X-Git-Url: https://git.whamcloud.com/?p=fs%2Flustre-release.git;a=blobdiff_plain;f=lustre%2Flmv%2Flmv_obd.c;h=8750215026e5d971a8d3be7120c3a8d749ad43b0;hp=88452beef4861ac4cbaf531e3fb300b4ac2fb0f3;hb=003df3c38fe74a092f75569793edd6ec5a387d5c;hpb=3f92a017ae2d97506db89b34c374bbcf9670db54;ds=sidebyside diff --git a/lustre/lmv/lmv_obd.c b/lustre/lmv/lmv_obd.c index 88452be..8750215 100644 --- a/lustre/lmv/lmv_obd.c +++ b/lustre/lmv/lmv_obd.c @@ -42,7 +42,7 @@ #include #include #include -#include +#include #include #include #else @@ -54,11 +54,74 @@ #include #include #include +#include #include +#include +#include #include #include +#include #include "lmv_internal.h" +/* This hash is only for testing purpose */ +static inline unsigned int +lmv_hash_all_chars(unsigned int count, const char *name, int namelen) +{ + unsigned int c = 0; + const unsigned char *p = (const unsigned char *)name; + + while (--namelen >= 0) + c += p[namelen]; + + c = c % count; + + return c; +} + +static inline unsigned int +lmv_hash_fnv1a(unsigned int count, const char *name, int namelen) +{ + __u64 hash; + + hash = lustre_hash_fnv_1a_64(name, namelen); + + hash = hash % count; + + return hash; +} + +int lmv_name_to_stripe_index(__u32 lmv_hash_type, unsigned int stripe_count, + const char *name, int namelen) +{ + int idx; + __u32 hash_type = lmv_hash_type & LMV_HASH_TYPE_MASK; + + LASSERT(namelen > 0); + if (stripe_count <= 1) + return 0; + + /* for migrating object, always start from 0 stripe */ + if (lmv_hash_type & LMV_HASH_FLAG_MIGRATION) + return 0; + + switch (hash_type) { + case LMV_HASH_TYPE_ALL_CHARS: + idx = lmv_hash_all_chars(stripe_count, name, namelen); + break; + case LMV_HASH_TYPE_FNV_1A_64: + idx = lmv_hash_fnv1a(stripe_count, name, namelen); + break; + default: + CERROR("Unknown hash type 0x%x\n", hash_type); + return -EINVAL; + } + + CDEBUG(D_INFO, "name %.*s hash_type %d idx %d\n", namelen, name, + hash_type, idx); + + return idx; +} + static void lmv_activate_target(struct lmv_obd *lmv, struct lmv_tgt_desc *tgt, int activate) @@ -133,9 +196,10 @@ static int lmv_set_mdc_active(struct lmv_obd *lmv, struct obd_uuid *lmv_get_uuid(struct obd_export *exp) { - struct lmv_obd *lmv = &exp->exp_obd->u.lmv; + struct lmv_obd *lmv = &exp->exp_obd->u.lmv; + struct lmv_tgt_desc *tgt = lmv->tgts[0]; - return obd_get_uuid(lmv->tgts[0]->ltd_exp); + return (tgt == NULL) ? NULL : obd_get_uuid(tgt->ltd_exp); } static int lmv_notify(struct obd_device *obd, struct obd_device *watched, @@ -203,9 +267,6 @@ static int lmv_connect(const struct lu_env *env, struct obd_uuid *cluuid, struct obd_connect_data *data, void *localdata) { -#ifdef __KERNEL__ - struct proc_dir_entry *lmv_proc_dir; -#endif struct lmv_obd *lmv = &obd->u.lmv; struct lustre_handle conn = { 0 }; int rc = 0; @@ -237,15 +298,16 @@ static int lmv_connect(const struct lu_env *env, if (data) lmv->conn_data = *data; -#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 + if (lmv->targets_proc_entry == NULL) { + lmv->targets_proc_entry = lprocfs_seq_register("target_obds", + obd->obd_proc_entry, + NULL, NULL); + if (IS_ERR(lmv->targets_proc_entry)) { + CERROR("could not register /proc/fs/lustre/%s/%s/target_obds.", + obd->obd_type->typ_name, obd->obd_name); + lmv->targets_proc_entry = NULL; + } + } /* * All real clients should perform actual connection right away, because @@ -253,22 +315,16 @@ static int lmv_connect(const struct lu_env *env, * and MDC stuff will be called directly, for instance while reading * ../mdc/../kbytesfree procfs file, etc. */ - if (data->ocd_connect_flags & OBD_CONNECT_REAL) + if (data != NULL && (data->ocd_connect_flags & OBD_CONNECT_REAL)) rc = lmv_check_connect(obd); -#ifdef __KERNEL__ - if (rc) { - if (lmv_proc_dir) - lprocfs_remove(&lmv_proc_dir); - } -#endif - - RETURN(rc); + if (rc && lmv->targets_proc_entry != NULL) + lprocfs_remove(&lmv->targets_proc_entry); + RETURN(rc); } static void lmv_set_timeouts(struct obd_device *obd) { - struct lmv_tgt_desc *tgt; struct lmv_obd *lmv; __u32 i; @@ -280,8 +336,9 @@ static void lmv_set_timeouts(struct obd_device *obd) return; for (i = 0; i < lmv->desc.ld_tgt_count; i++) { - tgt = lmv->tgts[i]; - if (tgt == NULL || tgt->ltd_exp == NULL || tgt->ltd_active == 0) + struct lmv_tgt_desc *tgt = lmv->tgts[i]; + + if (tgt == NULL || tgt->ltd_exp == NULL || !tgt->ltd_active) continue; obd_set_info_async(NULL, tgt->ltd_exp, sizeof(KEY_INTERMDS), @@ -290,7 +347,7 @@ static void lmv_set_timeouts(struct obd_device *obd) } static int lmv_init_ea_size(struct obd_export *exp, int easize, - int def_easize, int cookiesize) + int def_easize, int cookiesize, int def_cookiesize) { struct obd_device *obd = exp->exp_obd; struct lmv_obd *lmv = &obd->u.lmv; @@ -311,22 +368,26 @@ static int lmv_init_ea_size(struct obd_export *exp, int easize, lmv->max_cookiesize = cookiesize; change = 1; } - if (change == 0) - RETURN(0); + if (lmv->max_def_cookiesize < def_cookiesize) { + lmv->max_def_cookiesize = def_cookiesize; + change = 1; + } + if (change == 0) + RETURN(0); - if (lmv->connected == 0) - RETURN(0); + if (lmv->connected == 0) + RETURN(0); for (i = 0; i < lmv->desc.ld_tgt_count; i++) { - if (lmv->tgts[i] == NULL || - lmv->tgts[i]->ltd_exp == NULL || - lmv->tgts[i]->ltd_active == 0) { + struct lmv_tgt_desc *tgt = lmv->tgts[i]; + + if (tgt == NULL || tgt->ltd_exp == NULL || !tgt->ltd_active) { CWARN("%s: NULL export for %d\n", obd->obd_name, i); continue; } - rc = md_init_ea_size(lmv->tgts[i]->ltd_exp, easize, def_easize, - cookiesize); + rc = md_init_ea_size(tgt->ltd_exp, easize, def_easize, + cookiesize, def_cookiesize); if (rc) { CERROR("%s: obd_init_ea_size() failed on MDT target %d:" " rc = %d.\n", obd->obd_name, i, rc); @@ -340,9 +401,6 @@ static int lmv_init_ea_size(struct obd_export *exp, int easize, int lmv_connect_mdc(struct obd_device *obd, struct lmv_tgt_desc *tgt) { -#ifdef __KERNEL__ - struct proc_dir_entry *lmv_proc_dir; -#endif struct lmv_obd *lmv = &obd->u.lmv; struct obd_uuid *cluuid = &lmv->cluuid; struct obd_uuid lmv_mdc_uuid = { "LMV_MDC_UUID" }; @@ -410,40 +468,35 @@ int lmv_connect_mdc(struct obd_device *obd, struct lmv_tgt_desc *tgt) } } - tgt->ltd_active = 1; - tgt->ltd_exp = mdc_exp; - lmv->desc.ld_active_tgt_count++; - - md_init_ea_size(tgt->ltd_exp, lmv->max_easize, - lmv->max_def_easize, lmv->max_cookiesize); - - CDEBUG(D_CONFIG, "Connected to %s(%s) successfully (%d)\n", - mdc_obd->obd_name, mdc_obd->obd_uuid.uuid, - cfs_atomic_read(&obd->obd_refcount)); - -#ifdef __KERNEL__ - lmv_proc_dir = lprocfs_srch(obd->obd_proc_entry, "target_obds"); - if (lmv_proc_dir) { - struct proc_dir_entry *mdc_symlink; - - LASSERT(mdc_obd->obd_type != NULL); - LASSERT(mdc_obd->obd_type->typ_name != NULL); - mdc_symlink = lprocfs_add_symlink(mdc_obd->obd_name, - lmv_proc_dir, - "../../../%s/%s", - mdc_obd->obd_type->typ_name, - mdc_obd->obd_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 - RETURN(0); + tgt->ltd_active = 1; + tgt->ltd_exp = mdc_exp; + lmv->desc.ld_active_tgt_count++; + + md_init_ea_size(tgt->ltd_exp, lmv->max_easize, lmv->max_def_easize, + lmv->max_cookiesize, lmv->max_def_cookiesize); + + CDEBUG(D_CONFIG, "Connected to %s(%s) successfully (%d)\n", + mdc_obd->obd_name, mdc_obd->obd_uuid.uuid, + atomic_read(&obd->obd_refcount)); + + if (lmv->targets_proc_entry != NULL) { + struct proc_dir_entry *mdc_symlink; + + LASSERT(mdc_obd->obd_type != NULL); + LASSERT(mdc_obd->obd_type->typ_name != NULL); + mdc_symlink = lprocfs_add_symlink(mdc_obd->obd_name, + lmv->targets_proc_entry, + "../../../%s/%s", + mdc_obd->obd_type->typ_name, + mdc_obd->obd_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); + } + } + RETURN(0); } static void lmv_del_target(struct lmv_obd *lmv, int index) @@ -461,6 +514,7 @@ static int lmv_add_target(struct obd_device *obd, struct obd_uuid *uuidp, { struct lmv_obd *lmv = &obd->u.lmv; struct lmv_tgt_desc *tgt; + int orig_tgt_count = 0; int rc = 0; ENTRY; @@ -532,26 +586,28 @@ static int lmv_add_target(struct obd_device *obd, struct obd_uuid *uuidp, tgt->ltd_uuid = *uuidp; tgt->ltd_active = 0; lmv->tgts[index] = tgt; - if (index >= lmv->desc.ld_tgt_count) + if (index >= lmv->desc.ld_tgt_count) { + orig_tgt_count = lmv->desc.ld_tgt_count; lmv->desc.ld_tgt_count = index + 1; + } if (lmv->connected) { rc = lmv_connect_mdc(obd, tgt); - if (rc) { + if (rc != 0) { spin_lock(&lmv->lmv_lock); - lmv->desc.ld_tgt_count--; + if (lmv->desc.ld_tgt_count == index + 1) + lmv->desc.ld_tgt_count = orig_tgt_count; memset(tgt, 0, sizeof(*tgt)); spin_unlock(&lmv->lmv_lock); - } else { - int easize = sizeof(struct lmv_stripe_md) + - lmv->desc.ld_tgt_count * - sizeof(struct lu_fid); - lmv_init_ea_size(obd->obd_self_export, easize, 0, 0); - } - } + } else { + int easize = sizeof(struct lmv_stripe_md) + + lmv->desc.ld_tgt_count * sizeof(struct lu_fid); + lmv_init_ea_size(obd->obd_self_export, easize, 0, 0, 0); + } + } - lmv_init_unlock(lmv); - RETURN(rc); + lmv_init_unlock(lmv); + RETURN(rc); } int lmv_check_connect(struct obd_device *obd) @@ -578,11 +634,18 @@ int lmv_check_connect(struct obd_device *obd) RETURN(-EINVAL); } - CDEBUG(D_CONFIG, "Time to connect %s to %s\n", - lmv->cluuid.uuid, obd->obd_name); - LASSERT(lmv->tgts != NULL); + if (lmv->tgts[0] == NULL) { + lmv_init_unlock(lmv); + CERROR("%s: no target configured for index 0.\n", + obd->obd_name); + RETURN(-EINVAL); + } + + CDEBUG(D_CONFIG, "Time to connect %s to %s\n", + lmv->cluuid.uuid, obd->obd_name); + for (i = 0; i < lmv->desc.ld_tgt_count; i++) { tgt = lmv->tgts[i]; if (tgt == NULL) @@ -592,13 +655,13 @@ int lmv_check_connect(struct obd_device *obd) GOTO(out_disc, rc); } - lmv_set_timeouts(obd); - class_export_put(lmv->exp); - lmv->connected = 1; - easize = lmv_get_easize(lmv); - lmv_init_ea_size(obd->obd_self_export, easize, 0, 0); - lmv_init_unlock(lmv); - RETURN(0); + lmv_set_timeouts(obd); + class_export_put(lmv->exp); + lmv->connected = 1; + easize = lmv_mds_md_size(lmv->desc.ld_tgt_count, LMV_MAGIC); + lmv_init_ea_size(obd->obd_self_export, easize, 0, 0, 0); + lmv_init_unlock(lmv); + RETURN(0); out_disc: while (i-- > 0) { @@ -624,9 +687,6 @@ int lmv_check_connect(struct obd_device *obd) static int lmv_disconnect_mdc(struct obd_device *obd, struct lmv_tgt_desc *tgt) { -#ifdef __KERNEL__ - struct proc_dir_entry *lmv_proc_dir; -#endif struct lmv_obd *lmv = &obd->u.lmv; struct obd_device *mdc_obd; int rc; @@ -643,21 +703,10 @@ static int lmv_disconnect_mdc(struct obd_device *obd, struct lmv_tgt_desc *tgt) mdc_obd->obd_no_recov = obd->obd_no_recov; } -#ifdef __KERNEL__ - lmv_proc_dir = lprocfs_srch(obd->obd_proc_entry, "target_obds"); - if (lmv_proc_dir) { - struct proc_dir_entry *mdc_symlink; + if (lmv->targets_proc_entry != NULL) + lprocfs_remove_proc_entry(mdc_obd->obd_name, + lmv->targets_proc_entry); - 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 rc = obd_fid_fini(tgt->ltd_exp->exp_obd); if (rc) CERROR("Can't finanize fids factory\n"); @@ -683,9 +732,6 @@ static int lmv_disconnect_mdc(struct obd_device *obd, struct lmv_tgt_desc *tgt) static int lmv_disconnect(struct obd_export *exp) { struct obd_device *obd = class_exp2obd(exp); -#ifdef __KERNEL__ - struct proc_dir_entry *lmv_proc_dir; -#endif struct lmv_obd *lmv = &obd->u.lmv; int rc; __u32 i; @@ -708,15 +754,11 @@ static int lmv_disconnect(struct obd_export *exp) lmv_disconnect_mdc(obd, lmv->tgts[i]); } -#ifdef __KERNEL__ - lmv_proc_dir = lprocfs_srch(obd->obd_proc_entry, "target_obds"); - 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 + if (lmv->targets_proc_entry != NULL) + lprocfs_remove(&lmv->targets_proc_entry); + else + CERROR("/proc/fs/lustre/%s/%s/target_obds missing\n", + obd->obd_type->typ_name, obd->obd_name); out_local: /* @@ -858,55 +900,69 @@ static void lmv_hsm_req_build(struct lmv_obd *lmv, static int lmv_hsm_ct_unregister(struct lmv_obd *lmv, unsigned int cmd, int len, struct lustre_kernelcomm *lk, void *uarg) { - __u32 i; - int rc; + __u32 i; + int rc; + struct kkuc_ct_data *kcd = NULL; ENTRY; /* unregister request (call from llapi_hsm_copytool_fini) */ for (i = 0; i < lmv->desc.ld_tgt_count; i++) { + struct lmv_tgt_desc *tgt = lmv->tgts[i]; + + if (tgt == NULL || tgt->ltd_exp == NULL) + continue; /* best effort: try to clean as much as possible * (continue on error) */ - obd_iocontrol(cmd, lmv->tgts[i]->ltd_exp, len, lk, uarg); + obd_iocontrol(cmd, tgt->ltd_exp, len, lk, uarg); } /* Whatever the result, remove copytool from kuc groups. * Unreached coordinators will get EPIPE on next requests * and will unregister automatically. */ - rc = libcfs_kkuc_group_rem(lk->lk_uid, lk->lk_group); + rc = libcfs_kkuc_group_rem(lk->lk_uid, lk->lk_group, (void **)&kcd); + if (kcd != NULL) + OBD_FREE_PTR(kcd); + RETURN(rc); } static int lmv_hsm_ct_register(struct lmv_obd *lmv, unsigned int cmd, int len, struct lustre_kernelcomm *lk, void *uarg) { - struct file *filp; - __u32 i, j; - int err, rc; - bool any_set = false; + struct file *filp; + __u32 i, j; + int err, rc; + bool any_set = false; + struct kkuc_ct_data *kcd; ENTRY; /* All or nothing: try to register to all MDS. * In case of failure, unregister from previous MDS, * except if it because of inactive target. */ for (i = 0; i < lmv->desc.ld_tgt_count; i++) { - err = obd_iocontrol(cmd, lmv->tgts[i]->ltd_exp, - len, lk, uarg); + struct lmv_tgt_desc *tgt = lmv->tgts[i]; + + if (tgt == NULL || tgt->ltd_exp == NULL) + continue; + err = obd_iocontrol(cmd, tgt->ltd_exp, len, lk, uarg); if (err) { - if (lmv->tgts[i]->ltd_active) { + if (tgt->ltd_active) { /* permanent error */ CERROR("%s: iocontrol MDC %s on MDT" " idx %d cmd %x: err = %d\n", class_exp2obd(lmv->exp)->obd_name, - lmv->tgts[i]->ltd_uuid.uuid, - i, cmd, err); + tgt->ltd_uuid.uuid, i, cmd, err); rc = err; lk->lk_flags |= LK_FLG_STOP; /* unregister from previous MDS */ - for (j = 0; j < i; j++) - obd_iocontrol(cmd, - lmv->tgts[j]->ltd_exp, - len, lk, uarg); + for (j = 0; j < i; j++) { + tgt = lmv->tgts[j]; + if (tgt == NULL || tgt->ltd_exp == NULL) + continue; + obd_iocontrol(cmd, tgt->ltd_exp, len, + lk, uarg); + } RETURN(rc); } /* else: transient error. @@ -923,12 +979,25 @@ static int lmv_hsm_ct_register(struct lmv_obd *lmv, unsigned int cmd, int len, /* at least one registration done, with no failure */ filp = fget(lk->lk_wfd); - if (filp == NULL) { + if (filp == NULL) RETURN(-EBADF); - } - rc = libcfs_kkuc_group_add(filp, lk->lk_uid, lk->lk_group, lk->lk_data); - if (rc != 0 && filp != NULL) + + OBD_ALLOC_PTR(kcd); + if (kcd == NULL) { fput(filp); + RETURN(-ENOMEM); + } + kcd->kcd_magic = KKUC_CT_DATA_MAGIC; + kcd->kcd_uuid = lmv->cluuid; + kcd->kcd_archive = lk->lk_data; + + rc = libcfs_kkuc_group_add(filp, lk->lk_uid, lk->lk_group, kcd); + if (rc != 0) { + if (filp != NULL) + fput(filp); + OBD_FREE_PTR(kcd); + } + RETURN(rc); } @@ -940,6 +1009,7 @@ 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; + struct lmv_tgt_desc *tgt = NULL; __u32 i = 0; int rc = 0; int set = 0; @@ -960,11 +1030,11 @@ static int lmv_iocontrol(unsigned int cmd, struct obd_export *exp, if ((index >= count)) RETURN(-ENODEV); - if (lmv->tgts[index] == NULL || - lmv->tgts[index]->ltd_active == 0) + tgt = lmv->tgts[index]; + if (tgt == NULL || !tgt->ltd_active) RETURN(-ENODATA); - mdc_obd = class_exp2obd(lmv->tgts[index]->ltd_exp); + mdc_obd = class_exp2obd(tgt->ltd_exp); if (!mdc_obd) RETURN(-EINVAL); @@ -974,7 +1044,7 @@ static int lmv_iocontrol(unsigned int cmd, struct obd_export *exp, (int) sizeof(struct obd_uuid)))) RETURN(-EFAULT); - rc = obd_statfs(NULL, lmv->tgts[index]->ltd_exp, &stat_buf, + rc = obd_statfs(NULL, tgt->ltd_exp, &stat_buf, cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS), 0); if (rc) @@ -987,12 +1057,11 @@ static int lmv_iocontrol(unsigned int cmd, struct obd_export *exp, } case OBD_IOC_QUOTACTL: { struct if_quotactl *qctl = karg; - struct lmv_tgt_desc *tgt = NULL; struct obd_quotactl *oqctl; - if (qctl->qc_valid == QC_MDTIDX) { - if (qctl->qc_idx < 0 || count <= qctl->qc_idx) - RETURN(-EINVAL); + if (qctl->qc_valid == QC_MDTIDX) { + if (count <= qctl->qc_idx) + RETURN(-EINVAL); tgt = lmv->tgts[qctl->qc_idx]; if (tgt == NULL || tgt->ltd_exp == NULL) @@ -1018,7 +1087,7 @@ static int lmv_iocontrol(unsigned int cmd, struct obd_export *exp, if (i >= count) RETURN(-EAGAIN); - LASSERT(tgt && tgt->ltd_exp); + LASSERT(tgt != NULL && tgt->ltd_exp != NULL); OBD_ALLOC_PTR(oqctl); if (!oqctl) RETURN(-ENOMEM); @@ -1040,18 +1109,17 @@ static int lmv_iocontrol(unsigned int cmd, struct obd_export *exp, if (icc->icc_mdtindex >= count) RETURN(-ENODEV); - if (lmv->tgts[icc->icc_mdtindex] == NULL || - lmv->tgts[icc->icc_mdtindex]->ltd_exp == NULL || - lmv->tgts[icc->icc_mdtindex]->ltd_active == 0) + tgt = lmv->tgts[icc->icc_mdtindex]; + if (tgt == NULL || tgt->ltd_exp == NULL || !tgt->ltd_active) RETURN(-ENODEV); - rc = obd_iocontrol(cmd, lmv->tgts[icc->icc_mdtindex]->ltd_exp, - sizeof(*icc), icc, NULL); + rc = obd_iocontrol(cmd, tgt->ltd_exp, sizeof(*icc), icc, NULL); break; } case LL_IOC_GET_CONNECT_FLAGS: { - if (lmv->tgts[0] == NULL) + tgt = lmv->tgts[0]; + if (tgt == NULL || tgt->ltd_exp == NULL) RETURN(-ENODATA); - rc = obd_iocontrol(cmd, lmv->tgts[0]->ltd_exp, len, karg, uarg); + rc = obd_iocontrol(cmd, tgt->ltd_exp, len, karg, uarg); break; } case OBD_IOC_FID2PATH: { @@ -1062,7 +1130,6 @@ static int lmv_iocontrol(unsigned int cmd, struct obd_export *exp, case LL_IOC_HSM_STATE_SET: case LL_IOC_HSM_ACTION: { struct md_op_data *op_data = karg; - struct lmv_tgt_desc *tgt; tgt = lmv_find_target(lmv, &op_data->op_fid1); if (IS_ERR(tgt)) @@ -1076,7 +1143,6 @@ static int lmv_iocontrol(unsigned int cmd, struct obd_export *exp, } case LL_IOC_HSM_PROGRESS: { const struct hsm_progress_kernel *hpk = karg; - struct lmv_tgt_desc *tgt; tgt = lmv_find_target(lmv, &hpk->hpk_fid); if (IS_ERR(tgt)) @@ -1086,7 +1152,6 @@ static int lmv_iocontrol(unsigned int cmd, struct obd_export *exp, } case LL_IOC_HSM_REQUEST: { struct hsm_user_request *hur = karg; - struct lmv_tgt_desc *tgt; unsigned int reqcount = hur->hur_request.hr_itemcount; if (reqcount == 0) @@ -1108,7 +1173,11 @@ static int lmv_iocontrol(unsigned int cmd, struct obd_export *exp, int rc1; struct hsm_user_request *req; - nr = lmv_hsm_req_count(lmv, hur, lmv->tgts[i]); + tgt = lmv->tgts[i]; + if (tgt == NULL || tgt->ltd_exp == NULL) + continue; + + nr = lmv_hsm_req_count(lmv, hur, tgt); if (nr == 0) /* nothing for this MDS */ continue; @@ -1120,10 +1189,10 @@ static int lmv_iocontrol(unsigned int cmd, struct obd_export *exp, if (req == NULL) RETURN(-ENOMEM); - lmv_hsm_req_build(lmv, hur, lmv->tgts[i], req); + lmv_hsm_req_build(lmv, hur, tgt, req); - rc1 = obd_iocontrol(cmd, lmv->tgts[i]->ltd_exp, - reqlen, req, uarg); + rc1 = obd_iocontrol(cmd, tgt->ltd_exp, reqlen, + req, uarg); if (rc1 != 0 && rc == 0) rc = rc1; OBD_FREE_LARGE(req, reqlen); @@ -1166,23 +1235,21 @@ static int lmv_iocontrol(unsigned int cmd, struct obd_export *exp, struct obd_device *mdc_obd; int err; - if (lmv->tgts[i] == NULL || - lmv->tgts[i]->ltd_exp == NULL) + tgt = lmv->tgts[i]; + if (tgt == NULL || tgt->ltd_exp == NULL) continue; /* ll_umount_begin() sets force flag but for lmv, not * mdc. Let's pass it through */ - mdc_obd = class_exp2obd(lmv->tgts[i]->ltd_exp); + mdc_obd = class_exp2obd(tgt->ltd_exp); mdc_obd->obd_force = obddev->obd_force; - err = obd_iocontrol(cmd, lmv->tgts[i]->ltd_exp, len, - karg, uarg); + err = obd_iocontrol(cmd, tgt->ltd_exp, len, karg, uarg); if (err == -ENODATA && cmd == OBD_IOC_POLL_QUOTACHECK) { RETURN(err); } else if (err) { - if (lmv->tgts[i]->ltd_active) { + if (tgt->ltd_active) { CERROR("error: iocontrol MDC %s on MDT" " idx %d cmd %x: err = %d\n", - lmv->tgts[i]->ltd_uuid.uuid, - i, cmd, err); + tgt->ltd_uuid.uuid, i, cmd, err); if (!rc) rc = err; } @@ -1261,27 +1328,25 @@ static int lmv_placement_policy(struct obd_device *obd, * If stripe_offset is provided during setdirstripe * (setdirstripe -i xx), xx MDS will be choosen. */ - if (op_data->op_cli_flags & CLI_SET_MEA) { + if (op_data->op_cli_flags & CLI_SET_MEA && op_data->op_data != NULL) { struct lmv_user_md *lum; - lum = (struct lmv_user_md *)op_data->op_data; - if (lum->lum_type == LMV_STRIPE_TYPE && - lum->lum_stripe_offset != -1) { - if (lum->lum_stripe_offset >= lmv->desc.ld_tgt_count) { - CERROR("%s: Stripe_offset %d > MDT count %d:" - " rc = %d\n", obd->obd_name, - lum->lum_stripe_offset, - lmv->desc.ld_tgt_count, -ERANGE); - RETURN(-ERANGE); - } - *mds = lum->lum_stripe_offset; - RETURN(0); + lum = op_data->op_data; + + if (le32_to_cpu(lum->lum_stripe_offset) != (__u32)-1) { + *mds = le32_to_cpu(lum->lum_stripe_offset); + } else { + /* -1 means default, which will be in the same MDT with + * the stripe */ + *mds = op_data->op_mds; + lum->lum_stripe_offset = cpu_to_le32(op_data->op_mds); } + } else { + /* Allocate new fid on target according to operation type and + * parent home mds. */ + *mds = op_data->op_mds; } - /* Allocate new fid on target according to operation type and parent - * home mds. */ - *mds = op_data->op_mds; RETURN(0); } @@ -1292,7 +1357,7 @@ int __lmv_fid_alloc(struct lmv_obd *lmv, struct lu_fid *fid, int rc; ENTRY; - tgt = lmv_get_target(lmv, mds); + tgt = lmv_get_target(lmv, mds, NULL); if (IS_ERR(tgt)) RETURN(PTR_ERR(tgt)); @@ -1305,14 +1370,14 @@ int __lmv_fid_alloc(struct lmv_obd *lmv, struct lu_fid *fid, if (tgt->ltd_active == 0 || tgt->ltd_exp == NULL) GOTO(out, rc = -ENODEV); - /* - * Asking underlaying tgt layer to allocate new fid. - */ - rc = obd_fid_alloc(tgt->ltd_exp, fid, NULL); - if (rc > 0) { - LASSERT(fid_is_sane(fid)); - rc = 0; - } + /* + * Asking underlying tgt layer to allocate new fid. + */ + rc = obd_fid_alloc(NULL, tgt->ltd_exp, fid, NULL); + if (rc > 0) { + LASSERT(fid_is_sane(fid)); + rc = 0; + } EXIT; out: @@ -1320,8 +1385,8 @@ out: return rc; } -int lmv_fid_alloc(struct obd_export *exp, struct lu_fid *fid, - struct md_op_data *op_data) +int lmv_fid_alloc(const struct lu_env *env, struct obd_export *exp, + struct lu_fid *fid, struct md_op_data *op_data) { struct obd_device *obd = class_exp2obd(exp); struct lmv_obd *lmv = &obd->u.lmv; @@ -1350,11 +1415,10 @@ int lmv_fid_alloc(struct obd_export *exp, struct lu_fid *fid, static int lmv_setup(struct obd_device *obd, struct lustre_cfg *lcfg) { - struct lmv_obd *lmv = &obd->u.lmv; - struct lprocfs_static_vars lvars; - struct lmv_desc *desc; - int rc; - ENTRY; + struct lmv_obd *lmv = &obd->u.lmv; + struct lmv_desc *desc; + int rc; + ENTRY; if (LUSTRE_CFG_BUFLEN(lcfg, 1) < 1) { CERROR("LMV setup requires a descriptor\n"); @@ -1384,17 +1448,15 @@ static int lmv_setup(struct obd_device *obd, struct lustre_cfg *lcfg) spin_lock_init(&lmv->lmv_lock); mutex_init(&lmv->init_mutex); - lprocfs_lmv_init_vars(&lvars); - - lprocfs_obd_setup(obd, lvars.obd_vars); #ifdef LPROCFS - { - rc = lprocfs_seq_create(obd->obd_proc_entry, "target_obd", - 0444, &lmv_proc_target_fops, obd); - if (rc) - CWARN("%s: error adding LMV target_obd file: rc = %d\n", - obd->obd_name, rc); - } + obd->obd_vars = lprocfs_lmv_obd_vars; + lprocfs_seq_obd_setup(obd); + lprocfs_alloc_md_stats(obd, 0); + rc = lprocfs_seq_create(obd->obd_proc_entry, "target_obd", + 0444, &lmv_proc_target_fops, obd); + if (rc) + CWARN("%s: error adding LMV target_obd file: rc = %d\n", + obd->obd_name, rc); #endif rc = fld_client_init(&lmv->lmv_fld, obd->obd_name, LUSTRE_CLI_FLD_HASH_DHT); @@ -1446,7 +1508,7 @@ static int lmv_process_config(struct obd_device *obd, obd_count len, void *buf) obd_str2uuid(&obd_uuid, lustre_cfg_buf(lcfg, 1)); - if (sscanf(lustre_cfg_buf(lcfg, 2), "%d", &index) != 1) + if (sscanf(lustre_cfg_buf(lcfg, 2), "%u", &index) != 1) GOTO(out, rc = -EINVAL); if (sscanf(lustre_cfg_buf(lcfg, 3), "%d", &gen) != 1) GOTO(out, rc = -EINVAL); @@ -1642,27 +1704,39 @@ static int lmv_null_inode(struct obd_export *exp, const struct lu_fid *fid) static int lmv_find_cbdata(struct obd_export *exp, const struct lu_fid *fid, ldlm_iterator_t it, void *data) { - struct obd_device *obd = exp->exp_obd; - struct lmv_obd *lmv = &obd->u.lmv; - __u32 i; - int rc; - ENTRY; + struct obd_device *obd = exp->exp_obd; + struct lmv_obd *lmv = &obd->u.lmv; + int i; + int tgt; + int rc; + ENTRY; - rc = lmv_check_connect(obd); - if (rc) - RETURN(rc); + rc = lmv_check_connect(obd); + if (rc) + RETURN(rc); - CDEBUG(D_INODE, "CBDATA for "DFID"\n", PFID(fid)); + CDEBUG(D_INODE, "CBDATA for "DFID"\n", PFID(fid)); /* * With DNE every object can have two locks in different namespaces: * lookup lock in space of MDT storing direntry and update/open lock in - * space of MDT storing inode. + * space of MDT storing inode. Try the MDT that the FID maps to first, + * since this can be easily found, and only try others if that fails. */ - for (i = 0; i < lmv->desc.ld_tgt_count; i++) { - if (lmv->tgts[i] == NULL || lmv->tgts[i]->ltd_exp == NULL) + for (i = 0, tgt = lmv_find_target_index(lmv, fid); + i < lmv->desc.ld_tgt_count; + i++, tgt = (tgt + 1) % lmv->desc.ld_tgt_count) { + if (tgt < 0) { + CDEBUG(D_HA, "%s: "DFID" is inaccessible: rc = %d\n", + obd->obd_name, PFID(fid), tgt); + tgt = 0; + } + + if (lmv->tgts[tgt] == NULL || + lmv->tgts[tgt]->ltd_exp == NULL) continue; - rc = md_find_cbdata(lmv->tgts[i]->ltd_exp, fid, it, data); + + rc = md_find_cbdata(lmv->tgts[tgt]->ltd_exp, fid, it, data); if (rc) RETURN(rc); } @@ -1693,19 +1767,50 @@ static int lmv_close(struct obd_export *exp, struct md_op_data *op_data, RETURN(rc); } +/** + * Choosing the MDT by name or FID in @op_data. + * For non-striped directory, it will locate MDT by fid. + * For striped-directory, it will locate MDT by name. And also + * it will reset op_fid1 with the FID of the choosen stripe. + **/ +struct lmv_tgt_desc * +lmv_locate_target_for_name(struct lmv_obd *lmv, struct lmv_stripe_md *lsm, + const char *name, int namelen, struct lu_fid *fid, + mdsno_t *mds) +{ + struct lmv_tgt_desc *tgt; + const struct lmv_oinfo *oinfo; + + oinfo = lsm_name_to_stripe_info(lsm, name, namelen); + if (IS_ERR(oinfo)) + RETURN((void *)oinfo); + *fid = oinfo->lmo_fid; + *mds = oinfo->lmo_mds; + tgt = lmv_get_target(lmv, *mds, NULL); + + CDEBUG(D_INFO, "locate on mds %u "DFID"\n", *mds, PFID(fid)); + return tgt; +} + struct lmv_tgt_desc *lmv_locate_mds(struct lmv_obd *lmv, struct md_op_data *op_data, struct lu_fid *fid) { - struct lmv_tgt_desc *tgt; + struct lmv_stripe_md *lsm = op_data->op_mea1; + struct lmv_tgt_desc *tgt; - tgt = lmv_find_target(lmv, fid); - if (IS_ERR(tgt)) - return tgt; + if (lsm == NULL || op_data->op_namelen == 0) { + tgt = lmv_find_target(lmv, fid); + if (IS_ERR(tgt)) + return tgt; - op_data->op_mds = tgt->ltd_idx; + op_data->op_mds = tgt->ltd_idx; + return tgt; + } - return tgt; + return lmv_locate_target_for_name(lmv, lsm, op_data->op_name, + op_data->op_namelen, fid, + &op_data->op_mds); } int lmv_create(struct obd_export *exp, struct md_op_data *op_data, @@ -1730,18 +1835,28 @@ int lmv_create(struct obd_export *exp, struct md_op_data *op_data, if (IS_ERR(tgt)) RETURN(PTR_ERR(tgt)); - rc = lmv_fid_alloc(exp, &op_data->op_fid2, op_data); + CDEBUG(D_INODE, "CREATE name '%.*s' on "DFID" -> mds #%x\n", + op_data->op_namelen, op_data->op_name, PFID(&op_data->op_fid1), + op_data->op_mds); + + rc = lmv_fid_alloc(NULL, exp, &op_data->op_fid2, op_data); if (rc) RETURN(rc); - CDEBUG(D_INODE, "CREATE '%*s' on "DFID" -> mds #%x\n", - op_data->op_namelen, op_data->op_name, PFID(&op_data->op_fid1), - op_data->op_mds); + /* Send the create request to the MDT where the object + * will be located */ + tgt = lmv_find_target(lmv, &op_data->op_fid2); + if (IS_ERR(tgt)) + RETURN(PTR_ERR(tgt)); + + op_data->op_mds = tgt->ltd_idx; + + CDEBUG(D_INODE, "CREATE obj "DFID" -> mds #%x\n", + PFID(&op_data->op_fid2), op_data->op_mds); op_data->op_flags |= MF_MDC_CANCEL_FID1; rc = md_create(tgt->ltd_exp, op_data, data, datalen, mode, uid, gid, cap_effective, rdev, request); - if (rc == 0) { if (*request == NULL) RETURN(rc); @@ -1773,70 +1888,10 @@ static int lmv_done_writing(struct obd_export *exp, } static int -lmv_enqueue_remote(struct obd_export *exp, struct ldlm_enqueue_info *einfo, - struct lookup_intent *it, struct md_op_data *op_data, - struct lustre_handle *lockh, void *lmm, int lmmsize, - int extra_lock_flags) -{ - struct ptlrpc_request *req = it->d.lustre.it_data; - struct obd_device *obd = exp->exp_obd; - struct lmv_obd *lmv = &obd->u.lmv; - struct lustre_handle plock; - struct lmv_tgt_desc *tgt; - struct md_op_data *rdata; - struct lu_fid fid1; - struct mdt_body *body; - int rc = 0; - int pmode; - ENTRY; - - body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY); - LASSERT(body != NULL); - - if (!(body->valid & OBD_MD_MDS)) - RETURN(0); - - CDEBUG(D_INODE, "REMOTE_ENQUEUE '%s' on "DFID" -> "DFID"\n", - LL_IT2STR(it), PFID(&op_data->op_fid1), PFID(&body->fid1)); - - /* - * We got LOOKUP lock, but we really need attrs. - */ - pmode = it->d.lustre.it_lock_mode; - LASSERT(pmode != 0); - memcpy(&plock, lockh, sizeof(plock)); - it->d.lustre.it_lock_mode = 0; - it->d.lustre.it_data = NULL; - fid1 = body->fid1; - - it->d.lustre.it_disposition &= ~DISP_ENQ_COMPLETE; - ptlrpc_req_finished(req); - - tgt = lmv_find_target(lmv, &fid1); - if (IS_ERR(tgt)) - GOTO(out, rc = PTR_ERR(tgt)); - - OBD_ALLOC_PTR(rdata); - if (rdata == NULL) - GOTO(out, rc = -ENOMEM); - - rdata->op_fid1 = fid1; - rdata->op_bias = MDS_CROSS_REF; - - rc = md_enqueue(tgt->ltd_exp, einfo, it, rdata, lockh, - lmm, lmmsize, NULL, extra_lock_flags); - OBD_FREE_PTR(rdata); - EXIT; -out: - ldlm_lock_decref(&plock, pmode); - return rc; -} - -static int lmv_enqueue(struct obd_export *exp, struct ldlm_enqueue_info *einfo, - struct lookup_intent *it, struct md_op_data *op_data, - struct lustre_handle *lockh, void *lmm, int lmmsize, - struct ptlrpc_request **req, __u64 extra_lock_flags) + const union ldlm_policy_data *policy, + struct lookup_intent *it, struct md_op_data *op_data, + struct lustre_handle *lockh, __u64 extra_lock_flags) { struct obd_device *obd = exp->exp_obd; struct lmv_obd *lmv = &obd->u.lmv; @@ -1858,19 +1913,15 @@ lmv_enqueue(struct obd_export *exp, struct ldlm_enqueue_info *einfo, CDEBUG(D_INODE, "ENQUEUE '%s' on "DFID" -> mds #%d\n", LL_IT2STR(it), PFID(&op_data->op_fid1), tgt->ltd_idx); - rc = md_enqueue(tgt->ltd_exp, einfo, it, op_data, lockh, - lmm, lmmsize, req, extra_lock_flags); + rc = md_enqueue(tgt->ltd_exp, einfo, policy, it, op_data, lockh, + extra_lock_flags); - if (rc == 0 && it && it->it_op == IT_OPEN) { - rc = lmv_enqueue_remote(exp, einfo, it, op_data, lockh, - lmm, lmmsize, extra_lock_flags); - } RETURN(rc); } static int lmv_getattr_name(struct obd_export *exp,struct md_op_data *op_data, - struct ptlrpc_request **request) + struct ptlrpc_request **preq) { struct ptlrpc_request *req = NULL; struct obd_device *obd = exp->exp_obd; @@ -1892,22 +1943,22 @@ lmv_getattr_name(struct obd_export *exp,struct md_op_data *op_data, op_data->op_namelen, op_data->op_name, PFID(&op_data->op_fid1), tgt->ltd_idx); - rc = md_getattr_name(tgt->ltd_exp, op_data, request); + rc = md_getattr_name(tgt->ltd_exp, op_data, preq); if (rc != 0) RETURN(rc); - body = req_capsule_server_get(&(*request)->rq_pill, - &RMF_MDT_BODY); + body = req_capsule_server_get(&(*preq)->rq_pill, &RMF_MDT_BODY); LASSERT(body != NULL); - if (body->valid & OBD_MD_MDS) { - struct lu_fid rid = body->fid1; + if (body->mbo_valid & OBD_MD_MDS) { + struct lu_fid rid = body->mbo_fid1; CDEBUG(D_INODE, "Request attrs for "DFID"\n", PFID(&rid)); tgt = lmv_find_target(lmv, &rid); if (IS_ERR(tgt)) { - ptlrpc_req_finished(*request); + ptlrpc_req_finished(*preq); + preq = NULL; RETURN(PTR_ERR(tgt)); } @@ -1916,8 +1967,8 @@ lmv_getattr_name(struct obd_export *exp,struct md_op_data *op_data, op_data->op_namelen = 0; op_data->op_name = NULL; rc = md_getattr_name(tgt->ltd_exp, op_data, &req); - ptlrpc_req_finished(*request); - *request = req; + ptlrpc_req_finished(*preq); + *preq = req; } RETURN(rc); @@ -1930,23 +1981,25 @@ lmv_getattr_name(struct obd_export *exp,struct md_op_data *op_data, fl == MF_MDC_CANCEL_FID4 ? &op_data->op_fid4 : \ NULL) -static int lmv_early_cancel(struct obd_export *exp, struct md_op_data *op_data, - int op_tgt, ldlm_mode_t mode, int bits, int flag) +static int lmv_early_cancel(struct obd_export *exp, struct lmv_tgt_desc *tgt, + struct md_op_data *op_data, + int op_tgt, ldlm_mode_t mode, int bits, int flag) { - struct lu_fid *fid = md_op_data_fid(op_data, flag); - struct obd_device *obd = exp->exp_obd; - struct lmv_obd *lmv = &obd->u.lmv; - struct lmv_tgt_desc *tgt; - ldlm_policy_data_t policy = {{0}}; - int rc = 0; - ENTRY; + struct lu_fid *fid = md_op_data_fid(op_data, flag); + struct obd_device *obd = exp->exp_obd; + struct lmv_obd *lmv = &obd->u.lmv; + ldlm_policy_data_t policy = {{ 0 }}; + int rc = 0; + ENTRY; - if (!fid_is_sane(fid)) - RETURN(0); + if (!fid_is_sane(fid)) + RETURN(0); - tgt = lmv_find_target(lmv, fid); - if (IS_ERR(tgt)) - RETURN(PTR_ERR(tgt)); + if (tgt == NULL) { + tgt = lmv_find_target(lmv, fid); + if (IS_ERR(tgt)) + RETURN(PTR_ERR(tgt)); + } if (tgt->ltd_idx != op_tgt) { CDEBUG(D_INODE, "EARLY_CANCEL on "DFID"\n", PFID(fid)); @@ -1987,9 +2040,21 @@ static int lmv_link(struct obd_export *exp, struct md_op_data *op_data, PFID(&op_data->op_fid2), op_data->op_namelen, op_data->op_name, PFID(&op_data->op_fid1)); - op_data->op_fsuid = current_fsuid(); - op_data->op_fsgid = current_fsgid(); + op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid()); + op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid()); op_data->op_cap = cfs_curproc_cap_pack(); + if (op_data->op_mea2 != NULL) { + struct lmv_stripe_md *lsm = op_data->op_mea2; + const struct lmv_oinfo *oinfo; + + oinfo = lsm_name_to_stripe_info(lsm, op_data->op_name, + op_data->op_namelen); + if (IS_ERR(oinfo)) + RETURN(PTR_ERR(oinfo)); + + op_data->op_fid2 = oinfo->lmo_fid; + } + tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid2); if (IS_ERR(tgt)) RETURN(PTR_ERR(tgt)); @@ -1998,7 +2063,7 @@ static int lmv_link(struct obd_export *exp, struct md_op_data *op_data, * Cancel UPDATE lock on child (fid1). */ op_data->op_flags |= MF_MDC_CANCEL_FID2; - rc = lmv_early_cancel(exp, op_data, tgt->ltd_idx, LCK_EX, + rc = lmv_early_cancel(exp, NULL, op_data, tgt->ltd_idx, LCK_EX, MDS_INODELOCK_UPDATE, MF_MDC_CANCEL_FID1); if (rc != 0) RETURN(rc); @@ -2012,33 +2077,66 @@ static int lmv_rename(struct obd_export *exp, struct md_op_data *op_data, const char *old, int oldlen, const char *new, int newlen, struct ptlrpc_request **request) { - struct obd_device *obd = exp->exp_obd; - struct lmv_obd *lmv = &obd->u.lmv; - struct lmv_tgt_desc *src_tgt; - struct lmv_tgt_desc *tgt_tgt; + struct obd_device *obd = exp->exp_obd; + struct lmv_obd *lmv = &obd->u.lmv; + struct lmv_tgt_desc *src_tgt; int rc; ENTRY; - LASSERT(oldlen != 0); + LASSERT(oldlen != 0); - CDEBUG(D_INODE, "RENAME %*s in "DFID" to %*s in "DFID"\n", - oldlen, old, PFID(&op_data->op_fid1), - newlen, new, PFID(&op_data->op_fid2)); + CDEBUG(D_INODE, "RENAME %.*s in "DFID":%d to %.*s in "DFID":%d\n", + oldlen, old, PFID(&op_data->op_fid1), + op_data->op_mea1 ? op_data->op_mea1->lsm_md_stripe_count : 0, + newlen, new, PFID(&op_data->op_fid2), + op_data->op_mea2 ? op_data->op_mea2->lsm_md_stripe_count : 0); - rc = lmv_check_connect(obd); - if (rc) - RETURN(rc); + rc = lmv_check_connect(obd); + if (rc) + RETURN(rc); - op_data->op_fsuid = current_fsuid(); - op_data->op_fsgid = current_fsgid(); + op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid()); + op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid()); op_data->op_cap = cfs_curproc_cap_pack(); - src_tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid1); + if (op_data->op_cli_flags & CLI_MIGRATE) { + LASSERTF(fid_is_sane(&op_data->op_fid3), "invalid FID "DFID"\n", + PFID(&op_data->op_fid3)); + rc = lmv_fid_alloc(NULL, exp, &op_data->op_fid2, op_data); + if (rc) + RETURN(rc); + src_tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid3); + } else { + if (op_data->op_mea1 != NULL) { + struct lmv_stripe_md *lsm = op_data->op_mea1; + + src_tgt = lmv_locate_target_for_name(lmv, lsm, old, + oldlen, + &op_data->op_fid1, + &op_data->op_mds); + if (IS_ERR(src_tgt)) + RETURN(PTR_ERR(src_tgt)); + } else { + src_tgt = lmv_find_target(lmv, &op_data->op_fid1); + if (IS_ERR(src_tgt)) + RETURN(PTR_ERR(src_tgt)); + + op_data->op_mds = src_tgt->ltd_idx; + } + + if (op_data->op_mea2) { + struct lmv_stripe_md *lsm = op_data->op_mea2; + const struct lmv_oinfo *oinfo; + + oinfo = lsm_name_to_stripe_info(lsm, new, newlen); + if (IS_ERR(oinfo)) + RETURN(PTR_ERR(oinfo)); + + op_data->op_fid2 = oinfo->lmo_fid; + } + } if (IS_ERR(src_tgt)) RETURN(PTR_ERR(src_tgt)); - tgt_tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid2); - if (IS_ERR(tgt_tgt)) - RETURN(PTR_ERR(tgt_tgt)); /* * LOOKUP lock on src child (fid3) should also be cancelled for * src_tgt in mdc_rename. @@ -2049,30 +2147,50 @@ static int lmv_rename(struct obd_export *exp, struct md_op_data *op_data, * Cancel UPDATE locks on tgt parent (fid2), tgt_tgt is its * own target. */ - rc = lmv_early_cancel(exp, op_data, src_tgt->ltd_idx, + rc = lmv_early_cancel(exp, NULL, op_data, src_tgt->ltd_idx, LCK_EX, MDS_INODELOCK_UPDATE, MF_MDC_CANCEL_FID2); + if (rc != 0) + RETURN(rc); /* - * Cancel LOOKUP locks on tgt child (fid4) for parent tgt_tgt. + * Cancel LOOKUP locks on source child (fid3) for parent tgt_tgt. */ - if (rc == 0) { - rc = lmv_early_cancel(exp, op_data, src_tgt->ltd_idx, - LCK_EX, MDS_INODELOCK_LOOKUP, - MF_MDC_CANCEL_FID4); - } + if (fid_is_sane(&op_data->op_fid3)) { + struct lmv_tgt_desc *tgt; + + tgt = lmv_find_target(lmv, &op_data->op_fid1); + if (IS_ERR(tgt)) + RETURN(PTR_ERR(tgt)); + + /* Cancel LOOKUP lock on its parent */ + rc = lmv_early_cancel(exp, tgt, op_data, src_tgt->ltd_idx, + LCK_EX, MDS_INODELOCK_LOOKUP, + MF_MDC_CANCEL_FID3); + if (rc != 0) + RETURN(rc); + + rc = lmv_early_cancel(exp, NULL, op_data, src_tgt->ltd_idx, + LCK_EX, MDS_INODELOCK_FULL, + MF_MDC_CANCEL_FID3); + if (rc != 0) + RETURN(rc); + } /* * Cancel all the locks on tgt child (fid4). */ - if (rc == 0) - rc = lmv_early_cancel(exp, op_data, src_tgt->ltd_idx, + if (fid_is_sane(&op_data->op_fid4)) + rc = lmv_early_cancel(exp, NULL, op_data, src_tgt->ltd_idx, LCK_EX, MDS_INODELOCK_FULL, MF_MDC_CANCEL_FID4); - if (rc == 0) - rc = md_rename(src_tgt->ltd_exp, op_data, old, oldlen, - new, newlen, request); + CDEBUG(D_INODE, DFID":m%d to "DFID"\n", PFID(&op_data->op_fid1), + op_data->op_mds, PFID(&op_data->op_fid2)); + + rc = md_rename(src_tgt->ltd_exp, op_data, old, oldlen, new, newlen, + request); + RETURN(rc); } @@ -2105,25 +2223,25 @@ static int lmv_setattr(struct obd_export *exp, struct md_op_data *op_data, RETURN(rc); } -static int lmv_sync(struct obd_export *exp, const struct lu_fid *fid, - struct obd_capa *oc, struct ptlrpc_request **request) +static int lmv_fsync(struct obd_export *exp, const struct lu_fid *fid, + struct obd_capa *oc, struct ptlrpc_request **request) { - struct obd_device *obd = exp->exp_obd; - struct lmv_obd *lmv = &obd->u.lmv; - struct lmv_tgt_desc *tgt; - int rc; - ENTRY; + struct obd_device *obd = exp->exp_obd; + struct lmv_obd *lmv = &obd->u.lmv; + struct lmv_tgt_desc *tgt; + int rc; + ENTRY; - rc = lmv_check_connect(obd); - if (rc) - RETURN(rc); + rc = lmv_check_connect(obd); + if (rc != 0) + RETURN(rc); - tgt = lmv_find_target(lmv, fid); - if (IS_ERR(tgt)) - RETURN(PTR_ERR(tgt)); + tgt = lmv_find_target(lmv, fid); + if (IS_ERR(tgt)) + RETURN(PTR_ERR(tgt)); - rc = md_sync(tgt->ltd_exp, fid, oc, request); - RETURN(rc); + rc = md_fsync(tgt->ltd_exp, fid, oc, request); + RETURN(rc); } /* @@ -2229,44 +2347,212 @@ static void lmv_adjust_dirpages(struct page **pages, int ncfspgs, int nlupgs) #define lmv_adjust_dirpages(pages, ncfspgs, nlupgs) do {} while (0) #endif /* PAGE_CACHE_SIZE > LU_PAGE_SIZE */ -static int lmv_readpage(struct obd_export *exp, struct md_op_data *op_data, - struct page **pages, struct ptlrpc_request **request) +/** + * This function will read entry from a striped directory, bascially, it will + * read entries from all of stripes, and choose one closest to the required + * offset(&op_data->op_hash_offset). A few notes + * 1. skip . and .. for non-zero stripes, because there can only have one . + * and .. in a directory. + * 2. op_data will be shared by all of stripes, instead of allocating new + * one, so need to restore before reusing. + * 3. release the entry page if that is not being chosen. + * + * param[in]exp obd export refer to LMV + * param[in]op_data hold those MD parameters of read_entry. + * param[in]cb_op ldlm callback being used in enqueue in mdc_read_entry + * param[out]ldp the entry being read. + * param[out]ppage the page holding the entry, note: because the entry + * will be accessed in upper layer, so we need hold the + * page until the usages of entry is finished, see + * ll_dir_entry_next. + * + * retval =0 if get entry successfully + * <0 can not get entry. + */ +#define NORMAL_MAX_STRIPES 4 +static int lmv_read_striped_entry(struct obd_export *exp, + struct md_op_data *op_data, + struct md_callback *cb_op, + struct lu_dirent **ldp, + struct page **ppage) { struct obd_device *obd = exp->exp_obd; struct lmv_obd *lmv = &obd->u.lmv; - __u64 offset = op_data->op_offset; - int rc; - int ncfspgs; /* pages read in PAGE_CACHE_SIZE */ - int nlupgs; /* pages read in LU_PAGE_SIZE */ + struct lmv_stripe_md *lsm = op_data->op_mea1; struct lmv_tgt_desc *tgt; + struct lu_dirent *tmp_ents[NORMAL_MAX_STRIPES]; + struct lu_dirent **ents = NULL; + struct lu_fid master_fid = op_data->op_fid1; + void *master_data = op_data->op_data; + __u64 last_idx = op_data->op_stripe_offset; + __u64 hash_offset = op_data->op_hash_offset; + __u32 same_hash_offset = op_data->op_same_hash_offset; + __u32 cli_flags = op_data->op_cli_flags; + int stripe_count; + __u64 min_hash; + int min_same_hash_offset = 0; + int min_idx = 0; + struct page *min_page = NULL; + int i; + int rc; ENTRY; + LASSERT(lsm != NULL); + rc = lmv_check_connect(obd); if (rc) RETURN(rc); - CDEBUG(D_INODE, "READPAGE at "LPX64" from "DFID"\n", - offset, PFID(&op_data->op_fid1)); + /* . and .. will be stored on the master object, so we need iterate + * the master object as well */ + stripe_count = lsm->lsm_md_stripe_count; + if (stripe_count > NORMAL_MAX_STRIPES) { + OBD_ALLOC(ents, sizeof(ents[0]) * stripe_count); + if (ents == NULL) + GOTO(out, rc = -ENOMEM); + } else { + ents = tmp_ents; + memset(ents, 0, sizeof(ents[0]) * stripe_count); + } - tgt = lmv_find_target(lmv, &op_data->op_fid1); - if (IS_ERR(tgt)) - RETURN(PTR_ERR(tgt)); + min_hash = MDS_DIR_END_OFF; + for (i = 0; i < stripe_count; i++) { + struct page *page = NULL; - rc = md_readpage(tgt->ltd_exp, op_data, pages, request); + tgt = lmv_get_target(lmv, lsm->lsm_md_oinfo[i].lmo_mds, NULL); + if (IS_ERR(tgt)) + GOTO(out, rc = PTR_ERR(tgt)); + + if (last_idx != i) + op_data->op_same_hash_offset = 0; + else + op_data->op_same_hash_offset = same_hash_offset; + + /* op_data will be shared by each stripe, so we need + * reset these value for each stripe */ + op_data->op_stripe_offset = i; + op_data->op_hash_offset = hash_offset; + op_data->op_cli_flags = cli_flags; + op_data->op_fid1 = lsm->lsm_md_oinfo[i].lmo_fid; + op_data->op_fid2 = lsm->lsm_md_oinfo[i].lmo_fid; + op_data->op_data = lsm->lsm_md_oinfo[i].lmo_root; + +next: + rc = md_read_entry(tgt->ltd_exp, op_data, cb_op, &ents[i], + &page); + if (rc != 0) + GOTO(out, rc); + + if (ents[i] != NULL && + (strncmp(ents[i]->lde_name, ".", + le16_to_cpu(ents[i]->lde_namelen)) == 0 || + strncmp(ents[i]->lde_name, "..", + le16_to_cpu(ents[i]->lde_namelen)) == 0)) { + if (i == 0) { + /* replace . with master FID */ + if (le16_to_cpu(ents[i]->lde_namelen) == 1) + fid_cpu_to_le(&ents[i]->lde_fid, + &master_fid); + else + fid_cpu_to_le(&ents[i]->lde_fid, + &op_data->op_fid3); + } else { + /* skip . and .. for other stripes */ + op_data->op_cli_flags |= CLI_NEXT_ENTRY; + op_data->op_hash_offset = + le64_to_cpu(ents[i]->lde_hash); + kunmap(page); + page_cache_release(page); + goto next; + } + } + + if (ents[i] != NULL) { + /* If the hash value of read_entry is equal to the + * current min_hash, which is very rare and only + * happens if two entries have the same hash value + * but on different stripes, in this case, we need + * make sure these entries are being reading forward, + * not backward, i.e. only reset the min_entry, if + * current stripe is ahead of last entry. Note: if + * there are hash conflict inside the entry, MDC + * (see mdc_read_entry) will resolve them. */ + if (le64_to_cpu(ents[i]->lde_hash) < min_hash || + (le64_to_cpu(ents[i]->lde_hash) == min_hash && + i >= last_idx)) { + if (min_page != NULL) { + kunmap(min_page); + page_cache_release(min_page); + } + min_page = page; + min_hash = le64_to_cpu(ents[i]->lde_hash); + min_same_hash_offset = + op_data->op_same_hash_offset; + min_idx = i; + } else { + kunmap(page); + page_cache_release(page); + } + } + } + + if (min_hash != MDS_DIR_END_OFF) { + *ldp = ents[min_idx]; + op_data->op_stripe_offset = min_idx; + op_data->op_same_hash_offset = min_same_hash_offset; + *ppage = min_page; + } else { + *ldp = NULL; + *ppage = NULL; + } +out: + /* We do not want to allocate md_op_data during each + * dir entry reading, so op_data will be shared by every stripe, + * then we need to restore it back to original value before + * return to the upper layer */ + op_data->op_hash_offset = hash_offset; + op_data->op_fid1 = master_fid; + op_data->op_fid2 = master_fid; + op_data->op_data = master_data; + op_data->op_cli_flags = cli_flags; + if (stripe_count > NORMAL_MAX_STRIPES && ents != NULL) + OBD_FREE(ents, sizeof(ents[0]) * stripe_count); + + if (rc != 0 && min_page != NULL) { + kunmap(min_page); + page_cache_release(min_page); + } + + RETURN(rc); +} + +int lmv_read_entry(struct obd_export *exp, struct md_op_data *op_data, + struct md_callback *cb_op, struct lu_dirent **ldp, + struct page **ppage) +{ + struct obd_device *obd = exp->exp_obd; + struct lmv_obd *lmv = &obd->u.lmv; + struct lmv_stripe_md *lsm = op_data->op_mea1; + struct lmv_tgt_desc *tgt; + int rc; + ENTRY; + + rc = lmv_check_connect(obd); if (rc != 0) RETURN(rc); - ncfspgs = ((*request)->rq_bulk->bd_nob_transferred + - PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT; - nlupgs = (*request)->rq_bulk->bd_nob_transferred >> LU_PAGE_SHIFT; - LASSERT(!((*request)->rq_bulk->bd_nob_transferred & ~LU_PAGE_MASK)); - LASSERT(ncfspgs > 0 && ncfspgs <= op_data->op_npages); - - CDEBUG(D_INODE, "read %d(%d)/%d pages\n", ncfspgs, nlupgs, - op_data->op_npages); + if (unlikely(lsm != NULL)) { + rc = lmv_read_striped_entry(exp, op_data, cb_op, + ldp, ppage); + RETURN(rc); + } - lmv_adjust_dirpages(pages, ncfspgs, nlupgs); + tgt = lmv_find_target(lmv, &op_data->op_fid1); + if (IS_ERR(tgt)) + RETURN(PTR_ERR(tgt)); + rc = md_read_entry(tgt->ltd_exp, op_data, cb_op, ldp, + ppage); RETURN(rc); } @@ -2276,6 +2562,7 @@ static int lmv_unlink(struct obd_export *exp, struct md_op_data *op_data, struct obd_device *obd = exp->exp_obd; struct lmv_obd *lmv = &obd->u.lmv; struct lmv_tgt_desc *tgt = NULL; + struct lmv_tgt_desc *parent_tgt = NULL; struct mdt_body *body; int rc; ENTRY; @@ -2285,15 +2572,34 @@ static int lmv_unlink(struct obd_export *exp, struct md_op_data *op_data, RETURN(rc); retry: /* Send unlink requests to the MDT where the child is located */ - if (likely(!fid_is_zero(&op_data->op_fid2))) - tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid2); - else + if (likely(!fid_is_zero(&op_data->op_fid2))) { + tgt = lmv_find_target(lmv, &op_data->op_fid2); + if (IS_ERR(tgt)) + RETURN(PTR_ERR(tgt)); + + /* For striped dir, we need to locate the parent as well */ + if (op_data->op_mea1 != NULL) { + struct lmv_tgt_desc *tmp; + + LASSERT(op_data->op_name != NULL && + op_data->op_namelen != 0); + tmp = lmv_locate_target_for_name(lmv, + op_data->op_mea1, + op_data->op_name, + op_data->op_namelen, + &op_data->op_fid1, + &op_data->op_mds); + if (IS_ERR(tmp)) + RETURN(PTR_ERR(tmp)); + } + } else { tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid1); - if (IS_ERR(tgt)) - RETURN(PTR_ERR(tgt)); + if (IS_ERR(tgt)) + RETURN(PTR_ERR(tgt)); + } - op_data->op_fsuid = current_fsuid(); - op_data->op_fsgid = current_fsgid(); + op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid()); + op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid()); op_data->op_cap = cfs_curproc_cap_pack(); /* @@ -2308,9 +2614,18 @@ retry: /* * Cancel FULL locks on child (fid3). */ - rc = lmv_early_cancel(exp, op_data, tgt->ltd_idx, LCK_EX, - MDS_INODELOCK_FULL, MF_MDC_CANCEL_FID3); + parent_tgt = lmv_find_target(lmv, &op_data->op_fid1); + if (IS_ERR(parent_tgt)) + RETURN(PTR_ERR(parent_tgt)); + if (parent_tgt != tgt) { + rc = lmv_early_cancel(exp, parent_tgt, op_data, tgt->ltd_idx, + LCK_EX, MDS_INODELOCK_LOOKUP, + MF_MDC_CANCEL_FID3); + } + + rc = lmv_early_cancel(exp, NULL, op_data, tgt->ltd_idx, LCK_EX, + MDS_INODELOCK_FULL, MF_MDC_CANCEL_FID3); if (rc != 0) RETURN(rc); @@ -2326,11 +2641,11 @@ retry: RETURN(-EPROTO); /* Not cross-ref case, just get out of here. */ - if (likely(!(body->valid & OBD_MD_MDS))) + if (likely(!(body->mbo_valid & OBD_MD_MDS))) RETURN(0); CDEBUG(D_INODE, "%s: try unlink to another MDT for "DFID"\n", - exp->exp_obd->obd_name, PFID(&body->fid1)); + exp->exp_obd->obd_name, PFID(&body->mbo_fid1)); /* This is a remote object, try remote MDT, Note: it may * try more than 1 time here, Considering following case @@ -2351,7 +2666,7 @@ retry: * * In theory, it might try unlimited time here, but it should * be very rare case. */ - op_data->op_fid2 = body->fid1; + op_data->op_fid2 = body->mbo_fid1; ptlrpc_req_finished(*request); *request = NULL; @@ -2371,6 +2686,7 @@ static int lmv_precleanup(struct obd_device *obd, enum obd_cleanup_stage stage) case OBD_CLEANUP_EXPORTS: fld_client_proc_fini(&lmv->lmv_fld); lprocfs_obd_cleanup(obd); + lprocfs_free_md_stats(obd); break; default: break; @@ -2396,7 +2712,6 @@ static int lmv_get_info(const struct lu_env *env, struct obd_export *exp, lmv = &obd->u.lmv; if (keylen >= strlen("remote_flag") && !strcmp(key, "remote_flag")) { - struct lmv_tgt_desc *tgt; int i; rc = lmv_check_connect(obd); @@ -2405,7 +2720,7 @@ static int lmv_get_info(const struct lu_env *env, struct obd_export *exp, LASSERT(*vallen == sizeof(__u32)); for (i = 0; i < lmv->desc.ld_tgt_count; i++) { - tgt = lmv->tgts[i]; + struct lmv_tgt_desc *tgt = lmv->tgts[i]; /* * All tgts should be connected when this gets called. */ @@ -2415,12 +2730,16 @@ static int lmv_get_info(const struct lu_env *env, struct obd_export *exp, if (!obd_get_info(env, tgt->ltd_exp, keylen, key, vallen, val, NULL)) RETURN(0); - } - RETURN(-EINVAL); - } else if (KEY_IS(KEY_MAX_EASIZE) || KEY_IS(KEY_CONN_DATA)) { - rc = lmv_check_connect(obd); - if (rc) - RETURN(rc); + } + RETURN(-EINVAL); + } else if (KEY_IS(KEY_MAX_EASIZE) || + KEY_IS(KEY_DEFAULT_EASIZE) || + KEY_IS(KEY_MAX_COOKIESIZE) || + KEY_IS(KEY_DEFAULT_COOKIESIZE) || + KEY_IS(KEY_CONN_DATA)) { + rc = lmv_check_connect(obd); + if (rc) + RETURN(rc); /* * Forwarding this request to first MDS, it should know LOV @@ -2444,7 +2763,7 @@ int lmv_set_info_async(const struct lu_env *env, struct obd_export *exp, obd_count keylen, void *key, obd_count vallen, void *val, struct ptlrpc_request_set *set) { - struct lmv_tgt_desc *tgt; + struct lmv_tgt_desc *tgt = NULL; struct obd_device *obd; struct lmv_obd *lmv; int rc = 0; @@ -2479,108 +2798,263 @@ int lmv_set_info_async(const struct lu_env *env, struct obd_export *exp, RETURN(-EINVAL); } -int lmv_packmd(struct obd_export *exp, struct lov_mds_md **lmmp, - struct lov_stripe_md *lsm) +static int lmv_pack_md_v1(const struct lmv_stripe_md *lsm, + struct lmv_mds_md_v1 *lmm1) { - struct obd_device *obd = class_exp2obd(exp); - struct lmv_obd *lmv = &obd->u.lmv; - struct lmv_stripe_md *meap; - struct lmv_stripe_md *lsmp; - int mea_size; - __u32 i; - ENTRY; + int cplen; + int i; - mea_size = lmv_get_easize(lmv); - if (!lmmp) - RETURN(mea_size); + lmm1->lmv_magic = cpu_to_le32(lsm->lsm_md_magic); + lmm1->lmv_stripe_count = cpu_to_le32(lsm->lsm_md_stripe_count); + lmm1->lmv_master_mdt_index = cpu_to_le32(lsm->lsm_md_master_mdt_index); + lmm1->lmv_hash_type = cpu_to_le32(lsm->lsm_md_hash_type); + cplen = strlcpy(lmm1->lmv_pool_name, lsm->lsm_md_pool_name, + sizeof(lmm1->lmv_pool_name)); + if (cplen >= sizeof(lmm1->lmv_pool_name)) + return -E2BIG; - if (*lmmp && !lsm) { - OBD_FREE_LARGE(*lmmp, mea_size); - *lmmp = NULL; - RETURN(0); - } + for (i = 0; i < lsm->lsm_md_stripe_count; i++) + fid_cpu_to_le(&lmm1->lmv_stripe_fids[i], + &lsm->lsm_md_oinfo[i].lmo_fid); + return 0; +} - if (*lmmp == NULL) { - OBD_ALLOC_LARGE(*lmmp, mea_size); - if (*lmmp == NULL) - RETURN(-ENOMEM); - } +int lmv_pack_md(union lmv_mds_md **lmmp, const struct lmv_stripe_md *lsm, + int stripe_count) +{ + int lmm_size = 0; + bool allocated = false; + int rc = 0; + ENTRY; - if (!lsm) - RETURN(mea_size); + LASSERT(lmmp != NULL); + /* Free lmm */ + if (*lmmp != NULL && lsm == NULL) { + int stripe_count; - lsmp = (struct lmv_stripe_md *)lsm; - meap = (struct lmv_stripe_md *)*lmmp; + stripe_count = lmv_mds_md_stripe_count_get(*lmmp); + lmm_size = lmv_mds_md_size(stripe_count, + le32_to_cpu((*lmmp)->lmv_magic)); + if (lmm_size == 0) + RETURN(-EINVAL); + OBD_FREE(*lmmp, lmm_size); + *lmmp = NULL; + RETURN(0); + } - if (lsmp->mea_magic != MEA_MAGIC_LAST_CHAR && - lsmp->mea_magic != MEA_MAGIC_ALL_CHARS) - RETURN(-EINVAL); + /* Alloc lmm */ + if (*lmmp == NULL && lsm == NULL) { + lmm_size = lmv_mds_md_size(stripe_count, LMV_MAGIC); + LASSERT(lmm_size > 0); + OBD_ALLOC(*lmmp, lmm_size); + if (*lmmp == NULL) + RETURN(-ENOMEM); + lmv_mds_md_stripe_count_set(*lmmp, stripe_count); + (*lmmp)->lmv_magic = cpu_to_le32(LMV_MAGIC); + RETURN(lmm_size); + } - 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); + /* pack lmm */ + LASSERT(lsm != NULL); + lmm_size = lmv_mds_md_size(lsm->lsm_md_stripe_count, lsm->lsm_md_magic); + if (*lmmp == NULL) { + OBD_ALLOC(*lmmp, lmm_size); + if (*lmmp == NULL) + RETURN(-ENOMEM); + allocated = true; + } - for (i = 0; i < lmv->desc.ld_tgt_count; i++) { - meap->mea_ids[i] = lsmp->mea_ids[i]; - fid_cpu_to_le(&meap->mea_ids[i], &lsmp->mea_ids[i]); + switch (lsm->lsm_md_magic) { + case LMV_MAGIC_V1: + rc = lmv_pack_md_v1(lsm, &(*lmmp)->lmv_md_v1); + break; + default: + rc = -EINVAL; + break; } - RETURN(mea_size); + if (rc != 0 && allocated) { + OBD_FREE(*lmmp, lmm_size); + *lmmp = NULL; + } + + RETURN(lmm_size); } +EXPORT_SYMBOL(lmv_pack_md); -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 lmv_stripe_md **tmea = (struct lmv_stripe_md **)lsmp; - struct lmv_stripe_md *mea = (struct lmv_stripe_md *)lmm; - struct lmv_obd *lmv = &obd->u.lmv; - int mea_size; - __u32 i; - __u32 magic; - ENTRY; +static int lmv_unpack_md_v1(struct obd_export *exp, struct lmv_stripe_md *lsm, + const struct lmv_mds_md_v1 *lmm1) +{ + struct lmv_obd *lmv = &exp->exp_obd->u.lmv; + int stripe_count; + int cplen; + int i; + int rc = 0; + ENTRY; - mea_size = lmv_get_easize(lmv); - if (lsmp == NULL) - return mea_size; + lsm->lsm_md_magic = le32_to_cpu(lmm1->lmv_magic); + lsm->lsm_md_stripe_count = le32_to_cpu(lmm1->lmv_stripe_count); + lsm->lsm_md_master_mdt_index = le32_to_cpu(lmm1->lmv_master_mdt_index); + lsm->lsm_md_hash_type = le32_to_cpu(lmm1->lmv_hash_type); + lsm->lsm_md_layout_version = le32_to_cpu(lmm1->lmv_layout_version); + fid_le_to_cpu(&lsm->lsm_md_master_fid, &lmm1->lmv_master_fid); + cplen = strlcpy(lsm->lsm_md_pool_name, lmm1->lmv_pool_name, + sizeof(lsm->lsm_md_pool_name)); - if (*lsmp != NULL && lmm == NULL) { - OBD_FREE_LARGE(*tmea, mea_size); - *lsmp = NULL; - RETURN(0); - } + if (!fid_is_sane(&lsm->lsm_md_master_fid)) + RETURN(-EPROTO); - LASSERT(mea_size == lmm_size); + if (cplen >= sizeof(lsm->lsm_md_pool_name)) + RETURN(-E2BIG); + + CDEBUG(D_INFO, "unpack lsm count %d, master %d hash_type %d" + "layout_version %d\n", lsm->lsm_md_stripe_count, + lsm->lsm_md_master_mdt_index, lsm->lsm_md_hash_type, + lsm->lsm_md_layout_version); + + stripe_count = le32_to_cpu(lmm1->lmv_stripe_count); + for (i = 0; i < le32_to_cpu(stripe_count); i++) { + fid_le_to_cpu(&lsm->lsm_md_oinfo[i].lmo_fid, + &lmm1->lmv_stripe_fids[i]); + rc = lmv_fld_lookup(lmv, &lsm->lsm_md_oinfo[i].lmo_fid, + &lsm->lsm_md_oinfo[i].lmo_mds); + if (rc != 0) + RETURN(rc); + CDEBUG(D_INFO, "unpack fid #%d "DFID"\n", i, + PFID(&lsm->lsm_md_oinfo[i].lmo_fid)); + } - OBD_ALLOC_LARGE(*tmea, mea_size); - if (*tmea == NULL) - RETURN(-ENOMEM); + RETURN(rc); +} - if (!lmm) - RETURN(mea_size); +int lmv_unpack_md(struct obd_export *exp, struct lmv_stripe_md **lsmp, + const union lmv_mds_md *lmm, int stripe_count) +{ + struct lmv_stripe_md *lsm; + int lsm_size; + int rc; + bool allocated = false; + ENTRY; - if (mea->mea_magic == MEA_MAGIC_LAST_CHAR || - mea->mea_magic == MEA_MAGIC_ALL_CHARS || - mea->mea_magic == MEA_MAGIC_HASH_SEGMENT) - { - magic = le32_to_cpu(mea->mea_magic); - } else { - /* - * Old mea is not handled here. - */ - CERROR("Old not supportable EA is found\n"); - LBUG(); - } + LASSERT(lsmp != NULL); - (*tmea)->mea_magic = magic; - (*tmea)->mea_count = le32_to_cpu(mea->mea_count); - (*tmea)->mea_master = le32_to_cpu(mea->mea_master); + lsm = *lsmp; + /* Free memmd */ + if (lsm != NULL && lmm == NULL) { +#ifdef __KERNEL__ + int i; + for (i = 0; i < lsm->lsm_md_stripe_count; i++) { + /* For migrating inode, the master stripe and master + * object will be the same, so do not need iput, see + * ll_update_lsm_md */ + if (!(lsm->lsm_md_hash_type & LMV_HASH_FLAG_MIGRATION && + i == 0) && lsm->lsm_md_oinfo[i].lmo_root != NULL) + iput(lsm->lsm_md_oinfo[i].lmo_root); + } +#endif + lsm_size = lmv_stripe_md_size(lsm->lsm_md_stripe_count); + OBD_FREE(lsm, lsm_size); + *lsmp = NULL; + RETURN(0); + } - for (i = 0; i < (*tmea)->mea_count; i++) { - (*tmea)->mea_ids[i] = mea->mea_ids[i]; - fid_le_to_cpu(&(*tmea)->mea_ids[i], &(*tmea)->mea_ids[i]); - } - RETURN(mea_size); + /* Alloc memmd */ + if (lsm == NULL && lmm == NULL) { + lsm_size = lmv_stripe_md_size(stripe_count); + OBD_ALLOC(lsm, lsm_size); + if (lsm == NULL) + RETURN(-ENOMEM); + lsm->lsm_md_stripe_count = stripe_count; + *lsmp = lsm; + RETURN(0); + } + + if (le32_to_cpu(lmm->lmv_magic) == LMV_MAGIC_STRIPE) + RETURN(-EPERM); + + /* Unpack memmd */ + if (le32_to_cpu(lmm->lmv_magic) != LMV_MAGIC_V1 && + le32_to_cpu(lmm->lmv_magic) != LMV_USER_MAGIC) { + CERROR("%s: invalid lmv magic %x: rc = %d\n", + exp->exp_obd->obd_name, le32_to_cpu(lmm->lmv_magic), + -EIO); + RETURN(-EIO); + } + + if (le32_to_cpu(lmm->lmv_magic) == LMV_MAGIC_V1) + lsm_size = lmv_stripe_md_size(lmv_mds_md_stripe_count_get(lmm)); + else + /** + * Unpack default dirstripe(lmv_user_md) to lmv_stripe_md, + * stripecount should be 0 then. + */ + lsm_size = lmv_stripe_md_size(0); + + lsm_size = lmv_stripe_md_size(lmv_mds_md_stripe_count_get(lmm)); + if (lsm == NULL) { + OBD_ALLOC(lsm, lsm_size); + if (lsm == NULL) + RETURN(-ENOMEM); + allocated = true; + *lsmp = lsm; + } + + switch (le32_to_cpu(lmm->lmv_magic)) { + case LMV_MAGIC_V1: + rc = lmv_unpack_md_v1(exp, lsm, &lmm->lmv_md_v1); + break; + default: + CERROR("%s: unrecognized magic %x\n", exp->exp_obd->obd_name, + le32_to_cpu(lmm->lmv_magic)); + rc = -EINVAL; + break; + } + + if (rc != 0 && allocated) { + OBD_FREE(lsm, lsm_size); + *lsmp = NULL; + lsm_size = rc; + } + RETURN(lsm_size); +} + +int lmv_alloc_memmd(struct lmv_stripe_md **lsmp, int stripes) +{ + return lmv_unpack_md(NULL, lsmp, NULL, stripes); +} +EXPORT_SYMBOL(lmv_alloc_memmd); + +void lmv_free_memmd(struct lmv_stripe_md *lsm) +{ + lmv_unpack_md(NULL, &lsm, NULL, 0); +} +EXPORT_SYMBOL(lmv_free_memmd); + +int lmv_unpackmd(struct obd_export *exp, struct lov_stripe_md **lsmp, + struct lov_mds_md *lmm, int disk_len) +{ + return lmv_unpack_md(exp, (struct lmv_stripe_md **)lsmp, + (union lmv_mds_md *)lmm, disk_len); +} + +int lmv_packmd(struct obd_export *exp, struct lov_mds_md **lmmp, + struct lov_stripe_md *lsm) +{ + struct obd_device *obd = exp->exp_obd; + struct lmv_obd *lmv_obd = &obd->u.lmv; + const struct lmv_stripe_md *lmv = (struct lmv_stripe_md *)lsm; + int stripe_count; + + if (lmmp == NULL) { + if (lsm != NULL) + stripe_count = lmv->lsm_md_stripe_count; + else + stripe_count = lmv_obd->desc.ld_tgt_count; + + return lmv_mds_md_size(stripe_count, LMV_MAGIC_V1); + } + + return lmv_pack_md((union lmv_mds_md **)lmmp, lmv, 0); } static int lmv_cancel_unused(struct obd_export *exp, const struct lu_fid *fid, @@ -2596,27 +3070,31 @@ static int lmv_cancel_unused(struct obd_export *exp, const struct lu_fid *fid, LASSERT(fid != NULL); - for (i = 0; i < lmv->desc.ld_tgt_count; i++) { - if (lmv->tgts[i] == NULL || lmv->tgts[i]->ltd_exp == NULL || - lmv->tgts[i]->ltd_active == 0) + for (i = 0; i < lmv->desc.ld_tgt_count; i++) { + struct lmv_tgt_desc *tgt = lmv->tgts[i]; + + if (tgt == NULL || tgt->ltd_exp == NULL || !tgt->ltd_active) continue; - err = md_cancel_unused(lmv->tgts[i]->ltd_exp, fid, - policy, mode, flags, opaque); + err = md_cancel_unused(tgt->ltd_exp, fid, policy, mode, flags, + opaque); if (!rc) rc = err; - } + } RETURN(rc); } int lmv_set_lock_data(struct obd_export *exp, __u64 *lockh, void *data, __u64 *bits) { - struct lmv_obd *lmv = &exp->exp_obd->u.lmv; - int rc; + struct lmv_obd *lmv = &exp->exp_obd->u.lmv; + struct lmv_tgt_desc *tgt = lmv->tgts[0]; + int rc; ENTRY; - rc = md_set_lock_data(lmv->tgts[0]->ltd_exp, lockh, data, bits); + if (tgt == NULL || tgt->ltd_exp == NULL) + RETURN(-EINVAL); + rc = md_set_lock_data(tgt->ltd_exp, lockh, data, bits); RETURN(rc); } @@ -2625,33 +3103,42 @@ ldlm_mode_t lmv_lock_match(struct obd_export *exp, __u64 flags, ldlm_policy_data_t *policy, ldlm_mode_t mode, struct lustre_handle *lockh) { - struct obd_device *obd = exp->exp_obd; - struct lmv_obd *lmv = &obd->u.lmv; - ldlm_mode_t rc; - __u32 i; - ENTRY; + struct obd_device *obd = exp->exp_obd; + struct lmv_obd *lmv = &obd->u.lmv; + ldlm_mode_t rc; + int tgt; + int i; + ENTRY; - CDEBUG(D_INODE, "Lock match for "DFID"\n", PFID(fid)); + CDEBUG(D_INODE, "Lock match for "DFID"\n", PFID(fid)); /* - * With CMD every object can have two locks in different namespaces: - * lookup lock in space of mds storing direntry and update/open lock in - * space of mds storing inode. Thus we check all targets, not only that - * one fid was created in. - */ - for (i = 0; i < lmv->desc.ld_tgt_count; i++) { - if (lmv->tgts[i] == NULL || - lmv->tgts[i]->ltd_exp == NULL || - lmv->tgts[i]->ltd_active == 0) + * With DNE every object can have two locks in different namespaces: + * lookup lock in space of MDT storing direntry and update/open lock in + * space of MDT storing inode. Try the MDT that the FID maps to first, + * since this can be easily found, and only try others if that fails. + */ + for (i = 0, tgt = lmv_find_target_index(lmv, fid); + i < lmv->desc.ld_tgt_count; + i++, tgt = (tgt + 1) % lmv->desc.ld_tgt_count) { + if (tgt < 0) { + CDEBUG(D_HA, "%s: "DFID" is inaccessible: rc = %d\n", + obd->obd_name, PFID(fid), tgt); + tgt = 0; + } + + if (lmv->tgts[tgt] == NULL || + lmv->tgts[tgt]->ltd_exp == NULL || + lmv->tgts[tgt]->ltd_active == 0) continue; - rc = md_lock_match(lmv->tgts[i]->ltd_exp, flags, fid, - type, policy, mode, lockh); - if (rc) - RETURN(rc); - } + rc = md_lock_match(lmv->tgts[tgt]->ltd_exp, flags, fid, + type, policy, mode, lockh); + if (rc) + RETURN(rc); + } - RETURN(0); + RETURN(0); } int lmv_get_lustre_md(struct obd_export *exp, struct ptlrpc_request *req, @@ -2659,35 +3146,44 @@ int lmv_get_lustre_md(struct obd_export *exp, struct ptlrpc_request *req, struct lustre_md *md) { struct lmv_obd *lmv = &exp->exp_obd->u.lmv; + struct lmv_tgt_desc *tgt = lmv->tgts[0]; + + if (tgt == NULL || tgt->ltd_exp == NULL) + RETURN(-EINVAL); return md_get_lustre_md(lmv->tgts[0]->ltd_exp, req, dt_exp, md_exp, md); } int lmv_free_lustre_md(struct obd_export *exp, struct lustre_md *md) { - struct obd_device *obd = exp->exp_obd; - struct lmv_obd *lmv = &obd->u.lmv; + struct obd_device *obd = exp->exp_obd; + struct lmv_obd *lmv = &obd->u.lmv; + struct lmv_tgt_desc *tgt = lmv->tgts[0]; ENTRY; - if (md->mea) - obd_free_memmd(exp, (void *)&md->mea); + if (md->lmv != NULL) { + lmv_free_memmd(md->lmv); + md->lmv = NULL; + } + if (tgt == NULL || tgt->ltd_exp == NULL) + RETURN(-EINVAL); RETURN(md_free_lustre_md(lmv->tgts[0]->ltd_exp, md)); } int lmv_set_open_replay_data(struct obd_export *exp, - struct obd_client_handle *och, - struct ptlrpc_request *open_req) + struct obd_client_handle *och, + struct lookup_intent *it) { - struct obd_device *obd = exp->exp_obd; - struct lmv_obd *lmv = &obd->u.lmv; - struct lmv_tgt_desc *tgt; - ENTRY; + struct obd_device *obd = exp->exp_obd; + struct lmv_obd *lmv = &obd->u.lmv; + struct lmv_tgt_desc *tgt; + ENTRY; - tgt = lmv_find_target(lmv, &och->och_fid); - if (IS_ERR(tgt)) - RETURN(PTR_ERR(tgt)); + tgt = lmv_find_target(lmv, &och->och_fid); + if (IS_ERR(tgt)) + RETURN(PTR_ERR(tgt)); - RETURN(md_set_open_replay_data(tgt->ltd_exp, och, open_req)); + RETURN(md_set_open_replay_data(tgt->ltd_exp, och, it)); } int lmv_clear_open_replay_data(struct obd_export *exp, @@ -2752,9 +3248,12 @@ static int lmv_renew_capa(struct obd_export *exp, struct obd_capa *oc, int lmv_unpack_capa(struct obd_export *exp, struct ptlrpc_request *req, const struct req_msg_field *field, struct obd_capa **oc) { - struct lmv_obd *lmv = &exp->exp_obd->u.lmv; + struct lmv_obd *lmv = &exp->exp_obd->u.lmv; + struct lmv_tgt_desc *tgt = lmv->tgts[0]; - return md_unpack_capa(lmv->tgts[0]->ltd_exp, req, field, oc); + if (tgt == NULL || tgt->ltd_exp == NULL) + RETURN(-EINVAL); + return md_unpack_capa(tgt->ltd_exp, req, field, oc); } int lmv_intent_getattr_async(struct obd_export *exp, @@ -2772,7 +3271,7 @@ int lmv_intent_getattr_async(struct obd_export *exp, if (rc) RETURN(rc); - tgt = lmv_find_target(lmv, &op_data->op_fid1); + tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid1); if (IS_ERR(tgt)) RETURN(PTR_ERR(tgt)); @@ -2801,6 +3300,22 @@ int lmv_revalidate_lock(struct obd_export *exp, struct lookup_intent *it, RETURN(rc); } +int lmv_get_fid_from_lsm(struct obd_export *exp, + const struct lmv_stripe_md *lsm, + const char *name, int namelen, struct lu_fid *fid) +{ + const struct lmv_oinfo *oinfo; + + LASSERT(lsm != NULL); + oinfo = lsm_name_to_stripe_info(lsm, name, namelen); + if (IS_ERR(oinfo)) + return PTR_ERR(oinfo); + + *fid = oinfo->lmo_fid; + + RETURN(0); +} + /** * For lmv, only need to send request to master MDT, and the master MDT will * process with other slave MDTs. The only exception is Q_GETOQUOTA for which @@ -2817,10 +3332,13 @@ int lmv_quotactl(struct obd_device *unused, struct obd_export *exp, __u64 curspace, curinodes; ENTRY; - if (!lmv->desc.ld_tgt_count || !tgt->ltd_active) { - CERROR("master lmv inactive\n"); - RETURN(-EIO); - } + if (tgt == NULL || + tgt->ltd_exp == NULL || + !tgt->ltd_active || + lmv->desc.ld_tgt_count == 0) { + CERROR("master lmv inactive\n"); + RETURN(-EIO); + } if (oqctl->qc_cmd != Q_GETOQUOTA) { rc = obd_quotactl(tgt->ltd_exp, oqctl); @@ -2832,12 +3350,8 @@ int lmv_quotactl(struct obd_device *unused, struct obd_export *exp, int err; tgt = lmv->tgts[i]; - if (tgt == NULL || tgt->ltd_exp == NULL || tgt->ltd_active == 0) + if (tgt == NULL || tgt->ltd_exp == NULL || !tgt->ltd_active) continue; - if (!tgt->ltd_active) { - CDEBUG(D_HA, "mdt %d is inactive.\n", i); - continue; - } err = obd_quotactl(tgt->ltd_exp, oqctl); if (err) { @@ -2881,6 +3395,48 @@ int lmv_quotacheck(struct obd_device *unused, struct obd_export *exp, RETURN(rc); } +int lmv_update_lsm_md(struct obd_export *exp, struct lmv_stripe_md *lsm, + struct mdt_body *body, ldlm_blocking_callback cb_blocking) +{ + return lmv_revalidate_slaves(exp, body, lsm, cb_blocking, 0); +} + +int lmv_merge_attr(struct obd_export *exp, const struct lmv_stripe_md *lsm, + struct cl_attr *attr) +{ +#ifdef __KERNEL__ + int i; + + for (i = 0; i < lsm->lsm_md_stripe_count; i++) { + struct inode *inode = lsm->lsm_md_oinfo[i].lmo_root; + + CDEBUG(D_INFO, ""DFID" size %llu, nlink %u, atime %lu ctime" + "%lu, mtime %lu.\n", PFID(&lsm->lsm_md_oinfo[i].lmo_fid), + i_size_read(inode), inode->i_nlink, + LTIME_S(inode->i_atime), LTIME_S(inode->i_ctime), + LTIME_S(inode->i_mtime)); + + /* for slave stripe, it needs to subtract nlink for . and .. */ + if (i != 0) + attr->cat_nlink += inode->i_nlink - 2; + else + attr->cat_nlink = inode->i_nlink; + + attr->cat_size += i_size_read(inode); + + if (attr->cat_atime < LTIME_S(inode->i_atime)) + attr->cat_atime = LTIME_S(inode->i_atime); + + if (attr->cat_ctime < LTIME_S(inode->i_ctime)) + attr->cat_ctime = LTIME_S(inode->i_ctime); + + if (attr->cat_mtime < LTIME_S(inode->i_mtime)) + attr->cat_mtime = LTIME_S(inode->i_mtime); + } +#endif + return 0; +} + struct obd_ops lmv_obd_ops = { .o_owner = THIS_MODULE, .o_setup = lmv_setup, @@ -2917,34 +3473,34 @@ struct md_ops lmv_md_ops = { .m_rename = lmv_rename, .m_setattr = lmv_setattr, .m_setxattr = lmv_setxattr, - .m_sync = lmv_sync, - .m_readpage = lmv_readpage, + .m_fsync = lmv_fsync, + .m_read_entry = lmv_read_entry, .m_unlink = lmv_unlink, .m_init_ea_size = lmv_init_ea_size, .m_cancel_unused = lmv_cancel_unused, .m_set_lock_data = lmv_set_lock_data, .m_lock_match = lmv_lock_match, - .m_get_lustre_md = lmv_get_lustre_md, - .m_free_lustre_md = lmv_free_lustre_md, + .m_get_lustre_md = lmv_get_lustre_md, + .m_free_lustre_md = lmv_free_lustre_md, + .m_update_lsm_md = lmv_update_lsm_md, + .m_merge_attr = lmv_merge_attr, .m_set_open_replay_data = lmv_set_open_replay_data, .m_clear_open_replay_data = lmv_clear_open_replay_data, .m_renew_capa = lmv_renew_capa, .m_unpack_capa = lmv_unpack_capa, .m_get_remote_perm = lmv_get_remote_perm, .m_intent_getattr_async = lmv_intent_getattr_async, - .m_revalidate_lock = lmv_revalidate_lock + .m_revalidate_lock = lmv_revalidate_lock, + .m_get_fid_from_lsm = lmv_get_fid_from_lsm, }; int __init lmv_init(void) { - struct lprocfs_static_vars lvars; - int rc; - - lprocfs_lmv_init_vars(&lvars); - - rc = class_register_type(&lmv_obd_ops, &lmv_md_ops, - lvars.module_vars, LUSTRE_LMV_NAME, NULL); - return rc; + return class_register_type(&lmv_obd_ops, &lmv_md_ops, true, NULL, +#ifndef HAVE_ONLY_PROCFS_SEQ + NULL, +#endif + LUSTRE_LMV_NAME, NULL); } #ifdef __KERNEL__