X-Git-Url: https://git.whamcloud.com/?a=blobdiff_plain;f=lustre%2Fmds%2Fmds_lov.c;h=6d789075f02381eb4d266f3e54dd123051534ade;hb=ba2ecf04beab65a6a4c8399c6a740999de218ece;hp=4dd05a35f1c86100aef4c1733741a63b58ed52a9;hpb=090c677210ee2946d99c71412e4ff762bb300f4f;p=fs%2Flustre-release.git diff --git a/lustre/mds/mds_lov.c b/lustre/mds/mds_lov.c index 4dd05a3..6d78907 100644 --- a/lustre/mds/mds_lov.c +++ b/lustre/mds/mds_lov.c @@ -30,11 +30,13 @@ #include #include +#include #include #include #include #include #include +#include #include "mds_internal.h" @@ -54,279 +56,340 @@ 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; - lock_kernel(); - 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]; - unlock_kernel(); + 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); } -static 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); + /* + * 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); } -/* update the LOV-OSC knowledge of the last used object id's */ -int mds_lov_set_nextid(struct obd_device *obd) +/* tell the LOV-OSC by how much to pre-create */ +int mds_dt_set_growth(struct mds_obd *mds, int count) { - 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); - if (rc < 0) - GOTO(out, rc); - - rc = mds_lov_clearorphans(mds, NULL /* all OSTs */); + rc = obd_set_info(mds->mds_dt_exp, strlen("growth_count"), + "growth_count", sizeof(count), &count); -out: RETURN(rc); } -/* tell the LOV-OSC by how much to pre-create */ -int mds_lov_set_growth(struct mds_obd *mds, int count) +static int mds_dt_update_desc(struct obd_device *obd, struct obd_export *lov) { - int rc; + struct mds_obd *mds = &obd->u.mds; + __u32 valsize = sizeof(mds->mds_dt_desc); + int old_count, rc = 0, i; ENTRY; - rc = obd_set_info(mds->mds_osc_exp, strlen("growth_count"), - "growth_count", sizeof(count), &count); + 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); - RETURN(rc); + /* 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_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,}; - int valsize, rc, i; - __u32 group; + struct lustre_handle conn = { 0 }; + unsigned long sec_flags = PTLRPC_SEC_FL_MDS; + 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); - 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); } - rc = obd_connect(&conn, mds->mds_osc_obd, &obd->obd_uuid); + 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); + } + } + + rc = obd_set_info(mds->mds_dt_obd->obd_self_export, + strlen("sec_flags"), "sec_flags", + sizeof(sec_flags), &sec_flags); if (rc) { - CERROR("MDS cannot connect to LOV %s (%d)\n", - lov_name, rc); - mds->mds_osc_obd = ERR_PTR(rc); + mds->mds_dt_obd = ERR_PTR(rc); RETURN(rc); } - mds->mds_osc_exp = class_conn2export(&conn); - rc = obd_register_observer(mds->mds_osc_obd, obd); + CDEBUG(D_HA, "obd: %s osc: %s lov_name: %s\n", + obd->obd_name, mds->mds_dt_obd->obd_name, lov_name); + + 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_dt_obd = ERR_PTR(rc); + RETURN(rc); + } + mds->mds_dt_exp = class_conn2export(&conn); + + 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); } - valsize = sizeof(mds->mds_lov_desc); - rc = obd_get_info(mds->mds_osc_exp, strlen("lovdesc") + 1, "lovdesc", - &valsize, &mds->mds_lov_desc); + rc = mds_dt_update_desc(obd, mds->mds_dt_exp); if (rc) GOTO(err_reg, rc); - i = lov_mds_md_size(mds->mds_lov_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; - 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 = llog_cat_initialize(obd, &obd->obd_llogs, - mds->mds_lov_desc.ld_tgt_count); + rc = obd_llog_cat_initialize(obd, &obd->obd_llogs, + 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. */ + /* + * 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) { - struct llog_ctxt *ctxt; - ctxt = llog_get_context(&obd->obd_llogs, LLOG_UNLINK_ORIG_CTXT); - rc = llog_connect(ctxt, obd->u.mds.mds_lov_desc.ld_tgt_count, - NULL, NULL, NULL); - if (rc != 0) - CERROR("faild at llog_origin_connect: %d\n", rc); - - rc = mds_cleanup_orphans(obd); - if (rc > 0) - CERROR("Cleanup %d orphans while MDS isn't recovering\n", rc); + CDEBUG(D_OTHER, "call mds_postrecov_common()\n"); + rc = mds_postrecov_common(obd); + if (rc < 0) + GOTO(err_reg, rc); + rc = 0; + } - rc = mds_lov_set_nextid(obd); + for (i = 0; i < 2; i++) { + if (!mds->mds_capa_keys[i].k_key) + break; + rc = obd_set_info(mds->mds_dt_exp, strlen("capa_key"), + "capa_key", sizeof(struct lustre_capa_key), + mds->mds_capa_keys[i].k_key); if (rc) - GOTO(err_llog, rc); + GOTO(err_reg, rc); } RETURN(rc); -err_llog: - /* cleanup all llogging subsystems */ - rc = obd_llog_finish(obd, &obd->obd_llogs, - mds->mds_lov_desc.ld_tgt_count); - if (rc) - CERROR("failed to cleanup llogging subsystems\n"); 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); @@ -339,25 +402,28 @@ int mds_iocontrol(unsigned int cmd, struct obd_export *exp, int len, struct obd_device *obd = exp->exp_obd; struct mds_obd *mds = &obd->u.mds; struct obd_ioctl_data *data = karg; - struct obd_run_ctxt saved; - struct llog_ctxt *ctxt; + struct lvfs_run_ctxt saved; int rc = 0; + ENTRY; + CDEBUG(D_INFO, "ioctl cmd %x\n", cmd); switch (cmd) { case OBD_IOC_RECORD: { char *name = data->ioc_inlbuf1; if (mds->mds_cfg_llh) RETURN(-EBUSY); - push_ctxt(&saved, &obd->obd_ctxt, NULL); - ctxt = llog_get_context(&obd->obd_llogs, LLOG_CONFIG_ORIG_CTXT); - rc = llog_create(ctxt, &mds->mds_cfg_llh, NULL, name); + push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL); + rc = llog_open(llog_get_context(&obd->obd_llogs, + LLOG_CONFIG_ORIG_CTXT), + &mds->mds_cfg_llh, NULL, name, + OBD_LLOG_FL_CREATE); if (rc == 0) llog_init_handle(mds->mds_cfg_llh, LLOG_F_IS_PLAIN, &cfg_uuid); else mds->mds_cfg_llh = NULL; - pop_ctxt(&saved, &obd->obd_ctxt, NULL); + pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL); RETURN(rc); } @@ -366,9 +432,9 @@ int mds_iocontrol(unsigned int cmd, struct obd_export *exp, int len, if (!mds->mds_cfg_llh) RETURN(-EBADF); - push_ctxt(&saved, &obd->obd_ctxt, NULL); + push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL); rc = llog_close(mds->mds_cfg_llh); - pop_ctxt(&saved, &obd->obd_ctxt, NULL); + pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL); mds->mds_cfg_llh = NULL; RETURN(rc); @@ -379,9 +445,11 @@ int mds_iocontrol(unsigned int cmd, struct obd_export *exp, int len, if (mds->mds_cfg_llh) RETURN(-EBUSY); - push_ctxt(&saved, &obd->obd_ctxt, NULL); - ctxt = llog_get_context(&obd->obd_llogs, LLOG_CONFIG_ORIG_CTXT); - rc = llog_create(ctxt, &mds->mds_cfg_llh, NULL, name); + push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL); + rc = llog_open(llog_get_context(&obd->obd_llogs, + LLOG_CONFIG_ORIG_CTXT), + &mds->mds_cfg_llh, NULL, name, + OBD_LLOG_FL_CREATE); if (rc == 0) { llog_init_handle(mds->mds_cfg_llh, LLOG_F_IS_PLAIN, NULL); @@ -389,7 +457,7 @@ int mds_iocontrol(unsigned int cmd, struct obd_export *exp, int len, rc = llog_destroy(mds->mds_cfg_llh); llog_free_handle(mds->mds_cfg_llh); } - pop_ctxt(&saved, &obd->obd_ctxt, NULL); + pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL); mds->mds_cfg_llh = NULL; RETURN(rc); @@ -401,6 +469,8 @@ int mds_iocontrol(unsigned int cmd, struct obd_export *exp, int len, if (!mds->mds_cfg_llh) RETURN(-EBADF); + /* XXX - this probably should be a parameter to this ioctl. + * For now, just use llh_max_transno for expediency. */ rec.lrh_len = llog_data_len(data->ioc_plen1); if (data->ioc_type == LUSTRE_CFG_TYPE) { @@ -421,39 +491,34 @@ int mds_iocontrol(unsigned int cmd, struct obd_export *exp, int len, RETURN(rc); } - push_ctxt(&saved, &obd->obd_ctxt, NULL); + push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL); rc = llog_write_rec(mds->mds_cfg_llh, &rec, NULL, 0, cfg_buf, -1); - pop_ctxt(&saved, &obd->obd_ctxt, NULL); + pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL); OBD_FREE(cfg_buf, data->ioc_plen1); RETURN(rc); } - case OBD_IOC_PARSE: { struct llog_ctxt *ctxt = llog_get_context(&obd->obd_llogs, LLOG_CONFIG_ORIG_CTXT); - push_ctxt(&saved, &obd->obd_ctxt, NULL); - rc = class_config_parse_llog(ctxt, data->ioc_inlbuf1, NULL); - pop_ctxt(&saved, &obd->obd_ctxt, NULL); + push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL); + rc = class_config_process_llog(ctxt, data->ioc_inlbuf1, NULL); + pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL); if (rc) RETURN(rc); RETURN(rc); } - case OBD_IOC_DUMP_LOG: { struct llog_ctxt *ctxt = llog_get_context(&obd->obd_llogs, LLOG_CONFIG_ORIG_CTXT); - push_ctxt(&saved, &obd->obd_ctxt, NULL); + push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL); rc = class_config_dump_llog(ctxt, data->ioc_inlbuf1, NULL); - pop_ctxt(&saved, &obd->obd_ctxt, NULL); - if (rc) - RETURN(rc); + pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL); RETURN(rc); } - case OBD_IOC_SET_READONLY: { void *handle; struct inode *inode = obd->u.mds.mds_sb->s_root->d_inode; @@ -463,15 +528,15 @@ int mds_iocontrol(unsigned int cmd, struct obd_export *exp, int len, handle = fsfilt_start(obd, inode, FSFILT_OP_MKNOD, NULL); LASSERT(handle); - rc = fsfilt_commit(obd, inode, handle, 1); + 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; - rc = llog_catlog_list(obd, count, data); + int count = mds->mds_dt_desc.ld_tgt_count; + rc = llog_catalog_list(obd, count, data); RETURN(rc); } @@ -480,17 +545,21 @@ 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); - int rc2; + int rc2, valsize; + __u32 group; obd_llog_finish(obd, &obd->obd_llogs, - mds->mds_lov_desc.ld_tgt_count); - push_ctxt(&saved, &ctxt->loc_exp->exp_obd->obd_ctxt, NULL); + 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_exp->exp_obd->obd_ctxt, NULL); - llog_cat_initialize(obd, &obd->obd_llogs, - mds->mds_lov_desc.ld_tgt_count); - rc2 = obd_set_info(mds->mds_osc_exp, strlen("mds_conn"), "mds_conn", - 0, NULL); + pop_ctxt(&saved, ctxt->loc_lvfs_ctxt, NULL); + obd_llog_cat_initialize(obd, &obd->obd_llogs, + 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); @@ -500,66 +569,310 @@ int mds_iocontrol(unsigned int cmd, struct obd_export *exp, int len, struct llog_ctxt *ctxt = llog_get_context(&obd->obd_llogs, LLOG_CONFIG_ORIG_CTXT); - push_ctxt(&saved, &ctxt->loc_exp->exp_obd->obd_ctxt, NULL); + push_ctxt(&saved, ctxt->loc_lvfs_ctxt, NULL); rc = llog_ioctl(ctxt, cmd, data); - pop_ctxt(&saved, &ctxt->loc_exp->exp_obd->obd_ctxt, NULL); + pop_ctxt(&saved, ctxt->loc_lvfs_ctxt, NULL); RETURN(rc); } case OBD_IOC_ABORT_RECOVERY: - CERROR("aborting recovery for device %s\n", obd->obd_name); - target_abort_recovery(obd); + 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); + } -int mds_notify(struct obd_device *obd, struct obd_device *watched, int active) +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_dt_synchronize(void *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, group; + int old_count, count, index, i; + int rc; + char name[32] = "CATLIST"; + + lock_kernel(); + + ptlrpc_daemonize(); + + SIGNAL_MASK_LOCK(current, flags); + sigfillset(¤t->blocked); + RECALC_SIGPENDING; + SIGNAL_MASK_UNLOCK(current, flags); + + unlock_kernel(); + + obd = mlsi->mlsi_obd; + 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); + + uuid = &watched->u.cli.cl_import->imp_target_uuid; + CWARN("MDS %s: %s now active, repairing the connection\n", + obd->obd_name, uuid->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); + + for (i = 0; i < 2; i++) { + if (!mds->mds_capa_keys[i].k_key) + break; + rc = obd_set_info(mds->mds_dt_exp, strlen("capa_key"), + "capa_key", sizeof(struct lustre_capa_key), + mds->mds_capa_keys[i].k_key); + 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); + + ctxt = llog_get_context(&obd->obd_llogs, LLOG_UNLINK_ORIG_CTXT); + LASSERT(ctxt != NULL); + + rc = llog_connect(ctxt, count, NULL, NULL, uuid); + if (rc != 0) { + CERROR("%s: failed at llog_origin_connect: %d\n", + obd->obd_name, rc); + GOTO(cleanup, rc); + } + + CWARN("MDS %s: %s now active, resetting orphans\n", + obd->obd_name, uuid->uuid); + + rc = mds_dt_clear_orphans(&obd->u.mds, uuid); + if (rc != 0) { + CERROR("%s: failed at mds_dt_clear_orphans(): %d\n", + obd->obd_name, rc); + GOTO(cleanup, rc); + } + rc = 0; + + EXIT; +cleanup: + up(&mds->mds_orphan_recovery_sem); + return rc; +} + +int mds_dt_start_synchronize(struct obd_device *obd, + struct obd_device *watched, + void *data) +{ + struct mds_dt_sync_info *mlsi; + int rc; + + ENTRY; + + OBD_ALLOC(mlsi, sizeof(*mlsi)); + if (mlsi == NULL) + RETURN(-ENOMEM); + + mlsi->mlsi_obd = obd; + mlsi->mlsi_watched = watched; + mlsi->mlsi_index = (int)data; + + rc = kernel_thread(mds_dt_synchronize, mlsi, CLONE_VM | CLONE_FILES); + if (rc < 0) + CERROR("%s: error starting mds_dt_synchronize(): %d\n", + obd->obd_name, rc); + else { + CDEBUG(D_HA, "%s: mds_dt_synchronize() thread: %d\n", + obd->obd_name, rc); + rc = 0; + } + + RETURN(rc); +} + +int mds_notify(struct obd_device *obd, struct obd_device *watched, + int active, void *data) +{ int rc = 0; ENTRY; if (!active) RETURN(0); - if (strcmp(watched->obd_type->typ_name, "osc")) { + if (!strcmp(watched->obd_type->typ_name, OBD_MDC_DEVICENAME)) + RETURN(0); + + if (strcmp(watched->obd_type->typ_name, OBD_OSC_DEVICENAME)) { 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 { - struct llog_ctxt *ctxt; + rc = mds_dt_start_synchronize(obd, watched, data); + } + RETURN(rc); +} - ctxt = llog_get_context(&obd->obd_llogs, LLOG_UNLINK_ORIG_CTXT); - LASSERT(ctxt != NULL); +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; + ENTRY; - rc = obd_set_info(obd->u.mds.mds_osc_exp, strlen("mds_conn"), "mds_conn", - 0, uuid); - if (rc != 0) - RETURN(rc); +#define KEY_IS(str) \ + (keylen == strlen(str) && memcmp(key, str, keylen) == 0) - ctxt = llog_get_context(&obd->obd_llogs, LLOG_UNLINK_ORIG_CTXT); - rc = llog_connect(ctxt, obd->u.mds.mds_lov_desc.ld_tgt_count, - NULL, NULL, uuid); - if (rc != 0) { - CERROR("faild at llog_origin_connect: %d\n", rc); - RETURN(rc); - } + if (KEY_IS("next_id")) { + obd_id *id = (obd_id *)val; + int idx; - CWARN("MDS %s: %s now active, resetting orphans\n", - obd->obd_name, uuid->uuid); - rc = mds_lov_clearorphans(&obd->u.mds, uuid); + /* XXX - this really should be vallen != (2 * sizeof(*id)) * + * Just following the precedent set by lov_set_info. */ + if (vallen != 2) + RETURN(-EINVAL); + + 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)); + + 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_dt_write_objids(obd)); } + + RETURN(-EINVAL); +} + +int mds_dt_update_config(struct obd_device *obd, int clean) +{ + struct mds_obd *mds = &obd->u.mds; + struct lvfs_run_ctxt saved; + struct llog_ctxt *ctxt; + struct config_llog_instance cfg; + char *profile = mds->mds_profile, *name; + int rc, version, namelen, value; + __u32 valsize; + ENTRY; + + if (profile == NULL) + RETURN(0); + cfg.cfg_instance = NULL; + cfg.cfg_uuid = mds->mds_dt_uuid; + + namelen = strlen(profile) + 20; /* -clean-######### */ + OBD_ALLOC(name, namelen); + if (name == NULL) + RETURN(-ENOMEM); + if (clean) { + version = mds->mds_config_version - 1; + sprintf(name, "%s-clean-%d", profile, version); + } else { + version = mds->mds_config_version + 1; + sprintf(name, "%s-%d", profile, version); + } + CWARN("Applying configuration log %s\n", name); + + push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL); + 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); + CWARN("Finished applying configuration log %s: %d\n", name, rc); + if (rc != 0) { + GOTO(exit, rc); + } + /* retrieve size of EA */ + rc = obd_get_info(mds->mds_md_exp, strlen("mdsize"), + "mdsize", &valsize, &value); + + if (value > mds->mds_max_mdsize) + mds->mds_max_mdsize = value; + + CDEBUG(D_INFO, "mds max md size %d \n", mds->mds_max_mdsize); + + if (rc == 0) + mds->mds_config_version = version; +exit: + OBD_FREE(name, namelen); RETURN(rc); } @@ -586,13 +899,13 @@ 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; @@ -603,14 +916,100 @@ 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); - - err = fsfilt_commit(obd, inode, handle, 0); + 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; } + +/* Must be called with i_sem held */ +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 *dt_exp = mds->mds_dt_exp; + struct lov_mds_md *lmm= NULL; + struct lov_stripe_md *lsm = NULL; + struct obdo *oa = NULL; + struct obd_trans_info oti = {0}; + obd_valid valid = 0; + int lmm_size = 0, lsm_size = 0, err, rc; + void *handle; + ENTRY; + + LASSERT(down_trylock(&inode->i_sem) != 0); + + /* XXX - add way to know if EA is already up to date & return + * without doing anything. Easy to do since we get notified of + * LOV updates. */ + + lmm = lustre_msg_buf(msg, offset, 0); + if (lmm == NULL) { + CDEBUG(D_INFO, "no space reserved for inode %lu MD\n", + inode->i_ino); + RETURN(0); + } + lmm_size = msg->buflens[offset]; + + rc = obd_unpackmd(dt_exp, &lsm, lmm, lmm_size); + if (rc < 0) + RETURN(0); + + lsm_size = rc; + + LASSERT(lsm->lsm_magic == LOV_MAGIC); + + oa = obdo_alloc(); + if (oa == NULL) + GOTO(out_lsm, rc = -ENOMEM); + oa->o_mode = S_IFREG | 0600; + oa->o_id = inode->i_ino; + 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_FLGROUP; + valid = OBD_MD_FLTYPE | OBD_MD_FLATIME | OBD_MD_FLMTIME | + OBD_MD_FLCTIME; + obdo_from_inode(oa, inode, valid); + + rc = obd_revalidate_md(dt_exp, oa, lsm, &oti); + if (rc == 0) + GOTO(out_oa, rc); + if (rc < 0) { + CERROR("Error validating LOV EA on %lu/%u: %d\n", + inode->i_ino, inode->i_generation, rc); + GOTO(out_oa, rc); + } + + rc = obd_packmd(dt_exp, &lmm, lsm); + if (rc < 0) + GOTO(out_oa, rc); + lmm_size = rc; + + handle = fsfilt_start(obd, inode, FSFILT_OP_SETATTR, NULL); + if (IS_ERR(handle)) { + rc = PTR_ERR(handle); + GOTO(out_oa, rc); + } + + 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(dt_exp, &lsm); + return rc; +}