X-Git-Url: https://git.whamcloud.com/?p=fs%2Flustre-release.git;a=blobdiff_plain;f=lustre%2Fmds%2Fmds_lov.c;h=4dd05a35f1c86100aef4c1733741a63b58ed52a9;hp=ecca88c3cdd14c3eae26bba1588098a22c83d8b7;hb=090c677210ee2946d99c71412e4ff762bb300f4f;hpb=a2a0746305449dbd925879b14dc2c0d6040bb8bf diff --git a/lustre/mds/mds_lov.c b/lustre/mds/mds_lov.c index ecca88c..4dd05a3 100644 --- a/lustre/mds/mds_lov.c +++ b/lustre/mds/mds_lov.c @@ -23,7 +23,9 @@ * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA. */ -#define EXPORT_SYMTAB +#ifndef EXPORT_SYMTAB +# define EXPORT_SYMTAB +#endif #define DEBUG_SUBSYSTEM S_MDS #include @@ -52,233 +54,459 @@ void cpu_to_le_lov_desc (struct lov_desc *ld) ld->ld_pattern = cpu_to_le32 (ld->ld_pattern); } -int mds_set_lovdesc(struct obd_device *obd, struct lov_desc *desc, - struct obd_uuid *uuidarray) +void mds_lov_update_objids(struct obd_device *obd, obd_id *ids) { struct mds_obd *mds = &obd->u.mds; - struct obd_run_ctxt saved; - struct file *f; - int tgt_count; - int rc; int i; - struct lov_desc *disk_desc; ENTRY; - tgt_count = desc->ld_tgt_count; - if (desc->ld_default_stripe_count > desc->ld_tgt_count) { - CERROR("default stripe count %u > OST count %u\n", - desc->ld_default_stripe_count, desc->ld_tgt_count); - RETURN(-EINVAL); - } - if (desc->ld_default_stripe_size & (PAGE_SIZE - 1)) { - CERROR("default stripe size "LPU64" not a multiple of %lu\n", - desc->ld_default_stripe_size, PAGE_SIZE); - RETURN(-EINVAL); - } - if (desc->ld_default_stripe_offset > desc->ld_tgt_count) { - CERROR("default stripe offset "LPU64" > max OST index %u\n", - desc->ld_default_stripe_offset, desc->ld_tgt_count); - RETURN(-EINVAL); - } - if (desc->ld_pattern != 0) { - CERROR("stripe pattern %u unknown\n", - desc->ld_pattern); - RETURN(-EINVAL); - } - - OBD_ALLOC (disk_desc, sizeof (*disk_desc)); - if (disk_desc == NULL) { - CERROR ("Can't allocate disk_desc\n"); - RETURN (-ENOMEM); - } + 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(); + EXIT; +} - *disk_desc = *desc; - cpu_to_le_lov_desc (disk_desc); +static int mds_lov_read_objids(struct obd_device *obd) +{ + struct mds_obd *mds = &obd->u.mds; + obd_id *ids; + loff_t off = 0; + int i, rc, size = mds->mds_lov_desc.ld_tgt_count * sizeof(*ids); + ENTRY; - rc = 0; - push_ctxt(&saved, &mds->mds_ctxt, NULL); + if (mds->mds_lov_objids != NULL) + RETURN(0); - /* Bug 1186: FIXME: if there is an existing LOVDESC, verify new - * tgt_count > old */ - f = filp_open("LOVDESC", O_CREAT|O_RDWR, 0644); - if (IS_ERR(f)) { - CERROR("Cannot open/create LOVDESC file\n"); - GOTO(out, rc = PTR_ERR(f)); - } + OBD_ALLOC(ids, size); + if (ids == NULL) + RETURN(-ENOMEM); + mds->mds_lov_objids = ids; - rc = lustre_fwrite(f, (char *)disk_desc, sizeof(*disk_desc), &f->f_pos); - if (filp_close(f, 0)) - CERROR("Error closing LOVDESC file\n"); - if (rc != sizeof(*desc)) { - CERROR("Cannot open/create LOVDESC file\n"); - if (rc >= 0) - rc = -EIO; - GOTO(out, rc); + if (mds->mds_lov_objid_filp->f_dentry->d_inode->i_size == 0) + RETURN(0); + rc = fsfilt_read_record(obd, mds->mds_lov_objid_filp, ids, size, &off); + if (rc < 0) { + CERROR("Error reading objids %d\n", rc); + } else { + mds->mds_lov_objids_valid = 1; + rc = 0; } - /* Bug 1186: FIXME: if there is an existing LOVTGTS, verify - * existing UUIDs same */ - f = filp_open("LOVTGTS", O_CREAT|O_RDWR, 0644); - if (IS_ERR(f)) { - CERROR("Cannot open/create LOVTGTS file\n"); - GOTO(out, rc = PTR_ERR(f)); - } + 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); - rc = 0; - for (i = 0; i < tgt_count ; i++) { - rc = lustre_fwrite(f, uuidarray[i].uuid, - sizeof(uuidarray[i]), &f->f_pos); - if (rc != sizeof(uuidarray[i])) { - CERROR("cannot write LOV UUID %s (%d)\n", - uuidarray[i].uuid, i); - if (rc >= 0) - rc = -EIO; - break; - } - rc = 0; - } - if (filp_close(f, 0)) - CERROR("Error closing LOVTGTS file\n"); + RETURN(rc); +} - memcpy(&mds->mds_lov_desc, desc, sizeof *desc); - mds->mds_has_lov_desc = 1; - /* XXX the MDS should not really know about this */ - mds->mds_max_mdsize = lov_mds_md_size(desc->ld_tgt_count); - mds->mds_max_cookiesize = desc->ld_tgt_count*sizeof(struct llog_cookie); +int mds_lov_write_objids(struct obd_device *obd) +{ + struct mds_obd *mds = &obd->u.mds; + loff_t off = 0; + int i, rc, size = mds->mds_lov_desc.ld_tgt_count * sizeof(obd_id); + ENTRY; -out: - pop_ctxt(&saved, &mds->mds_ctxt, NULL); - OBD_FREE (disk_desc, sizeof (*disk_desc)); + for (i = 0; i < mds->mds_lov_desc.ld_tgt_count; i++) + CDEBUG(D_INFO, "writing last object "LPU64" for idx %d\n", + mds->mds_lov_objids[i], i); + rc = fsfilt_write_record(obd, mds->mds_lov_objid_filp, + mds->mds_lov_objids, size, &off, 0); RETURN(rc); } -int mds_get_lovdesc(struct mds_obd *mds, struct lov_desc *desc) +static int mds_lov_clearorphans(struct mds_obd *mds, struct obd_uuid *ost_uuid) { - struct obd_run_ctxt saved; - struct file *f; int rc; + struct obdo oa; + struct obd_trans_info oti = {0}; + struct lov_stripe_md *empty_ea = NULL; ENTRY; - push_ctxt(&saved, &mds->mds_ctxt, NULL); - f = filp_open("LOVDESC", O_RDONLY, 0644); - if (IS_ERR(f)) { - CERROR("Cannot open LOVDESC file\n"); - GOTO(out, rc = PTR_ERR(f)); + LASSERT(mds->mds_lov_objids != NULL); + + /* 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; + if (ost_uuid != NULL) { + 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); - rc = lustre_fread(f, (char *)desc, sizeof(*desc), &f->f_pos); - if (filp_close(f, 0)) - CERROR("Error closing LOVDESC file\n"); + RETURN(rc); +} - if (rc != sizeof(*desc)) { - CERROR("Cannot read LOVDESC file: rc = %d\n", rc); - GOTO(out, rc = -EIO); - } else - rc = 0; +/* 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; - le_lov_desc_to_cpu (desc); /* convert to my byte order */ + LASSERT(!obd->obd_recovering); - EXIT; -out: - pop_ctxt(&saved, &mds->mds_ctxt, NULL); + LASSERT(mds->mds_lov_objids != NULL); - return rc; + 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 */); + +out: + RETURN(rc); } -int mds_get_lovtgts(struct mds_obd *mds, int tgt_count, - struct obd_uuid *uuidarray) +/* tell the LOV-OSC by how much to pre-create */ +int mds_lov_set_growth(struct mds_obd *mds, int count) { - struct obd_run_ctxt saved; - struct file *f; int rc; - int rc2; + ENTRY; + + rc = obd_set_info(mds->mds_osc_exp, strlen("growth_count"), + "growth_count", sizeof(count), &count); + + RETURN(rc); +} + +int mds_lov_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; + ENTRY; + + if (IS_ERR(mds->mds_osc_obd)) + RETURN(PTR_ERR(mds->mds_osc_obd)); - push_ctxt(&saved, &mds->mds_ctxt, NULL); - f = filp_open("LOVTGTS", O_RDONLY, 0644); - if (IS_ERR(f)) { - CERROR("Cannot open LOVTGTS file\n"); - GOTO(out, rc = PTR_ERR(f)); + if (mds->mds_osc_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); + RETURN(-ENOTCONN); } - rc = lustre_fread(f, (char *)uuidarray, tgt_count * sizeof(*uuidarray), - &f->f_pos); - rc2 = filp_close(f, 0); - if (rc2) - CERROR("Error closing LOVTGTS file: rc = %d\n", rc2); + rc = obd_connect(&conn, mds->mds_osc_obd, &obd->obd_uuid); + if (rc) { + CERROR("MDS cannot connect to LOV %s (%d)\n", + lov_name, rc); + mds->mds_osc_obd = ERR_PTR(rc); + RETURN(rc); + } + mds->mds_osc_exp = class_conn2export(&conn); - if (rc != tgt_count * sizeof(*uuidarray)) { - CERROR("Error reading LOVTGTS file: rc = %d\n", rc); - if (rc >= 0) - rc = -EIO; - GOTO(out, rc); - } else - rc = 0; - EXIT; -out: - pop_ctxt(&saved, &mds->mds_ctxt, NULL); + rc = obd_register_observer(mds->mds_osc_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); + 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); + 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); + 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 (!rc) { + for (i = 0; i < mds->mds_lov_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); + 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) { + 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); + + rc = mds_lov_set_nextid(obd); + if (rc) + GOTO(err_llog, 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); +err_discon: + obd_disconnect(mds->mds_osc_exp, 0); + mds->mds_osc_exp = NULL; + mds->mds_osc_obd = ERR_PTR(rc); + RETURN(rc); +} + +int mds_lov_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) { + /* 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"); + + obd_register_observer(mds->mds_osc_obd, NULL); + + rc = obd_disconnect(mds->mds_osc_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; + } RETURN(rc); } -int mds_iocontrol(unsigned int cmd, struct lustre_handle *conn, - int len, void *karg, void *uarg) +int mds_iocontrol(unsigned int cmd, struct obd_export *exp, int len, + void *karg, void *uarg) { - struct obd_device *obd = class_conn2obd(conn); + static struct obd_uuid cfg_uuid = { .uuid = "config_uuid" }; + struct obd_device *obd = exp->exp_obd; + struct mds_obd *mds = &obd->u.mds; struct obd_ioctl_data *data = karg; - struct lov_desc *desc; - struct obd_uuid *uuidarray; - int count; - int rc; + struct obd_run_ctxt saved; + struct llog_ctxt *ctxt; + int rc = 0; switch (cmd) { - case OBD_IOC_LOV_SET_CONFIG: - desc = (struct lov_desc *)data->ioc_inlbuf1; - if (sizeof(*desc) > data->ioc_inllen1) { - CERROR("descriptor size wrong\n"); - RETURN(-EINVAL); - } + 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); + 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); - count = desc->ld_tgt_count; - uuidarray = (struct obd_uuid *)data->ioc_inlbuf2; - if (sizeof(*uuidarray) * count != data->ioc_inllen2) { - CERROR("UUID array size wrong\n"); - RETURN(-EINVAL); + RETURN(rc); + } + + case OBD_IOC_ENDRECORD: { + if (!mds->mds_cfg_llh) + RETURN(-EBADF); + + push_ctxt(&saved, &obd->obd_ctxt, NULL); + rc = llog_close(mds->mds_cfg_llh); + pop_ctxt(&saved, &obd->obd_ctxt, NULL); + + mds->mds_cfg_llh = NULL; + RETURN(rc); + } + + case OBD_IOC_CLEAR_LOG: { + 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); + if (rc == 0) { + llog_init_handle(mds->mds_cfg_llh, LLOG_F_IS_PLAIN, + NULL); + + rc = llog_destroy(mds->mds_cfg_llh); + llog_free_handle(mds->mds_cfg_llh); } - rc = mds_set_lovdesc(obd, desc, uuidarray); + pop_ctxt(&saved, &obd->obd_ctxt, NULL); + mds->mds_cfg_llh = NULL; RETURN(rc); - case OBD_IOC_LOV_GET_CONFIG: - desc = (struct lov_desc *)data->ioc_inlbuf1; - if (sizeof(*desc) > data->ioc_inllen1) { - CERROR("descriptor size wrong\n"); + } + + case OBD_IOC_DORECORD: { + char *cfg_buf; + struct llog_rec_hdr rec; + if (!mds->mds_cfg_llh) + RETURN(-EBADF); + + rec.lrh_len = llog_data_len(data->ioc_plen1); + + if (data->ioc_type == LUSTRE_CFG_TYPE) { + rec.lrh_type = OBD_CFG_REC; + } else if (data->ioc_type == PORTALS_CFG_TYPE) { + rec.lrh_type = PTL_CFG_REC; + } else { + CERROR("unknown cfg record type:%d \n", data->ioc_type); RETURN(-EINVAL); } - count = desc->ld_tgt_count; - uuidarray = (struct obd_uuid *)data->ioc_inlbuf2; - if (sizeof(*uuidarray) * count != data->ioc_inllen2) { - CERROR("UUID array size wrong\n"); + OBD_ALLOC(cfg_buf, data->ioc_plen1); + if (cfg_buf == NULL) RETURN(-EINVAL); + rc = copy_from_user(cfg_buf, data->ioc_pbuf1, data->ioc_plen1); + if (rc) { + OBD_FREE(cfg_buf, data->ioc_plen1); + RETURN(rc); } - rc = mds_get_lovdesc(&obd->u.mds, desc); - if (desc->ld_tgt_count > count) { - CERROR("UUID array size too small\n"); - RETURN(-ENOSPC); - } - rc = mds_get_lovtgts(&obd->u.mds, desc->ld_tgt_count, - uuidarray); + + push_ctxt(&saved, &obd->obd_ctxt, NULL); + rc = llog_write_rec(mds->mds_cfg_llh, &rec, NULL, 0, + cfg_buf, -1); + pop_ctxt(&saved, &obd->obd_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); + 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); + rc = class_config_dump_llog(ctxt, data->ioc_inlbuf1, NULL); + pop_ctxt(&saved, &obd->obd_ctxt, NULL); + if (rc) + RETURN(rc); + + RETURN(rc); + } case OBD_IOC_SET_READONLY: { + void *handle; + struct inode *inode = obd->u.mds.mds_sb->s_root->d_inode; BDEVNAME_DECLARE_STORAGE(tmp); CERROR("setting device %s read-only\n", - ll_bdevname(obd->u.mds.mds_sb->s_dev, tmp)); - dev_set_rdonly(obd->u.mds.mds_sb->s_dev, 2); + ll_bdevname(obd->u.mds.mds_sb, tmp)); + + handle = fsfilt_start(obd, inode, FSFILT_OP_MKNOD, NULL); + LASSERT(handle); + rc = fsfilt_commit(obd, inode, handle, 1); + + dev_set_rdonly(ll_sbdev(obd->u.mds.mds_sb), 2); RETURN(0); } + case OBD_IOC_CATLOGLIST: { + int count = mds->mds_lov_desc.ld_tgt_count; + rc = llog_catlog_list(obd, count, data); + RETURN(rc); + + } + case OBD_IOC_LLOG_CHECK: + case OBD_IOC_LLOG_CANCEL: + case OBD_IOC_LLOG_REMOVE: { + struct llog_ctxt *ctxt = + llog_get_context(&obd->obd_llogs, LLOG_CONFIG_ORIG_CTXT); + int rc2; + + 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); + 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); + if (!rc) + rc = rc2; + RETURN(rc); + } + case OBD_IOC_LLOG_INFO: + case OBD_IOC_LLOG_PRINT: { + 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); + rc = llog_ioctl(ctxt, cmd, data); + pop_ctxt(&saved, &ctxt->loc_exp->exp_obd->obd_ctxt, NULL); + + RETURN(rc); + } + case OBD_IOC_ABORT_RECOVERY: CERROR("aborting recovery for device %s\n", obd->obd_name); target_abort_recovery(obd); @@ -289,3 +517,100 @@ int mds_iocontrol(unsigned int cmd, struct lustre_handle *conn, } RETURN(0); } + +int mds_notify(struct obd_device *obd, struct obd_device *watched, int active) +{ + struct obd_uuid *uuid; + int rc = 0; + ENTRY; + + if (!active) + RETURN(0); + + if (strcmp(watched->obd_type->typ_name, "osc")) { + 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) { + CWARN("MDS %s: in recovery, not resetting orphans on %s\n", + obd->obd_name, uuid->uuid); + } else { + struct llog_ctxt *ctxt; + + ctxt = llog_get_context(&obd->obd_llogs, LLOG_UNLINK_ORIG_CTXT); + LASSERT(ctxt != NULL); + + rc = obd_set_info(obd->u.mds.mds_osc_exp, strlen("mds_conn"), "mds_conn", + 0, uuid); + if (rc != 0) + RETURN(rc); + + 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); + } + + CWARN("MDS %s: %s now active, resetting orphans\n", + obd->obd_name, uuid->uuid); + rc = mds_lov_clearorphans(&obd->u.mds, uuid); + } + RETURN(rc); +} + +/* Convert the on-disk LOV EA structre. + * We always try to convert from an old LOV EA format to the common in-memory + * (lsm) format (obd_unpackmd() understands the old on-disk (lmm) format) and + * then convert back to the new on-disk format and save it back to disk + * (obd_packmd() only ever saves to the new on-disk format) so we don't have + * to convert it each time this inode is accessed. + * + * This function is a bit interesting in the error handling. We can safely + * ship the old lmm to the client in case of failure, since it uses the same + * obd_unpackmd() code and can do the conversion if the MDS fails for some + * reason. We will not delete the old lmm data until we have written the + * new format lmm data in fsfilt_set_md(). */ +int mds_convert_lov_ea(struct obd_device *obd, struct inode *inode, + struct lov_mds_md *lmm, int lmm_size) +{ + struct lov_stripe_md *lsm = NULL; + void *handle; + int rc, err; + ENTRY; + + 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); + if (rc < 0) + GOTO(conv_end, rc); + + rc = obd_packmd(obd->u.mds.mds_osc_exp, &lmm, lsm); + if (rc < 0) + GOTO(conv_free, rc); + lmm_size = rc; + + handle = fsfilt_start(obd, inode, FSFILT_OP_SETATTR, NULL); + if (IS_ERR(handle)) { + rc = PTR_ERR(handle); + GOTO(conv_free, rc); + } + + rc = fsfilt_set_md(obd, inode, handle, lmm, lmm_size); + + err = fsfilt_commit(obd, 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); +conv_end: + return rc; +}