X-Git-Url: https://git.whamcloud.com/?p=fs%2Flustre-release.git;a=blobdiff_plain;f=lustre%2Fmds%2Fmds_lov.c;h=2270fd7562e90b8b5062686966ce2e808ab8575c;hp=031ec325512eb9d5f09a745a1935bdc7b35f670d;hb=87c86d444e61e38d6454bba5700ba966dc1ac83d;hpb=6987e0666ba1e3f1e9490c3ca2478cd40d917749 diff --git a/lustre/mds/mds_lov.c b/lustre/mds/mds_lov.c index 031ec32..2270fd7 100644 --- a/lustre/mds/mds_lov.c +++ b/lustre/mds/mds_lov.c @@ -55,270 +55,321 @@ void cpu_to_le_lov_desc (struct lov_desc *ld) ld->ld_pattern = cpu_to_le32 (ld->ld_pattern); } -void mds_lov_update_objids(struct obd_device *obd, obd_id *ids) +void mds_dt_save_objids(struct obd_device *obd, obd_id *ids) { struct mds_obd *mds = &obd->u.mds; int i; ENTRY; - spin_lock(&mds->mds_lov_lock); - for (i = 0; i < mds->mds_lov_desc.ld_tgt_count; i++) - if (ids[i] > (mds->mds_lov_objids)[i]) - (mds->mds_lov_objids)[i] = ids[i]; - spin_unlock(&mds->mds_lov_lock); + spin_lock(&mds->mds_dt_lock); + for (i = 0; i < mds->mds_dt_desc.ld_tgt_count; i++) + ids[i] = mds->mds_dt_objids[i]; + spin_unlock(&mds->mds_dt_lock); EXIT; } -static int mds_lov_read_objids(struct obd_device *obd) +void mds_dt_update_objids(struct obd_device *obd, obd_id *ids) { struct mds_obd *mds = &obd->u.mds; - obd_id *ids; + int i; + ENTRY; + + spin_lock(&mds->mds_dt_lock); + for (i = 0; i < mds->mds_dt_desc.ld_tgt_count; i++) + if (ids[i] > mds->mds_dt_objids[i]) + mds->mds_dt_objids[i] = ids[i]; + spin_unlock(&mds->mds_dt_lock); + EXIT; +} + +static int mds_dt_read_objids(struct obd_device *obd) +{ + struct mds_obd *mds = &obd->u.mds; + int i, rc, size; loff_t off = 0; - int i, rc, size = mds->mds_lov_desc.ld_tgt_count * sizeof(*ids); + obd_id *ids; ENTRY; - if (mds->mds_lov_objids != NULL) + if (mds->mds_dt_objids != NULL) RETURN(0); + size = mds->mds_dt_desc.ld_tgt_count * sizeof(*ids); OBD_ALLOC(ids, size); if (ids == NULL) RETURN(-ENOMEM); - mds->mds_lov_objids = ids; + mds->mds_dt_objids = ids; - if (mds->mds_lov_objid_filp->f_dentry->d_inode->i_size == 0) + if (mds->mds_dt_objid_filp->f_dentry->d_inode->i_size == 0) RETURN(0); - rc = fsfilt_read_record(obd, mds->mds_lov_objid_filp, ids, size, &off); + + rc = fsfilt_read_record(obd, mds->mds_dt_objid_filp, ids, size, &off); if (rc < 0) { - CERROR("Error reading objids %d\n", rc); + CERROR("error reading objids %d\n", rc); } else { - mds->mds_lov_objids_valid = 1; + mds->mds_dt_objids_valid = 1; rc = 0; } - for (i = 0; i < mds->mds_lov_desc.ld_tgt_count; i++) - CDEBUG(D_INFO, "read last object "LPU64" for idx %d\n", - mds->mds_lov_objids[i], i); + for (i = 0; i < mds->mds_dt_desc.ld_tgt_count; i++) { + CDEBUG(D_INFO, "read last object "LPU64 + " for idx %d\n", mds->mds_dt_objids[i], i); + } RETURN(rc); } -int mds_lov_write_objids(struct obd_device *obd) +int mds_dt_write_objids(struct obd_device *obd) { struct mds_obd *mds = &obd->u.mds; + int i, rc, size; loff_t off = 0; - int i, rc, size = mds->mds_lov_desc.ld_tgt_count * sizeof(obd_id); ENTRY; - for (i = 0; i < mds->mds_lov_desc.ld_tgt_count; i++) + for (i = 0; i < mds->mds_dt_desc.ld_tgt_count; i++) CDEBUG(D_INFO, "writing last object "LPU64" for idx %d\n", - mds->mds_lov_objids[i], i); + mds->mds_dt_objids[i], i); - rc = fsfilt_write_record(obd, mds->mds_lov_objid_filp, - mds->mds_lov_objids, size, &off, 0); + size = mds->mds_dt_desc.ld_tgt_count * sizeof(obd_id); + rc = fsfilt_write_record(obd, mds->mds_dt_objid_filp, + mds->mds_dt_objids, size, &off, 0); RETURN(rc); } -int mds_lov_clearorphans(struct mds_obd *mds, struct obd_uuid *ost_uuid) +int mds_dt_clear_orphans(struct mds_obd *mds, struct obd_uuid *ost_uuid) { + struct lov_stripe_md *empty_ea = NULL; + struct obd_trans_info oti = { 0 }; + struct obdo *oa = NULL; int rc; - struct obdo oa; - struct obd_trans_info oti = {0}; - struct lov_stripe_md *empty_ea = NULL; ENTRY; - LASSERT(mds->mds_lov_objids != NULL); + LASSERT(mds->mds_dt_objids != NULL); - /* This create will in fact either create or destroy: If the OST is + /* + * this create will in fact either create or destroy: If the OST is * missing objects below this ID, they will be created. If it finds - * objects above this ID, they will be removed. */ - memset(&oa, 0, sizeof(oa)); - oa.o_gr = FILTER_GROUP_FIRST_MDS + mds->mds_num; - oa.o_valid = OBD_MD_FLFLAGS | OBD_MD_FLGROUP; - oa.o_flags = OBD_FL_DELORPHAN; + * objects above this ID, they will be removed. + */ + oa = obdo_alloc(); + if (oa == NULL) + RETURN(-ENOMEM); + + memset(oa, 0, sizeof(*oa)); + + oa->o_gr = FILTER_GROUP_FIRST_MDS + mds->mds_num; + oa->o_valid = OBD_MD_FLFLAGS | OBD_MD_FLGROUP; + oa->o_flags = OBD_FL_DELORPHAN; + if (ost_uuid != NULL) { - memcpy(&oa.o_inline, ost_uuid, sizeof(*ost_uuid)); - oa.o_valid |= OBD_MD_FLINLINE; + memcpy(&oa->o_inline, ost_uuid, + sizeof(*ost_uuid)); + oa->o_valid |= OBD_MD_FLINLINE; } - rc = obd_create(mds->mds_osc_exp, &oa, &empty_ea, &oti); - - RETURN(rc); -} - -/* update the LOV-OSC knowledge of the last used object id's */ -int mds_lov_set_nextid(struct obd_device *obd) -{ - struct mds_obd *mds = &obd->u.mds; - int rc; - ENTRY; - - LASSERT(!obd->obd_recovering); - - LASSERT(mds->mds_lov_objids != NULL); - rc = obd_set_info(mds->mds_osc_exp, strlen("next_id"), "next_id", - mds->mds_lov_desc.ld_tgt_count, mds->mds_lov_objids); + /* + * passing current objids for letting data layer know last objids MDS + * knows about and do appropriate. --umka + */ + oti.oti_objid = mds->mds_dt_objids; + + rc = obd_create(mds->mds_dt_exp, oa, + NULL, 0, &empty_ea, &oti); + + obdo_free(oa); RETURN(rc); } /* tell the LOV-OSC by how much to pre-create */ -int mds_lov_set_growth(struct mds_obd *mds, int count) +int mds_dt_set_growth(struct mds_obd *mds, int count) { int rc; ENTRY; - rc = obd_set_info(mds->mds_osc_exp, strlen("growth_count"), + rc = obd_set_info(mds->mds_dt_exp, strlen("growth_count"), "growth_count", sizeof(count), &count); RETURN(rc); } -static int mds_lov_update_desc(struct obd_device *obd, struct obd_export *lov) +static int mds_dt_update_desc(struct obd_device *obd, struct obd_export *lov) { struct mds_obd *mds = &obd->u.mds; - int valsize = sizeof(mds->mds_lov_desc), rc, i; + __u32 valsize = sizeof(mds->mds_dt_desc); + int old_count, rc = 0, i; ENTRY; - rc = obd_get_info(lov, strlen("lovdesc") + 1, "lovdesc", &valsize, - &mds->mds_lov_desc); + old_count = mds->mds_dt_desc.ld_tgt_count; + + rc = obd_get_info(lov, strlen("lovdesc") + 1, "lovdesc", + &valsize, &mds->mds_dt_desc); if (rc) RETURN(rc); - i = lov_mds_md_size(mds->mds_lov_desc.ld_tgt_count); + /* The size of the LOV target table may have increased. */ + if (old_count >= mds->mds_dt_desc.ld_tgt_count) { + obd_id *ids; + int size; + + size = mds->mds_dt_desc.ld_tgt_count * sizeof(*ids); + OBD_ALLOC(ids, size); + if (ids == NULL) + RETURN(-ENOMEM); + + memset(ids, 0, size); + + if (mds->mds_dt_objids != NULL) { + int oldsize = old_count * sizeof(*ids); + + memcpy(ids, mds->mds_dt_objids, oldsize); + OBD_FREE(mds->mds_dt_objids, oldsize); + } + mds->mds_dt_objids = ids; + } + + i = lov_mds_md_size(mds->mds_dt_desc.ld_tgt_count); if (i > mds->mds_max_mdsize) mds->mds_max_mdsize = i; - mds->mds_max_cookiesize = mds->mds_lov_desc.ld_tgt_count * - sizeof(struct llog_cookie); - mds->mds_has_lov_desc = 1; + mds->mds_max_cookiesize = mds->mds_dt_desc.ld_tgt_count * + sizeof(struct llog_cookie); + mds->mds_has_dt_desc = 1; RETURN(0); } -int mds_lov_connect(struct obd_device *obd, char * lov_name) +int mds_dt_connect(struct obd_device *obd, char *lov_name) { struct mds_obd *mds = &obd->u.mds; - struct lustre_handle conn = {0,}; - char name[32] = "CATLIST"; - int rc, i, valsize; - __u32 group; + struct lustre_handle conn = { 0 }; + int i, rc = 0; ENTRY; - if (IS_ERR(mds->mds_osc_obd)) - RETURN(PTR_ERR(mds->mds_osc_obd)); + if (IS_ERR(mds->mds_dt_obd)) + RETURN(PTR_ERR(mds->mds_dt_obd)); - if (mds->mds_osc_obd) + if (mds->mds_dt_obd) RETURN(0); - spin_lock_init(&mds->mds_lov_lock); - mds->mds_osc_obd = class_name2obd(lov_name); - if (!mds->mds_osc_obd) { - CERROR("MDS cannot locate LOV %s\n", - lov_name); - mds->mds_osc_obd = ERR_PTR(-ENOTCONN); + spin_lock_init(&mds->mds_dt_lock); + mds->mds_dt_obd = class_name2obd(lov_name); + if (!mds->mds_dt_obd) { + CERROR("MDS cannot locate LOV %s\n", lov_name); + mds->mds_dt_obd = ERR_PTR(-ENOTCONN); RETURN(-ENOTCONN); } + if (mds->mds_ost_sec) { + rc = obd_set_info(mds->mds_dt_obd->obd_self_export, + strlen("sec"), "sec", + strlen(mds->mds_ost_sec), + mds->mds_ost_sec); + if (rc) { + mds->mds_dt_obd = ERR_PTR(rc); + RETURN(rc); + } + } + CDEBUG(D_HA, "obd: %s osc: %s lov_name: %s\n", - obd->obd_name, mds->mds_osc_obd->obd_name, lov_name); + obd->obd_name, mds->mds_dt_obd->obd_name, lov_name); - rc = obd_connect(&conn, mds->mds_osc_obd, &obd->obd_uuid); + rc = obd_connect(&conn, mds->mds_dt_obd, &obd->obd_uuid, NULL, + mds->mds_num + FILTER_GROUP_FIRST_MDS); if (rc) { - CERROR("MDS cannot connect to LOV %s (%d)\n", - lov_name, rc); - mds->mds_osc_obd = ERR_PTR(rc); + CERROR("MDS cannot connect to LOV %s (%d)\n", lov_name, rc); + mds->mds_dt_obd = ERR_PTR(rc); RETURN(rc); } - mds->mds_osc_exp = class_conn2export(&conn); + mds->mds_dt_exp = class_conn2export(&conn); - rc = obd_register_observer(mds->mds_osc_obd, obd); + rc = obd_register_observer(mds->mds_dt_obd, obd); if (rc) { CERROR("MDS cannot register as observer of LOV %s (%d)\n", lov_name, rc); GOTO(err_discon, rc); } - rc = mds_lov_update_desc(obd, mds->mds_osc_exp); + rc = mds_dt_update_desc(obd, mds->mds_dt_exp); if (rc) GOTO(err_reg, rc); - rc = mds_lov_read_objids(obd); + rc = mds_dt_read_objids(obd); if (rc) { CERROR("cannot read %s: rc = %d\n", "lov_objids", rc); GOTO(err_reg, rc); } rc = obd_llog_cat_initialize(obd, &obd->obd_llogs, - mds->mds_lov_desc.ld_tgt_count, name); + mds->mds_dt_desc.ld_tgt_count, CATLIST); if (rc) { CERROR("failed to initialize catalog %d\n", rc); GOTO(err_reg, rc); } - /* FIXME before set info call is made, we must initialize logging */ - group = FILTER_GROUP_FIRST_MDS + mds->mds_num; - valsize = sizeof(group); - rc = obd_set_info(mds->mds_osc_exp, strlen("mds_conn"), "mds_conn", - valsize, &group); - if (rc) - GOTO(err_reg, rc); - - /* If we're mounting this code for the first time on an existing FS, - * we need to populate the objids array from the real OST values */ - if (!mds->mds_lov_objids_valid) { - int size = sizeof(obd_id) * mds->mds_lov_desc.ld_tgt_count; - rc = obd_get_info(mds->mds_osc_exp, strlen("last_id"), - "last_id", &size, mds->mds_lov_objids); + /* + * If we're mounting this code for the first time on an existing FS, we + * need to populate the objids array from the real OST values. + */ + if (!mds->mds_dt_objids_valid) { + __u32 size = sizeof(obd_id) * mds->mds_dt_desc.ld_tgt_count; + + rc = obd_get_info(mds->mds_dt_exp, strlen("last_id"), + "last_id", &size, mds->mds_dt_objids); if (!rc) { - for (i = 0; i < mds->mds_lov_desc.ld_tgt_count; i++) + for (i = 0; i < mds->mds_dt_desc.ld_tgt_count; i++) CWARN("got last object "LPU64" from OST %d\n", - mds->mds_lov_objids[i], i); - mds->mds_lov_objids_valid = 1; - rc = mds_lov_write_objids(obd); + mds->mds_dt_objids[i], i); + mds->mds_dt_objids_valid = 1; + rc = mds_dt_write_objids(obd); if (rc) CERROR("got last objids from OSTs, but error " "writing objids file: %d\n", rc); } } - /* I want to see a callback happen when the OBD moves to a - * "For General Use" state, and that's when we'll call - * set_nextid(). The class driver can help us here, because - * it can use the obd_recovering flag to determine when the - * the OBD is full available. */ - if (!obd->obd_recovering) - rc = mds_postrecov(obd); + /* + * I want to see a callback happen when the OBD moves to a "For General + * Use" state, and that's when we'll call set_nextid(). The class driver + * can help us here, because it can use the obd_recovering flag to + * determine when the the OBD is full available. + */ + if (!obd->obd_recovering) { + CDEBUG(D_OTHER, "call mds_postrecov_common()\n"); + rc = mds_postrecov_common(obd); + if (rc > 0) + rc = 0; + } RETURN(rc); err_reg: - obd_register_observer(mds->mds_osc_obd, NULL); + obd_register_observer(mds->mds_dt_obd, NULL); err_discon: - obd_disconnect(mds->mds_osc_exp, 0); - mds->mds_osc_exp = NULL; - mds->mds_osc_obd = ERR_PTR(rc); - RETURN(rc); + obd_disconnect(mds->mds_dt_exp, 0); + mds->mds_dt_obd = ERR_PTR(rc); + mds->mds_dt_exp = NULL; + return rc; } -int mds_lov_disconnect(struct obd_device *obd, int flags) +int mds_dt_disconnect(struct obd_device *obd, int flags) { struct mds_obd *mds = &obd->u.mds; int rc = 0; ENTRY; - if (!IS_ERR(mds->mds_osc_obd) && mds->mds_osc_exp != NULL) { + if (!IS_ERR(mds->mds_dt_obd) && mds->mds_dt_exp != NULL) { /* cleanup all llogging subsystems */ rc = obd_llog_finish(obd, &obd->obd_llogs, - mds->mds_lov_desc.ld_tgt_count); + mds->mds_dt_desc.ld_tgt_count); if (rc) CERROR("failed to cleanup llogging subsystems\n"); - obd_register_observer(mds->mds_osc_obd, NULL); + obd_register_observer(mds->mds_dt_obd, NULL); - rc = obd_disconnect(mds->mds_osc_exp, flags); + rc = obd_disconnect(mds->mds_dt_exp, flags); /* if obd_disconnect fails (probably because the * export was disconnected by class_disconnect_exports) * then we just need to drop our ref. */ if (rc != 0) - class_export_put(mds->mds_osc_exp); - mds->mds_osc_exp = NULL; - mds->mds_osc_obd = NULL; + class_export_put(mds->mds_dt_exp); + mds->mds_dt_exp = NULL; + mds->mds_dt_obd = NULL; } RETURN(rc); @@ -335,6 +386,7 @@ int mds_iocontrol(unsigned int cmd, struct obd_export *exp, int len, int rc = 0; ENTRY; + CDEBUG(D_INFO, "ioctl cmd %x\n", cmd); switch (cmd) { case OBD_IOC_RECORD: { char *name = data->ioc_inlbuf1; @@ -427,13 +479,6 @@ int mds_iocontrol(unsigned int cmd, struct obd_export *exp, int len, OBD_FREE(cfg_buf, data->ioc_plen1); RETURN(rc); } - case OBD_IOC_SNAP_ADD: { - char *name = data->ioc_inlbuf1; - if (name) { - rc = fsfilt_set_snap_item(obd, mds->mds_sb, name); - } - RETURN(rc); - } case OBD_IOC_PARSE: { struct llog_ctxt *ctxt = llog_get_context(&obd->obd_llogs, LLOG_CONFIG_ORIG_CTXT); @@ -445,7 +490,6 @@ int mds_iocontrol(unsigned int cmd, struct obd_export *exp, int len, RETURN(rc); } - case OBD_IOC_DUMP_LOG: { struct llog_ctxt *ctxt = llog_get_context(&obd->obd_llogs, LLOG_CONFIG_ORIG_CTXT); @@ -455,7 +499,6 @@ int mds_iocontrol(unsigned int cmd, struct obd_export *exp, int len, RETURN(rc); } - case OBD_IOC_SET_READONLY: { void *handle; struct inode *inode = obd->u.mds.mds_sb->s_root->d_inode; @@ -467,12 +510,12 @@ int mds_iocontrol(unsigned int cmd, struct obd_export *exp, int len, LASSERT(handle); rc = fsfilt_commit(obd, obd->u.mds.mds_sb, inode, handle, 1); - dev_set_rdonly(ll_sbdev(obd->u.mds.mds_sb), 2); + ll_set_rdonly(ll_sbdev(obd->u.mds.mds_sb)); RETURN(0); } case OBD_IOC_CATLOGLIST: { - int count = mds->mds_lov_desc.ld_tgt_count; + int count = mds->mds_dt_desc.ld_tgt_count; rc = llog_catalog_list(obd, count, data); RETURN(rc); @@ -482,19 +525,21 @@ int mds_iocontrol(unsigned int cmd, struct obd_export *exp, int len, case OBD_IOC_LLOG_REMOVE: { struct llog_ctxt *ctxt = llog_get_context(&obd->obd_llogs, LLOG_CONFIG_ORIG_CTXT); - char name[32] = "CATLIST"; - int rc2; + int rc2, valsize; + __u32 group; obd_llog_finish(obd, &obd->obd_llogs, - mds->mds_lov_desc.ld_tgt_count); + mds->mds_dt_desc.ld_tgt_count); push_ctxt(&saved, ctxt->loc_lvfs_ctxt, NULL); rc = llog_ioctl(ctxt, cmd, data); pop_ctxt(&saved, ctxt->loc_lvfs_ctxt, NULL); obd_llog_cat_initialize(obd, &obd->obd_llogs, - mds->mds_lov_desc.ld_tgt_count, - name); - rc2 = obd_set_info(mds->mds_osc_exp, strlen("mds_conn"), - "mds_conn", 0, NULL); + mds->mds_dt_desc.ld_tgt_count, + CATLIST); + group = FILTER_GROUP_FIRST_MDS + mds->mds_num; + valsize = sizeof(group); + rc2 = obd_set_info(mds->mds_dt_exp, strlen("mds_conn"), + "mds_conn", valsize, &group); if (!rc) rc = rc2; RETURN(rc); @@ -514,83 +559,152 @@ int mds_iocontrol(unsigned int cmd, struct obd_export *exp, int len, case OBD_IOC_ABORT_RECOVERY: target_stop_recovery_thread(obd); RETURN(0); + case OBD_IOC_ROOT_SQUASH: { + __u32 *p = (__u32 *) data->ioc_inlbuf1; + if (data->ioc_inllen1 != + (sizeof(__u32) * 4 + sizeof(ptl_nid_t))) + RETURN(-EINVAL); + + if (*p == 0) { /* get */ + p += 2; + *p++ = mds->mds_squash_uid; + *p++ = mds->mds_squash_gid; + *((ptl_nid_t*) p) = mds->mds_nosquash_nid; + } else { /* set */ + p += 2; + mds->mds_squash_uid = *p++; + mds->mds_squash_gid = *p++; + mds->mds_nosquash_nid = *((ptl_nid_t*) p); + CWARN("MDS: squash root to %d:%d, except nid 0x%llx\n", + mds->mds_squash_uid, mds->mds_squash_gid, + mds->mds_nosquash_nid); + } + RETURN(0); + } default: + CDEBUG(D_INFO, "unknown command %x\n", cmd); RETURN(-EINVAL); } RETURN(0); } -struct mds_lov_sync_info { - struct obd_device *mlsi_obd; /* the lov device to sync */ - struct obd_uuid *mlsi_uuid; /* target to sync */ +struct mds_dt_sync_info { + struct obd_device *mlsi_obd; /* the mds to sync */ + struct obd_device *mlsi_watched; /* new lov target */ + int mlsi_index; /* index into mds_dt_objids */ }; -int mds_lov_synchronize(void *data) +int mds_dt_synchronize(void *data) { - struct mds_lov_sync_info *mlsi = data; + struct mds_dt_sync_info *mlsi = data; struct llog_ctxt *ctxt; struct obd_device *obd; + struct obd_device *watched; + struct mds_obd *mds; struct obd_uuid *uuid; + obd_id vals[2]; unsigned long flags; + __u32 vallen; + __u32 group; + int old_count; + int count; + int index; int rc; - int valsize; - __u32 group; + char name[32] = "CATLIST"; lock_kernel(); + ptlrpc_daemonize(); - snprintf (current->comm, sizeof (current->comm), "%s", "mds_lov_sync"); SIGNAL_MASK_LOCK(current, flags); sigfillset(¤t->blocked); RECALC_SIGPENDING; SIGNAL_MASK_UNLOCK(current, flags); + unlock_kernel(); + obd = mlsi->mlsi_obd; - uuid = mlsi->mlsi_uuid; + watched = mlsi->mlsi_watched; + index = mlsi->mlsi_index; + + LASSERT(obd != NULL); + LASSERT(watched != NULL); OBD_FREE(mlsi, sizeof(*mlsi)); + mds = &obd->u.mds; + down(&mds->mds_orphan_recovery_sem); - LASSERT(obd != NULL); - LASSERT(uuid != NULL); + uuid = &watched->u.cli.cl_import->imp_target_uuid; + + group = FILTER_GROUP_FIRST_MDS + mds->mds_num; + rc = obd_set_info(watched->obd_self_export, strlen("mds_conn"), + "mds_conn", sizeof(group), &group); + if (rc) + GOTO(cleanup, rc); + + old_count = mds->mds_dt_desc.ld_tgt_count; + + rc = mds_dt_update_desc(obd, mds->mds_dt_exp); + if (rc) + GOTO(cleanup, rc); + + count = mds->mds_dt_desc.ld_tgt_count; + LASSERT(count >= old_count); + + vallen = sizeof(vals[1]); + rc = obd_get_info(watched->obd_self_export, strlen("last_id"), + "last_id", &vallen, &vals[1]); + if (rc) + GOTO(cleanup, rc); + + /* we don't set next id manually, instead OSCs will set them + * during own recovery from DELORPHAN reply -bzzz */ +#if 0 + vals[0] = index; + rc = mds_dt_set_info(obd->obd_self_export, strlen("next_id"), + "next_id", 2, vals); + if (rc) + GOTO(cleanup, rc); +#endif + + obd_llog_finish(obd, &obd->obd_llogs, old_count); + obd_llog_cat_initialize(obd, &obd->obd_llogs, count, name); - group = FILTER_GROUP_FIRST_MDS + obd->u.mds.mds_num; - valsize = sizeof(group); - rc = obd_set_info(obd->u.mds.mds_osc_exp, strlen("mds_conn"), - "mds_conn", valsize, &group); - if (rc != 0) { - CERROR("obd_set_info(mds_conn) failed %d\n", rc); - RETURN(rc); - } - ctxt = llog_get_context(&obd->obd_llogs, LLOG_UNLINK_ORIG_CTXT); LASSERT(ctxt != NULL); - rc = llog_connect(ctxt, obd->u.mds.mds_lov_desc.ld_tgt_count, - NULL, NULL, uuid); + rc = llog_connect(ctxt, count, NULL, NULL, uuid); if (rc != 0) { CERROR("%s: failed at llog_origin_connect: %d\n", obd->obd_name, rc); - RETURN(rc); + GOTO(cleanup, rc); } CWARN("MDS %s: %s now active, resetting orphans\n", obd->obd_name, uuid->uuid); - rc = mds_lov_clearorphans(&obd->u.mds, uuid); + + rc = mds_dt_clear_orphans(&obd->u.mds, uuid); if (rc != 0) { - CERROR("%s: failed at mds_lov_clearorphans: %d\n", + CERROR("%s: failed at mds_dt_clear_orphans(): %d\n", obd->obd_name, rc); - RETURN(rc); + GOTO(cleanup, rc); } + rc = 0; - RETURN(0); + EXIT; +cleanup: + up(&mds->mds_orphan_recovery_sem); + return rc; } -int mds_lov_start_synchronize(struct obd_device *obd, struct obd_uuid *uuid) +int mds_dt_start_synchronize(struct obd_device *obd, + struct obd_device *watched, + void *data) { - struct mds_lov_sync_info *mlsi; + struct mds_dt_sync_info *mlsi; int rc; ENTRY; @@ -600,14 +714,15 @@ int mds_lov_start_synchronize(struct obd_device *obd, struct obd_uuid *uuid) RETURN(-ENOMEM); mlsi->mlsi_obd = obd; - mlsi->mlsi_uuid = uuid; + mlsi->mlsi_watched = watched; + mlsi->mlsi_index = (int)data; - rc = kernel_thread(mds_lov_synchronize, mlsi, CLONE_VM | CLONE_FILES); + rc = kernel_thread(mds_dt_synchronize, mlsi, CLONE_VM | CLONE_FILES); if (rc < 0) - CERROR("%s: error starting mds_lov_synchronize: %d\n", + CERROR("%s: error starting mds_dt_synchronize(): %d\n", obd->obd_name, rc); else { - CDEBUG(D_HA, "%s: mds_lov_synchronize thread: %d\n", + CDEBUG(D_HA, "%s: mds_dt_synchronize() thread: %d\n", obd->obd_name, rc); rc = 0; } @@ -615,33 +730,38 @@ int mds_lov_start_synchronize(struct obd_device *obd, struct obd_uuid *uuid) RETURN(rc); } -int mds_notify(struct obd_device *obd, struct obd_device *watched, int active) +int mds_notify(struct obd_device *obd, struct obd_device *watched, + int active, void *data) { - struct obd_uuid *uuid; int rc = 0; ENTRY; if (!active) RETURN(0); + if (!strcmp(watched->obd_type->typ_name, LUSTRE_MDC_NAME)) + RETURN(0); + if (strcmp(watched->obd_type->typ_name, LUSTRE_OSC_NAME)) { CERROR("unexpected notification of %s %s!\n", watched->obd_type->typ_name, watched->obd_name); RETURN(-EINVAL); } - uuid = &watched->u.cli.cl_import->imp_target_uuid; if (obd->obd_recovering) { + struct obd_uuid *uuid; + uuid = &watched->u.cli.cl_import->imp_target_uuid; + CWARN("MDS %s: in recovery, not resetting orphans on %s\n", obd->obd_name, uuid->uuid); } else { - rc = mds_lov_start_synchronize(obd, uuid); + rc = mds_dt_start_synchronize(obd, watched, data); } RETURN(rc); } -int mds_set_info(struct obd_export *exp, obd_count keylen, - void *key, obd_count vallen, void *val) +int mds_dt_set_info(struct obd_export *exp, obd_count keylen, + void *key, obd_count vallen, void *val) { struct obd_device *obd = class_exp2obd(exp); struct mds_obd *mds = &obd->u.mds; @@ -652,51 +772,29 @@ int mds_set_info(struct obd_export *exp, obd_count keylen, if (KEY_IS("next_id")) { obd_id *id = (obd_id *)val; - int idx, rc; + int idx; /* XXX - this really should be vallen != (2 * sizeof(*id)) * * Just following the precedent set by lov_set_info. */ if (vallen != 2) RETURN(-EINVAL); - if ((idx = *id) != *id) + idx = *id; + if ((idx != *id) || (idx >= mds->mds_dt_desc.ld_tgt_count)) RETURN(-EINVAL); CDEBUG(D_CONFIG, "idx: %d id: %llu\n", idx, *(id + 1)); - /* The size of the LOV target table may have increased. */ - if (idx >= mds->mds_lov_desc.ld_tgt_count) { - obd_id *ids; - int oldsize, size; - - oldsize = mds->mds_lov_desc.ld_tgt_count * sizeof(*ids); - rc = mds_lov_update_desc(obd, mds->mds_osc_exp); - if (rc) - RETURN(rc); - if (idx >= mds->mds_lov_desc.ld_tgt_count) - RETURN(-EINVAL); - size = mds->mds_lov_desc.ld_tgt_count * sizeof(*ids); - OBD_ALLOC(ids, size); - if (ids == NULL) - RETURN(-ENOMEM); - memset(ids, 0, size); - if (mds->mds_lov_objids != NULL) { - memcpy(ids, mds->mds_lov_objids, oldsize); - OBD_FREE(mds->mds_lov_objids, oldsize); - } - mds->mds_lov_objids = ids; - } - - mds->mds_lov_objids[idx] = *++id; + mds->mds_dt_objids[idx] = *++id; CDEBUG(D_CONFIG, "objid: %d: %lld\n", idx, *id); /* XXX - should we be writing this out here ? */ - RETURN(mds_lov_write_objids(obd)); + RETURN(mds_dt_write_objids(obd)); } RETURN(-EINVAL); } -int mds_lov_update_config(struct obd_device *obd, int clean) +int mds_dt_update_config(struct obd_device *obd, int clean) { struct mds_obd *mds = &obd->u.mds; struct lvfs_run_ctxt saved; @@ -710,7 +808,7 @@ int mds_lov_update_config(struct obd_device *obd, int clean) RETURN(0); cfg.cfg_instance = NULL; - cfg.cfg_uuid = mds->mds_lov_uuid; + cfg.cfg_uuid = mds->mds_dt_uuid; namelen = strlen(profile) + 20; /* -clean-######### */ OBD_ALLOC(name, namelen); @@ -731,10 +829,8 @@ int mds_lov_update_config(struct obd_device *obd, int clean) ctxt = llog_get_context(&obd->obd_llogs, LLOG_CONFIG_ORIG_CTXT); rc = class_config_process_llog(ctxt, name, &cfg); pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL); - if (rc == 0) { + if (rc == 0) mds->mds_config_version = version; - rc = mds_lov_update_desc(obd, mds->mds_osc_exp); - } CWARN("Finished applying configuration log %s: %d\n", name, rc); OBD_FREE(name, namelen); @@ -764,13 +860,13 @@ int mds_convert_lov_ea(struct obd_device *obd, struct inode *inode, if (le32_to_cpu(lmm->lmm_magic) == LOV_MAGIC) RETURN(0); - CWARN("converting LOV EA on %lu/%u from V0 to V1\n", - inode->i_ino, inode->i_generation); - rc = obd_unpackmd(obd->u.mds.mds_osc_exp, &lsm, lmm, lmm_size); + CDEBUG(D_INODE, "converting LOV EA on %lu/%u from V0 to V1\n", + inode->i_ino, inode->i_generation); + rc = obd_unpackmd(obd->u.mds.mds_dt_exp, &lsm, lmm, lmm_size); if (rc < 0) GOTO(conv_end, rc); - rc = obd_packmd(obd->u.mds.mds_osc_exp, &lmm, lsm); + rc = obd_packmd(obd->u.mds.mds_dt_exp, &lmm, lsm); if (rc < 0) GOTO(conv_free, rc); lmm_size = rc; @@ -781,14 +877,13 @@ int mds_convert_lov_ea(struct obd_device *obd, struct inode *inode, GOTO(conv_free, rc); } - rc = fsfilt_set_md(obd, inode, handle, lmm, lmm_size); - + rc = fsfilt_set_md(obd, inode, handle, lmm, lmm_size, EA_LOV); err = fsfilt_commit(obd, obd->u.mds.mds_sb, inode, handle, 0); if (!rc) rc = err ? err : lmm_size; GOTO(conv_free, rc); conv_free: - obd_free_memmd(obd->u.mds.mds_osc_exp, &lsm); + obd_free_memmd(obd->u.mds.mds_dt_exp, &lsm); conv_end: return rc; } @@ -798,12 +893,12 @@ int mds_revalidate_lov_ea(struct obd_device *obd, struct inode *inode, struct lustre_msg *msg, int offset) { struct mds_obd *mds = &obd->u.mds; - struct obd_export *osc_exp = mds->mds_osc_exp; + struct obd_export *dt_exp = mds->mds_dt_exp; struct lov_mds_md *lmm= NULL; struct lov_stripe_md *lsm = NULL; - struct obdo *oa; + struct obdo *oa = NULL; struct obd_trans_info oti = {0}; - unsigned valid = 0; + obd_valid valid = 0; int lmm_size = 0, lsm_size = 0, err, rc; void *handle; ENTRY; @@ -822,7 +917,7 @@ int mds_revalidate_lov_ea(struct obd_device *obd, struct inode *inode, } lmm_size = msg->buflens[offset]; - rc = obd_unpackmd(osc_exp, &lsm, lmm, lmm_size); + rc = obd_unpackmd(dt_exp, &lsm, lmm, lmm_size); if (rc < 0) RETURN(0); @@ -838,14 +933,16 @@ int mds_revalidate_lov_ea(struct obd_device *obd, struct inode *inode, oa->o_generation = inode->i_generation; oa->o_uid = 0; /* must have 0 uid / gid on OST */ oa->o_gid = 0; + oa->o_gr = FILTER_GROUP_FIRST_MDS + mds->mds_num; oa->o_valid = OBD_MD_FLID | OBD_MD_FLGENER | OBD_MD_FLTYPE | - OBD_MD_FLMODE | OBD_MD_FLUID | OBD_MD_FLGID; + OBD_MD_FLMODE | OBD_MD_FLUID | OBD_MD_FLGID | + OBD_MD_FLGROUP; valid = OBD_MD_FLTYPE | OBD_MD_FLATIME | OBD_MD_FLMTIME | OBD_MD_FLCTIME; obdo_from_inode(oa, inode, valid); - rc = obd_revalidate_md(osc_exp, oa, lsm, &oti); + rc = obd_revalidate_md(dt_exp, oa, lsm, &oti); if (rc == 0) GOTO(out_oa, rc); if (rc < 0) { @@ -854,7 +951,7 @@ int mds_revalidate_lov_ea(struct obd_device *obd, struct inode *inode, GOTO(out_oa, rc); } - rc = obd_packmd(osc_exp, &lmm, lsm); + rc = obd_packmd(dt_exp, &lmm, lsm); if (rc < 0) GOTO(out_oa, rc); lmm_size = rc; @@ -865,13 +962,15 @@ int mds_revalidate_lov_ea(struct obd_device *obd, struct inode *inode, GOTO(out_oa, rc); } - rc = fsfilt_set_md(obd, inode, handle, lmm, lmm_size); + rc = fsfilt_set_md(obd, inode, handle, lmm, lmm_size, EA_LOV); err = fsfilt_commit(obd, inode->i_sb, inode, handle, 0); if (!rc) rc = err; + + EXIT; out_oa: obdo_free(oa); out_lsm: - obd_free_memmd(osc_exp, &lsm); - RETURN(rc); + obd_free_memmd(dt_exp, &lsm); + return rc; }