X-Git-Url: https://git.whamcloud.com/?a=blobdiff_plain;f=lustre%2Fmdt%2Fmdt_recovery.c;h=8ae36313456c33b32187a36f0d551708c9d316af;hb=0e67cdb9bd986e1b4dfa915ae8c2e7c2687f3df2;hp=4853c3eaa3a26ffb8c51721007286924ae0e6ce1;hpb=fbf5870b9848929d352460f1f005b79c0b5ccc5a;p=fs%2Flustre-release.git diff --git a/lustre/mdt/mdt_recovery.c b/lustre/mdt/mdt_recovery.c index 4853c3e..8ae3631 100644 --- a/lustre/mdt/mdt_recovery.c +++ b/lustre/mdt/mdt_recovery.c @@ -26,8 +26,11 @@ * GPL HEADER END */ /* - * Copyright 2008 Sun Microsystems, Inc. All rights reserved + * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved. * Use is subject to license terms. + * + * Copyright (c) 2011 Whamcloud, Inc. + * */ /* * This file is part of Lustre, http://www.lustre.org/ @@ -77,119 +80,62 @@ const struct lu_buf *mdt_buf_const(const struct lu_env *env, return buf; } -int mdt_record_read(const struct lu_env *env, - struct dt_object *dt, struct lu_buf *buf, loff_t *pos) +static inline int mdt_trans_credit_get(const struct lu_env *env, + struct mdt_device *mdt, + enum mdt_txn_op op) { - int rc; - - LASSERTF(dt != NULL, "dt is NULL when we want to read record\n"); - - rc = dt->do_body_ops->dbo_read(env, dt, buf, pos, BYPASS_CAPA); - - if (rc == buf->lb_len) - rc = 0; - else if (rc >= 0) - rc = -EFAULT; - return rc; + struct dt_device *dev = mdt->mdt_bottom; + int cr; + switch (op) { + case MDT_TXN_CAPA_KEYS_WRITE_OP: + case MDT_TXN_LAST_RCVD_WRITE_OP: + cr = dev->dd_ops->dt_credit_get(env, + dev, + DTO_WRITE_BLOCK); + break; + default: + LBUG(); + } + return cr; } -int mdt_record_write(const struct lu_env *env, - struct dt_object *dt, const struct lu_buf *buf, - loff_t *pos, struct thandle *th) +void mdt_trans_credit_init(const struct lu_env *env, + struct mdt_device *mdt, + enum mdt_txn_op op) { - int rc; + struct mdt_thread_info *mti; + struct txn_param *p; + int cr; - LASSERTF(dt != NULL, "dt is NULL when we want to write record\n"); - LASSERT(th != NULL); - rc = dt->do_body_ops->dbo_write(env, dt, buf, pos, th, BYPASS_CAPA); - if (rc == buf->lb_len) - rc = 0; - else if (rc >= 0) - rc = -EFAULT; - return rc; -} -/* only one record write */ + mti = lu_context_key_get(&env->le_ctx, &mdt_thread_key); + p = &mti->mti_txn_param; -enum { - MDT_TXN_LAST_RCVD_WRITE_CREDITS = 3 -}; + cr = mdt_trans_credit_get(env, mdt, op); + txn_param_init(p, cr); +} struct thandle* mdt_trans_start(const struct lu_env *env, - struct mdt_device *mdt, int credits) + struct mdt_device *mdt) { struct mdt_thread_info *mti; struct txn_param *p; mti = lu_context_key_get(&env->le_ctx, &mdt_thread_key); p = &mti->mti_txn_param; - txn_param_init(p, credits); - /* export can require sync operations */ - if (mti->mti_exp != NULL) - p->tp_sync = mti->mti_exp->exp_need_sync; - - return mdt->mdt_bottom->dd_ops->dt_trans_start(env, mdt->mdt_bottom, p); + return dt_trans_start(env, mdt->mdt_bottom, p); } void mdt_trans_stop(const struct lu_env *env, struct mdt_device *mdt, struct thandle *th) { - mdt->mdt_bottom->dd_ops->dt_trans_stop(env, th); -} - -/* last_rcvd handling */ -static inline void lsd_le_to_cpu(struct lr_server_data *buf, - struct lr_server_data *lsd) -{ - memcpy(lsd->lsd_uuid, buf->lsd_uuid, sizeof (lsd->lsd_uuid)); - lsd->lsd_last_transno = le64_to_cpu(buf->lsd_last_transno); - lsd->lsd_mount_count = le64_to_cpu(buf->lsd_mount_count); - lsd->lsd_feature_compat = le32_to_cpu(buf->lsd_feature_compat); - lsd->lsd_feature_rocompat = le32_to_cpu(buf->lsd_feature_rocompat); - lsd->lsd_feature_incompat = le32_to_cpu(buf->lsd_feature_incompat); - lsd->lsd_server_size = le32_to_cpu(buf->lsd_server_size); - lsd->lsd_client_start = le32_to_cpu(buf->lsd_client_start); - lsd->lsd_client_size = le16_to_cpu(buf->lsd_client_size); -} - -static inline void lsd_cpu_to_le(struct lr_server_data *lsd, - struct lr_server_data *buf) -{ - memcpy(buf->lsd_uuid, lsd->lsd_uuid, sizeof (lsd->lsd_uuid)); - buf->lsd_last_transno = cpu_to_le64(lsd->lsd_last_transno); - buf->lsd_mount_count = cpu_to_le64(lsd->lsd_mount_count); - buf->lsd_feature_compat = cpu_to_le32(lsd->lsd_feature_compat); - buf->lsd_feature_rocompat = cpu_to_le32(lsd->lsd_feature_rocompat); - buf->lsd_feature_incompat = cpu_to_le32(lsd->lsd_feature_incompat); - buf->lsd_server_size = cpu_to_le32(lsd->lsd_server_size); - buf->lsd_client_start = cpu_to_le32(lsd->lsd_client_start); - buf->lsd_client_size = cpu_to_le16(lsd->lsd_client_size); -} - -static inline void lcd_le_to_cpu(struct lsd_client_data *buf, - struct lsd_client_data *lcd) -{ - memcpy(lcd->lcd_uuid, buf->lcd_uuid, sizeof (lcd->lcd_uuid)); - lcd->lcd_last_transno = le64_to_cpu(buf->lcd_last_transno); - lcd->lcd_last_xid = le64_to_cpu(buf->lcd_last_xid); - lcd->lcd_last_result = le32_to_cpu(buf->lcd_last_result); - lcd->lcd_last_data = le32_to_cpu(buf->lcd_last_data); - lcd->lcd_last_close_transno = le64_to_cpu(buf->lcd_last_close_transno); - lcd->lcd_last_close_xid = le64_to_cpu(buf->lcd_last_close_xid); - lcd->lcd_last_close_result = le32_to_cpu(buf->lcd_last_close_result); -} + struct mdt_thread_info *mti; -static inline void lcd_cpu_to_le(struct lsd_client_data *lcd, - struct lsd_client_data *buf) -{ - memcpy(buf->lcd_uuid, lcd->lcd_uuid, sizeof (lcd->lcd_uuid)); - buf->lcd_last_transno = cpu_to_le64(lcd->lcd_last_transno); - buf->lcd_last_xid = cpu_to_le64(lcd->lcd_last_xid); - buf->lcd_last_result = cpu_to_le32(lcd->lcd_last_result); - buf->lcd_last_data = cpu_to_le32(lcd->lcd_last_data); - buf->lcd_last_close_transno = cpu_to_le64(lcd->lcd_last_close_transno); - buf->lcd_last_close_xid = cpu_to_le64(lcd->lcd_last_close_xid); - buf->lcd_last_close_result = cpu_to_le32(lcd->lcd_last_close_result); + mti = lu_context_key_get(&env->le_ctx, &mdt_thread_key); + /* export can require sync operations */ + if (mti->mti_exp != NULL) + th->th_sync |= mti->mti_exp->exp_need_sync; + dt_trans_stop(env, mdt->mdt_bottom, th); } static inline int mdt_last_rcvd_header_read(const struct lu_env *env, @@ -201,54 +147,46 @@ static inline int mdt_last_rcvd_header_read(const struct lu_env *env, mti = lu_context_key_get(&env->le_ctx, &mdt_thread_key); mti->mti_off = 0; - rc = mdt_record_read(env, mdt->mdt_last_rcvd, - mdt_buf(env, &mti->mti_lsd, sizeof(mti->mti_lsd)), - &mti->mti_off); + rc = dt_record_read(env, mdt->mdt_lut.lut_last_rcvd, + mdt_buf(env, &mti->mti_lsd, sizeof(mti->mti_lsd)), + &mti->mti_off); if (rc == 0) - lsd_le_to_cpu(&mti->mti_lsd, &mdt->mdt_lsd); + lsd_le_to_cpu(&mti->mti_lsd, &mdt->mdt_lut.lut_lsd); - CDEBUG(D_INFO, "read last_rcvd header rc = %d:\n" - "uuid = %s\n" - "last_transno = "LPU64"\n", - rc, mdt->mdt_lsd.lsd_uuid, - mdt->mdt_lsd.lsd_last_transno); + CDEBUG(D_INFO, "read last_rcvd header rc = %d, uuid = %s, " + "last_transno = "LPU64"\n", rc, mdt->mdt_lut.lut_lsd.lsd_uuid, + mdt->mdt_lut.lut_lsd.lsd_last_transno); return rc; } -static inline int mdt_last_rcvd_header_write(const struct lu_env *env, - struct mdt_device *mdt) +static int mdt_last_rcvd_header_write(const struct lu_env *env, + struct mdt_device *mdt, + struct thandle *th) { struct mdt_thread_info *mti; - struct thandle *th; int rc; ENTRY; mti = lu_context_key_get(&env->le_ctx, &mdt_thread_key); - th = mdt_trans_start(env, mdt, MDT_TXN_LAST_RCVD_WRITE_CREDITS); - if (IS_ERR(th)) - RETURN(PTR_ERR(th)); - mti->mti_off = 0; - lsd_cpu_to_le(&mdt->mdt_lsd, &mti->mti_lsd); + lsd_cpu_to_le(&mdt->mdt_lut.lut_lsd, &mti->mti_lsd); - rc = mdt_record_write(env, mdt->mdt_last_rcvd, - mdt_buf_const(env, &mti->mti_lsd, - sizeof(mti->mti_lsd)), - &mti->mti_off, th); + rc = dt_record_write(env, mdt->mdt_lut.lut_last_rcvd, + mdt_buf_const(env, &mti->mti_lsd, + sizeof(mti->mti_lsd)), + &mti->mti_off, th); - mdt_trans_stop(env, mdt, th); - - CDEBUG(D_INFO, "write last_rcvd header rc = %d:\n" - "uuid = %s\nlast_transno = "LPU64"\n", - rc, mdt->mdt_lsd.lsd_uuid, mdt->mdt_lsd.lsd_last_transno); + CDEBUG(D_INFO, "write last_rcvd header rc = %d, uuid = %s, " + "last_transno = "LPU64"\n", rc, mdt->mdt_lut.lut_lsd.lsd_uuid, + mdt->mdt_lut.lut_lsd.lsd_last_transno); RETURN(rc); } -static int mdt_last_rcvd_read(const struct lu_env *env, - struct mdt_device *mdt, - struct lsd_client_data *lcd, loff_t *off) +static int mdt_last_rcvd_read(const struct lu_env *env, struct mdt_device *mdt, + struct lsd_client_data *lcd, loff_t *off, + int index) { struct mdt_thread_info *mti; struct lsd_client_data *tmp; @@ -256,30 +194,21 @@ static int mdt_last_rcvd_read(const struct lu_env *env, mti = lu_context_key_get(&env->le_ctx, &mdt_thread_key); tmp = &mti->mti_lcd; - rc = mdt_record_read(env, mdt->mdt_last_rcvd, - mdt_buf(env, tmp, sizeof(*tmp)), off); - if (rc == 0) + rc = dt_record_read(env, mdt->mdt_lut.lut_last_rcvd, + mdt_buf(env, tmp, sizeof(*tmp)), off); + if (rc == 0) { + check_lcd(mdt2obd_dev(mdt)->obd_name, index, tmp); lcd_le_to_cpu(tmp, lcd); + } - CDEBUG(D_INFO, "read lcd @%d rc = %d:\n" - "uuid = %s\n" - "last_transno = "LPU64"\n" - "last_xid = "LPU64"\n" - "last_result = %u\n" - "last_data = %u\n" - "last_close_transno = "LPU64"\n" - "last_close_xid = "LPU64"\n" - "last_close_result = %u\n", - (int)(*off - sizeof(*tmp)), - rc, - lcd->lcd_uuid, - lcd->lcd_last_transno, - lcd->lcd_last_xid, - lcd->lcd_last_result, - lcd->lcd_last_data, - lcd->lcd_last_close_transno, - lcd->lcd_last_close_xid, - lcd->lcd_last_close_result); + CDEBUG(D_INFO, "read lcd @%d rc = %d, uuid = %s, last_transno = "LPU64 + ", last_xid = "LPU64", last_result = %u, last_data = %u, " + "last_close_transno = "LPU64", last_close_xid = "LPU64", " + "last_close_result = %u\n", (int)(*off - sizeof(*tmp)), + rc, lcd->lcd_uuid, lcd->lcd_last_transno, lcd->lcd_last_xid, + lcd->lcd_last_result, lcd->lcd_last_data, + lcd->lcd_last_close_transno, lcd->lcd_last_close_xid, + lcd->lcd_last_close_result); return rc; } @@ -298,28 +227,17 @@ static int mdt_last_rcvd_write(const struct lu_env *env, lcd_cpu_to_le(lcd, tmp); - rc = mdt_record_write(env, mdt->mdt_last_rcvd, - mdt_buf_const(env, tmp, sizeof(*tmp)), off, th); - - CDEBUG(D_INFO, "write lcd @%d rc = %d:\n" - "uuid = %s\n" - "last_transno = "LPU64"\n" - "last_xid = "LPU64"\n" - "last_result = %u\n" - "last_data = %u\n" - "last_close_transno = "LPU64"\n" - "last_close_xid = "LPU64"\n" - "last_close_result = %u\n", - (int)(*off - sizeof(*tmp)), - rc, - lcd->lcd_uuid, - lcd->lcd_last_transno, - lcd->lcd_last_xid, - lcd->lcd_last_result, - lcd->lcd_last_data, - lcd->lcd_last_close_transno, - lcd->lcd_last_close_xid, - lcd->lcd_last_close_result); + rc = dt_record_write(env, mdt->mdt_lut.lut_last_rcvd, + mdt_buf_const(env, tmp, sizeof(*tmp)), off, th); + + CDEBUG(D_INFO, "write lcd @%d rc = %d, uuid = %s, last_transno = "LPU64 + ", last_xid = "LPU64", last_result = %u, last_data = %u, " + "last_close_transno = "LPU64", last_close_xid = "LPU64" ," + "last_close_result = %u\n", (int)(*off - sizeof(*tmp)), + rc, lcd->lcd_uuid, lcd->lcd_last_transno, lcd->lcd_last_xid, + lcd->lcd_last_result, lcd->lcd_last_data, + lcd->lcd_last_close_transno, lcd->lcd_last_close_xid, + lcd->lcd_last_close_result); return rc; } @@ -327,33 +245,32 @@ static int mdt_clients_data_init(const struct lu_env *env, struct mdt_device *mdt, unsigned long last_size) { - struct lr_server_data *lsd = &mdt->mdt_lsd; - struct lsd_client_data *lcd = NULL; - struct obd_device *obd = mdt->mdt_md_dev.md_lu_dev.ld_obd; + struct lr_server_data *lsd = &mdt->mdt_lut.lut_lsd; + struct lsd_client_data *lcd; + struct obd_device *obd = mdt2obd_dev(mdt); loff_t off; int cl_idx; int rc = 0; ENTRY; + OBD_ALLOC_PTR(lcd); + if (!lcd) + RETURN(-ENOMEM); + /* When we do a clean MDS shutdown, we save the last_transno into * the header. If we find clients with higher last_transno values * then those clients may need recovery done. */ - LASSERT(atomic_read(&obd->obd_req_replay_clients) == 0); + LASSERT(cfs_atomic_read(&obd->obd_req_replay_clients) == 0); for (cl_idx = 0, off = lsd->lsd_client_start; off < last_size; cl_idx++) { __u64 last_transno; struct obd_export *exp; - - if (!lcd) { - OBD_ALLOC_PTR(lcd); - if (!lcd) - RETURN(-ENOMEM); - } + struct mdt_thread_info *mti; off = lsd->lsd_client_start + cl_idx * lsd->lsd_client_size; - rc = mdt_last_rcvd_read(env, mdt, lcd, &off); + rc = mdt_last_rcvd_read(env, mdt, lcd, &off, cl_idx); if (rc) { CERROR("error reading MDS %s idx %d, off %llu: rc %d\n", LAST_RCVD, cl_idx, off, rc); @@ -382,51 +299,54 @@ static int mdt_clients_data_init(const struct lu_env *env, if (IS_ERR(exp)) { if (PTR_ERR(exp) == -EALREADY) { /* export already exists, zero out this one */ - lcd->lcd_uuid[0] = '\0'; - } else - GOTO(err_client, rc = PTR_ERR(exp)); - } else { - struct mdt_thread_info *mti; - mti = lu_context_key_get(&env->le_ctx, &mdt_thread_key); - LASSERT(mti != NULL); - mti->mti_exp = exp; - exp->exp_mdt_data.med_lcd = lcd; - rc = mdt_client_add(env, mdt, cl_idx); - /* can't fail existing */ - LASSERTF(rc == 0, "rc = %d\n", rc); - lcd = NULL; - spin_lock(&exp->exp_lock); - exp->exp_connecting = 0; - exp->exp_in_recovery = 0; - spin_unlock(&exp->exp_lock); - obd->obd_max_recoverable_clients++; - class_export_put(exp); + CERROR("Duplicate export %s!\n", lcd->lcd_uuid); + continue; + } + GOTO(err_client, rc = PTR_ERR(exp)); } + mti = lu_context_key_get(&env->le_ctx, &mdt_thread_key); + LASSERT(mti != NULL); + mti->mti_exp = exp; + /* copy on-disk lcd to the export */ + *exp->exp_target_data.ted_lcd = *lcd; + rc = mdt_client_add(env, mdt, cl_idx); + /* can't fail existing */ + LASSERTF(rc == 0, "rc = %d\n", rc); + /* VBR: set export last committed version */ + exp->exp_last_committed = last_transno; + cfs_spin_lock(&exp->exp_lock); + exp->exp_connecting = 0; + exp->exp_in_recovery = 0; + cfs_spin_unlock(&exp->exp_lock); + obd->obd_max_recoverable_clients++; + class_export_put(exp); + CDEBUG(D_OTHER, "client at idx %d has last_transno="LPU64"\n", cl_idx, last_transno); /* protect __u64 value update */ - spin_lock(&mdt->mdt_transno_lock); - mdt->mdt_last_transno = max(last_transno, - mdt->mdt_last_transno); - spin_unlock(&mdt->mdt_transno_lock); + cfs_spin_lock(&mdt->mdt_lut.lut_translock); + mdt->mdt_lut.lut_last_transno = max(last_transno, + mdt->mdt_lut.lut_last_transno); + cfs_spin_unlock(&mdt->mdt_lut.lut_translock); } err_client: - if (lcd) - OBD_FREE_PTR(lcd); + OBD_FREE_PTR(lcd); RETURN(rc); } static int mdt_server_data_init(const struct lu_env *env, - struct mdt_device *mdt) + struct mdt_device *mdt, + struct lustre_sb_info *lsi) { - struct lr_server_data *lsd = &mdt->mdt_lsd; + struct lr_server_data *lsd = &mdt->mdt_lut.lut_lsd; struct lsd_client_data *lcd = NULL; - struct obd_device *obd = mdt->mdt_md_dev.md_lu_dev.ld_obd; + struct obd_device *obd = mdt2obd_dev(mdt); struct mdt_thread_info *mti; struct dt_object *obj; struct lu_attr *la; + struct lustre_disk_data *ldd; unsigned long last_rcvd_size; __u64 mount_count; int rc; @@ -442,8 +362,8 @@ static int mdt_server_data_init(const struct lu_env *env, LASSERT(mti != NULL); la = &mti->mti_attr.ma_attr; - obj = mdt->mdt_last_rcvd; - rc = obj->do_ops->do_attr_get(env, mdt->mdt_last_rcvd, la, BYPASS_CAPA); + obj = mdt->mdt_lut.lut_last_rcvd; + rc = obj->do_ops->do_attr_get(env, obj, la, BYPASS_CAPA); if (rc) RETURN(rc); @@ -459,9 +379,10 @@ static int mdt_server_data_init(const struct lu_env *env, lsd->lsd_server_size = LR_SERVER_SIZE; lsd->lsd_client_start = LR_CLIENT_START; lsd->lsd_client_size = LR_CLIENT_SIZE; + lsd->lsd_feature_compat = OBD_COMPAT_MDT; lsd->lsd_feature_rocompat = OBD_ROCOMPAT_LOVOBJID; lsd->lsd_feature_incompat = OBD_INCOMPAT_MDT | - OBD_INCOMPAT_COMMON_LR; + OBD_INCOMPAT_COMMON_LR; } else { LCONSOLE_WARN("%s: used disk, loading\n", obd->obd_name); rc = mdt_last_rcvd_header_read(env, mdt); @@ -476,18 +397,54 @@ static int mdt_server_data_init(const struct lu_env *env, obd->obd_uuid.uuid, lsd->lsd_uuid); GOTO(out, rc = -EINVAL); } + lsd->lsd_feature_compat |= OBD_COMPAT_MDT; + lsd->lsd_feature_incompat |= OBD_INCOMPAT_MDT | + OBD_INCOMPAT_COMMON_LR; } mount_count = lsd->lsd_mount_count; - lsd->lsd_feature_compat = OBD_COMPAT_MDT; + ldd = lsi->lsi_ldd; - spin_lock(&mdt->mdt_transno_lock); - mdt->mdt_last_transno = lsd->lsd_last_transno; - spin_unlock(&mdt->mdt_transno_lock); + if (lsd->lsd_feature_incompat & ~MDT_INCOMPAT_SUPP) { + CERROR("%s: unsupported incompat filesystem feature(s) %x\n", + obd->obd_name, + lsd->lsd_feature_incompat & ~MDT_INCOMPAT_SUPP); + GOTO(out, rc = -EINVAL); + } + if (lsd->lsd_feature_rocompat & ~MDT_ROCOMPAT_SUPP) { + CERROR("%s: unsupported read-only filesystem feature(s) %x\n", + obd->obd_name, + lsd->lsd_feature_rocompat & ~MDT_ROCOMPAT_SUPP); + /* XXX: Do something like remount filesystem read-only */ + GOTO(out, rc = -EINVAL); + } + /** Interop: evict all clients at first boot with 1.8 last_rcvd */ + if (!(lsd->lsd_feature_compat & OBD_COMPAT_20)) { + if (last_rcvd_size > lsd->lsd_client_start) { + LCONSOLE_WARN("Mounting %s at first time on 1.8 FS, " + "remove all clients for interop needs\n", + obd->obd_name); + simple_truncate(lsi->lsi_srv_mnt->mnt_sb->s_root, + lsi->lsi_srv_mnt, LAST_RCVD, + lsd->lsd_client_start); + last_rcvd_size = lsd->lsd_client_start; + } + /** set 2.0 flag to upgrade/downgrade between 1.8 and 2.0 */ + lsd->lsd_feature_compat |= OBD_COMPAT_20; + } + + if (ldd->ldd_flags & LDD_F_IAM_DIR) + lsd->lsd_feature_incompat |= OBD_INCOMPAT_IAM_DIR; + + lsd->lsd_feature_incompat |= OBD_INCOMPAT_FID; + + cfs_spin_lock(&mdt->mdt_lut.lut_translock); + mdt->mdt_lut.lut_last_transno = lsd->lsd_last_transno; + cfs_spin_unlock(&mdt->mdt_lut.lut_translock); CDEBUG(D_INODE, "========BEGIN DUMPING LAST_RCVD========\n"); CDEBUG(D_INODE, "%s: server last_transno: "LPU64"\n", - obd->obd_name, mdt->mdt_last_transno); + obd->obd_name, mdt->mdt_lut.lut_last_transno); CDEBUG(D_INODE, "%s: server mount_count: "LPU64"\n", obd->obd_name, mount_count + 1); CDEBUG(D_INODE, "%s: server data size: %u\n", @@ -514,14 +471,15 @@ static int mdt_server_data_init(const struct lu_env *env, if (rc) GOTO(err_client, rc); - spin_lock(&mdt->mdt_transno_lock); + cfs_spin_lock(&mdt->mdt_lut.lut_translock); /* obd_last_committed is used for compatibility * with other lustre recovery code */ - obd->obd_last_committed = mdt->mdt_last_transno; - spin_unlock(&mdt->mdt_transno_lock); + obd->obd_last_committed = mdt->mdt_lut.lut_last_transno; + cfs_spin_unlock(&mdt->mdt_lut.lut_translock); - mdt->mdt_mount_count++; - lsd->lsd_mount_count = mdt->mdt_mount_count; + obd->u.obt.obt_mount_count = mount_count + 1; + obd->u.obt.obt_instance = (__u32)obd->u.obt.obt_mount_count; + lsd->lsd_mount_count = obd->u.obt.obt_mount_count; /* save it, so mount count and last_transno is current */ rc = mdt_server_data_update(env, mdt); @@ -531,7 +489,7 @@ static int mdt_server_data_init(const struct lu_env *env, RETURN(0); err_client: - target_recovery_fini(obd); + class_disconnect_exports(obd); out: return rc; } @@ -539,41 +497,38 @@ out: static int mdt_server_data_update(const struct lu_env *env, struct mdt_device *mdt) { - int rc = 0; - ENTRY; + struct mdt_thread_info *mti; + struct thandle *th; + int rc; - CDEBUG(D_SUPER, "MDS mount_count is "LPU64", last_transno is "LPU64"\n", - mdt->mdt_mount_count, mdt->mdt_last_transno); + mti = lu_context_key_get(&env->le_ctx, &mdt_thread_key); - spin_lock(&mdt->mdt_transno_lock); - mdt->mdt_lsd.lsd_last_transno = mdt->mdt_last_transno; - spin_unlock(&mdt->mdt_transno_lock); + mdt_trans_credit_init(env, mdt, MDT_TXN_LAST_RCVD_WRITE_OP); + th = mdt_trans_start(env, mdt); + if (IS_ERR(th)) + RETURN(PTR_ERR(th)); - /* - * This may be called from difficult reply handler and - * mdt->mdt_last_rcvd may be NULL that time. - */ - if (mdt->mdt_last_rcvd != NULL) - rc = mdt_last_rcvd_header_write(env, mdt); - RETURN(rc); -} + CDEBUG(D_SUPER, "MDS mount_count is "LPU64", last_transno is "LPU64"\n", + mdt->mdt_lut.lut_obd->u.obt.obt_mount_count, + mdt->mdt_lut.lut_last_transno); -void mdt_cb_new_client(const struct mdt_device *mdt, __u64 transno, - void *data, int err) -{ - struct obd_device *obd = mdt->mdt_md_dev.md_lu_dev.ld_obd; + cfs_spin_lock(&mdt->mdt_lut.lut_translock); + mdt->mdt_lut.lut_lsd.lsd_last_transno = mdt->mdt_lut.lut_last_transno; + cfs_spin_unlock(&mdt->mdt_lut.lut_translock); - target_client_add_cb(obd, transno, data, err); + rc = mdt_last_rcvd_header_write(env, mdt, th); + mdt_trans_stop(env, mdt, th); + return rc; } + int mdt_client_new(const struct lu_env *env, struct mdt_device *mdt) { - unsigned long *bitmap = mdt->mdt_client_bitmap; + unsigned long *bitmap = mdt->mdt_lut.lut_client_bitmap; struct mdt_thread_info *mti; - struct mdt_export_data *med; - struct lsd_client_data *lcd; - struct lr_server_data *lsd = &mdt->mdt_lsd; - struct obd_device *obd = mdt->mdt_md_dev.md_lu_dev.ld_obd; + struct tg_export_data *ted; + struct lr_server_data *lsd = &mdt->mdt_lut.lut_lsd; + struct obd_device *obd = mdt2obd_dev(mdt); struct thandle *th; loff_t off; int rc; @@ -583,53 +538,68 @@ int mdt_client_new(const struct lu_env *env, struct mdt_device *mdt) mti = lu_context_key_get(&env->le_ctx, &mdt_thread_key); LASSERT(mti != NULL); - med = &mti->mti_exp->exp_mdt_data; - lcd = med->med_lcd; + ted = &mti->mti_exp->exp_target_data; LASSERT(bitmap != NULL); - if (!strcmp(med->med_lcd->lcd_uuid, obd->obd_uuid.uuid)) + if (!strcmp(ted->ted_lcd->lcd_uuid, obd->obd_uuid.uuid)) RETURN(0); /* the bitmap operations can handle cl_idx > sizeof(long) * 8, so * there's no need for extra complication here */ - spin_lock(&mdt->mdt_client_bitmap_lock); - cl_idx = find_first_zero_bit(bitmap, LR_MAX_CLIENTS); + cfs_spin_lock(&mdt->mdt_lut.lut_client_bitmap_lock); + cl_idx = cfs_find_first_zero_bit(bitmap, LR_MAX_CLIENTS); if (cl_idx >= LR_MAX_CLIENTS || OBD_FAIL_CHECK(OBD_FAIL_MDS_CLIENT_ADD)) { CERROR("no room for %u clients - fix LR_MAX_CLIENTS\n", cl_idx); - spin_unlock(&mdt->mdt_client_bitmap_lock); + cfs_spin_unlock(&mdt->mdt_lut.lut_client_bitmap_lock); RETURN(-EOVERFLOW); } - set_bit(cl_idx, bitmap); - spin_unlock(&mdt->mdt_client_bitmap_lock); + cfs_set_bit(cl_idx, bitmap); + cfs_spin_unlock(&mdt->mdt_lut.lut_client_bitmap_lock); CDEBUG(D_INFO, "client at idx %d with UUID '%s' added\n", - cl_idx, med->med_lcd->lcd_uuid); + cl_idx, ted->ted_lcd->lcd_uuid); - med->med_lr_idx = cl_idx; - med->med_lr_off = lsd->lsd_client_start + + ted->ted_lr_idx = cl_idx; + ted->ted_lr_off = lsd->lsd_client_start + (cl_idx * lsd->lsd_client_size); - init_mutex(&med->med_lcd_lock); + cfs_init_mutex(&ted->ted_lcd_lock); + + LASSERTF(ted->ted_lr_off > 0, "ted_lr_off = %llu\n", ted->ted_lr_off); - LASSERTF(med->med_lr_off > 0, "med_lr_off = %llu\n", med->med_lr_off); - /* write new client data */ - off = med->med_lr_off; - th = mdt_trans_start(env, mdt, MDT_TXN_LAST_RCVD_WRITE_CREDITS); + /* Write new client data. */ + off = ted->ted_lr_off; + + if (OBD_FAIL_CHECK(OBD_FAIL_TGT_CLIENT_ADD)) + RETURN(-ENOSPC); + + mdt_trans_credit_init(env, mdt, MDT_TXN_LAST_RCVD_WRITE_OP); + + th = mdt_trans_start(env, mdt); if (IS_ERR(th)) RETURN(PTR_ERR(th)); - /* until this operations will be committed the sync is needed for this - * export */ - mdt_trans_add_cb(th, mdt_cb_new_client, mti->mti_exp); - spin_lock(&mti->mti_exp->exp_lock); - mti->mti_exp->exp_need_sync = 1; - spin_unlock(&mti->mti_exp->exp_lock); + /* + * Until this operations will be committed the sync is needed + * for this export. This should be done _after_ starting the + * transaction so that many connecting clients will not bring + * server down with lots of sync writes. + */ + rc = lut_new_client_cb_add(th, mti->mti_exp); + if (rc) { + /* can't add callback, do sync now */ + th->th_sync = 1; + } else { + cfs_spin_lock(&mti->mti_exp->exp_lock); + mti->mti_exp->exp_need_sync = 1; + cfs_spin_unlock(&mti->mti_exp->exp_lock); + } - rc = mdt_last_rcvd_write(env, mdt, lcd, &off, th); - CDEBUG(D_INFO, "wrote client lcd at idx %u off %llu (len "LPSZ")\n", - cl_idx, med->med_lr_off, sizeof(*lcd)); + rc = mdt_last_rcvd_write(env, mdt, ted->ted_lcd, &off, th); + CDEBUG(D_INFO, "wrote client lcd at idx %u off %llu (len %u)\n", + cl_idx, ted->ted_lr_off, (int)sizeof(*(ted->ted_lcd))); mdt_trans_stop(env, mdt, th); RETURN(rc); @@ -647,41 +617,41 @@ int mdt_client_add(const struct lu_env *env, struct mdt_device *mdt, int cl_idx) { struct mdt_thread_info *mti; - struct mdt_export_data *med; - unsigned long *bitmap = mdt->mdt_client_bitmap; - struct obd_device *obd = mdt->mdt_md_dev.md_lu_dev.ld_obd; - struct lr_server_data *lsd = &mdt->mdt_lsd; + struct tg_export_data *ted; + unsigned long *bitmap = mdt->mdt_lut.lut_client_bitmap; + struct obd_device *obd = mdt2obd_dev(mdt); + struct lr_server_data *lsd = &mdt->mdt_lut.lut_lsd; int rc = 0; ENTRY; mti = lu_context_key_get(&env->le_ctx, &mdt_thread_key); LASSERT(mti != NULL); - med = &mti->mti_exp->exp_mdt_data; + ted = &mti->mti_exp->exp_target_data; LASSERT(bitmap != NULL); LASSERTF(cl_idx >= 0, "%d\n", cl_idx); - if (!strcmp(med->med_lcd->lcd_uuid, obd->obd_uuid.uuid)) + if (!strcmp(ted->ted_lcd->lcd_uuid, obd->obd_uuid.uuid)) RETURN(0); - spin_lock(&mdt->mdt_client_bitmap_lock); - if (test_and_set_bit(cl_idx, bitmap)) { + cfs_spin_lock(&mdt->mdt_lut.lut_client_bitmap_lock); + if (cfs_test_and_set_bit(cl_idx, bitmap)) { CERROR("MDS client %d: bit already set in bitmap!!\n", cl_idx); LBUG(); } - spin_unlock(&mdt->mdt_client_bitmap_lock); + cfs_spin_unlock(&mdt->mdt_lut.lut_client_bitmap_lock); CDEBUG(D_INFO, "client at idx %d with UUID '%s' added\n", - cl_idx, med->med_lcd->lcd_uuid); + cl_idx, ted->ted_lcd->lcd_uuid); - med->med_lr_idx = cl_idx; - med->med_lr_off = lsd->lsd_client_start + + ted->ted_lr_idx = cl_idx; + ted->ted_lr_off = lsd->lsd_client_start + (cl_idx * lsd->lsd_client_size); - init_mutex(&med->med_lcd_lock); + cfs_init_mutex(&ted->ted_lcd_lock); - LASSERTF(med->med_lr_off > 0, "med_lr_off = %llu\n", med->med_lr_off); + LASSERTF(ted->ted_lr_off > 0, "ted_lr_off = %llu\n", ted->ted_lr_off); RETURN(rc); } @@ -689,38 +659,38 @@ int mdt_client_add(const struct lu_env *env, int mdt_client_del(const struct lu_env *env, struct mdt_device *mdt) { struct mdt_thread_info *mti; - struct mdt_export_data *med; - struct lsd_client_data *lcd; - struct obd_device *obd = mdt->mdt_md_dev.md_lu_dev.ld_obd; - struct thandle *th; - loff_t off; - int rc = 0; + struct tg_export_data *ted; + struct obd_device *obd = mdt2obd_dev(mdt); + struct obd_export *exp; + struct thandle *th; + loff_t off; + int rc = 0; ENTRY; mti = lu_context_key_get(&env->le_ctx, &mdt_thread_key); LASSERT(mti != NULL); - med = &mti->mti_exp->exp_mdt_data; - lcd = med->med_lcd; - if (!lcd) + exp = mti->mti_exp; + ted = &exp->exp_target_data; + if (!ted->ted_lcd) RETURN(0); /* XXX: If lcd_uuid were a real obd_uuid, I could use obd_uuid_equals */ - if (!strcmp(med->med_lcd->lcd_uuid, obd->obd_uuid.uuid)) + if (!strcmp(ted->ted_lcd->lcd_uuid, obd->obd_uuid.uuid)) GOTO(free, 0); CDEBUG(D_INFO, "freeing client at idx %u, offset %lld\n", - med->med_lr_idx, med->med_lr_off); + ted->ted_lr_idx, ted->ted_lr_off); - off = med->med_lr_off; + off = ted->ted_lr_off; /* - * Don't clear med_lr_idx here as it is likely also unset. At worst we + * Don't clear ted_lr_idx here as it is likely also unset. At worst we * leak a client slot that will be cleaned on the next recovery. */ if (off <= 0) { CERROR("client idx %d has offset %lld\n", - med->med_lr_idx, off); + ted->ted_lr_idx, off); GOTO(free, rc = -EINVAL); } @@ -728,46 +698,32 @@ int mdt_client_del(const struct lu_env *env, struct mdt_device *mdt) * Clear the bit _after_ zeroing out the client so we don't race with * mdt_client_add and zero out new clients. */ - if (!test_bit(med->med_lr_idx, mdt->mdt_client_bitmap)) { + if (!cfs_test_bit(ted->ted_lr_idx, mdt->mdt_lut.lut_client_bitmap)) { CERROR("MDT client %u: bit already clear in bitmap!!\n", - med->med_lr_idx); + ted->ted_lr_idx); LBUG(); } - /* - * This may be called from difficult reply handler path and - * mdt->mdt_last_rcvd may be NULL that time. - */ - if (mdt->mdt_last_rcvd != NULL) { - th = mdt_trans_start(env, mdt, MDT_TXN_LAST_RCVD_WRITE_CREDITS); - if (IS_ERR(th)) - GOTO(free, rc = PTR_ERR(th)); + /* Make sure the server's last_transno is up to date. + * This should be done before zeroing client slot so last_transno will + * be in server data or in client data in case of failure */ + mdt_server_data_update(env, mdt); - mutex_down(&med->med_lcd_lock); - memset(lcd, 0, sizeof *lcd); + mdt_trans_credit_init(env, mdt, MDT_TXN_LAST_RCVD_WRITE_OP); + th = mdt_trans_start(env, mdt); + if (IS_ERR(th)) + GOTO(free, rc = PTR_ERR(th)); - rc = mdt_last_rcvd_write(env, mdt, lcd, &off, th); - mutex_up(&med->med_lcd_lock); - mdt_trans_stop(env, mdt, th); - } + cfs_mutex_down(&ted->ted_lcd_lock); + memset(ted->ted_lcd->lcd_uuid, 0, sizeof ted->ted_lcd->lcd_uuid); + rc = mdt_last_rcvd_write(env, mdt, ted->ted_lcd, &off, th); + cfs_mutex_up(&ted->ted_lcd_lock); + mdt_trans_stop(env, mdt, th); CDEBUG(rc == 0 ? D_INFO : D_ERROR, "Zeroing out client idx %u in " - "%s rc %d\n", med->med_lr_idx, LAST_RCVD, rc); - - spin_lock(&mdt->mdt_client_bitmap_lock); - clear_bit(med->med_lr_idx, mdt->mdt_client_bitmap); - spin_unlock(&mdt->mdt_client_bitmap_lock); - - /* - * Make sure the server's last_transno is up to date. Do this after the - * client is freed so we know all the client's transactions have been - * committed. - */ - mdt_server_data_update(env, mdt); - EXIT; + "%s, rc %d\n", ted->ted_lr_idx, LAST_RCVD, rc); + RETURN(0); free: - OBD_FREE_PTR(lcd); - med->med_lcd = NULL; return 0; } @@ -779,38 +735,80 @@ static int mdt_last_rcvd_update(struct mdt_thread_info *mti, { struct mdt_device *mdt = mti->mti_mdt; struct ptlrpc_request *req = mdt_info_req(mti); - struct mdt_export_data *med; + struct tg_export_data *ted; struct lsd_client_data *lcd; loff_t off; int err; __s32 rc = th->th_result; - __u64 *transno_p; ENTRY; LASSERT(req); LASSERT(req->rq_export); LASSERT(mdt); - med = &req->rq_export->exp_mdt_data; - LASSERT(med); - lcd = med->med_lcd; - /* if the export has already been failed, we have no last_rcvd slot */ - if (req->rq_export->exp_failed) { + ted = &req->rq_export->exp_target_data; + LASSERT(ted); + + cfs_mutex_down(&ted->ted_lcd_lock); + lcd = ted->ted_lcd; + /* if the export has already been disconnected, we have no last_rcvd slot, + * update server data with latest transno then */ + if (lcd == NULL) { + cfs_mutex_up(&ted->ted_lcd_lock); CWARN("commit transaction for disconnected client %s: rc %d\n", req->rq_export->exp_client_uuid.uuid, rc); - if (rc == 0) - rc = -ENOTCONN; - RETURN(rc); + err = mdt_last_rcvd_header_write(mti->mti_env, mdt, th); + RETURN(err); } - off = med->med_lr_off; - mutex_down(&med->med_lcd_lock); + off = ted->ted_lr_off; + LASSERT(ergo(mti->mti_transno == 0, rc != 0)); if (lustre_msg_get_opc(req->rq_reqmsg) == MDS_CLOSE || lustre_msg_get_opc(req->rq_reqmsg) == MDS_DONE_WRITING) { - transno_p = &lcd->lcd_last_close_transno; + if (mti->mti_transno != 0) { + if (lcd->lcd_last_close_transno > mti->mti_transno) { + CERROR("Trying to overwrite bigger transno:" + "on-disk: "LPU64", new: "LPU64" " + "replay: %d. see LU-617.\n", + lcd->lcd_last_close_transno, + mti->mti_transno, req_is_replay(req)); + if (req_is_replay(req)) { + cfs_spin_lock(&req->rq_export->exp_lock); + req->rq_export->exp_vbr_failed = 1; + cfs_spin_unlock(&req->rq_export->exp_lock); + } + cfs_mutex_up(&ted->ted_lcd_lock); + RETURN(req_is_replay(req) ? -EOVERFLOW : 0); + } + lcd->lcd_last_close_transno = mti->mti_transno; + } lcd->lcd_last_close_xid = req->rq_xid; lcd->lcd_last_close_result = rc; } else { - transno_p = &lcd->lcd_last_transno; + /* VBR: save versions in last_rcvd for reconstruct. */ + __u64 *pre_versions = lustre_msg_get_versions(req->rq_repmsg); + if (pre_versions) { + lcd->lcd_pre_versions[0] = pre_versions[0]; + lcd->lcd_pre_versions[1] = pre_versions[1]; + lcd->lcd_pre_versions[2] = pre_versions[2]; + lcd->lcd_pre_versions[3] = pre_versions[3]; + } + if (mti->mti_transno != 0) { + if (lcd->lcd_last_transno > mti->mti_transno) { + CERROR("Trying to overwrite bigger transno:" + "on-disk: "LPU64", new: "LPU64" " + "replay: %d. see LU-617.\n", + lcd->lcd_last_transno, + mti->mti_transno, req_is_replay(req)); + if (req_is_replay(req)) { + cfs_spin_lock(&req->rq_export->exp_lock); + req->rq_export->exp_vbr_failed = 1; + cfs_spin_unlock(&req->rq_export->exp_lock); + } + cfs_mutex_up(&ted->ted_lcd_lock); + RETURN(req_is_replay(req) ? -EOVERFLOW : 0); + } + lcd->lcd_last_transno = mti->mti_transno; + } lcd->lcd_last_xid = req->rq_xid; lcd->lcd_last_result = rc; /*XXX: save intent_disposition in mdt_thread_info? @@ -819,25 +817,13 @@ static int mdt_last_rcvd_update(struct mdt_thread_info *mti, lcd->lcd_last_data = mti->mti_opdata; } - /* - * When we store zero transno in lcd we can lost last transno value - * because lcd contains 0, but lsd is not yet written - * The server data should be updated also if the latest - * transno is rewritten by zero. See the bug 11125 for details. - */ - if (mti->mti_transno == 0 && - *transno_p == mdt->mdt_last_transno) - mdt_server_data_update(mti->mti_env, mdt); - - *transno_p = mti->mti_transno; - if (off <= 0) { - CERROR("client idx %d has offset %lld\n", med->med_lr_idx, off); + CERROR("client idx %d has offset %lld\n", ted->ted_lr_idx, off); err = -EINVAL; } else { err = mdt_last_rcvd_write(mti->mti_env, mdt, lcd, &off, th); } - mutex_up(&med->med_lcd_lock); + cfs_mutex_up(&ted->ted_lcd_lock); RETURN(err); } @@ -847,26 +833,35 @@ extern struct lu_context_key mdt_thread_key; static int mdt_txn_start_cb(const struct lu_env *env, struct txn_param *param, void *cookie) { - param->tp_credits += MDT_TXN_LAST_RCVD_WRITE_CREDITS; + struct mdt_device *mdt = cookie; + + param->tp_credits += mdt_trans_credit_get(env, mdt, + MDT_TXN_LAST_RCVD_WRITE_OP); return 0; } +/* Set new object versions */ +static void mdt_version_set(struct mdt_thread_info *info) +{ + if (info->mti_mos != NULL) { + mo_version_set(info->mti_env, mdt_object_child(info->mti_mos), + info->mti_transno); + info->mti_mos = NULL; + } +} + /* Update last_rcvd records with latests transaction data */ static int mdt_txn_stop_cb(const struct lu_env *env, struct thandle *txn, void *cookie) { struct mdt_device *mdt = cookie; - struct mdt_txn_info *txi; struct mdt_thread_info *mti; struct ptlrpc_request *req; - /* transno in two contexts - for commit_cb and for thread */ - txi = lu_context_key_get(&txn->th_ctx, &mdt_txn_key); mti = lu_context_key_get(&env->le_ctx, &mdt_thread_key); req = mdt_info_req(mti); if (mti->mti_mdt == NULL || req == NULL || mti->mti_no_need_trans) { - txi->txi_transno = 0; mti->mti_no_need_trans = 0; return 0; } @@ -880,73 +875,43 @@ static int mdt_txn_stop_cb(const struct lu_env *env, } mti->mti_has_trans = 1; - spin_lock(&mdt->mdt_transno_lock); + cfs_spin_lock(&mdt->mdt_lut.lut_translock); if (txn->th_result != 0) { if (mti->mti_transno != 0) { - CERROR("Replay transno "LPU64" failed: rc %i\n", + CERROR("Replay transno "LPU64" failed: rc %d\n", mti->mti_transno, txn->th_result); - mti->mti_transno = 0; } } else if (mti->mti_transno == 0) { - mti->mti_transno = ++ mdt->mdt_last_transno; + mti->mti_transno = ++ mdt->mdt_lut.lut_last_transno; } else { /* should be replay */ - if (mti->mti_transno > mdt->mdt_last_transno) - mdt->mdt_last_transno = mti->mti_transno; + if (mti->mti_transno > mdt->mdt_lut.lut_last_transno) + mdt->mdt_lut.lut_last_transno = mti->mti_transno; } - + cfs_spin_unlock(&mdt->mdt_lut.lut_translock); /* sometimes the reply message has not been successfully packed */ LASSERT(req != NULL && req->rq_repmsg != NULL); + /** VBR: set new versions */ + if (txn->th_result == 0) + mdt_version_set(mti); + /* filling reply data */ - CDEBUG(D_INODE, "transno = %llu, last_committed = %llu\n", + CDEBUG(D_INODE, "transno = "LPU64", last_committed = "LPU64"\n", mti->mti_transno, req->rq_export->exp_obd->obd_last_committed); req->rq_transno = mti->mti_transno; lustre_msg_set_transno(req->rq_repmsg, mti->mti_transno); - lustre_msg_set_last_xid(req->rq_repmsg, - lcd_last_xid(req->rq_export->exp_mdt_data.med_lcd)); - /* save transno for the commit callback */ - txi->txi_transno = mti->mti_transno; - spin_unlock(&mdt->mdt_transno_lock); - + /* if can't add callback, do sync write */ + txn->th_sync = !!lut_last_commit_cb_add(txn, &mdt->mdt_lut, + mti->mti_exp, + mti->mti_transno); return mdt_last_rcvd_update(mti, txn); } -/* commit callback, need to update last_committed value */ -static int mdt_txn_commit_cb(const struct lu_env *env, - struct thandle *txn, void *cookie) -{ - struct mdt_device *mdt = cookie; - struct obd_device *obd = md2lu_dev(&mdt->mdt_md_dev)->ld_obd; - struct mdt_txn_info *txi; - int i; - - txi = lu_context_key_get(&txn->th_ctx, &mdt_txn_key); - - /* copy of obd_transno_commit_cb() but with locking */ - spin_lock(&mdt->mdt_transno_lock); - if (txi->txi_transno > obd->obd_last_committed) { - obd->obd_last_committed = txi->txi_transno; - spin_unlock(&mdt->mdt_transno_lock); - ptlrpc_commit_replies(obd); - } else - spin_unlock(&mdt->mdt_transno_lock); - - if (txi->txi_transno) - CDEBUG(D_HA, "%s: transno "LPD64" is committed\n", - obd->obd_name, txi->txi_transno); - - /* iterate through all additional callbacks */ - for (i = 0; i < txi->txi_cb_count; i++) { - txi->txi_cb[i].mdt_cb_func(mdt, txi->txi_transno, - txi->txi_cb[i].mdt_cb_data, 0); - } - return 0; -} - int mdt_fs_setup(const struct lu_env *env, struct mdt_device *mdt, - struct obd_device *obd) + struct obd_device *obd, + struct lustre_sb_info *lsi) { struct lu_fid fid; struct dt_object *o; @@ -959,25 +924,18 @@ int mdt_fs_setup(const struct lu_env *env, struct mdt_device *mdt, /* prepare transactions callbacks */ mdt->mdt_txn_cb.dtc_txn_start = mdt_txn_start_cb; mdt->mdt_txn_cb.dtc_txn_stop = mdt_txn_stop_cb; - mdt->mdt_txn_cb.dtc_txn_commit = mdt_txn_commit_cb; + mdt->mdt_txn_cb.dtc_txn_commit = NULL; mdt->mdt_txn_cb.dtc_cookie = mdt; + mdt->mdt_txn_cb.dtc_tag = LCT_MD_THREAD; CFS_INIT_LIST_HEAD(&mdt->mdt_txn_cb.dtc_linkage); dt_txn_callback_add(mdt->mdt_bottom, &mdt->mdt_txn_cb); - o = dt_store_open(env, mdt->mdt_bottom, LAST_RCVD, &fid); - if (!IS_ERR(o)) { - mdt->mdt_last_rcvd = o; - rc = mdt_server_data_init(env, mdt); - if (rc) - GOTO(put_last_rcvd, rc); - } else { - rc = PTR_ERR(o); - CERROR("cannot open %s: rc = %d\n", LAST_RCVD, rc); + rc = mdt_server_data_init(env, mdt, lsi); + if (rc) RETURN(rc); - } - o = dt_store_open(env, mdt->mdt_bottom, CAPA_KEYS, &fid); + o = dt_store_open(env, mdt->mdt_bottom, "", CAPA_KEYS, &fid); if (!IS_ERR(o)) { mdt->mdt_ck_obj = o; rc = mdt_capa_keys_init(env, mdt); @@ -986,16 +944,15 @@ int mdt_fs_setup(const struct lu_env *env, struct mdt_device *mdt, } else { rc = PTR_ERR(o); CERROR("cannot open %s: rc = %d\n", CAPA_KEYS, rc); - GOTO(put_last_rcvd, rc); + GOTO(disconnect_exports, rc); } RETURN(0); put_ck_object: lu_object_put(env, &o->do_lu); mdt->mdt_ck_obj = NULL; -put_last_rcvd: - lu_object_put(env, &mdt->mdt_last_rcvd->do_lu); - mdt->mdt_last_rcvd = NULL; +disconnect_exports: + class_disconnect_exports(obd); return rc; } @@ -1005,9 +962,6 @@ void mdt_fs_cleanup(const struct lu_env *env, struct mdt_device *mdt) /* Remove transaction callback */ dt_txn_callback_del(mdt->mdt_bottom, &mdt->mdt_txn_cb); - if (mdt->mdt_last_rcvd) - lu_object_put(env, &mdt->mdt_last_rcvd->do_lu); - mdt->mdt_last_rcvd = NULL; if (mdt->mdt_ck_obj) lu_object_put(env, &mdt->mdt_ck_obj->do_lu); mdt->mdt_ck_obj = NULL; @@ -1018,36 +972,35 @@ void mdt_fs_cleanup(const struct lu_env *env, struct mdt_device *mdt) static void mdt_steal_ack_locks(struct ptlrpc_request *req) { struct obd_export *exp = req->rq_export; - struct list_head *tmp; + cfs_list_t *tmp; struct ptlrpc_reply_state *oldrep; struct ptlrpc_service *svc; int i; /* CAVEAT EMPTOR: spinlock order */ - spin_lock(&exp->exp_lock); - list_for_each (tmp, &exp->exp_outstanding_replies) { - oldrep = list_entry(tmp, struct ptlrpc_reply_state,rs_exp_list); + cfs_spin_lock(&exp->exp_lock); + cfs_list_for_each (tmp, &exp->exp_outstanding_replies) { + oldrep = cfs_list_entry(tmp, struct ptlrpc_reply_state, + rs_exp_list); if (oldrep->rs_xid != req->rq_xid) continue; - if (lustre_msg_get_opc(oldrep->rs_msg) != - lustre_msg_get_opc(req->rq_reqmsg)) - CERROR ("Resent req xid "LPX64" has mismatched opc: " + if (oldrep->rs_opc != lustre_msg_get_opc(req->rq_reqmsg)) + CERROR ("Resent req xid "LPU64" has mismatched opc: " "new %d old %d\n", req->rq_xid, lustre_msg_get_opc(req->rq_reqmsg), - lustre_msg_get_opc(oldrep->rs_msg)); + oldrep->rs_opc); svc = oldrep->rs_service; - spin_lock (&svc->srv_lock); + cfs_spin_lock (&svc->srv_rs_lock); - list_del_init (&oldrep->rs_exp_list); + cfs_list_del_init (&oldrep->rs_exp_list); CWARN("Stealing %d locks from rs %p x"LPD64".t"LPD64 " o%d NID %s\n", oldrep->rs_nlocks, oldrep, - oldrep->rs_xid, oldrep->rs_transno, - lustre_msg_get_opc(oldrep->rs_msg), + oldrep->rs_xid, oldrep->rs_transno, oldrep->rs_opc, libcfs_nid2str(exp->exp_connection->c_peer.nid)); for (i = 0; i < oldrep->rs_nlocks; i++) @@ -1056,12 +1009,28 @@ static void mdt_steal_ack_locks(struct ptlrpc_request *req) oldrep->rs_nlocks = 0; DEBUG_REQ(D_HA, req, "stole locks for"); + cfs_spin_lock(&oldrep->rs_lock); ptlrpc_schedule_difficult_reply (oldrep); + cfs_spin_unlock(&oldrep->rs_lock); - spin_unlock (&svc->srv_lock); + cfs_spin_unlock (&svc->srv_rs_lock); break; } - spin_unlock(&exp->exp_lock); + cfs_spin_unlock(&exp->exp_lock); +} + +/** + * VBR: restore versions + */ +void mdt_vbr_reconstruct(struct ptlrpc_request *req, + struct lsd_client_data *lcd) +{ + __u64 pre_versions[4] = {0}; + pre_versions[0] = lcd->lcd_pre_versions[0]; + pre_versions[1] = lcd->lcd_pre_versions[1]; + pre_versions[2] = lcd->lcd_pre_versions[2]; + pre_versions[3] = lcd->lcd_pre_versions[3]; + lustre_msg_set_versions(req->rq_repmsg, pre_versions); } void mdt_req_from_lcd(struct ptlrpc_request *req, @@ -1074,14 +1043,18 @@ void mdt_req_from_lcd(struct ptlrpc_request *req, lustre_msg_get_opc(req->rq_repmsg) == MDS_DONE_WRITING) { req->rq_transno = lcd->lcd_last_close_transno; req->rq_status = lcd->lcd_last_close_result; - lustre_msg_set_transno(req->rq_repmsg, req->rq_transno); - lustre_msg_set_status(req->rq_repmsg, req->rq_status); } else { req->rq_transno = lcd->lcd_last_transno; req->rq_status = lcd->lcd_last_result; - lustre_msg_set_transno(req->rq_repmsg, req->rq_transno); - lustre_msg_set_status(req->rq_repmsg, req->rq_status); + mdt_vbr_reconstruct(req, lcd); } + if (req->rq_status != 0) + req->rq_transno = 0; + lustre_msg_set_transno(req->rq_repmsg, req->rq_transno); + lustre_msg_set_status(req->rq_repmsg, req->rq_status); + DEBUG_REQ(D_RPCTRACE, req, "restoring transno "LPD64"/status %d", + req->rq_transno, req->rq_status); + mdt_steal_ack_locks(req); } @@ -1089,9 +1062,9 @@ void mdt_reconstruct_generic(struct mdt_thread_info *mti, struct mdt_lock_handle *lhc) { struct ptlrpc_request *req = mdt_info_req(mti); - struct mdt_export_data *med = &req->rq_export->exp_mdt_data; + struct tg_export_data *ted = &req->rq_export->exp_target_data; - return mdt_req_from_lcd(req, med->med_lcd); + return mdt_req_from_lcd(req, ted->ted_lcd); } static void mdt_reconstruct_create(struct mdt_thread_info *mti, @@ -1099,13 +1072,13 @@ static void mdt_reconstruct_create(struct mdt_thread_info *mti, { struct ptlrpc_request *req = mdt_info_req(mti); struct obd_export *exp = req->rq_export; - struct mdt_export_data *med = &exp->exp_mdt_data; + struct tg_export_data *ted = &exp->exp_target_data; struct mdt_device *mdt = mti->mti_mdt; struct mdt_object *child; struct mdt_body *body; int rc; - mdt_req_from_lcd(req, med->med_lcd); + mdt_req_from_lcd(req, ted->ted_lcd); if (req->rq_status) return; @@ -1147,7 +1120,7 @@ static void mdt_reconstruct_setattr(struct mdt_thread_info *mti, struct mdt_object *obj; struct mdt_body *body; - mdt_req_from_lcd(req, med->med_lcd); + mdt_req_from_lcd(req, med->med_ted.ted_lcd); if (req->rq_status) return; @@ -1169,19 +1142,19 @@ static void mdt_reconstruct_setattr(struct mdt_thread_info *mti, mo_attr_get(mti->mti_env, mdt_object_child(obj), &mti->mti_attr); mdt_pack_attr2body(mti, body, &mti->mti_attr.ma_attr, mdt_object_fid(obj)); - if (mti->mti_epoch && (mti->mti_epoch->flags & MF_EPOCH_OPEN)) { + if (mti->mti_ioepoch && (mti->mti_ioepoch->flags & MF_EPOCH_OPEN)) { struct mdt_file_data *mfd; struct mdt_body *repbody; repbody = req_capsule_server_get(mti->mti_pill, &RMF_MDT_BODY); repbody->ioepoch = obj->mot_ioepoch; - spin_lock(&med->med_open_lock); - list_for_each_entry(mfd, &med->med_open_head, mfd_list) { + cfs_spin_lock(&med->med_open_lock); + cfs_list_for_each_entry(mfd, &med->med_open_head, mfd_list) { if (mfd->mfd_xid == req->rq_xid) break; } LASSERT(&mfd->mfd_list != &med->med_open_head); - spin_unlock(&med->med_open_lock); + cfs_spin_unlock(&med->med_open_lock); repbody->handle.cookie = mfd->mfd_handle.h_cookie; }