X-Git-Url: https://git.whamcloud.com/?p=fs%2Flustre-release.git;a=blobdiff_plain;f=lustre%2Flmv%2Flmv_obd.c;h=bd74795355f3015da6e973741491f6e1a7248ec6;hp=3fe8196c21a5ccfc3321e3dffbed82de69720eb3;hb=57b9e88690b51cd8ccd9b2fd8a09f533ca09a7bd;hpb=89891a7b931071eccb8c877bcbcd05c4635076e9 diff --git a/lustre/lmv/lmv_obd.c b/lustre/lmv/lmv_obd.c index 3fe8196..bd74795 100644 --- a/lustre/lmv/lmv_obd.c +++ b/lustre/lmv/lmv_obd.c @@ -31,6 +31,7 @@ #include #include #include +#include #else #include #endif @@ -47,8 +48,29 @@ #include #include #include +#include #include "lmv_internal.h" +/* not defined for liblustre building */ +#if !defined(ATOMIC_INIT) +#define ATOMIC_INIT(val) { (val) } +#endif + +/* object cache. */ +kmem_cache_t *obj_cache; +atomic_t obj_cache_count = ATOMIC_INIT(0); + +static void lmv_activate_target(struct lmv_obd *lmv, + struct lmv_tgt_desc *tgt, + int activate) +{ + if (tgt->active == activate) + return; + + tgt->active = activate; + lmv->desc.ld_active_tgt_count += (activate ? 1 : -1); +} + /* Error codes: * * -EINVAL : UUID can't be found in the LMV's target list @@ -58,8 +80,8 @@ static int lmv_set_mdc_active(struct lmv_obd *lmv, struct obd_uuid *uuid, int activate) { - struct obd_device *obd; struct lmv_tgt_desc *tgt; + struct obd_device *obd; int i, rc = 0; ENTRY; @@ -68,18 +90,22 @@ static int lmv_set_mdc_active(struct lmv_obd *lmv, struct obd_uuid *uuid, spin_lock(&lmv->lmv_lock); for (i = 0, tgt = lmv->tgts; i < lmv->desc.ld_tgt_count; i++, tgt++) { + if (tgt->ltd_exp == NULL) + continue; + CDEBUG(D_INFO, "lmv idx %d is %s conn "LPX64"\n", i, tgt->uuid.uuid, tgt->ltd_exp->exp_handle.h_cookie); - if (strncmp(uuid->uuid, tgt->uuid.uuid, sizeof uuid->uuid) == 0) + + if (obd_uuid_equals(uuid, &tgt->uuid)) break; } if (i == lmv->desc.ld_tgt_count) - GOTO(out, rc = -EINVAL); + GOTO(out_lmv_lock, rc = -EINVAL); obd = class_exp2obd(tgt->ltd_exp); if (obd == NULL) - GOTO(out, rc = -ENOTCONN); + GOTO(out_lmv_lock, rc = -ENOTCONN); CDEBUG(D_INFO, "Found OBD %s=%s device %d (%p) type %s at LMV idx %d\n", obd->obd_name, obd->obd_uuid.uuid, obd->obd_minor, obd, @@ -89,39 +115,38 @@ static int lmv_set_mdc_active(struct lmv_obd *lmv, struct obd_uuid *uuid, if (tgt->active == activate) { CDEBUG(D_INFO, "OBD %p already %sactive!\n", obd, activate ? "" : "in"); - GOTO(out, rc); + GOTO(out_lmv_lock, rc); } - CDEBUG(D_INFO, "Marking OBD %p %sactive\n", obd, activate ? "" : "in"); + CDEBUG(D_INFO, "Marking OBD %p %sactive\n", + obd, activate ? "" : "in"); - tgt->active = activate; - if (activate) - lmv->desc.ld_active_tgt_count++; - else - lmv->desc.ld_active_tgt_count--; + lmv_activate_target(lmv, tgt, activate); EXIT; - out: + + out_lmv_lock: spin_unlock(&lmv->lmv_lock); return rc; } static int lmv_notify(struct obd_device *obd, struct obd_device *watched, - int active) + int active, void *data) { - int rc; struct obd_uuid *uuid; + int rc; + ENTRY; if (strcmp(watched->obd_type->typ_name, LUSTRE_MDC_NAME)) { CERROR("unexpected notification of %s %s!\n", watched->obd_type->typ_name, watched->obd_name); - return -EINVAL; + RETURN(-EINVAL); } uuid = &watched->u.cli.cl_import->imp_target_uuid; - /* Set MDC as active before notifying the observer, so the - * observer can use the MDC normally. + /* Set MDC as active before notifying the observer, so the observer can + * use the MDC normally. */ rc = lmv_set_mdc_active(&obd->u.lmv, uuid, active); if (rc) { @@ -132,7 +157,7 @@ static int lmv_notify(struct obd_device *obd, struct obd_device *watched, if (obd->obd_observer) /* Pass the notification up the chain. */ - rc = obd_notify(obd->obd_observer, watched, active); + rc = obd_notify(obd->obd_observer, watched, active, data); RETURN(rc); } @@ -145,17 +170,18 @@ int lmv_attach(struct obd_device *dev, obd_count len, void *data) lprocfs_init_vars(lmv, &lvars); rc = lprocfs_obd_attach(dev, lvars.obd_vars); - if (rc == 0) { #ifdef __KERNEL__ + if (rc == 0) { struct proc_dir_entry *entry; - entry = create_proc_entry("target_obd", 0444, dev->obd_proc_entry); + entry = create_proc_entry("target_obd_status", 0444, + dev->obd_proc_entry); if (entry == NULL) RETURN(-ENOMEM); - /* entry->proc_fops = &lmv_proc_target_fops; */ + entry->proc_fops = &lmv_proc_target_fops; entry->data = dev; -#endif } +#endif RETURN (rc); } @@ -164,15 +190,18 @@ int lmv_detach(struct obd_device *dev) return lprocfs_obd_detach(dev); } -/* This is fake connect function. Its purpose is to initialize lmv and - * say caller that everything is okay. Real connection will be performed - * later. */ +/* this is fake connect function. Its purpose is to initialize lmv and say + * caller that everything is okay. Real connection will be performed later. */ static int lmv_connect(struct lustre_handle *conn, struct obd_device *obd, - struct obd_uuid *cluuid) + struct obd_uuid *cluuid, struct obd_connect_data *data, + unsigned long flags) { +#ifdef __KERNEL__ + struct proc_dir_entry *lmv_proc_dir; +#endif struct lmv_obd *lmv = &obd->u.lmv; struct obd_export *exp; - int rc; + int rc = 0; ENTRY; rc = class_connect(conn, obd, cluuid); @@ -182,7 +211,8 @@ static int lmv_connect(struct lustre_handle *conn, struct obd_device *obd, } exp = class_conn2export(conn); - /* We don't want to actually do the underlying connections more than + + /* we don't want to actually do the underlying connections more than * once, so keep track. */ lmv->refcount++; if (lmv->refcount > 1) { @@ -190,11 +220,41 @@ static int lmv_connect(struct lustre_handle *conn, struct obd_device *obd, RETURN(0); } - lmv->cluuid = *cluuid; - lmv->connected = 0; lmv->exp = exp; + lmv->connected = 0; + lmv->cluuid = *cluuid; + lmv->connect_flags = flags; + sema_init(&lmv->init_sem, 1); + if (data) + memcpy(&lmv->conn_data, data, sizeof(*data)); - RETURN(0); +#ifdef __KERNEL__ + lmv_proc_dir = lprocfs_register("target_obds", obd->obd_proc_entry, + NULL, NULL); + if (IS_ERR(lmv_proc_dir)) { + CERROR("could not register /proc/fs/lustre/%s/%s/target_obds.", + obd->obd_type->typ_name, obd->obd_name); + lmv_proc_dir = NULL; + } +#endif + + /* + * all real clients shouls perform actual connection rightaway, because + * it is possible, that LMV will not have opportunity to connect + * targets, as MDC stuff will bit called directly, for instance while + * reading ../mdc/../kbytesfree procfs file, etc. + */ + if (flags & OBD_OPT_REAL_CLIENT) + rc = lmv_check_connect(obd); + +#ifdef __KERNEL__ + if (rc) { + if (lmv_proc_dir) + lprocfs_remove(lmv_proc_dir); + } +#endif + + RETURN(rc); } void lmv_set_timeouts(struct obd_device *obd) @@ -213,23 +273,36 @@ void lmv_set_timeouts(struct obd_device *obd) for (i = 0, tgts = lmv->tgts; i < lmv->desc.ld_tgt_count; i++, tgts++) { if (tgts->ltd_exp == NULL) continue; + obd_set_info(tgts->ltd_exp, strlen("inter_mds"), "inter_mds", 0, NULL); } } -/* Performs a check if passed obd is connected. If no - connect it. */ -int lmv_check_connect(struct obd_device *obd) { +#define MAX_STRING_SIZE 128 + +/* performs a check if passed obd is connected. If no - connect it. */ +int lmv_check_connect(struct obd_device *obd) +{ +#ifdef __KERNEL__ + struct proc_dir_entry *lmv_proc_dir; +#endif struct lmv_obd *lmv = &obd->u.lmv; - struct obd_uuid *cluuid; struct lmv_tgt_desc *tgts; + struct obd_uuid *cluuid; struct obd_export *exp; int rc, rc2, i; + ENTRY; if (lmv->connected) - return 0; - - lmv->connected = 1; + RETURN(0); + + down(&lmv->init_sem); + if (lmv->connected) { + up(&lmv->init_sem); + RETURN(0); + } + cluuid = &lmv->cluuid; exp = lmv->exp; @@ -237,16 +310,16 @@ int lmv_check_connect(struct obd_device *obd) { cluuid->uuid, obd->obd_name); for (i = 0, tgts = lmv->tgts; i < lmv->desc.ld_tgt_count; i++, tgts++) { - struct obd_device *tgt_obd; - struct obd_uuid lmv_osc_uuid = { "LMV_OSC_UUID" }; + struct obd_uuid lmv_mdc_uuid = { "LMV_MDC_UUID" }; struct lustre_handle conn = {0, }; + struct obd_device *tgt_obd; LASSERT(tgts != NULL); tgt_obd = class_find_client_obd(&tgts->uuid, LUSTRE_MDC_NAME, &obd->obd_uuid); if (!tgt_obd) { - CERROR("Target %s not attached\n", tgts->uuid.uuid); + CERROR("target %s not attached\n", tgts->uuid.uuid); GOTO(out_disc, rc = -EINVAL); } @@ -264,13 +337,14 @@ int lmv_check_connect(struct obd_device *obd) { cluuid->uuid); if (!tgt_obd->obd_set_up) { - CERROR("Target %s not set up\n", tgts->uuid.uuid); + CERROR("target %s not set up\n", tgts->uuid.uuid); GOTO(out_disc, rc = -EINVAL); } - rc = obd_connect(&conn, tgt_obd, &lmv_osc_uuid); + rc = obd_connect(&conn, tgt_obd, &lmv_mdc_uuid, &lmv->conn_data, + lmv->connect_flags); if (rc) { - CERROR("Target %s connect error %d\n", + CERROR("target %s connect error %d\n", tgts->uuid.uuid, rc); GOTO(out_disc, rc); } @@ -278,10 +352,10 @@ int lmv_check_connect(struct obd_device *obd) { obd_init_ea_size(tgts->ltd_exp, lmv->max_easize, lmv->max_cookiesize); - + rc = obd_register_observer(tgt_obd, obd); if (rc) { - CERROR("Target %s register_observer error %d\n", + CERROR("target %s register_observer error %d\n", tgts->uuid.uuid, rc); obd_disconnect(tgts->ltd_exp, 0); GOTO(out_disc, rc); @@ -289,16 +363,44 @@ int lmv_check_connect(struct obd_device *obd) { lmv->desc.ld_active_tgt_count++; tgts->active = 1; - + CDEBUG(D_OTHER, "connected to %s(%s) successfully (%d)\n", tgt_obd->obd_name, tgt_obd->obd_uuid.uuid, atomic_read(&obd->obd_refcount)); + +#ifdef __KERNEL__ + lmv_proc_dir = lprocfs_srch(obd->obd_proc_entry, "target_obds"); + if (lmv_proc_dir) { + struct obd_device *mdc_obd = class_conn2obd(&conn); + struct proc_dir_entry *mdc_symlink; + char name[MAX_STRING_SIZE + 1]; + + LASSERT(mdc_obd != NULL); + LASSERT(mdc_obd->obd_type != NULL); + LASSERT(mdc_obd->obd_type->typ_name != NULL); + name[MAX_STRING_SIZE] = '\0'; + snprintf(name, MAX_STRING_SIZE, "../../../%s/%s", + mdc_obd->obd_type->typ_name, + mdc_obd->obd_name); + mdc_symlink = proc_symlink(mdc_obd->obd_name, + lmv_proc_dir, name); + if (mdc_symlink == NULL) { + CERROR("could not register LMV target " + "/proc/fs/lustre/%s/%s/target_obds/%s.", + obd->obd_type->typ_name, obd->obd_name, + mdc_obd->obd_name); + lprocfs_remove(lmv_proc_dir); + lmv_proc_dir = NULL; + } + } +#endif } lmv_set_timeouts(obd); - class_export_put(exp); - return 0; + lmv->connected = 1; + up(&lmv->init_sem); + RETURN(0); out_disc: while (i-- > 0) { @@ -310,17 +412,22 @@ int lmv_check_connect(struct obd_device *obd) { uuid = tgts->uuid; rc2 = obd_disconnect(tgts->ltd_exp, 0); if (rc2) - CERROR("error: LMV target %s disconnect on MDT idx %d: " - "rc = %d\n", uuid.uuid, i, rc2); + CERROR("error: LMV target %s disconnect on MDC idx %d: " + "error %d\n", uuid.uuid, i, rc2); } class_disconnect(exp, 0); - RETURN (rc); + up(&lmv->init_sem); + return rc; } -static int lmv_disconnect(struct obd_export *exp, int flags) +static int lmv_disconnect(struct obd_export *exp, unsigned long flags) { struct obd_device *obd = class_exp2obd(exp); struct lmv_obd *lmv = &obd->u.lmv; + +#ifdef __KERNEL__ + struct proc_dir_entry *lmv_proc_dir; +#endif int rc, i; ENTRY; @@ -332,27 +439,40 @@ static int lmv_disconnect(struct obd_export *exp, int flags) if (lmv->refcount != 0) goto out_local; +#ifdef __KERNEL__ + lmv_proc_dir = lprocfs_srch(obd->obd_proc_entry, "target_obds"); +#endif + for (i = 0; i < lmv->desc.ld_tgt_count; i++) { + struct obd_device *mdc_obd; + if (lmv->tgts[i].ltd_exp == NULL) continue; - if (obd->obd_no_recov) { - /* Pass it on to our clients. - * XXX This should be an argument to disconnect, - * XXX not a back-door flag on the OBD. Ah well. - */ - struct obd_device *mdc_obd; - mdc_obd = class_exp2obd(lmv->tgts[i].ltd_exp); - if (mdc_obd) - mdc_obd->obd_no_recov = 1; - } + mdc_obd = class_exp2obd(lmv->tgts[i].ltd_exp); + + if (mdc_obd) + mdc_obd->obd_no_recov = obd->obd_no_recov; + +#ifdef __KERNEL__ + if (lmv_proc_dir) { + struct proc_dir_entry *mdc_symlink; + mdc_symlink = lprocfs_srch(lmv_proc_dir, mdc_obd->obd_name); + if (mdc_symlink) { + lprocfs_remove(mdc_symlink); + } else { + CERROR("/proc/fs/lustre/%s/%s/target_obds/%s missing\n", + obd->obd_type->typ_name, obd->obd_name, + mdc_obd->obd_name); + } + } +#endif CDEBUG(D_OTHER, "disconnected from %s(%s) successfully\n", lmv->tgts[i].ltd_exp->exp_obd->obd_name, lmv->tgts[i].ltd_exp->exp_obd->obd_uuid.uuid); obd_register_observer(lmv->tgts[i].ltd_exp->exp_obd, NULL); - rc = obd_disconnect(lmv->tgts[i].ltd_exp, flags); if (rc) { if (lmv->tgts[i].active) { @@ -361,19 +481,28 @@ static int lmv_disconnect(struct obd_export *exp, int flags) } rc = 0; } - if (lmv->tgts[i].active) { - lmv->desc.ld_active_tgt_count--; - lmv->tgts[i].active = 0; - } + + lmv_activate_target(lmv, &lmv->tgts[i], 0); lmv->tgts[i].ltd_exp = NULL; } +#ifdef __KERNEL__ + if (lmv_proc_dir) { + lprocfs_remove(lmv_proc_dir); + } else { + CERROR("/proc/fs/lustre/%s/%s/target_obds missing\n", + obd->obd_type->typ_name, obd->obd_name); + } +#endif + out_local: - /* This is the case when no real connection is established by + /* this is the case when no real connection is established by * lmv_check_connect(). */ if (!lmv->connected) class_export_put(exp); rc = class_disconnect(exp, 0); + if (lmv->refcount == 0) + lmv->connected = 0; RETURN(rc); } @@ -383,7 +512,6 @@ static int lmv_iocontrol(unsigned int cmd, struct obd_export *exp, struct obd_device *obddev = class_exp2obd(exp); struct lmv_obd *lmv = &obddev->u.lmv; int i, rc = 0, set = 0; - ENTRY; if (lmv->desc.ld_tgt_count == 0) @@ -392,8 +520,10 @@ static int lmv_iocontrol(unsigned int cmd, struct obd_export *exp, for (i = 0; i < lmv->desc.ld_tgt_count; i++) { int err; - err = obd_iocontrol(cmd, lmv->tgts[i].ltd_exp, - len, karg, uarg); + if (lmv->tgts[i].ltd_exp == NULL) + continue; + + err = obd_iocontrol(cmd, lmv->tgts[i].ltd_exp, len, karg, uarg); if (err) { if (lmv->tgts[i].active) { CERROR("error: iocontrol MDC %s on MDT" @@ -413,44 +543,44 @@ static int lmv_iocontrol(unsigned int cmd, struct obd_export *exp, static int lmv_setup(struct obd_device *obd, obd_count len, void *buf) { - struct lustre_cfg *lcfg = buf; + int i, rc = 0; struct lmv_desc *desc; - struct lmv_obd *lmv = &obd->u.lmv; struct obd_uuid *uuids; struct lmv_tgt_desc *tgts; - int i; - int rc = 0; + struct obd_device *tgt_obd; + struct lustre_cfg *lcfg = buf; + struct lmv_obd *lmv = &obd->u.lmv; ENTRY; - if (lcfg->lcfg_inllen1 < 1) { + if (LUSTRE_CFG_BUFLEN(lcfg, 1) < 1) { CERROR("LMV setup requires a descriptor\n"); RETURN(-EINVAL); } - if (lcfg->lcfg_inllen2 < 1) { - CERROR("LMV setup requires an OST UUID list\n"); + if (LUSTRE_CFG_BUFLEN(lcfg, 2) < 1) { + CERROR("LMV setup requires an MDT UUID list\n"); RETURN(-EINVAL); } - desc = (struct lmv_desc *)lcfg->lcfg_inlbuf1; - if (sizeof(*desc) > lcfg->lcfg_inllen1) { + desc = (struct lmv_desc *)lustre_cfg_buf(lcfg, 1); + if (sizeof(*desc) > LUSTRE_CFG_BUFLEN(lcfg, 1)) { CERROR("descriptor size wrong: %d > %d\n", - (int)sizeof(*desc), lcfg->lcfg_inllen1); + (int)sizeof(*desc), LUSTRE_CFG_BUFLEN(lcfg, 1)); RETURN(-EINVAL); } - uuids = (struct obd_uuid *)lcfg->lcfg_inlbuf2; - if (sizeof(*uuids) * desc->ld_tgt_count != lcfg->lcfg_inllen2) { + uuids = (struct obd_uuid *)lustre_cfg_buf(lcfg, 2); + if (sizeof(*uuids) * desc->ld_tgt_count != LUSTRE_CFG_BUFLEN(lcfg, 2)) { CERROR("UUID array size wrong: %u * %u != %u\n", - sizeof(*uuids), desc->ld_tgt_count, lcfg->lcfg_inllen2); + sizeof(*uuids), desc->ld_tgt_count, LUSTRE_CFG_BUFLEN(lcfg, 2)); RETURN(-EINVAL); } - lmv->bufsize = sizeof(struct lmv_tgt_desc) * desc->ld_tgt_count; - OBD_ALLOC(lmv->tgts, lmv->bufsize); + lmv->tgts_size = sizeof(struct lmv_tgt_desc) * desc->ld_tgt_count; + OBD_ALLOC(lmv->tgts, lmv->tgts_size); if (lmv->tgts == NULL) { CERROR("Out of memory\n"); - RETURN(-EINVAL); + RETURN(-ENOMEM); } lmv->desc = *desc; @@ -459,102 +589,173 @@ static int lmv_setup(struct obd_device *obd, obd_count len, void *buf) for (i = 0, tgts = lmv->tgts; i < desc->ld_tgt_count; i++, tgts++) tgts->uuid = uuids[i]; - lmv->max_easize = sizeof(struct ll_fid) * + lmv->max_cookiesize = 0; + + lmv->max_easize = sizeof(struct lustre_id) * desc->ld_tgt_count + sizeof(struct mea); - lmv->max_cookiesize = 0; + rc = lmv_setup_mgr(obd); + if (rc) { + CERROR("Can't setup LMV object manager, " + "error %d.\n", rc); + OBD_FREE(lmv->tgts, lmv->tgts_size); + } + + tgt_obd = class_find_client_obd(&lmv->tgts->uuid, LUSTRE_MDC_NAME, + &obd->obd_uuid); + if (!tgt_obd) { + CERROR("Target %s not attached\n", lmv->tgts->uuid.uuid); + RETURN(-EINVAL); + } RETURN(rc); } +static int lmv_cleanup(struct obd_device *obd, int flags) +{ + struct lmv_obd *lmv = &obd->u.lmv; + ENTRY; + + lmv_cleanup_mgr(obd); + OBD_FREE(lmv->tgts, lmv->tgts_size); + + RETURN(0); +} + static int lmv_statfs(struct obd_device *obd, struct obd_statfs *osfs, unsigned long max_age) { struct lmv_obd *lmv = &obd->u.lmv; - struct obd_statfs temp; + struct obd_statfs *temp; int rc = 0, i; ENTRY; rc = lmv_check_connect(obd); if (rc) RETURN(rc); + + OBD_ALLOC(temp, sizeof(*temp)); + if (temp == NULL) + RETURN(-ENOMEM); for (i = 0; i < lmv->desc.ld_tgt_count; i++) { - rc = obd_statfs(lmv->tgts[i].ltd_exp->exp_obd, &temp, max_age); + if (lmv->tgts[i].ltd_exp == NULL) + continue; + + rc = obd_statfs(lmv->tgts[i].ltd_exp->exp_obd, temp, max_age); if (rc) { - CERROR("can't stat MDS #%d (%s)\n", i, - lmv->tgts[i].ltd_exp->exp_obd->obd_name); - RETURN(rc); + CERROR("can't stat MDS #%d (%s), error %d\n", i, + lmv->tgts[i].ltd_exp->exp_obd->obd_name, + rc); + GOTO(out_free_temp, rc); } if (i == 0) { - memcpy(osfs, &temp, sizeof(temp)); + memcpy(osfs, temp, sizeof(*temp)); } else { - osfs->os_bavail += temp.os_bavail; - osfs->os_blocks += temp.os_blocks; - osfs->os_ffree += temp.os_ffree; - osfs->os_files += temp.os_files; + osfs->os_bavail += temp->os_bavail; + osfs->os_blocks += temp->os_blocks; + osfs->os_ffree += temp->os_ffree; + osfs->os_files += temp->os_files; } } - RETURN(rc); -} -static int lmv_cleanup(struct obd_device *obd, int flags) -{ - struct lmv_obd *lmv = &obd->u.lmv; - ENTRY; - lmv_cleanup_objs(obd); - OBD_FREE(lmv->tgts, lmv->bufsize); - RETURN(0); + EXIT; +out_free_temp: + OBD_FREE(temp, sizeof(*temp)); + return rc; } -static int lmv_getstatus(struct obd_export *exp, struct ll_fid *fid) +static int lmv_getstatus(struct obd_export *exp, struct lustre_id *id) { struct obd_device *obd = exp->exp_obd; struct lmv_obd *lmv = &obd->u.lmv; int rc; ENTRY; + rc = lmv_check_connect(obd); if (rc) RETURN(rc); - rc = md_getstatus(lmv->tgts[0].ltd_exp, fid); - fid->mds = 0; + + rc = md_getstatus(lmv->tgts[0].ltd_exp, id); + id_group(id) = 0; + RETURN(rc); } -static int lmv_getattr(struct obd_export *exp, struct ll_fid *fid, - unsigned long valid, unsigned int ea_size, - struct ptlrpc_request **request) +static int lmv_getattr(struct obd_export *exp, struct lustre_id *id, + __u64 valid, const char *ea_name, int ea_namelen, + unsigned int ea_size, struct ptlrpc_request **request) { struct obd_device *obd = exp->exp_obd; struct lmv_obd *lmv = &obd->u.lmv; - int rc, i = fid->mds; + int rc, i = id_group(id); struct lmv_obj *obj; ENTRY; + rc = lmv_check_connect(obd); if (rc) RETURN(rc); - obj = lmv_grab_obj(obd, fid, 0); - CDEBUG(D_OTHER, "GETATTR for %lu/%lu/%lu %s\n", - (unsigned long) fid->mds, - (unsigned long) fid->id, - (unsigned long) fid->generation, - obj ? "(splitted)" : ""); - - LASSERT(fid->mds < lmv->desc.ld_tgt_count); - rc = md_getattr(lmv->tgts[i].ltd_exp, fid, - valid, ea_size, request); - if (rc == 0 && obj) { - /* we have to loop over dirobjs here and gather attrs - * for all the slaves */ -#warning "attrs gathering here" + + LASSERT(i < lmv->desc.ld_tgt_count); + + + rc = md_getattr(lmv->tgts[i].ltd_exp, id, valid, + ea_name, ea_namelen, ea_size, request); + if (rc) + RETURN(rc); + + obj = lmv_grab_obj(obd, id); + + CDEBUG(D_OTHER, "GETATTR for "DLID4" %s\n", + OLID4(id), obj ? "(splitted)" : ""); + + /* + * if object is splitted, then we loop over all the slaves and gather + * size attribute. In ideal world we would have to gather also mds field + * from all slaves, as object is spread over the cluster and this is + * definitely interesting information and it is not good to loss it, + * but... + */ + if (obj) { + struct mds_body *body; + + if (*request == NULL) { + lmv_put_obj(obj); + RETURN(rc); + } + + body = lustre_msg_buf((*request)->rq_repmsg, 0, + sizeof(*body)); + LASSERT(body != NULL); + + lmv_lock_obj(obj); + + for (i = 0; i < obj->objcount; i++) { + + if (lmv->tgts[i].ltd_exp == NULL) { + CWARN("%s: NULL export for %d\n", + obd->obd_name, i); + continue; + } + + /* skip master obj. */ + if (id_equal_fid(&obj->id, &obj->objs[i].id)) + continue; + + body->size += obj->objs[i].size; + } + + lmv_unlock_obj(obj); + lmv_put_obj(obj); } - lmv_put_obj(obj); + RETURN(rc); } static int lmv_change_cbdata(struct obd_export *exp, - struct ll_fid *fid, - ldlm_iterator_t it, void *data) + struct lustre_id *id, + ldlm_iterator_t it, + void *data) { struct obd_device *obd = exp->exp_obd; struct lmv_obd *lmv = &obd->u.lmv; @@ -564,121 +765,137 @@ static int lmv_change_cbdata(struct obd_export *exp, rc = lmv_check_connect(obd); if (rc) RETURN(rc); - CDEBUG(D_OTHER, "CBDATA for %lu/%lu/%lu\n", - (unsigned long) fid->mds, - (unsigned long) fid->id, - (unsigned long) fid->generation); - LASSERT(fid->mds < lmv->desc.ld_tgt_count); - rc = md_change_cbdata(lmv->tgts[fid->mds].ltd_exp, fid, it, data); + + CDEBUG(D_OTHER, "CBDATA for "DLID4"\n", OLID4(id)); + LASSERT(id_group(id) < lmv->desc.ld_tgt_count); + + rc = md_change_cbdata(lmv->tgts[id_group(id)].ltd_exp, + id, it, data); + RETURN(rc); } -static int lmv_change_cbdata_name(struct obd_export *exp, struct ll_fid *pfid, - char *name, int len, struct ll_fid *cfid, - ldlm_iterator_t it, void *data) +static int lmv_change_cbdata_name(struct obd_export *exp, + struct lustre_id *pid, + char *name, int len, + struct lustre_id *cid, + ldlm_iterator_t it, + void *data) { struct obd_device *obd = exp->exp_obd; struct lmv_obd *lmv = &obd->u.lmv; + struct lustre_id rcid = *cid; struct lmv_obj *obj; int rc = 0, mds; ENTRY; + rc = lmv_check_connect(obd); if (rc) RETURN(rc); - LASSERT(pfid->mds < lmv->desc.ld_tgt_count); - LASSERT(cfid->mds < lmv->desc.ld_tgt_count); - CDEBUG(D_OTHER, "CBDATA for %lu/%lu/%lu:%*s -> %lu/%lu/%lu\n", - (unsigned long) pfid->mds, (unsigned long) pfid->id, - (unsigned long) pfid->generation, len, name, - (unsigned long) cfid->mds, (unsigned long) cfid->id, - (unsigned long) cfid->generation); - - /* this is default mds for directory name belongs to */ - mds = pfid->mds; - obj = lmv_grab_obj(obd, pfid, 0); + + LASSERT(id_group(pid) < lmv->desc.ld_tgt_count); + LASSERT(id_group(cid) < lmv->desc.ld_tgt_count); + + CDEBUG(D_OTHER, "CBDATA for "DLID4":%*s -> "DLID4"\n", + OLID4(pid), len, name, OLID4(cid)); + + /* this is default mds for directory name belongs to. */ + mds = id_group(pid); + obj = lmv_grab_obj(obd, pid); if (obj) { - /* directory is splitted. look for right mds for this name */ - mds = raw_name2idx(obj->objcount, name, len); + /* directory is splitted. look for right mds for this name. */ + mds = raw_name2idx(obj->hashtype, obj->objcount, name, len); + rcid = obj->objs[mds].id; + mds = id_group(&rcid); lmv_put_obj(obj); } - rc = md_change_cbdata(lmv->tgts[mds].ltd_exp, cfid, it, data); + rc = md_change_cbdata(lmv->tgts[mds].ltd_exp, &rcid, it, data); RETURN(rc); } -static int lmv_valid_attrs(struct obd_export *exp, struct ll_fid *fid) +static int lmv_valid_attrs(struct obd_export *exp, struct lustre_id *id) { struct obd_device *obd = exp->exp_obd; struct lmv_obd *lmv = &obd->u.lmv; int rc = 0; ENTRY; + rc = lmv_check_connect(obd); if (rc) RETURN(rc); - CDEBUG(D_OTHER, "validate %lu/%lu/%lu\n", - (unsigned long) fid->mds, - (unsigned long) fid->id, - (unsigned long) fid->generation); - LASSERT(fid->mds < lmv->desc.ld_tgt_count); - rc = md_valid_attrs(lmv->tgts[fid->mds].ltd_exp, fid); + + CDEBUG(D_OTHER, "validate "DLID4"\n", OLID4(id)); + LASSERT(id_group(id) < lmv->desc.ld_tgt_count); + rc = md_valid_attrs(lmv->tgts[id_group(id)].ltd_exp, id); RETURN(rc); } int lmv_close(struct obd_export *exp, struct obdo *obdo, - struct obd_client_handle *och, - struct ptlrpc_request **request) + struct obd_client_handle *och, + struct ptlrpc_request **request) { struct obd_device *obd = exp->exp_obd; struct lmv_obd *lmv = &obd->u.lmv; int rc, i = obdo->o_mds; ENTRY; + rc = lmv_check_connect(obd); if (rc) RETURN(rc); + LASSERT(i < lmv->desc.ld_tgt_count); - CDEBUG(D_OTHER, "CLOSE %lu/%lu/%lu\n", (unsigned long) obdo->o_mds, - (unsigned long) obdo->o_id, (unsigned long) obdo->o_generation); + CDEBUG(D_OTHER, "CLOSE %lu/%lu/%lu\n", (unsigned long)obdo->o_mds, + (unsigned long)obdo->o_id, (unsigned long)obdo->o_generation); rc = md_close(lmv->tgts[i].ltd_exp, obdo, och, request); RETURN(rc); } -int lmv_get_mea_and_update_object(struct obd_export *exp, struct ll_fid *fid) +int lmv_get_mea_and_update_object(struct obd_export *exp, + struct lustre_id *id) { struct obd_device *obd = exp->exp_obd; struct lmv_obd *lmv = &obd->u.lmv; struct ptlrpc_request *req = NULL; + struct lmv_obj *obj; struct lustre_md md; - unsigned long valid; int mealen, rc; + __u64 valid; + ENTRY; md.mea = NULL; mealen = MEA_SIZE_LMV(lmv); - valid = OBD_MD_FLEASIZE | OBD_MD_FLDIREA; + valid = OBD_MD_FLEASIZE | OBD_MD_FLDIREA | OBD_MD_MEA; - /* time to update mea of parent fid */ - rc = md_getattr(lmv->tgts[fid->mds].ltd_exp, fid, - valid, mealen, &req); + /* time to update mea of parent id */ + rc = md_getattr(lmv->tgts[id_group(id)].ltd_exp, + id, valid, NULL, 0, mealen, &req); if (rc) { - CERROR("md_getattr() failed, rc = %d\n", rc); + CERROR("md_getattr() failed, error %d\n", rc); GOTO(cleanup, rc); } rc = mdc_req2lustre_md(exp, req, 0, NULL, &md); if (rc) { - CERROR("mdc_req2lustre_md() failed, rc = %d\n", rc); + CERROR("mdc_req2lustre_md() failed, error %d\n", rc); GOTO(cleanup, rc); } if (md.mea == NULL) GOTO(cleanup, rc = -ENODATA); - rc = lmv_create_obj_from_attrs(exp, fid, md.mea); - obd_free_memmd(exp, (struct lov_stripe_md **) &md.mea); + obj = lmv_create_obj(exp, id, md.mea); + if (IS_ERR(obj)) + rc = PTR_ERR(obj); + + lmv_put_obj(obj); + obd_free_memmd(exp, (struct lov_stripe_md **)&md.mea); + EXIT; cleanup: if (req) ptlrpc_req_finished(req); - RETURN(rc); + return rc; } int lmv_create(struct obd_export *exp, struct mdc_op_data *op_data, @@ -687,9 +904,9 @@ int lmv_create(struct obd_export *exp, struct mdc_op_data *op_data, { struct obd_device *obd = exp->exp_obd; struct lmv_obd *lmv = &obd->u.lmv; - struct mds_body *mds_body; + struct mds_body *body; struct lmv_obj *obj; - int rc, mds; + int rc, mds, loop = 0; ENTRY; rc = lmv_check_connect(obd); @@ -699,37 +916,37 @@ int lmv_create(struct obd_export *exp, struct mdc_op_data *op_data, if (!lmv->desc.ld_active_tgt_count) RETURN(-EIO); repeat: - obj = lmv_grab_obj(obd, &op_data->fid1, 0); + LASSERT(++loop <= 2); + obj = lmv_grab_obj(obd, &op_data->id1); if (obj) { - mds = raw_name2idx(obj->objcount, op_data->name, - op_data->namelen); - op_data->fid1 = obj->objs[mds].fid; + mds = raw_name2idx(obj->hashtype, obj->objcount, + op_data->name, op_data->namelen); + op_data->id1 = obj->objs[mds].id; lmv_put_obj(obj); } - CDEBUG(D_OTHER, "CREATE '%*s' on %lu/%lu/%lu\n", - op_data->namelen, op_data->name, - (unsigned long) op_data->fid1.mds, - (unsigned long) op_data->fid1.id, - (unsigned long) op_data->fid1.generation); - rc = md_create(lmv->tgts[op_data->fid1.mds].ltd_exp, op_data, data, - datalen, mode, uid, gid, rdev, request); + CDEBUG(D_OTHER, "CREATE '%*s' on "DLID4"\n", op_data->namelen, + op_data->name, OLID4(&op_data->id1)); + + rc = md_create(lmv->tgts[id_group(&op_data->id1)].ltd_exp, + op_data, data, datalen, mode, uid, gid, rdev, + request); if (rc == 0) { if (*request == NULL) - RETURN(rc); - mds_body = lustre_msg_buf((*request)->rq_repmsg, 0, - sizeof(*mds_body)); - LASSERT(mds_body != NULL); - CDEBUG(D_OTHER, "created. id = %lu, generation = %lu, mds = %d\n", - (unsigned long) mds_body->fid1.id, - (unsigned long) mds_body->fid1.generation, - op_data->fid1.mds); - LASSERT(mds_body->valid & OBD_MD_MDS || - mds_body->mds == op_data->fid1.mds); + RETURN(rc); + + body = lustre_msg_buf((*request)->rq_repmsg, 0, + sizeof(*body)); + if (body == NULL) + RETURN(-ENOMEM); + + CDEBUG(D_OTHER, "created. "DLID4"\n", OLID4(&op_data->id1)); } else if (rc == -ERESTART) { - /* directory got splitted. time to update local object - * and repeat the request with proper MDS */ - rc = lmv_get_mea_and_update_object(exp, &op_data->fid1); + /* + * directory got splitted. time to update local object and + * repeat the request with proper MDS. + */ + rc = lmv_get_mea_and_update_object(exp, &op_data->id1); if (rc == 0) { ptlrpc_req_finished(*request); goto repeat; @@ -744,6 +961,7 @@ int lmv_done_writing(struct obd_export *exp, struct obdo *obdo) struct lmv_obd *lmv = &obd->u.lmv; int rc; ENTRY; + rc = lmv_check_connect(obd); if (rc) RETURN(rc); @@ -755,63 +973,114 @@ int lmv_done_writing(struct obd_export *exp, struct obdo *obdo) } int lmv_enqueue_slaves(struct obd_export *exp, int locktype, - struct lookup_intent *it, int lockmode, - struct mdc_op_data *data, struct lustre_handle *lockh, - void *lmm, int lmmsize, - ldlm_completion_callback cb_completion, - ldlm_blocking_callback cb_blocking, void *cb_data) + struct lookup_intent *it, int lockmode, + struct mdc_op_data *data, struct lustre_handle *lockh, + void *lmm, int lmmsize, ldlm_completion_callback cb_compl, + ldlm_blocking_callback cb_blocking, void *cb_data) { struct obd_device *obd = exp->exp_obd; struct lmv_obd *lmv = &obd->u.lmv; struct mea *mea = data->mea1; - struct mdc_op_data data2; + struct mdc_op_data *data2; int i, rc, mds; ENTRY; + OBD_ALLOC(data2, sizeof(*data2)); + if (data2 == NULL) + RETURN(-ENOMEM); + LASSERT(mea != NULL); for (i = 0; i < mea->mea_count; i++) { - if (lmv->tgts[i].ltd_exp == NULL) + memset(data2, 0, sizeof(*data2)); + data2->id1 = mea->mea_ids[i]; + mds = id_group(&data2->id1); + + if (lmv->tgts[mds].ltd_exp == NULL) continue; - memset(&data2, 0, sizeof(data2)); - data2.fid1 = mea->mea_fids[i]; - mds = data2.fid1.mds; - rc = md_enqueue(lmv->tgts[mds].ltd_exp, locktype, it, lockmode, - &data2, lockh + i, lmm, lmmsize, cb_completion, - cb_blocking, cb_data); - CDEBUG(D_OTHER, "take lock on slave %lu/%lu/%lu -> %d/%d\n", - (unsigned long) mea->mea_fids[i].mds, - (unsigned long) mea->mea_fids[i].id, - (unsigned long) mea->mea_fids[i].generation, - rc, it->d.lustre.it_status); + rc = md_enqueue(lmv->tgts[mds].ltd_exp, locktype, it, + lockmode, data2, lockh + i, lmm, lmmsize, + cb_compl, cb_blocking, cb_data); + + CDEBUG(D_OTHER, "take lock on slave "DLID4" -> %d/%d\n", + OLID4(&mea->mea_ids[i]), rc, LUSTRE_IT(it)->it_status); if (rc) GOTO(cleanup, rc); - if (it->d.lustre.it_data) { + if (LUSTRE_IT(it)->it_data) { struct ptlrpc_request *req; - req = (struct ptlrpc_request *) it->d.lustre.it_data; + req = (struct ptlrpc_request *) LUSTRE_IT(it)->it_data; ptlrpc_req_finished(req); } - if (it->d.lustre.it_status) - GOTO(cleanup, rc = it->d.lustre.it_status); + if (LUSTRE_IT(it)->it_status) + GOTO(cleanup, rc = LUSTRE_IT(it)->it_status); } - RETURN(0); + OBD_FREE(data2, sizeof(*data2)); + RETURN(0); cleanup: + OBD_FREE(data2, sizeof(*data2)); + /* drop all taken locks */ while (--i >= 0) { if (lockh[i].cookie) ldlm_lock_decref(lockh + i, lockmode); lockh[i].cookie = 0; } + return rc; +} + +int lmv_enqueue_remote(struct obd_export *exp, int lock_type, + struct lookup_intent *it, int lock_mode, + struct mdc_op_data *data, struct lustre_handle *lockh, + void *lmm, int lmmsize, ldlm_completion_callback cb_compl, + ldlm_blocking_callback cb_blocking, void *cb_data) +{ + struct ptlrpc_request *req = LUSTRE_IT(it)->it_data; + struct obd_device *obd = exp->exp_obd; + struct lmv_obd *lmv = &obd->u.lmv; + struct lustre_handle plock; + struct mdc_op_data rdata; + struct mds_body *body = NULL; + int rc = 0, pmode; + ENTRY; + + body = lustre_msg_buf(req->rq_repmsg, 1, sizeof(*body)); + LASSERT(body != NULL); + + if (!(body->valid & OBD_MD_MDS)) + RETURN(0); + + CDEBUG(D_OTHER, "ENQUEUE '%s' on "DLID4" -> "DLID4"\n", + LL_IT2STR(it), OLID4(&data->id1), OLID4(&body->id1)); + + /* we got LOOKUP lock, but we really need attrs */ + pmode = LUSTRE_IT(it)->it_lock_mode; + LASSERT(pmode != 0); + memcpy(&plock, lockh, sizeof(plock)); + LUSTRE_IT(it)->it_lock_mode = 0; + LUSTRE_IT(it)->it_data = NULL; + LASSERT((body->valid & OBD_MD_FID) != 0); + + memcpy(&rdata, data, sizeof(rdata)); + rdata.id1 = body->id1; + rdata.name = NULL; + rdata.namelen = 0; + + LUSTRE_IT(it)->it_disposition &= ~DISP_ENQ_COMPLETE; + ptlrpc_req_finished(req); + + rc = md_enqueue(lmv->tgts[id_group(&rdata.id1)].ltd_exp, + lock_type, it, lock_mode, &rdata, lockh, lmm, + lmmsize, cb_compl, cb_blocking, cb_data); + ldlm_lock_decref(&plock, pmode); RETURN(rc); } int lmv_enqueue(struct obd_export *exp, int lock_type, struct lookup_intent *it, int lock_mode, struct mdc_op_data *data, struct lustre_handle *lockh, - void *lmm, int lmmsize, - ldlm_completion_callback cb_completion, + void *lmm, int lmmsize, ldlm_completion_callback cb_compl, ldlm_blocking_callback cb_blocking, void *cb_data) { struct obd_device *obd = exp->exp_obd; @@ -824,85 +1093,96 @@ int lmv_enqueue(struct obd_export *exp, int lock_type, if (rc) RETURN(rc); - if (it->it_op == IT_UNLINK) { + if (data->mea1 && it->it_op == IT_UNLINK) { rc = lmv_enqueue_slaves(exp, lock_type, it, lock_mode, data, lockh, lmm, lmmsize, - cb_completion, cb_blocking, cb_data); + cb_compl, cb_blocking, cb_data); RETURN(rc); } if (data->namelen) { - obj = lmv_grab_obj(obd, &data->fid1, 0); + obj = lmv_grab_obj(obd, &data->id1); if (obj) { - /* directory is splitted. look for - * right mds for this name */ - mds = raw_name2idx(obj->objcount, (char *)data->name, - data->namelen); - data->fid1 = obj->objs[mds].fid; + /* directory is splitted. look for right mds for this + * name */ + mds = raw_name2idx(obj->hashtype, obj->objcount, + (char *)data->name, data->namelen); + data->id1 = obj->objs[mds].id; lmv_put_obj(obj); } } - CDEBUG(D_OTHER, "ENQUEUE '%s' on %lu/%lu\n", - LL_IT2STR(it), (unsigned long) data->fid1.id, - (unsigned long) data->fid1.generation); - rc = md_enqueue(lmv->tgts[data->fid1.mds].ltd_exp, lock_type, it, - lock_mode, data, lockh, lmm, lmmsize, cb_completion, - cb_blocking, cb_data); - + CDEBUG(D_OTHER, "ENQUEUE '%s' on "DLID4"\n", LL_IT2STR(it), + OLID4(&data->id1)); + + rc = md_enqueue(lmv->tgts[id_group(&data->id1)].ltd_exp, + lock_type, it, lock_mode, data, lockh, lmm, + lmmsize, cb_compl, cb_blocking, cb_data); + if (rc == 0 && it->it_op == IT_OPEN) + rc = lmv_enqueue_remote(exp, lock_type, it, lock_mode, + data, lockh, lmm, lmmsize, + cb_compl, cb_blocking, cb_data); RETURN(rc); } -int lmv_getattr_name(struct obd_export *exp, struct ll_fid *fid, - char *filename, int namelen, unsigned long valid, - unsigned int ea_size, struct ptlrpc_request **request) +int lmv_getattr_lock(struct obd_export *exp, struct lustre_id *id, + char *filename, int namelen, __u64 valid, + unsigned int ea_size, struct ptlrpc_request **request) { + int rc, mds = id_group(id), loop = 0; struct obd_device *obd = exp->exp_obd; struct lmv_obd *lmv = &obd->u.lmv; - struct ll_fid rfid = *fid; - int rc, mds = fid->mds; + struct lustre_id rid = *id; struct mds_body *body; struct lmv_obj *obj; ENTRY; + rc = lmv_check_connect(obd); if (rc) RETURN(rc); repeat: - obj = lmv_grab_obj(obd, fid, 0); + LASSERT(++loop <= 2); + obj = lmv_grab_obj(obd, id); if (obj) { /* directory is splitted. look for right mds for this name */ - mds = raw_name2idx(obj->objcount, filename, namelen - 1); - rfid = obj->objs[mds].fid; + mds = raw_name2idx(obj->hashtype, obj->objcount, + filename, namelen - 1); + rid = obj->objs[mds].id; lmv_put_obj(obj); } - CDEBUG(D_OTHER, "getattr_name for %*s on %lu/%lu/%lu -> %lu/%lu/%lu\n", - namelen, filename, (unsigned long) fid->mds, - (unsigned long) fid->id, (unsigned long) fid->generation, - (unsigned long) rfid.mds, (unsigned long) rfid.id, - (unsigned long) rfid.generation); - rc = md_getattr_name(lmv->tgts[mds].ltd_exp, &rfid, filename, namelen, - valid, ea_size, request); + + CDEBUG(D_OTHER, "getattr_lock for %*s on "DLID4" -> "DLID4"\n", + namelen, filename, OLID4(id), OLID4(&rid)); + + rc = md_getattr_lock(lmv->tgts[id_group(&rid)].ltd_exp, + &rid, filename, namelen, + valid == OBD_MD_FLID ? valid : valid | OBD_MD_FID, + ea_size, request); if (rc == 0) { - /* this could be cross-node reference. in this case all - * we have right now is mds/ino/generation triple. we'd - * like to find other attributes */ + /* + * this could be cross-node reference. in this case all we have + * right now is lustre_id triple. we'd like to find other + * attributes. + */ body = lustre_msg_buf((*request)->rq_repmsg, 0, sizeof(*body)); LASSERT(body != NULL); + LASSERT((body->valid & OBD_MD_FID) != 0 + || body->valid == OBD_MD_FLID); + if (body->valid & OBD_MD_MDS) { struct ptlrpc_request *req = NULL; - rfid = body->fid1; - CDEBUG(D_OTHER, "request attrs for %lu/%lu/%lu\n", - (unsigned long) rfid.mds, - (unsigned long) rfid.id, - (unsigned long) rfid.generation); - rc = md_getattr_name(lmv->tgts[rfid.mds].ltd_exp, &rfid, - NULL, 1, valid, ea_size, &req); + + rid = body->id1; + CDEBUG(D_OTHER, "request attrs for "DLID4"\n", OLID4(&rid)); + + rc = md_getattr_lock(lmv->tgts[id_group(&rid)].ltd_exp, + &rid, NULL, 1, valid, ea_size, &req); ptlrpc_req_finished(*request); *request = req; } } else if (rc == -ERESTART) { - /* directory got splitted. time to update local object - * and repeat the request with proper MDS */ - rc = lmv_get_mea_and_update_object(exp, &rfid); + /* directory got splitted. time to update local object and + * repeat the request with proper MDS */ + rc = lmv_get_mea_and_update_object(exp, &rid); if (rc == 0) { ptlrpc_req_finished(*request); goto repeat; @@ -911,10 +1191,9 @@ repeat: RETURN(rc); } - /* - * llite passes fid of an target inode in data->fid1 and - * fid of directory in data->fid2 + * llite passes id of an target inode in data->id1 and id of directory in + * data->id2 */ int lmv_link(struct obd_export *exp, struct mdc_op_data *data, struct ptlrpc_request **request) @@ -922,34 +1201,40 @@ int lmv_link(struct obd_export *exp, struct mdc_op_data *data, struct obd_device *obd = exp->exp_obd; struct lmv_obd *lmv = &obd->u.lmv; struct lmv_obj *obj; - int rc; + int rc, mds; ENTRY; + rc = lmv_check_connect(obd); if (rc) RETURN(rc); + if (data->namelen != 0) { /* usual link request */ - obj = lmv_grab_obj(obd, &data->fid1, 0); + obj = lmv_grab_obj(obd, &data->id2); if (obj) { - rc = raw_name2idx(obj->objcount, data->name, - data->namelen); - data->fid1 = obj->objs[rc].fid; + rc = raw_name2idx(obj->hashtype, obj->objcount, + data->name, data->namelen); + data->id2 = obj->objs[rc].id; lmv_put_obj(obj); } - CDEBUG(D_OTHER,"link %u/%u/%u:%*s to %u/%u/%u mds %d\n", - (unsigned) data->fid2.mds, (unsigned) data->fid2.id, - (unsigned) data->fid2.generation, data->namelen, - data->name, (unsigned) data->fid1.mds, - (unsigned) data->fid1.id, - (unsigned) data->fid1.generation, data->fid1.mds); + + mds = id_group(&data->id2); + + CDEBUG(D_OTHER,"link "DLID4":%*s to "DLID4"\n", + OLID4(&data->id2), data->namelen, data->name, + OLID4(&data->id1)); } else { - /* request from MDS to acquire i_links for inode by fid1 */ - CDEBUG(D_OTHER, "inc i_nlinks for %u/%u/%u\n", - (unsigned) data->fid1.mds, (unsigned) data->fid1.id, - (unsigned) data->fid1.generation); + mds = id_group(&data->id1); + + /* request from MDS to acquire i_links for inode by id1 */ + CDEBUG(D_OTHER, "inc i_nlinks for "DLID4"\n", + OLID4(&data->id1)); } - - rc = md_link(lmv->tgts[data->fid1.mds].ltd_exp, data, request); + + CDEBUG(D_OTHER, "forward to MDS #%u ("DLID4")\n", + mds, OLID4(&data->id1)); + rc = md_link(lmv->tgts[mds].ltd_exp, data, request); + RETURN(rc); } @@ -963,128 +1248,143 @@ int lmv_rename(struct obd_export *exp, struct mdc_op_data *data, int rc, mds; ENTRY; - CDEBUG(D_OTHER, "rename %*s in %lu/%lu/%lu to %*s in %lu/%lu/%lu\n", - oldlen, old, (unsigned long) data->fid1.mds, - (unsigned long) data->fid1.id, - (unsigned long) data->fid1.generation, - newlen, new, (unsigned long) data->fid2.mds, - (unsigned long) data->fid2.id, - (unsigned long) data->fid2.generation); - if (!fid_equal(&data->fid1, &data->fid2)) - CWARN("cross-node rename %lu/%lu/%lu:%*s to %lu/%lu/%lu:%*s\n", - (unsigned long) data->fid1.mds, - (unsigned long) data->fid1.id, - (unsigned long) data->fid1.generation, oldlen, old, - (unsigned long) data->fid2.mds, - (unsigned long) data->fid2.id, - (unsigned long) data->fid2.generation, newlen, new); + CDEBUG(D_OTHER, "rename %*s in "DLID4" to %*s in "DLID4"\n", + oldlen, old, OLID4(&data->id1), newlen, new, + OLID4(&data->id2)); rc = lmv_check_connect(obd); if (rc) RETURN(rc); if (oldlen == 0) { - /* MDS with old dir entry is asking another MDS - * to create name there */ + /* + * MDS with old dir entry is asking another MDS to create name + * there. + */ CDEBUG(D_OTHER, - "create %*s(%d/%d) in %lu/%lu/%lu pointing to %lu/%lu/%lu\n", - newlen, new, oldlen, newlen, - (unsigned long) data->fid2.mds, - (unsigned long) data->fid2.id, - (unsigned long) data->fid2.generation, - (unsigned long) data->fid1.mds, - (unsigned long) data->fid1.id, - (unsigned long) data->fid1.generation); - mds = data->fid2.mds; + "create %*s(%d/%d) in "DLID4" pointing " + "to "DLID4"\n", newlen, new, oldlen, newlen, + OLID4(&data->id2), OLID4(&data->id1)); + + mds = id_group(&data->id2); + + /* + * target directory can be splitted, sowe should forward request + * to the right MDS. + */ + obj = lmv_grab_obj(obd, &data->id2); + if (obj) { + mds = raw_name2idx(obj->hashtype, obj->objcount, + (char *)new, newlen); + data->id2 = obj->objs[mds].id; + CDEBUG(D_OTHER, "forward to MDS #%u ("DLID4")\n", mds, + OLID4(&data->id2)); + lmv_put_obj(obj); + } goto request; } - obj = lmv_grab_obj(obd, &data->fid1, 0); + obj = lmv_grab_obj(obd, &data->id1); if (obj) { - /* directory is already splitted, so we have to forward - * request to the right MDS */ - mds = raw_name2idx(obj->objcount, (char *)old, oldlen); - data->fid1 = obj->objs[mds].fid; - CDEBUG(D_OTHER, "forward to MDS #%u (%lu/%lu/%lu)\n", mds, - (unsigned long) obj->objs[mds].fid.mds, - (unsigned long) obj->objs[mds].fid.id, - (unsigned long) obj->objs[mds].fid.generation); + /* + * directory is already splitted, so we have to forward request + * to the right MDS. + */ + mds = raw_name2idx(obj->hashtype, obj->objcount, + (char *)old, oldlen); + data->id1 = obj->objs[mds].id; + CDEBUG(D_OTHER, "forward to MDS #%u ("DLID4")\n", mds, + OLID4(&data->id1)); + lmv_put_obj(obj); } - lmv_put_obj(obj); - obj = lmv_grab_obj(obd, &data->fid2, 0); + obj = lmv_grab_obj(obd, &data->id2); if (obj) { - /* directory is already splitted, so we have to forward - * request to the right MDS */ - mds = raw_name2idx(obj->objcount, (char *)new, newlen); - data->fid2 = obj->objs[mds].fid; - CDEBUG(D_OTHER, "forward to MDS #%u (%lu/%lu/%lu)\n", mds, - (unsigned long) obj->objs[mds].fid.mds, - (unsigned long) obj->objs[mds].fid.id, - (unsigned long) obj->objs[mds].fid.generation); + /* + * directory is already splitted, so we have to forward request + * to the right MDS. + */ + mds = raw_name2idx(obj->hashtype, obj->objcount, + (char *)new, newlen); + + data->id2 = obj->objs[mds].id; + CDEBUG(D_OTHER, "forward to MDS #%u ("DLID4")\n", mds, + OLID4(&data->id2)); + lmv_put_obj(obj); } - lmv_put_obj(obj); - mds = data->fid1.mds; + mds = id_group(&data->id1); request: + if (id_group(&data->id1) != id_group(&data->id2)) { + CDEBUG(D_OTHER,"cross-node rename "DLID4"/%*s to "DLID4"/%*s\n", + OLID4(&data->id1), oldlen, old, OLID4(&data->id2), + newlen, new); + } + rc = md_rename(lmv->tgts[mds].ltd_exp, data, old, oldlen, new, newlen, request); RETURN(rc); } int lmv_setattr(struct obd_export *exp, struct mdc_op_data *data, - struct iattr *iattr, void *ea, int ealen, void *ea2, int ea2len, - struct ptlrpc_request **request) + struct iattr *iattr, void *ea, int ealen, void *ea2, + int ea2len, struct ptlrpc_request **request) { struct obd_device *obd = exp->exp_obd; struct lmv_obd *lmv = &obd->u.lmv; - int rc = 0, i = data->fid1.mds; struct ptlrpc_request *req; - struct mds_body *mds_body; + struct mds_body *body; struct lmv_obj *obj; + int rc = 0, i; ENTRY; rc = lmv_check_connect(obd); if (rc) RETURN(rc); - obj = lmv_grab_obj(obd, &data->fid1, 0); - CDEBUG(D_OTHER, "SETATTR for %lu/%lu/%lu, valid 0x%x%s\n", - (unsigned long) data->fid1.mds, - (unsigned long) data->fid1.id, - (unsigned long) data->fid1.generation, iattr->ia_valid, - obj ? ", splitted" : ""); + obj = lmv_grab_obj(obd, &data->id1); + + CDEBUG(D_OTHER, "SETATTR for "DLID4", valid 0x%x%s\n", + OLID4(&data->id1), iattr->ia_valid, obj ? ", splitted" : ""); + if (obj) { for (i = 0; i < obj->objcount; i++) { - data->fid1 = obj->objs[i].fid; - rc = md_setattr(lmv->tgts[i].ltd_exp, data, iattr, ea, - ealen, ea2, ea2len, &req); - LASSERT(rc == 0); - if (fid_equal(&obj->fid, &obj->objs[i].fid)) { - /* this is master object and this request - * should be returned back to llite */ + data->id1 = obj->objs[i].id; + + rc = md_setattr(lmv->tgts[id_group(&data->id1)].ltd_exp, + data, iattr, ea, ealen, ea2, ea2len, &req); + + if (id_equal_fid(&obj->id, &obj->objs[i].id)) { + /* + * this is master object and this request should + * be returned back to llite. + */ *request = req; } else { ptlrpc_req_finished(req); } + + if (rc) + break; } lmv_put_obj(obj); } else { - LASSERT(data->fid1.mds < lmv->desc.ld_tgt_count); - rc = md_setattr(lmv->tgts[i].ltd_exp, data, iattr, ea, ealen, - ea2, ea2len, request); + LASSERT(id_group(&data->id1) < lmv->desc.ld_tgt_count); + rc = md_setattr(lmv->tgts[id_group(&data->id1)].ltd_exp, + data, iattr, ea, ealen, ea2, ea2len, request); if (rc == 0) { - mds_body = lustre_msg_buf((*request)->rq_repmsg, 0, - sizeof(*mds_body)); - LASSERT(mds_body != NULL); - LASSERT(mds_body->mds == i); + body = lustre_msg_buf((*request)->rq_repmsg, 0, + sizeof(*body)); + LASSERT(body != NULL); + LASSERT((body->valid & OBD_MD_FID) != 0); + LASSERT(id_group(&body->id1) == id_group(&data->id1)); } } RETURN(rc); } -int lmv_sync(struct obd_export *exp, struct ll_fid *fid, +int lmv_sync(struct obd_export *exp, struct lustre_id *id, struct ptlrpc_request **request) { struct obd_device *obd = exp->exp_obd; @@ -1096,12 +1396,14 @@ int lmv_sync(struct obd_export *exp, struct ll_fid *fid, if (rc) RETURN(rc); - rc = md_sync(lmv->tgts[0].ltd_exp, fid, request); + rc = md_sync(lmv->tgts[id_group(id)].ltd_exp, + id, request); RETURN(rc); } -int lmv_dirobj_blocking_ast(struct ldlm_lock *lock, - struct ldlm_lock_desc *desc, void *data, int flag) +int lmv_dirobj_blocking_ast(struct ldlm_lock *lock, + struct ldlm_lock_desc *desc, + void *data, int flag) { struct lustre_handle lockh; struct lmv_obj *obj; @@ -1120,17 +1422,16 @@ int lmv_dirobj_blocking_ast(struct ldlm_lock *lock, case LDLM_CB_CANCELING: /* time to drop cached attrs for dirobj */ obj = lock->l_ast_data; - if (!obj) - break; - - CDEBUG(D_OTHER, "cancel %s on %lu/%lu, master %lu/%lu/%lu\n", - lock->l_resource->lr_name.name[3] == 1 ? - "LOOKUP" : "UPDATE", - (unsigned long) lock->l_resource->lr_name.name[0], - (unsigned long) lock->l_resource->lr_name.name[1], - (unsigned long) obj->fid.mds, - (unsigned long) obj->fid.id, - (unsigned long) obj->fid.generation); + if (obj) { + CDEBUG(D_OTHER, "cancel %s on "LPU64"/"LPU64 + ", master "DLID4"\n", + lock->l_resource->lr_name.name[3] == 1 ? + "LOOKUP" : "UPDATE", + lock->l_resource->lr_name.name[0], + lock->l_resource->lr_name.name[1], + OLID4(&obj->id)); + lmv_put_obj(obj); + } break; default: LBUG(); @@ -1140,10 +1441,10 @@ int lmv_dirobj_blocking_ast(struct ldlm_lock *lock, void lmv_remove_dots(struct page *page) { - char *kaddr = page_address(page); unsigned limit = PAGE_CACHE_SIZE; - unsigned offs, rec_len; + char *kaddr = page_address(page); struct ext2_dir_entry_2 *p; + unsigned offs, rec_len; for (offs = 0; offs <= limit - EXT2_DIR_REC_LEN(1); offs += rec_len) { p = (struct ext2_dir_entry_2 *)(kaddr + offs); @@ -1155,53 +1456,53 @@ void lmv_remove_dots(struct page *page) } } -int lmv_readpage(struct obd_export *exp, struct ll_fid *mdc_fid, +int lmv_readpage(struct obd_export *exp, struct lustre_id *id, __u64 offset, struct page *page, struct ptlrpc_request **request) { struct obd_device *obd = exp->exp_obd; struct lmv_obd *lmv = &obd->u.lmv; - struct ll_fid rfid = *mdc_fid; + struct lustre_id rid = *id; struct lmv_obj *obj; int rc, i; ENTRY; +#warning "we need well-desgined readdir() implementation" rc = lmv_check_connect(obd); if (rc) RETURN(rc); - LASSERT(mdc_fid->mds < lmv->desc.ld_tgt_count); - CDEBUG(D_OTHER, "READPAGE at %llu from %lu/%lu/%lu\n", - offset, (unsigned long) rfid.mds, - (unsigned long) rfid.id, - (unsigned long) rfid.generation); + LASSERT(id_group(id) < lmv->desc.ld_tgt_count); + CDEBUG(D_OTHER, "READPAGE at %llu from "DLID4"\n", + offset, OLID4(&rid)); - obj = lmv_grab_obj(obd, mdc_fid, 0); + obj = lmv_grab_obj(obd, id); if (obj) { - /* find dirobj containing page with requested offset */ - /* FIXME: what about protecting cached attrs here? */ + lmv_lock_obj(obj); + + /* find dirobj containing page with requested offset. */ for (i = 0; i < obj->objcount; i++) { if (offset < obj->objs[i].size) break; offset -= obj->objs[i].size; } - rfid = obj->objs[i].fid; - CDEBUG(D_OTHER, "forward to %lu/%lu/%lu with offset %lu\n", - (unsigned long) rfid.mds, - (unsigned long) rfid.id, - (unsigned long) rfid.generation, - (unsigned long) offset); - } - rc = md_readpage(lmv->tgts[rfid.mds].ltd_exp, &rfid, offset, page, request); - if (rc == 0 && !fid_equal(&rfid, mdc_fid)) { - /* this page isn't from master object. to avoid - * ./.. duplication in directory, we have to remove them - * from all slave objects */ - lmv_remove_dots(page); + rid = obj->objs[i].id; + + lmv_unlock_obj(obj); + lmv_put_obj(obj); + + CDEBUG(D_OTHER, "forward to "DLID4" with offset %lu\n", + OLID4(&rid), (unsigned long)offset); } - - lmv_put_obj(obj); - + rc = md_readpage(lmv->tgts[id_group(&rid)].ltd_exp, &rid, + offset, page, request); + + if (rc == 0 && !id_equal_fid(&rid, id)) + /* this page isn't from master object. To avoid "." and ".." + * duplication in directory, we have to remove them from all + * slave objects */ + lmv_remove_dots(page); + RETURN(rc); } @@ -1211,34 +1512,52 @@ int lmv_unlink_slaves(struct obd_export *exp, struct mdc_op_data *data, struct obd_device *obd = exp->exp_obd; struct lmv_obd *lmv = &obd->u.lmv; struct mea *mea = data->mea1; - struct mdc_op_data data2; - int i, rc = 0, mds; + struct mdc_op_data *data2; + int i, rc = 0; ENTRY; + OBD_ALLOC(data2, sizeof(*data2)); + if (data2 == NULL) + RETURN(-ENOMEM); + LASSERT(mea != NULL); for (i = 0; i < mea->mea_count; i++) { - if (lmv->tgts[i].ltd_exp == NULL) + memset(data2, 0, sizeof(*data2)); + data2->id1 = mea->mea_ids[i]; + data2->create_mode = MDS_MODE_DONT_LOCK | S_IFDIR; + + if (lmv->tgts[id_group(&data2->id1)].ltd_exp == NULL) continue; - memset(&data2, 0, sizeof(data2)); - data2.fid1 = mea->mea_fids[i]; - data2.create_mode = MDS_MODE_DONT_LOCK | S_IFDIR; - mds = data2.fid1.mds; - rc = md_unlink(lmv->tgts[mds].ltd_exp, &data2, req); - CDEBUG(D_OTHER, "unlink slave %lu/%lu/%lu -> %d\n", - (unsigned long) mea->mea_fids[i].mds, - (unsigned long) mea->mea_fids[i].id, - (unsigned long) mea->mea_fids[i].generation, rc); + rc = md_unlink(lmv->tgts[id_group(&data2->id1)].ltd_exp, + data2, req); + + CDEBUG(D_OTHER, "unlink slave "DLID4" -> %d\n", + OLID4(&mea->mea_ids[i]), rc); + if (*req) { ptlrpc_req_finished(*req); *req = NULL; } if (rc) - break; + RETURN(rc); } + OBD_FREE(data2, sizeof(*data2)); RETURN(rc); } +int lmv_delete_inode(struct obd_export *exp, struct lustre_id *id) +{ + ENTRY; + + LASSERT(exp && id); + if (lmv_delete_obj(exp, id)) { + CDEBUG(D_OTHER, "lmv object "DLID4" is destroyed.\n", + OLID4(id)); + } + RETURN(0); +} + int lmv_unlink(struct obd_export *exp, struct mdc_op_data *data, struct ptlrpc_request **request) { @@ -1246,6 +1565,7 @@ int lmv_unlink(struct obd_export *exp, struct mdc_op_data *data, struct lmv_obd *lmv = &obd->u.lmv; int rc, i = 0; ENTRY; + rc = lmv_check_connect(obd); if (rc) RETURN(rc); @@ -1254,32 +1574,32 @@ int lmv_unlink(struct obd_export *exp, struct mdc_op_data *data, /* mds asks to remove slave objects */ rc = lmv_unlink_slaves(exp, data, request); RETURN(rc); - } else if (data->namelen != 0) { + } + + if (data->namelen != 0) { struct lmv_obj *obj; - obj = lmv_grab_obj(obd, &data->fid1, 0); + + obj = lmv_grab_obj(obd, &data->id1); if (obj) { - i = raw_name2idx(obj->objcount, data->name, - data->namelen); - data->fid1 = obj->objs[i].fid; + i = raw_name2idx(obj->hashtype, obj->objcount, + data->name, data->namelen); + data->id1 = obj->objs[i].id; lmv_put_obj(obj); } - CDEBUG(D_OTHER, "unlink '%*s' in %lu/%lu/%lu -> %u\n", - data->namelen, data->name, - (unsigned long) data->fid1.mds, - (unsigned long) data->fid1.id, - (unsigned long) data->fid1.generation, i); + CDEBUG(D_OTHER, "unlink '%*s' in "DLID4" -> %u\n", + data->namelen, data->name, OLID4(&data->id1), + i); } else { - CDEBUG(D_OTHER, "drop i_nlink on %lu/%lu/%lu\n", - (unsigned long) data->fid1.mds, - (unsigned long) data->fid1.id, - (unsigned long) data->fid1.generation); + CDEBUG(D_OTHER, "drop i_nlink on "DLID4"\n", + OLID4(&data->id1)); } - rc = md_unlink(lmv->tgts[data->fid1.mds].ltd_exp, data, request); + rc = md_unlink(lmv->tgts[id_group(&data->id1)].ltd_exp, + data, request); RETURN(rc); } struct obd_device *lmv_get_real_obd(struct obd_export *exp, - char *name, int len) + struct lustre_id *id) { struct obd_device *obd = exp->exp_obd; struct lmv_obd *lmv = &obd->u.lmv; @@ -1289,12 +1609,14 @@ struct obd_device *lmv_get_real_obd(struct obd_export *exp, rc = lmv_check_connect(obd); if (rc) RETURN(ERR_PTR(rc)); - obd = lmv->tgts[0].ltd_exp->exp_obd; + obd = lmv->tgts[id_group(id)].ltd_exp->exp_obd; EXIT; + return obd; } -int lmv_init_ea_size(struct obd_export *exp, int easize, int cookiesize) +int lmv_init_ea_size(struct obd_export *exp, int easize, + int cookiesize) { struct obd_device *obd = exp->exp_obd; struct lmv_obd *lmv = &obd->u.lmv; @@ -1315,13 +1637,24 @@ int lmv_init_ea_size(struct obd_export *exp, int easize, int cookiesize) if (lmv->connected == 0) RETURN(0); - /* FIXME: error handling? */ - for (i = 0; i < lmv->desc.ld_tgt_count; i++) + for (i = 0; i < lmv->desc.ld_tgt_count; i++) { + if (lmv->tgts[i].ltd_exp == NULL) { + CWARN("%s: NULL export for %d\n", obd->obd_name, i); + continue; + } + rc = obd_init_ea_size(lmv->tgts[i].ltd_exp, easize, cookiesize); + if (rc) { + CERROR("obd_init_ea_size() failed on MDT target %d, " + "error %d.\n", i, rc); + break; + } + } RETURN(rc); } int lmv_obd_create_single(struct obd_export *exp, struct obdo *oa, + void *acl, int acl_size, struct lov_stripe_md **ea, struct obd_trans_info *oti) { struct obd_device *obd = exp->exp_obd; @@ -1331,30 +1664,38 @@ int lmv_obd_create_single(struct obd_export *exp, struct obdo *oa, int rc = 0; ENTRY; - rc = lmv_check_connect(obd); - if (rc) - RETURN(rc); - LASSERT(ea == NULL); LASSERT(oa->o_mds < lmv->desc.ld_tgt_count); - rc = obd_create(lmv->tgts[oa->o_mds].ltd_exp, oa, &obj_mdp, oti); - LASSERT(rc == 0); + rc = obd_create(lmv->tgts[oa->o_mds].ltd_exp, oa, + acl, acl_size, &obj_mdp, oti); RETURN(rc); } +int lmv_getready(struct obd_export *exp) +{ + struct obd_device *obd = exp->exp_obd; + int rc = 0; + + ENTRY; + rc = lmv_check_connect(obd); + RETURN(rc); +} + /* - * to be called from MDS only + * to be called from MDS only. @oa should have correct store cookie and o_fid + * values for "master" object, as it will be used. */ int lmv_obd_create(struct obd_export *exp, struct obdo *oa, + void *acl, int acl_size, struct lov_stripe_md **ea, struct obd_trans_info *oti) { struct obd_device *obd = exp->exp_obd; struct lmv_obd *lmv = &obd->u.lmv; - struct mea *mea; + struct lustre_id mid; int i, c, rc = 0; - struct ll_fid mfid; + struct mea *mea; ENTRY; rc = lmv_check_connect(obd); @@ -1364,70 +1705,137 @@ int lmv_obd_create(struct obd_export *exp, struct obdo *oa, LASSERT(oa != NULL); if (ea == NULL) { - rc = lmv_obd_create_single(exp, oa, NULL, oti); + rc = lmv_obd_create_single(exp, oa, acl, acl_size, NULL, oti); + if (rc) + CERROR("Can't create object, rc = %d\n", rc); RETURN(rc); } + /* acl is only suppied when mds create single remote obj */ + LASSERT(acl == NULL && acl_size == 0); + if (*ea == NULL) { rc = obd_alloc_diskmd(exp, (struct lov_mds_md **)ea); - LASSERT(*ea != NULL); + if (rc < 0) { + CERROR("obd_alloc_diskmd() failed, error %d\n", + rc); + RETURN(rc); + } else + rc = 0; + + if (*ea == NULL) + RETURN(-ENOMEM); } + /* + * here we should take care about splitted dir, so store cookie and fid + * for "master" object should already be allocated and passed in @oa. + */ + LASSERT(oa->o_id != 0); + LASSERT(oa->o_fid != 0); + + /* save "master" object id */ + obdo2id(&mid, oa); + mea = (struct mea *)*ea; - mfid.id = oa->o_id; - mfid.generation = oa->o_generation; - rc = 0; + mea->mea_master = -1; + mea->mea_magic = MEA_MAGIC_ALL_CHARS; + if (!mea->mea_count || mea->mea_count > lmv->desc.ld_tgt_count) mea->mea_count = lmv->desc.ld_tgt_count; - mea->mea_master = -1; - - /* FIXME: error handling? */ - for (i = 0, c = 0; c < mea->mea_count && - i < lmv->desc.ld_tgt_count; i++) { + for (i = 0, c = 0; c < mea->mea_count && i < lmv->desc.ld_tgt_count; i++) { struct lov_stripe_md obj_md; struct lov_stripe_md *obj_mdp = &obj_md; if (lmv->tgts[i].ltd_exp == NULL) { - /* this is master MDS */ - mea->mea_fids[c].id = mfid.id; - mea->mea_fids[c].generation = mfid.generation; - mea->mea_fids[c].mds = i; + /* this is "master" MDS */ mea->mea_master = i; + mea->mea_ids[c] = mid; c++; continue; } - /* "Master" MDS should always be part of stripped dir, so - scan for it */ + /* + * "master" MDS should always be part of stripped dir, + * so scan for it. + */ if (mea->mea_master == -1 && c == mea->mea_count - 1) continue; - oa->o_valid = OBD_MD_FLGENER | OBD_MD_FLTYPE | OBD_MD_FLMODE - | OBD_MD_FLUID | OBD_MD_FLGID | OBD_MD_FLID; + oa->o_valid = OBD_MD_FLGENER | OBD_MD_FLTYPE | OBD_MD_FLMODE | + OBD_MD_FLUID | OBD_MD_FLGID | OBD_MD_FLID; - rc = obd_create(lmv->tgts[c].ltd_exp, oa, &obj_mdp, oti); - /* FIXME: error handling here */ - LASSERT(rc == 0); + rc = obd_create(lmv->tgts[c].ltd_exp, oa, NULL, 0, + &obj_mdp, oti); + if (rc) { + CERROR("obd_create() failed on MDT target %d, " + "error %d\n", c, rc); + RETURN(rc); + } - mea->mea_fids[c].id = oa->o_id; - mea->mea_fids[c].generation = oa->o_generation; - mea->mea_fids[c].mds = i; - c++; CDEBUG(D_OTHER, "dirobj at mds %d: "LPU64"/%u\n", i, oa->o_id, oa->o_generation); + + + /* + * here, when object is created (or it is master and was passed + * from caller) on desired MDS we save its fid to local mea_ids. + */ + LASSERT(oa->o_fid); + + /* + * store cookie should be defined here for both cases (master + * object and not master), because master is already created. + */ + LASSERT(oa->o_id); + + /* fill mea by store cookie and fid */ + obdo2id(&mea->mea_ids[c], oa); + c++; } LASSERT(c == mea->mea_count); - CDEBUG(D_OTHER, "%d dirobjects created\n", (int) mea->mea_count); + + CDEBUG(D_OTHER, "%d dirobjects created\n", + (int)mea->mea_count); + + RETURN(rc); +} + +static int lmv_llog_init(struct obd_device *obd, struct obd_llogs *llogs, + struct obd_device *tgt, int count, + struct llog_catid *logid) +{ + struct llog_ctxt *ctxt; + int rc; + ENTRY; + + rc = obd_llog_setup(obd, llogs, LLOG_CONFIG_REPL_CTXT, tgt, 0, NULL, + &llog_client_ops); + if (rc == 0) { + ctxt = llog_get_context(llogs, LLOG_CONFIG_REPL_CTXT); + ctxt->loc_imp = tgt->u.cli.cl_import; + } RETURN(rc); } +static int lmv_llog_finish(struct obd_device *obd, + struct obd_llogs *llogs, int count) +{ + int rc; + ENTRY; + + rc = obd_llog_cleanup(llog_get_context(llogs, LLOG_CONFIG_REPL_CTXT)); + RETURN(rc); +} + static int lmv_get_info(struct obd_export *exp, __u32 keylen, void *key, __u32 *vallen, void *val) { struct obd_device *obd; struct lmv_obd *lmv; + int rc = 0; ENTRY; obd = class_exp2obd(exp); @@ -1441,8 +1849,8 @@ static int lmv_get_info(struct obd_export *exp, __u32 keylen, if (keylen == 6 && memcmp(key, "mdsize", 6) == 0) { __u32 *mdsize = val; *vallen = sizeof(__u32); - *mdsize = sizeof(struct ll_fid) * lmv->desc.ld_tgt_count - + sizeof(struct mea); + *mdsize = sizeof(struct lustre_id) * lmv->desc.ld_tgt_count + + sizeof(struct mea); RETURN(0); } else if (keylen == 6 && memcmp(key, "mdsnum", 6) == 0) { struct obd_uuid *cluuid = &lmv->cluuid; @@ -1458,6 +1866,35 @@ static int lmv_get_info(struct obd_export *exp, __u32 keylen, } } LASSERT(0); + } else if (keylen == 6 && memcmp(key, "rootid", 6) == 0) { + /* getting rootid from first MDS. */ + rc = obd_get_info(lmv->tgts[0].ltd_exp, keylen, key, + vallen, val); + RETURN(rc); + } else if (keylen >= strlen("lmvdesc") && strcmp(key, "lmvdesc") == 0) { + struct lmv_desc *desc_ret = val; + *desc_ret = lmv->desc; + RETURN(0); + } else if (keylen == strlen("remote_flag") && + !strcmp(key, "remote_flag")) { + struct lmv_tgt_desc *tgts; + int i; + + LASSERT(*vallen == sizeof(__u32)); + for (i = 0, tgts = lmv->tgts; i < lmv->desc.ld_tgt_count; + i++, tgts++) { + + /* all tgts should be connected when this get called. */ + if (!tgts || !tgts->ltd_exp) { + CERROR("target not setup?\n"); + continue; + } + + if (!obd_get_info(tgts->ltd_exp, keylen, key, + vallen, val)) + RETURN(0); + } + RETURN(-EINVAL); } CDEBUG(D_IOCTL, "invalid key\n"); @@ -1467,8 +1904,9 @@ static int lmv_get_info(struct obd_export *exp, __u32 keylen, int lmv_set_info(struct obd_export *exp, obd_count keylen, void *key, obd_count vallen, void *val) { - struct obd_device *obd; - struct lmv_obd *lmv; + struct lmv_tgt_desc *tgt; + struct obd_device *obd; + struct lmv_obd *lmv; ENTRY; obd = class_exp2obd(exp); @@ -1479,27 +1917,67 @@ int lmv_set_info(struct obd_export *exp, obd_count keylen, } lmv = &obd->u.lmv; - if (keylen >= strlen("client") && strcmp(key, "client") == 0) { - struct lmv_tgt_desc *tgts; - int i, rc; + if (keylen >= strlen("inter_mds") && strcmp(key, "inter_mds") == 0) { + lmv->server_timeout = 1; + lmv_set_timeouts(obd); + RETURN(0); + } - rc = lmv_check_connect(obd); - if (rc) - RETURN(rc); + /* maybe this could be default */ + if ((keylen == strlen("sec") && strcmp(key, "sec") == 0) || + (keylen == strlen("nllu") && strcmp(key, "nllu") == 0)) { + struct obd_export *exp; + int rc = 0, err, i; + + spin_lock(&lmv->lmv_lock); + for (i = 0, tgt = lmv->tgts; i < lmv->desc.ld_tgt_count; + i++, tgt++) { + exp = tgt->ltd_exp; + /* during setup time the connections to mdc might + * haven't been established. + */ + if (exp == NULL) { + struct obd_device *tgt_obd; + + tgt_obd = class_find_client_obd(&tgt->uuid, + LUSTRE_MDC_NAME, + &obd->obd_uuid); + if (!tgt_obd) { + CERROR("can't set info %s, " + "device %s not attached?\n", + (char *) key, tgt->uuid.uuid); + rc = -EINVAL; + continue; + } + exp = tgt_obd->obd_self_export; + } + + err = obd_set_info(exp, keylen, key, vallen, val); + if (!rc) + rc = err; + } + spin_unlock(&lmv->lmv_lock); + + RETURN(rc); + } + + if ((keylen == strlen("flush_cred") && + strcmp(key, "flush_cred") == 0)) { + int rc = 0, i; - for (i = 0, tgts = lmv->tgts; - i < lmv->desc.ld_tgt_count; i++, tgts++) { - rc = obd_set_info(tgts->ltd_exp, keylen, key, vallen, val); + for (i = 0, tgt = lmv->tgts; i < lmv->desc.ld_tgt_count; + i++, tgt++) { + if (!tgt->ltd_exp) + continue; + rc = obd_set_info(tgt->ltd_exp, + keylen, key, vallen, val); if (rc) RETURN(rc); } - RETURN(0); - } else if (keylen >= strlen("inter_mds") && strcmp(key, "inter_mds") == 0) { - lmv->server_timeout = 1; - lmv_set_timeouts(obd); + RETURN(0); } - + RETURN(-EINVAL); } @@ -1508,11 +1986,12 @@ int lmv_packmd(struct obd_export *exp, struct lov_mds_md **lmmp, { struct obd_device *obd = class_exp2obd(exp); struct lmv_obd *lmv = &obd->u.lmv; - int mea_size; + struct mea *meap, *lsmp; + int mea_size, i; ENTRY; - mea_size = sizeof(struct ll_fid) * - lmv->desc.ld_tgt_count + sizeof(struct mea); + mea_size = (sizeof(struct lustre_id) * + lmv->desc.ld_tgt_count) + sizeof(struct mea); if (!lmmp) RETURN(mea_size); @@ -1522,57 +2001,102 @@ int lmv_packmd(struct obd_export *exp, struct lov_mds_md **lmmp, RETURN(0); } - if (!*lmmp) { + if (*lmmp == NULL) { OBD_ALLOC(*lmmp, mea_size); - if (!*lmmp) + if (*lmmp == NULL) RETURN(-ENOMEM); } if (!lsm) RETURN(mea_size); -#warning "MEA packing/convertation must be here! -bzzz" - memcpy(*lmmp, lsm, mea_size); + lsmp = (struct mea *)lsm; + meap = (struct mea *)*lmmp; + + if (lsmp->mea_magic != MEA_MAGIC_LAST_CHAR && + lsmp->mea_magic != MEA_MAGIC_ALL_CHARS) + RETURN(-EINVAL); + + meap->mea_magic = cpu_to_le32(lsmp->mea_magic); + meap->mea_count = cpu_to_le32(lsmp->mea_count); + meap->mea_master = cpu_to_le32(lsmp->mea_master); + + for (i = 0; i < lmv->desc.ld_tgt_count; i++) { + meap->mea_ids[i] = meap->mea_ids[i]; + id_cpu_to_le(&meap->mea_ids[i]); + } + RETURN(mea_size); } -int lmv_unpackmd(struct obd_export *exp, struct lov_stripe_md **mem_tgt, - struct lov_mds_md *disk_src, int mdsize) +int lmv_unpackmd(struct obd_export *exp, struct lov_stripe_md **lsmp, + struct lov_mds_md *lmm, int lmm_size) { struct obd_device *obd = class_exp2obd(exp); + struct mea **tmea = (struct mea **)lsmp; + struct mea *mea = (struct mea *)lmm; struct lmv_obd *lmv = &obd->u.lmv; - struct mea **tmea = (struct mea **) mem_tgt; - struct mea *mea = (void *) disk_src; - int mea_size; + int mea_size, i, rc = 0; + __u32 magic; ENTRY; - mea_size = sizeof(struct ll_fid) * + mea_size = sizeof(struct lustre_id) * lmv->desc.ld_tgt_count + sizeof(struct mea); - if (mem_tgt == NULL) + + if (lsmp == NULL) return mea_size; - if (*mem_tgt != NULL && disk_src == NULL) { + if (*lsmp != NULL && lmm == NULL) { OBD_FREE(*tmea, mea_size); RETURN(0); } - LASSERT(mea_size == mdsize); + LASSERT(mea_size == lmm_size); OBD_ALLOC(*tmea, mea_size); - /* FIXME: error handling here */ - LASSERT(*tmea != NULL); + if (*tmea == NULL) + RETURN(-ENOMEM); - if (!disk_src) + if (!lmm) RETURN(mea_size); -#warning "MEA unpacking/convertation must be here! -bzzz" - memcpy(*tmea, mea, mdsize); + if (mea->mea_magic == MEA_MAGIC_LAST_CHAR || + mea->mea_magic == MEA_MAGIC_ALL_CHARS) + { + magic = le32_to_cpu(mea->mea_magic); + } else { + struct mea_old *old = (struct mea_old *)lmm; + + mea_size = sizeof(struct lustre_id) * old->mea_count + + sizeof(struct mea_old); + + if (old->mea_count > 256 || old->mea_master > 256 || + lmm_size < mea_size || old->mea_master > old->mea_count) { + CWARN("bad MEA: count %u, master %u, size %u\n", + old->mea_count, old->mea_master, mea_size); + GOTO(out_free_mea, rc = -EINVAL); + } + magic = MEA_MAGIC_LAST_CHAR; + } + + (*tmea)->mea_magic = magic; + (*tmea)->mea_count = le32_to_cpu(mea->mea_count); + (*tmea)->mea_master = le32_to_cpu(mea->mea_master); + + for (i = 0; i < (*tmea)->mea_count; i++) { + (*tmea)->mea_ids[i] = mea->mea_ids[i]; + id_le_to_cpu(&(*tmea)->mea_ids[i]); + } RETURN(mea_size); + +out_free_mea: + OBD_FREE(*tmea, mea_size); + return rc; } int lmv_brw(int rw, struct obd_export *exp, struct obdo *oa, - struct lov_stripe_md *ea, obd_count oa_bufs, - struct brw_page *pgarr, struct obd_trans_info *oti) + struct lov_stripe_md *ea, obd_count oa_bufs, + struct brw_page *pgarr, struct obd_trans_info *oti) { struct obd_device *obd = exp->exp_obd; struct lmv_obd *lmv = &obd->u.lmv; @@ -1584,14 +2108,38 @@ int lmv_brw(int rw, struct obd_export *exp, struct obdo *oa, LASSERT(pgarr != NULL); LASSERT(oa->o_mds < lmv->desc.ld_tgt_count); - oa->o_gr = mea->mea_fids[oa->o_mds].generation; - oa->o_id = mea->mea_fids[oa->o_mds].id; - oa->o_valid = OBD_MD_FLID | OBD_MD_FLGROUP; - err = obd_brw(rw, lmv->tgts[oa->o_mds].ltd_exp, oa, - NULL, oa_bufs, pgarr, oti); + oa->o_gr = id_gen(&mea->mea_ids[oa->o_mds]); + oa->o_id = id_ino(&mea->mea_ids[oa->o_mds]); + oa->o_valid = OBD_MD_FLID | OBD_MD_FLGROUP; + + err = obd_brw(rw, lmv->tgts[oa->o_mds].ltd_exp, + oa, NULL, oa_bufs, pgarr, oti); RETURN(err); } +static int lmv_cancel_unused(struct obd_export *exp, + struct lov_stripe_md *lsm, + int flags, void *opaque) +{ + struct obd_device *obd = exp->exp_obd; + struct lmv_obd *lmv = &obd->u.lmv; + int rc = 0, err, i; + ENTRY; + + LASSERT(lsm == NULL); + + for (i = 0; i < lmv->desc.ld_tgt_count; i++) { + if (!lmv->tgts[i].ltd_exp || !lmv->tgts[i].active) + continue; + + err = obd_cancel_unused(lmv->tgts[i].ltd_exp, + NULL, flags, opaque); + if (!rc) + rc = err; + } + RETURN(rc); +} + struct obd_ops lmv_obd_ops = { .o_owner = THIS_MODULE, .o_attach = lmv_attach, @@ -1601,6 +2149,8 @@ struct obd_ops lmv_obd_ops = { .o_connect = lmv_connect, .o_disconnect = lmv_disconnect, .o_statfs = lmv_statfs, + .o_llog_init = lmv_llog_init, + .o_llog_finish = lmv_llog_finish, .o_get_info = lmv_get_info, .o_set_info = lmv_set_info, .o_create = lmv_obd_create, @@ -1610,27 +2160,30 @@ struct obd_ops lmv_obd_ops = { .o_init_ea_size = lmv_init_ea_size, .o_notify = lmv_notify, .o_iocontrol = lmv_iocontrol, + .o_getready = lmv_getready, + .o_cancel_unused = lmv_cancel_unused, }; struct md_ops lmv_md_ops = { - .m_getstatus = lmv_getstatus, - .m_getattr = lmv_getattr, - .m_change_cbdata = lmv_change_cbdata, - .m_change_cbdata_name = lmv_change_cbdata_name, - .m_close = lmv_close, - .m_create = lmv_create, - .m_done_writing = lmv_done_writing, - .m_enqueue = lmv_enqueue, - .m_getattr_name = lmv_getattr_name, - .m_intent_lock = lmv_intent_lock, - .m_link = lmv_link, - .m_rename = lmv_rename, - .m_setattr = lmv_setattr, - .m_sync = lmv_sync, - .m_readpage = lmv_readpage, - .m_unlink = lmv_unlink, - .m_get_real_obd = lmv_get_real_obd, - .m_valid_attrs = lmv_valid_attrs, + .m_getstatus = lmv_getstatus, + .m_getattr = lmv_getattr, + .m_change_cbdata = lmv_change_cbdata, + .m_change_cbdata_name = lmv_change_cbdata_name, + .m_close = lmv_close, + .m_create = lmv_create, + .m_done_writing = lmv_done_writing, + .m_enqueue = lmv_enqueue, + .m_getattr_lock = lmv_getattr_lock, + .m_intent_lock = lmv_intent_lock, + .m_link = lmv_link, + .m_rename = lmv_rename, + .m_setattr = lmv_setattr, + .m_sync = lmv_sync, + .m_readpage = lmv_readpage, + .m_unlink = lmv_unlink, + .m_get_real_obd = lmv_get_real_obd, + .m_valid_attrs = lmv_valid_attrs, + .m_delete_inode = lmv_delete_inode, }; int __init lmv_init(void) @@ -1638,16 +2191,32 @@ int __init lmv_init(void) struct lprocfs_static_vars lvars; int rc; + obj_cache = kmem_cache_create("lmv_objects", + sizeof(struct lmv_obj), + 0, 0, NULL, NULL); + if (!obj_cache) { + CERROR("error allocating lmv objects cache\n"); + return -ENOMEM; + } + lprocfs_init_vars(lmv, &lvars); rc = class_register_type(&lmv_obd_ops, &lmv_md_ops, - lvars.module_vars, OBD_LMV_DEVICENAME); - RETURN(rc); + lvars.module_vars, + OBD_LMV_DEVICENAME); + if (rc) + kmem_cache_destroy(obj_cache); + + return rc; } #ifdef __KERNEL__ static void lmv_exit(void) { class_unregister_type(OBD_LMV_DEVICENAME); + + LASSERTF(kmem_cache_destroy(obj_cache) == 0, + "can't free lmv objects cache, %d object(s)" + "still in use\n", atomic_read(&obj_cache_count)); } MODULE_AUTHOR("Cluster File Systems, Inc. ");