - int rc; + 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_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; rc = mdd_connect_to_next(env, mdd, lustre_cfg_string(lcfg, 3)); - if (rc) + if (rc != 0) RETURN(rc); mdd->mdd_atime_diff = MAX_ATIME_DIFF; - /* sync permission changes */ - mdd->mdd_sync_permission = 1; + /* sync permission changes */ + mdd->mdd_sync_permission = 1; + /* enable changelog garbage collection */ + mdd->mdd_changelog_gc = 1; + /* with a significant amount of idle time */ + mdd->mdd_changelog_max_idle_time = CHLOG_MAX_IDLE_TIME; + /* or a significant amount of late indexes */ + mdd->mdd_changelog_max_idle_indexes = CHLOG_MAX_IDLE_INDEXES; + /* with a reasonable interval between each check */ + mdd->mdd_changelog_min_gc_interval = CHLOG_MIN_GC_INTERVAL; + /* with a very few number of free entries */ + mdd->mdd_changelog_min_free_cat_entries = CHLOG_MIN_FREE_CAT_ENTRIES; dt_conf_get(env, mdd->mdd_child, &mdd->mdd_dt_conf); @@ -147,17 +169,12 @@ static int mdd_init0(const struct lu_env *env, struct mdd_device *mdd, static struct lu_device *mdd_device_fini(const struct lu_env *env, struct lu_device *d) { - struct mdd_device *mdd = lu2mdd_dev(d); - int rc; + struct mdd_device *mdd = lu2mdd_dev(d); 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); - } + mdd_procfs_fini(mdd); return NULL; } @@ -171,10 +188,10 @@ static int changelog_init_cb(const struct lu_env *env, struct llog_handle *llh, LASSERT(rec->cr_hdr.lrh_type == CHANGELOG_REC); 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, + "seeing record at index %d/%d/%llu t=%x %.*s in log" + DFID"\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, POSTID(&llh->lgh_id.lgl_oi)); + changelog_rec_name(&rec->cr), PFID(&llh->lgh_id.lgl_oi.oi_fid)); mdd->mdd_cl.mc_index = rec->cr.cr_index; return LLOG_PROC_BREAK; @@ -184,16 +201,16 @@ 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; + struct mdd_device *mdd = (struct mdd_device *)data; + struct llog_changelog_user_rec *rec = + (struct llog_changelog_user_rec *)hdr; - LASSERT(llh->lgh_hdr->llh_flags & LLOG_F_IS_PLAIN); - LASSERT(rec->cur_hdr.lrh_type == CHANGELOG_USER_REC); + 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 "DOSTID"\n", hdr->lrh_index, rec->cur_hdr.lrh_index, - rec->cur_id, rec->cur_endrec, POSTID(&llh->lgh_id.lgl_oi)); + CDEBUG(D_INFO, "seeing user at index %d/%d id=%d endrec=%llu" + " in log "DFID"\n", hdr->lrh_index, rec->cur_hdr.lrh_index, + rec->cur_id, rec->cur_endrec, PFID(&llh->lgh_id.lgl_oi.oi_fid)); spin_lock(&mdd->mdd_cl.mc_user_lock); mdd->mdd_cl.mc_lastuser = rec->cur_id; @@ -251,16 +268,58 @@ static int llog_changelog_cancel(const struct lu_env *env, /* 0 or 1 means we're done */ rc = 0; else - CERROR("%s: cancel idx %u of catalog "DOSTID" rc=%d\n", + CERROR("%s: cancel idx %u of catalog "DFID": rc = %d\n", ctxt->loc_obd->obd_name, cathandle->lgh_last_idx, - POSTID(&cathandle->lgh_id.lgl_oi), rc); + PFID(&cathandle->lgh_id.lgl_oi.oi_fid), rc); RETURN(rc); } 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_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 = 0; + + if ((mdd->mdd_cl.mc_flags & CLM_ON) != 0) + 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 int +mdd_changelog_off(const struct lu_env *env, struct mdd_device *mdd) +{ + 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) @@ -293,7 +352,7 @@ static int mdd_changelog_llog_init(const struct lu_env *env, if (rc) GOTO(out_cleanup, rc); - rc = llog_cat_init_and_process(env, ctxt->loc_handle); + rc = llog_init_handle(env, ctxt->loc_handle, LLOG_F_IS_CAT, NULL); if (rc) GOTO(out_close, rc); @@ -306,7 +365,7 @@ static int mdd_changelog_llog_init(const struct lu_env *env, GOTO(out_close, rc); } - CDEBUG(D_IOCTL, "changelog starting index="LPU64"\n", + CDEBUG(D_IOCTL, "changelog starting index=%llu\n", mdd->mdd_cl.mc_index); /* setup user changelog */ @@ -329,7 +388,7 @@ static int mdd_changelog_llog_init(const struct lu_env *env, 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); + rc = llog_init_handle(env, uctxt->loc_handle, LLOG_F_IS_CAT, NULL); if (rc) GOTO(out_uclose, rc); @@ -343,7 +402,7 @@ static int mdd_changelog_llog_init(const struct lu_env *env, /* If we have registered users, assume we want changelogs on */ if (mdd->mdd_cl.mc_lastuser > 0) { - rc = mdd_changelog_on(env, mdd, 1); + rc = mdd_changelog_on(env, mdd); if (rc < 0) GOTO(out_uclose, rc); } @@ -368,9 +427,7 @@ static int mdd_changelog_init(const struct lu_env *env, struct mdd_device *mdd) 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; + mdd->mdd_cl.mc_starttime = ktime_get(); spin_lock_init(&mdd->mdd_cl.mc_user_lock); mdd->mdd_cl.mc_lastuser = 0; @@ -404,45 +461,15 @@ static void mdd_changelog_fini(const struct lu_env *env, } } -int mdd_changelog_write_header(const struct lu_env *env, - struct mdd_device *mdd, int markerflags); - -/* Start / stop recording */ -int mdd_changelog_on(const struct lu_env *env, struct mdd_device *mdd, int on) -{ - int rc = 0; - - 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 * changelog * \param mdd * \param endrec * \retval 0 ok */ -int mdd_changelog_llog_cancel(const struct lu_env *env, - 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; @@ -477,7 +504,7 @@ int mdd_changelog_llog_cancel(const struct lu_env *env, /* 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(); + mdd->mdd_cl.mc_starttime = ktime_get(); rc = llog_cancel(env, ctxt, (struct llog_cookie *)&endrec, 0); out: @@ -504,7 +531,7 @@ int mdd_changelog_write_header(const struct lu_env *env, ENTRY; if (mdd->mdd_cl.mc_mask & (1 << CL_MARK)) { - mdd->mdd_cl.mc_starttime = cfs_time_current_64(); + mdd->mdd_cl.mc_starttime = ktime_get(); RETURN(0); } @@ -517,10 +544,11 @@ int mdd_changelog_write_header(const struct lu_env *env, 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_hdr.lrh_len = llog_data_len(sizeof(*rec) + rec->cr.cr_namelen); + 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); @@ -530,13 +558,13 @@ int mdd_changelog_write_header(const struct lu_env *env, ctxt = llog_get_context(obd, LLOG_CHANGELOG_ORIG_CTXT); LASSERT(ctxt); - rc = llog_cat_add(env, ctxt->loc_handle, &rec->cr_hdr, NULL, NULL); + rc = llog_cat_add(env, ctxt->loc_handle, &rec->cr_hdr, 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(); + mdd->mdd_cl.mc_starttime = ktime_get(); RETURN(rc); } @@ -558,26 +586,25 @@ static int obf_lookup(const struct lu_env *env, struct md_object *p, sscanf(name, SFID, RFID(f)); if (!fid_is_sane(f)) { - CWARN("%s: bad FID format [%s], should be "DFID"\n", - mdd2obd_dev(mdd)->obd_name, lname->ln_name, + 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); } if (!fid_is_norm(f) && !fid_is_igif(f) && !fid_is_root(f) && !fid_seq_is_dot(f->f_seq)) { - CWARN("%s: "DFID" is invalid, sequence should be " - ">= "LPX64" or within ["LPX64","LPX64"].\n", - mdd2obd_dev(mdd)->obd_name, PFID(f), - (__u64)FID_SEQ_NORMAL, (__u64)FID_SEQ_IGIF, - (__u64)FID_SEQ_IGIF_MAX); + CWARN("%s: Trying to lookup invalid FID "DFID" in %s/%s, " + "sequence should be >= %#llx or within [%#llx," + "%#llx].\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); } /* 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)); @@ -819,7 +846,7 @@ static int mdd_hsm_actions_llog_init(const struct lu_env *env, GOTO(out_cleanup, rc); } - rc = llog_cat_init_and_process(env, ctxt->loc_handle); + rc = llog_init_handle(env, ctxt->loc_handle, LLOG_F_IS_CAT, NULL); if (rc) GOTO(out_close, rc); @@ -858,13 +885,21 @@ static int mdd_hsm_actions_llog_fini(const struct lu_env *env, static void mdd_device_shutdown(const struct lu_env *env, struct mdd_device *m, struct lustre_cfg *cfg) { + barrier_deregister(m->mdd_bottom); 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); - if (m->mdd_los != NULL) + if (mdd2obd_dev(m)->u.obt.obt_nodemap_config_file) { + nm_config_file_deregister_tgt(env, + mdd2obd_dev(m)->u.obt.obt_nodemap_config_file); + mdd2obd_dev(m)->u.obt.obt_nodemap_config_file = NULL; + } + 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); if (m->mdd_child_exp) @@ -881,23 +916,26 @@ 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_conf_get(env, dt, &m->mdd_dt_conf); break; - case LCFG_CLEANUP: + case LCFG_PRE_CLEANUP: + rc = next->ld_ops->ldo_process_config(env, next, cfg); + mdd_generic_thread_stop(&m->mdd_orph_cleanup_thread); + break; + case LCFG_CLEANUP: rc = next->ld_ops->ldo_process_config(env, next, cfg); lu_dev_del_linkage(d->ld_site, d); mdd_device_shutdown(env, m, cfg); @@ -922,18 +960,19 @@ static int mdd_recovery_complete(const struct lu_env *env, next = &mdd->mdd_child->dd_lu_dev; /* XXX: orphans handling. */ - __mdd_orphan_cleanup(env, mdd); + if (!mdd->mdd_bottom->dd_rdonly) + mdd_orphan_cleanup(env, mdd); rc = next->ld_ops->ldo_recovery_complete(env, next); RETURN(rc); } 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) + const struct lu_fid *pfid, const char *name, + __u32 mode, struct lu_fid *fid) { - struct dt_object *parent, *dto; - int rc; + struct dt_object *parent, *dto; + int rc = 0; ENTRY; @@ -955,11 +994,11 @@ int mdd_local_file_create(const struct lu_env *env, struct mdd_device *mdd, *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); + dt_object_put_nocache(env, dto); EXIT; out_put: - lu_object_put(env, &parent->do_lu); - return 0; + dt_object_put(env, parent); + return rc; } static int mdd_lfsck_out_notify(const struct lu_env *env, void *data, @@ -972,10 +1011,12 @@ 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; + struct mdd_device *mdd = lu2mdd_dev(cdev); + struct lu_device *next = &mdd->mdd_child->dd_lu_dev; + struct nm_config_file *nodemap_config; + struct obd_device_target *obt = &mdd2obd_dev(mdd)->u.obt; + struct lu_fid fid; + int rc; ENTRY; @@ -1014,10 +1055,6 @@ static int mdd_prepare(const struct lu_env *env, mdd2obd_dev(mdd)->obd_name, rc); GOTO(out_los, rc); } - - rc = mdd_compat_fixes(env, mdd); - if (rc != 0) - GOTO(out_dot, rc); } else { /* Normal client usually send root access to MDT0 directly, * the root FID on non-MDT0 will only be used by echo client. */ @@ -1039,15 +1076,39 @@ static int mdd_prepare(const struct lu_env *env, if (rc != 0) GOTO(out_changelog, rc); + nodemap_config = nm_config_file_register_tgt(env, mdd->mdd_bottom, + mdd->mdd_los); + if (IS_ERR(nodemap_config)) { + rc = PTR_ERR(nodemap_config); + if (rc != -EROFS) + GOTO(out_hsm, rc); + } else { + obt->obt_nodemap_config_file = nodemap_config; + } + 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_hsm, rc); + GOTO(out_nodemap, rc); } + + rc = barrier_register(mdd->mdd_bottom, mdd->mdd_child); + if (rc) { + CERROR("%s: failed to register to barrier: rc = %d\n", + mdd2obd_dev(mdd)->obd_name, rc); + GOTO(out_lfsck, rc); + } + RETURN(0); + +out_lfsck: + lfsck_degister(env, mdd->mdd_bottom); +out_nodemap: + nm_config_file_deregister_tgt(env, obt->obt_nodemap_config_file); + obt->obt_nodemap_config_file = NULL; out_hsm: mdd_hsm_actions_llog_fini(env, mdd); out_changelog: @@ -1099,41 +1160,12 @@ static int mdd_statfs(const struct lu_env *env, struct md_device *m, RETURN(rc); } -/* - * No permission check is needed. - */ -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); - int rc; - ENTRY; - - /* 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); -} - -static int mdd_maxeasize_get(const struct lu_env *env, struct md_device *m, - int *easize) +static const struct dt_device_param *mdd_dtconf_get(const struct lu_env *env, + struct md_device *m) { struct mdd_device *mdd = lu2mdd_dev(&m->md_lu_dev); - ENTRY; - - *easize = mdd->mdd_dt_conf.ddp_max_ea_size; - - RETURN(0); -} -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; + return &mdd->mdd_dt_conf; } static int mdd_llog_ctxt_get(const struct lu_env *env, struct md_device *m, @@ -1232,8 +1264,7 @@ static int mdd_obd_disconnect(struct obd_export *exp) } static int mdd_obd_get_info(const struct lu_env *env, struct obd_export *exp, - __u32 keylen, void *key, __u32 *vallen, void *val, - struct lov_stripe_md *lsm) + __u32 keylen, void *key, __u32 *vallen, void *val) { int rc = -EINVAL; @@ -1247,18 +1278,39 @@ static int mdd_obd_get_info(const struct lu_env *env, struct obd_export *exp, mdd = lu2mdd_dev(obd->obd_lu_dev); LASSERT(mdd); rc = obd_get_info(env, mdd->mdd_child_exp, keylen, key, vallen, - val, lsm); + val); RETURN(rc); } 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) +{ + 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); +} + 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, @@ -1280,10 +1332,10 @@ static int mdd_changelog_user_register(const struct lu_env *env, RETURN(-ENOMEM); } - /* Assume we want it on since somebody registered */ - rc = mdd_changelog_on(env, 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; @@ -1295,9 +1347,14 @@ static int mdd_changelog_user_register(const struct lu_env *env, } *id = rec->cur_id = ++mdd->mdd_cl.mc_lastuser; rec->cur_endrec = mdd->mdd_cl.mc_index; + + rec->cur_time = (__u32)get_seconds(); + if (OBD_FAIL_CHECK(OBD_FAIL_TIME_IN_CHLOG_USER)) + rec->cur_time = 0; + spin_unlock(&mdd->mdd_cl.mc_user_lock); - rc = llog_cat_add(env, ctxt->loc_handle, &rec->cur_hdr, NULL, NULL); + rc = llog_cat_add(env, ctxt->loc_handle, &rec->cur_hdr, NULL); CDEBUG(D_IOCTL, "Registered changelog user %d\n", *id); out: @@ -1306,138 +1363,258 @@ out: 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; - unsigned int mcud_found:1; +struct mdd_changelog_user_purge { + __u32 mcup_id; + __u32 mcup_usercount; + __u64 mcup_minrec; + bool mcup_found; }; -#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 +/** + * changelog_user_purge callback + * + * Is called once per user. + * + * Check to see the user requested is available from rec. + * Truncate the changelog. + * Keep track of the total number of users (calls). */ 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; + struct mdd_changelog_user_purge *mcup = data; + struct llog_cookie cookie; int rc; - ENTRY; + ENTRY; - LASSERT(llh->lgh_hdr->llh_flags & LLOG_F_IS_PLAIN); + if ((llh->lgh_hdr->llh_flags & LLOG_F_IS_PLAIN) == 0) + RETURN(-ENXIO); - rec = (struct llog_changelog_user_rec *)hdr; + rec = container_of(hdr, struct llog_changelog_user_rec, cur_hdr); - mcud->mcud_usercount++; + mcup->mcup_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); + if (rec->cur_id != mcup->mcup_id) { + /* truncate to the lowest endrec that is not this user */ + mcup->mcup_minrec = min(mcup->mcup_minrec, + rec->cur_endrec); + RETURN(0); + } - /* 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; - } + /* Unregister this user */ + cookie.lgc_lgl = llh->lgh_id; + cookie.lgc_index = hdr->lrh_index; - if (rec->cur_id != mcud->mcud_id) - RETURN(0); + rc = llog_cat_cancel_records(env, llh->u.phd.phd_cat_handle, + 1, &cookie); + if (rc == 0) { + mcup->mcup_found = true; + mcup->mcup_usercount--; + } - /* Update this user's record */ - mcud->mcud_found = 1; + RETURN(rc); +} - /* Special case: unregister this user */ - if (mcud->mcud_endrec == MCUD_UNREGISTER) { - struct llog_cookie cookie; +int mdd_changelog_user_purge(const struct lu_env *env, + struct mdd_device *mdd, __u32 id) +{ + struct mdd_changelog_user_purge mcup = { + .mcup_id = id, + .mcup_found = false, + .mcup_usercount = 0, + .mcup_minrec = ULLONG_MAX, + }; + struct llog_ctxt *ctxt; + int rc; - cookie.lgc_lgl = llh->lgh_id; - cookie.lgc_index = hdr->lrh_index; + ENTRY; - rc = llog_cat_cancel_records(env, llh->u.phd.phd_cat_handle, - 1, &cookie); - if (rc == 0) - mcud->mcud_usercount--; + CDEBUG(D_IOCTL, "%s: Purge request: id=%u\n", + mdd2obd_dev(mdd)->obd_name, id); - RETURN(rc); - } + ctxt = llog_get_context(mdd2obd_dev(mdd), + LLOG_CHANGELOG_USER_ORIG_CTXT); + if (ctxt == NULL || + (ctxt->loc_handle->lgh_hdr->llh_flags & LLOG_F_IS_CAT) == 0) + GOTO(out, rc = -ENXIO); - /* Update the endrec */ - CDEBUG(D_IOCTL, "Rewriting changelog user %d endrec to "LPU64"\n", - mcud->mcud_id, rec->cur_endrec); + rc = llog_cat_process(env, ctxt->loc_handle, + mdd_changelog_user_purge_cb, &mcup, + 0, 0); - /* 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); + if ((rc == 0) && mcup.mcup_found) { + CDEBUG(D_IOCTL, "%s: Purging changelog entries for user %d " + "record=%llu\n", + mdd2obd_dev(mdd)->obd_name, id, mcup.mcup_minrec); + /* Cancelling record 0 destroys the entire changelog, make sure + we don't do that unless we mean it. */ + if (mcup.mcup_minrec != 0 || mcup.mcup_usercount == 0) { + rc = mdd_changelog_llog_cancel(env, mdd, + mcup.mcup_minrec); + } + } else { + CWARN("%s: No changelog for user %u; rc=%d\n", + mdd2obd_dev(mdd)->obd_name, id, rc); + GOTO(out, rc = -ENOENT); + } - RETURN(rc); + if ((rc == 0) && (mcup.mcup_usercount == 0)) { + /* No more users; turn changelogs off */ + CDEBUG(D_IOCTL, "turning off changelogs\n"); + rc = mdd_changelog_off(env, mdd); + } + + EXIT; +out: + if (ctxt != NULL) + llog_ctxt_put(ctxt); + + 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_clear { + __u64 mcuc_endrec; + __u64 mcuc_minrec; + __u32 mcuc_id; + bool mcuc_flush; +}; + +/** + * changelog_clear callback + * + * Is called once per user. + * + * Check to see the user requested is available from rec. + * Check the oldest (smallest) record for boundary conditions. + * Truncate the changelog. + */ +static int mdd_changelog_clear_cb(const struct lu_env *env, + struct llog_handle *llh, + struct llog_rec_hdr *hdr, + void *data) { - struct mdd_changelog_user_data data; - struct llog_ctxt *ctxt; - int rc; - ENTRY; + struct llog_changelog_user_rec *rec; + struct mdd_changelog_user_clear *mcuc = data; + int rc; + + ENTRY; + + if ((llh->lgh_hdr->llh_flags & LLOG_F_IS_PLAIN) == 0) + RETURN(-ENXIO); + + rec = container_of(hdr, struct llog_changelog_user_rec, cur_hdr); + + /* Does the changelog id match the requested id? */ + if (rec->cur_id != mcuc->mcuc_id) { + mcuc->mcuc_minrec = min(mcuc->mcuc_minrec, + rec->cur_endrec); + RETURN(0); + } + + /* cur_endrec is the oldest purgeable record, make sure we're newer */ + if (rec->cur_endrec > mcuc->mcuc_endrec) { + CDEBUG(D_IOCTL, "Request %llu out of range: %llu\n", + mcuc->mcuc_endrec, rec->cur_endrec); + RETURN(-EINVAL); + } + + /* Flag that we've met all the range and user checks. + * We now know the record to flush. + */ + rec->cur_endrec = mcuc->mcuc_endrec; + + rec->cur_time = (__u32)get_seconds(); + if (OBD_FAIL_CHECK(OBD_FAIL_TIME_IN_CHLOG_USER)) + rec->cur_time = 0; - CDEBUG(D_IOCTL, "Purge request: id=%d, endrec=%lld\n", id, endrec); + mcuc->mcuc_flush = true; - data.mcud_id = id; - data.mcud_minid = 0; - data.mcud_minrec = 0; - data.mcud_usercount = 0; - data.mcud_endrec = endrec; + CDEBUG(D_IOCTL, "Rewriting changelog user %u endrec to %llu\n", + mcuc->mcuc_id, rec->cur_endrec); + + /* Update the endrec */ + rc = llog_write(env, llh, hdr, hdr->lrh_index); + + RETURN(rc); +} + +/** + * Clear a changelog up to entry specified by endrec for user id. + */ +static int mdd_changelog_clear(const struct lu_env *env, + struct mdd_device *mdd, __u32 id, + __u64 endrec) +{ + struct mdd_changelog_user_clear mcuc = { + .mcuc_id = id, + .mcuc_minrec = endrec, + .mcuc_flush = false, + }; + struct llog_ctxt *ctxt; + __u64 start_rec; + int rc; + + ENTRY; + + CDEBUG(D_IOCTL, "%s: Purge request: id=%u, endrec=%llu\n", + mdd2obd_dev(mdd)->obd_name, id, endrec); + /* start_rec is the newest (largest value) entry in the changelogs*/ spin_lock(&mdd->mdd_cl.mc_lock); - endrec = mdd->mdd_cl.mc_index; + start_rec = 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; + if (start_rec < endrec) { + CDEBUG(D_IOCTL, "%s: Could not clear changelog, requested "\ + "address out of range\n", mdd2obd_dev(mdd)->obd_name); + RETURN(-EINVAL); + } + + if (endrec == 0) { + mcuc.mcuc_endrec = start_rec; + mcuc.mcuc_minrec = ULLONG_MAX; + } else { + mcuc.mcuc_endrec = endrec; + } - LASSERT(ctxt->loc_handle->lgh_hdr->llh_flags & LLOG_F_IS_CAT); + ctxt = llog_get_context(mdd2obd_dev(mdd), + LLOG_CHANGELOG_USER_ORIG_CTXT); + if (ctxt == NULL || + (ctxt->loc_handle->lgh_hdr->llh_flags & LLOG_F_IS_CAT) == 0) + GOTO(out, rc = -ENXIO); rc = llog_cat_process(env, ctxt->loc_handle, - mdd_changelog_user_purge_cb, (void *)&data, + mdd_changelog_clear_cb, (void *)&mcuc, 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 < 0) { + CWARN("%s: Failure to clear the changelog for user %d: %d\n", + mdd2obd_dev(mdd)->obd_name, id, rc); + } else if (mcuc.mcuc_flush) { + /* Cancelling record 0 destroys the entire changelog, make sure + we don't do that unless we mean it. */ + if (mcuc.mcuc_minrec != 0) { + CDEBUG(D_IOCTL, "%s: Purging changelog entries up "\ + "to %llu\n", mdd2obd_dev(mdd)->obd_name, + mcuc.mcuc_minrec); + + rc = mdd_changelog_llog_cancel(env, mdd, + mcuc.mcuc_minrec); + } + } else { + CWARN("%s: No entry for user %d\n", + mdd2obd_dev(mdd)->obd_name, id); + rc = -ENOENT; + } - if (!rc && data.mcud_usercount == 0) - /* No more users; turn changelogs off */ - rc = mdd_changelog_on(env, mdd, 0); + EXIT; +out: + if (ctxt != NULL) + llog_ctxt_put(ctxt); - RETURN (rc); + return rc; } /** mdd_iocontrol @@ -1451,26 +1628,24 @@ static int mdd_changelog_user_purge(const struct lu_env *env, 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; + struct mdd_device *mdd = lu2mdd_dev(&m->md_lu_dev); + 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; - /* 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); - } + if (unlikely(!barrier_entry(mdd->mdd_bottom))) + RETURN(-EINPROGRESS); + + rc = mdd_changelog_clear(env, mdd, cs->cs_id, + cs->cs_recno); + barrier_exit(mdd->mdd_bottom); + RETURN(rc); + } case OBD_IOC_START_LFSCK: { rc = lfsck_start(env, mdd->mdd_bottom, (struct lfsck_start_param *)karg); @@ -1481,45 +1656,55 @@ static int mdd_iocontrol(const struct lu_env *env, struct md_device *m, (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)) { - 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); - } + /* 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; - } + switch (cmd) { + case OBD_IOC_CHANGELOG_REG: + if (unlikely(!barrier_entry(mdd->mdd_bottom))) + RETURN(-EINPROGRESS); + + rc = mdd_changelog_user_register(env, mdd, &data->ioc_u32_1); + barrier_exit(mdd->mdd_bottom); + break; + case OBD_IOC_CHANGELOG_DEREG: + if (unlikely(!barrier_entry(mdd->mdd_bottom))) + RETURN(-EINPROGRESS); - RETURN (rc); + rc = mdd_changelog_user_purge(env, mdd, data->ioc_u32_1); + barrier_exit(mdd->mdd_bottom); + break; + default: + rc = -ENOTTY; + } + + RETURN(rc); } /* type constructor/destructor: mdd_type_init, mdd_type_fini */ LU_TYPE_INIT_FINI(mdd, &mdd_thread_key); -const struct md_device_operations mdd_ops = { +static const struct md_device_operations mdd_ops = { .mdo_statfs = mdd_statfs, .mdo_root_get = mdd_root_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, - .mdo_maxeasize_get = mdd_maxeasize_get, + .mdo_dtconf_get = mdd_dtconf_get, }; static struct lu_device_type_operations mdd_device_type_ops = { @@ -1546,25 +1731,52 @@ static struct lu_device_type mdd_device_type = { LU_KEY_INIT(mdd, struct mdd_thread_info); static void mdd_key_fini(const struct lu_context *ctx, - struct lu_context_key *key, void *data) + struct lu_context_key *key, void *data) { - struct mdd_thread_info *info = data; + struct mdd_thread_info *info = data; lu_buf_free(&info->mti_big_buf); lu_buf_free(&info->mti_link_buf); + lu_buf_free(&info->mti_xattr_buf); - OBD_FREE_PTR(info); + OBD_FREE_PTR(info); } /* context key: mdd_thread_key */ LU_CONTEXT_KEY_DEFINE(mdd, LCT_MD_THREAD); -static int __init mdd_mod_init(void) +int mdd_generic_thread_start(struct mdd_generic_thread *thread, + int (*func)(void *), void *data, char *name) { - struct lprocfs_static_vars lvars; - int rc; + struct task_struct *task; + + 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; + + 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; +} - lprocfs_mdd_init_vars(&lvars); +void mdd_generic_thread_stop(struct mdd_generic_thread *thread) +{ + if (thread->mgt_init == true) { + thread->mgt_abort = true; + wait_for_completion(&thread->mgt_finished); + } +} + +static int __init mdd_init(void) +{ + int rc; rc = lu_kmem_init(mdd_caches); if (rc) @@ -1579,24 +1791,23 @@ static int __init mdd_mod_init(void) 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, NULL, -#ifndef HAVE_ONLY_PROCFS_SEQ - lvars.module_vars, -#endif - LUSTRE_MDD_NAME, &mdd_device_type); + 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) { 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);