X-Git-Url: https://git.whamcloud.com/?p=fs%2Flustre-release.git;a=blobdiff_plain;f=lustre%2Fmgs%2Fmgs_llog.c;h=99a58c59151547c3b0d7262a506956dfe1b64cbb;hp=d0ff8f3d7392aeb3643ab6ac94d49ec6b3121c49;hb=e78df92a33634b78df30733deb8123598ef55d35;hpb=aa37f159e721723c3a719592c4d75df4c774b46c diff --git a/lustre/mgs/mgs_llog.c b/lustre/mgs/mgs_llog.c index d0ff8f3..99a58c5 100644 --- a/lustre/mgs/mgs_llog.c +++ b/lustre/mgs/mgs_llog.c @@ -27,7 +27,7 @@ * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved. * Use is subject to license terms. * - * Copyright (c) 2011, Whamcloud, Inc. + * Copyright (c) 2011, 2012, Intel Corporation. */ /* * This file is part of Lustre, http://www.lustre.org/ @@ -38,6 +38,8 @@ * Lustre Management Server (mgs) config llog creation * * Author: Nathan Rutman + * Author: Alex Zhuravlev + * Author: Mikhail Pershin */ #define DEBUG_SUBSYSTEM S_MGS @@ -47,48 +49,84 @@ #include #include #include -#include -#include +#include #include "mgs_internal.h" /********************** Class functions ********************/ -/* Caller must list_del and OBD_FREE each dentry from the list */ int class_dentry_readdir(const struct lu_env *env, - struct mgs_device *mgs, cfs_list_t *dentry_list) -{ - /* see mds_cleanup_pending */ - struct lvfs_run_ctxt saved; - struct file *file; - struct dentry *dentry; - struct vfsmount *mnt; - int rc = 0; - ENTRY; + struct mgs_device *mgs, cfs_list_t *list) +{ + struct dt_object *dir = mgs->mgs_configs_dir; + const struct dt_it_ops *iops; + struct dt_it *it; + struct mgs_direntry *de; + char *key; + int rc, key_sz; + + CFS_INIT_LIST_HEAD(list); + + if (!dt_try_as_dir(env, dir)) + GOTO(out, rc = -ENOTDIR); + + LASSERT(dir); + LASSERT(dir->do_index_ops); + + iops = &dir->do_index_ops->dio_it; + it = iops->init(env, dir, LUDA_64BITHASH, BYPASS_CAPA); + if (IS_ERR(it)) + RETURN(PTR_ERR(it)); + + rc = iops->load(env, it, 0); + if (rc <= 0) + GOTO(fini, rc = 0); + + /* main cycle */ + do { + key = (void *)iops->key(env, it); + if (IS_ERR(key)) { + CERROR("%s: key failed when listing %s: rc = %d\n", + mgs->mgs_obd->obd_name, MOUNT_CONFIGS_DIR, + (int) PTR_ERR(key)); + goto next; + } + key_sz = iops->key_size(env, it); + LASSERT(key_sz > 0); + + /* filter out "." and ".." entries */ + if (key[0] == '.') { + if (key_sz == 1) + goto next; + if (key_sz == 2 && key[1] == '.') + goto next; + } - push_ctxt(&saved, &mgs->mgs_obd->obd_lvfs_ctxt, NULL); - dentry = dget(mgs->mgs_configs_dir); - if (IS_ERR(dentry)) - GOTO(out_pop, rc = PTR_ERR(dentry)); - mnt = mntget(mgs->mgs_vfsmnt); - if (IS_ERR(mnt)) { - l_dput(dentry); - GOTO(out_pop, rc = PTR_ERR(mnt)); - } + de = mgs_direntry_alloc(key_sz + 1); + if (de == NULL) { + rc = -ENOMEM; + break; + } - file = ll_dentry_open(dentry, mnt, O_RDONLY, current_cred()); - if (IS_ERR(file)) - /* dentry_open_it() drops the dentry, mnt refs */ - GOTO(out_pop, rc = PTR_ERR(file)); + memcpy(de->name, key, key_sz); + de->name[key_sz] = 0; - CFS_INIT_LIST_HEAD(dentry_list); - rc = l_readdir(file, dentry_list); - filp_close(file, 0); - /* filp_close->fput() drops the dentry, mnt refs */ + cfs_list_add(&de->list, list); -out_pop: - pop_ctxt(&saved, &mgs->mgs_obd->obd_lvfs_ctxt, NULL); - RETURN(rc); +next: + rc = iops->next(env, it); + } while (rc == 0); + rc = 0; + + iops->put(env, it); + +fini: + iops->fini(env, it); +out: + if (rc) + CERROR("%s: key failed when listing %s: rc = %d\n", + mgs->mgs_obd->obd_name, MOUNT_CONFIGS_DIR, rc); + RETURN(rc); } /******************** DB functions *********************/ @@ -160,7 +198,7 @@ static int mgs_fsdb_handler(const struct lu_env *env, struct llog_handle *llh, CDEBUG(D_MGS, "OST index for %s is %u (%s)\n", lustre_cfg_string(lcfg, 1), index, lustre_cfg_string(lcfg, 2)); - cfs_set_bit(index, fsdb->fsdb_ost_index_map); + set_bit(index, fsdb->fsdb_ost_index_map); } /* Figure out mdt indicies */ @@ -176,7 +214,7 @@ static int mgs_fsdb_handler(const struct lu_env *env, struct llog_handle *llh, } rc = 0; CDEBUG(D_MGS, "MDT index is %u\n", index); - cfs_set_bit(index, fsdb->fsdb_mdt_index_map); + set_bit(index, fsdb->fsdb_mdt_index_map); fsdb->fsdb_mdt_count ++; } @@ -192,13 +230,13 @@ static int mgs_fsdb_handler(const struct lu_env *env, struct llog_handle *llh, /* * compat to 1.8, check osc name used by MDT0 to OSTs, bz18548. */ - if (!cfs_test_bit(FSDB_OSCNAME18, &fsdb->fsdb_flags) && + if (!test_bit(FSDB_OSCNAME18, &fsdb->fsdb_flags) && lcfg->lcfg_command == LCFG_ATTACH && strcmp(lustre_cfg_string(lcfg, 1), LUSTRE_OSC_NAME) == 0) { if (OBD_OCD_VERSION_MAJOR(d->ver) == 1 && OBD_OCD_VERSION_MINOR(d->ver) <= 8) { CWARN("MDT using 1.8 OSC name scheme\n"); - cfs_set_bit(FSDB_OSCNAME18, &fsdb->fsdb_flags); + set_bit(FSDB_OSCNAME18, &fsdb->fsdb_flags); } } @@ -220,42 +258,35 @@ static int mgs_get_fsdb_from_llog(const struct lu_env *env, struct mgs_device *mgs, struct fs_db *fsdb) { - char *logname; - struct llog_handle *loghandle; - struct lvfs_run_ctxt saved; - struct llog_ctxt *ctxt; - struct mgs_fsdb_handler_data d = { fsdb, 0 }; - int rc, rc2; - ENTRY; + char *logname; + struct llog_handle *loghandle; + struct llog_ctxt *ctxt; + struct mgs_fsdb_handler_data d = { fsdb, 0 }; + int rc; + + ENTRY; ctxt = llog_get_context(mgs->mgs_obd, LLOG_CONFIG_ORIG_CTXT); LASSERT(ctxt != NULL); rc = name_create(&logname, fsdb->fsdb_name, "-client"); if (rc) GOTO(out_put, rc); - cfs_mutex_lock(&fsdb->fsdb_mutex); - push_ctxt(&saved, &mgs->mgs_obd->obd_lvfs_ctxt, NULL); - rc = llog_open_create(NULL, ctxt, &loghandle, NULL, logname); + rc = llog_open_create(env, ctxt, &loghandle, NULL, logname); if (rc) GOTO(out_pop, rc); - rc = llog_init_handle(NULL, loghandle, LLOG_F_IS_PLAIN, NULL); + rc = llog_init_handle(env, loghandle, LLOG_F_IS_PLAIN, NULL); if (rc) GOTO(out_close, rc); if (llog_get_size(loghandle) <= 1) - cfs_set_bit(FSDB_LOG_EMPTY, &fsdb->fsdb_flags); + set_bit(FSDB_LOG_EMPTY, &fsdb->fsdb_flags); - rc = llog_process_or_fork(env, loghandle, mgs_fsdb_handler, (void *)&d, - NULL, false); + rc = llog_process(env, loghandle, mgs_fsdb_handler, (void *)&d, NULL); CDEBUG(D_INFO, "get_db = %d\n", rc); out_close: - rc2 = llog_close(NULL, loghandle); - if (!rc) - rc = rc2; + llog_close(env, loghandle); out_pop: - pop_ctxt(&saved, &mgs->mgs_obd->obd_lvfs_ctxt, NULL); - cfs_mutex_unlock(&fsdb->fsdb_mutex); name_destroy(&logname); out_put: llog_ctxt_put(ctxt); @@ -314,12 +345,12 @@ static struct fs_db *mgs_new_fsdb(const struct lu_env *env, RETURN(NULL); strcpy(fsdb->fsdb_name, fsname); - cfs_mutex_init(&fsdb->fsdb_mutex); - cfs_set_bit(FSDB_UDESC, &fsdb->fsdb_flags); + mutex_init(&fsdb->fsdb_mutex); + set_bit(FSDB_UDESC, &fsdb->fsdb_flags); fsdb->fsdb_gen = 1; if (strcmp(fsname, MGSSELF_NAME) == 0) { - cfs_set_bit(FSDB_MGS_SELF, &fsdb->fsdb_flags); + set_bit(FSDB_MGS_SELF, &fsdb->fsdb_flags); } else { OBD_ALLOC(fsdb->fsdb_ost_index_map, INDEX_MAP_SIZE); OBD_ALLOC(fsdb->fsdb_mdt_index_map, INDEX_MAP_SIZE); @@ -358,7 +389,7 @@ err: static void mgs_free_fsdb(struct mgs_device *mgs, struct fs_db *fsdb) { /* wait for anyone with the sem */ - cfs_mutex_lock(&fsdb->fsdb_mutex); + mutex_lock(&fsdb->fsdb_mutex); lproc_mgs_del_live(mgs, fsdb); cfs_list_del(&fsdb->fsdb_list); @@ -372,7 +403,7 @@ static void mgs_free_fsdb(struct mgs_device *mgs, struct fs_db *fsdb) name_destroy(&fsdb->fsdb_clilov); name_destroy(&fsdb->fsdb_clilmv); mgs_free_fsdb_srpc(fsdb); - cfs_mutex_unlock(&fsdb->fsdb_mutex); + mutex_unlock(&fsdb->fsdb_mutex); OBD_FREE_PTR(fsdb); } @@ -386,12 +417,12 @@ int mgs_cleanup_fsdb_list(struct mgs_device *mgs) { struct fs_db *fsdb; cfs_list_t *tmp, *tmp2; - cfs_mutex_lock(&mgs->mgs_mutex); + mutex_lock(&mgs->mgs_mutex); cfs_list_for_each_safe(tmp, tmp2, &mgs->mgs_fs_db_list) { fsdb = cfs_list_entry(tmp, struct fs_db, fsdb_list); mgs_free_fsdb(mgs, fsdb); } - cfs_mutex_unlock(&mgs->mgs_mutex); + mutex_unlock(&mgs->mgs_mutex); return 0; } @@ -402,27 +433,30 @@ int mgs_find_or_make_fsdb(const struct lu_env *env, struct fs_db *fsdb; int rc = 0; - cfs_mutex_lock(&mgs->mgs_mutex); + ENTRY; + mutex_lock(&mgs->mgs_mutex); fsdb = mgs_find_fsdb(mgs, name); if (fsdb) { - cfs_mutex_unlock(&mgs->mgs_mutex); + mutex_unlock(&mgs->mgs_mutex); *dbh = fsdb; - return 0; + RETURN(0); } CDEBUG(D_MGS, "Creating new db\n"); fsdb = mgs_new_fsdb(env, mgs, name); - cfs_mutex_unlock(&mgs->mgs_mutex); + /* lock fsdb_mutex until the db is loaded from llogs */ + if (fsdb) + mutex_lock(&fsdb->fsdb_mutex); + mutex_unlock(&mgs->mgs_mutex); if (!fsdb) - return -ENOMEM; + RETURN(-ENOMEM); - if (!cfs_test_bit(FSDB_MGS_SELF, &fsdb->fsdb_flags)) { + if (!test_bit(FSDB_MGS_SELF, &fsdb->fsdb_flags)) { /* populate the db from the client llog */ rc = mgs_get_fsdb_from_llog(env, mgs, fsdb); if (rc) { CERROR("Can't get db from client log %d\n", rc); - mgs_free_fsdb(mgs, fsdb); - return rc; + GOTO(out_free, rc); } } @@ -430,13 +464,18 @@ int mgs_find_or_make_fsdb(const struct lu_env *env, rc = mgs_get_fsdb_srpc_from_llog(env, mgs, fsdb); if (rc) { CERROR("Can't get db from params log %d\n", rc); - mgs_free_fsdb(mgs, fsdb); - return rc; + GOTO(out_free, rc); } + mutex_unlock(&fsdb->fsdb_mutex); *dbh = fsdb; - return 0; + RETURN(0); + +out_free: + mutex_unlock(&fsdb->fsdb_mutex); + mgs_free_fsdb(mgs, fsdb); + return rc; } /* 1 = index in use @@ -459,7 +498,7 @@ int mgs_check_index(const struct lu_env *env, RETURN(rc); } - if (cfs_test_bit(FSDB_LOG_EMPTY, &fsdb->fsdb_flags)) + if (test_bit(FSDB_LOG_EMPTY, &fsdb->fsdb_flags)) RETURN(-1); if (mti->mti_flags & LDD_F_SV_TYPE_OST) @@ -469,7 +508,7 @@ int mgs_check_index(const struct lu_env *env, else RETURN(-EINVAL); - if (cfs_test_bit(mti->mti_stripe_index, imap)) + if (test_bit(mti->mti_stripe_index, imap)) RETURN(1); RETURN(0); } @@ -478,7 +517,7 @@ static __inline__ int next_index(void *index_map, int map_len) { int i; for (i = 0; i < map_len * 8; i++) - if (!cfs_test_bit(i, index_map)) { + if (!test_bit(i, index_map)) { return i; } CERROR("max index %d exceeded.\n", i); @@ -504,23 +543,19 @@ static int mgs_set_index(const struct lu_env *env, RETURN(rc); } + mutex_lock(&fsdb->fsdb_mutex); if (mti->mti_flags & LDD_F_SV_TYPE_OST) { imap = fsdb->fsdb_ost_index_map; } else if (mti->mti_flags & LDD_F_SV_TYPE_MDT) { imap = fsdb->fsdb_mdt_index_map; - if (fsdb->fsdb_mdt_count >= MAX_MDT_COUNT) { - LCONSOLE_ERROR_MSG(0x13f, "The max mdt count" - "is %d\n", (int)MAX_MDT_COUNT); - RETURN(-ERANGE); - } } else { - RETURN(-EINVAL); + GOTO(out_up, rc = -EINVAL); } if (mti->mti_flags & LDD_F_NEED_INDEX) { rc = next_index(imap, INDEX_MAP_SIZE); if (rc == -1) - RETURN(-ERANGE); + GOTO(out_up, rc = -ERANGE); mti->mti_stripe_index = rc; if (mti->mti_flags & LDD_F_SV_TYPE_MDT) fsdb->fsdb_mdt_count ++; @@ -531,10 +566,10 @@ static int mgs_set_index(const struct lu_env *env, "but the max index is %d.\n", mti->mti_svname, mti->mti_stripe_index, INDEX_MAP_SIZE * 8); - RETURN(-ERANGE); + GOTO(out_up, rc = -ERANGE); } - if (cfs_test_bit(mti->mti_stripe_index, imap)) { + if (test_bit(mti->mti_stripe_index, imap)) { if ((mti->mti_flags & LDD_F_VIRGIN) && !(mti->mti_flags & LDD_F_WRITECONF)) { LCONSOLE_ERROR_MSG(0x140, "Server %s requested index " @@ -542,16 +577,17 @@ static int mgs_set_index(const struct lu_env *env, "use. Use --writeconf to force\n", mti->mti_svname, mti->mti_stripe_index); - RETURN(-EADDRINUSE); + GOTO(out_up, rc = -EADDRINUSE); } else { CDEBUG(D_MGS, "Server %s updating index %d\n", mti->mti_svname, mti->mti_stripe_index); - RETURN(EALREADY); + GOTO(out_up, rc = EALREADY); } } - cfs_set_bit(mti->mti_stripe_index, imap); - cfs_clear_bit(FSDB_LOG_EMPTY, &fsdb->fsdb_flags); + set_bit(mti->mti_stripe_index, imap); + clear_bit(FSDB_LOG_EMPTY, &fsdb->fsdb_flags); + mutex_unlock(&fsdb->fsdb_mutex); server_make_name(mti->mti_flags & ~(LDD_F_VIRGIN | LDD_F_WRITECONF), mti->mti_stripe_index, mti->mti_fsname, mti->mti_svname); @@ -559,6 +595,9 @@ static int mgs_set_index(const struct lu_env *env, mti->mti_stripe_index); RETURN(0); +out_up: + mutex_unlock(&fsdb->fsdb_mutex); + return rc; } struct mgs_modify_lookup { @@ -572,9 +611,8 @@ static int mgs_modify_handler(const struct lu_env *env, { struct mgs_modify_lookup *mml = data; struct cfg_marker *marker; - struct lustre_cfg *lcfg = (struct lustre_cfg *)(rec + 1); - int cfg_len = rec->lrh_len - sizeof(struct llog_rec_hdr) - - sizeof(struct llog_rec_tail); + struct lustre_cfg *lcfg = REC_DATA(rec); + int cfg_len = REC_DATA_LEN(rec); int rc; ENTRY; @@ -609,8 +647,8 @@ static int mgs_modify_handler(const struct lu_env *env, /* Header and tail are added back to lrh_len in llog_lvfs_write_rec */ rec->lrh_len = cfg_len; - rc = llog_write_rec(NULL, llh, rec, NULL, 0, (void *)lcfg, - rec->lrh_index); + rc = llog_write(env, llh, rec, NULL, 0, (void *)lcfg, + rec->lrh_index); if (!rc) mml->mml_modified++; } @@ -618,33 +656,38 @@ static int mgs_modify_handler(const struct lu_env *env, RETURN(rc); } -/* Modify an existing config log record (for CM_SKIP or CM_EXCLUDE) */ +/** + * Modify an existing config log record (for CM_SKIP or CM_EXCLUDE) + * Return code: + * 0 - modified successfully, + * 1 - no modification was done + * negative - error + */ static int mgs_modify(const struct lu_env *env, struct mgs_device *mgs, struct fs_db *fsdb, struct mgs_target_info *mti, char *logname, char *devname, char *comment, int flags) { struct llog_handle *loghandle; - struct lvfs_run_ctxt saved; struct llog_ctxt *ctxt; struct mgs_modify_lookup *mml; - int rc, rc2; + int rc; + ENTRY; - CDEBUG(D_MGS, "modify %s/%s/%s fl=%x\n", logname, devname, comment, - flags); + LASSERT(mutex_is_locked(&fsdb->fsdb_mutex)); + CDEBUG(D_MGS, "modify %s/%s/%s fl=%x\n", logname, devname, comment, + flags); ctxt = llog_get_context(mgs->mgs_obd, LLOG_CONFIG_ORIG_CTXT); LASSERT(ctxt != NULL); - push_ctxt(&saved, &mgs->mgs_obd->obd_lvfs_ctxt, NULL); - rc = llog_open(NULL, ctxt, &loghandle, NULL, logname, - LLOG_OPEN_EXISTS); + rc = llog_open(env, ctxt, &loghandle, NULL, logname, LLOG_OPEN_EXISTS); if (rc < 0) { if (rc == -ENOENT) rc = 0; GOTO(out_pop, rc); } - rc = llog_init_handle(NULL, loghandle, LLOG_F_IS_PLAIN, NULL); + rc = llog_init_handle(env, loghandle, LLOG_F_IS_PLAIN, NULL); if (rc) GOTO(out_close, rc); @@ -660,34 +703,88 @@ static int mgs_modify(const struct lu_env *env, struct mgs_device *mgs, mml->mml_marker.cm_flags = flags; mml->mml_marker.cm_canceltime = flags ? cfs_time_current_sec() : 0; mml->mml_modified = 0; - rc = llog_process_or_fork(env, loghandle, mgs_modify_handler, - (void *)mml, NULL, false); - if (!rc && !mml->mml_modified) + rc = llog_process(env, loghandle, mgs_modify_handler, (void *)mml, + NULL); + if (!rc && !mml->mml_modified) rc = 1; OBD_FREE_PTR(mml); out_close: - rc2 = llog_close(NULL, loghandle); - if (!rc) - rc = rc2; + llog_close(env, loghandle); out_pop: - pop_ctxt(&saved, &mgs->mgs_obd->obd_lvfs_ctxt, NULL); if (rc < 0) - CERROR("modify %s/%s failed %d\n", - mti->mti_svname, comment, rc); + CERROR("%s: modify %s/%s failed: rc = %d\n", + mgs->mgs_obd->obd_name, mti->mti_svname, comment, rc); llog_ctxt_put(ctxt); RETURN(rc); } -/******************** config log recording functions *********************/ +/** This structure is passed to mgs_replace_handler */ +struct mgs_replace_uuid_lookup { + /* Nids are replaced for this target device */ + struct mgs_target_info target; + /* Temporary modified llog */ + struct llog_handle *temp_llh; + /* Flag is set if in target block*/ + int in_target_device; + /* Nids already added. Just skip (multiple nids) */ + int device_nids_added; + /* Flag is set if this block should not be copied */ + int skip_it; +}; + +/** + * Check: a) if block should be skipped + * b) is it target block + * + * \param[in] lcfg + * \param[in] mrul + * + * \retval 0 should not to be skipped + * \retval 1 should to be skipped + */ +static int check_markers(struct lustre_cfg *lcfg, + struct mgs_replace_uuid_lookup *mrul) +{ + struct cfg_marker *marker; + + /* Track markers. Find given device */ + if (lcfg->lcfg_command == LCFG_MARKER) { + marker = lustre_cfg_buf(lcfg, 1); + /* Clean llog from records marked as CM_EXCLUDE. + CM_SKIP records are used for "active" command + and can be restored if needed */ + if ((marker->cm_flags & (CM_EXCLUDE | CM_START)) == + (CM_EXCLUDE | CM_START)) { + mrul->skip_it = 1; + return 1; + } + + if ((marker->cm_flags & (CM_EXCLUDE | CM_END)) == + (CM_EXCLUDE | CM_END)) { + mrul->skip_it = 0; + return 1; + } + + if (strcmp(mrul->target.mti_svname, marker->cm_tgtname) == 0) { + LASSERT(!(marker->cm_flags & CM_START) || + !(marker->cm_flags & CM_END)); + if (marker->cm_flags & CM_START) { + mrul->in_target_device = 1; + mrul->device_nids_added = 0; + } else if (marker->cm_flags & CM_END) + mrul->in_target_device = 0; + } + } + + return 0; +} static int record_lcfg(const struct lu_env *env, struct llog_handle *llh, - struct lustre_cfg *lcfg) + struct lustre_cfg *lcfg) { - struct lvfs_run_ctxt saved; - struct llog_rec_hdr rec; - int buflen, rc; - struct obd_device *obd = llh->lgh_ctxt->loc_obd; + struct llog_rec_hdr rec; + int buflen, rc; if (!lcfg || !llh) return -ENOMEM; @@ -699,10 +796,8 @@ static int record_lcfg(const struct lu_env *env, struct llog_handle *llh, rec.lrh_len = llog_data_len(buflen); rec.lrh_type = OBD_CFG_REC; - push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL); /* idx = -1 means append */ - rc = llog_write_rec(NULL, llh, &rec, NULL, 0, (void *)lcfg, -1); - pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL); + rc = llog_write(env, llh, &rec, NULL, 0, (void *)lcfg, -1); if (rc) CERROR("failed %d\n", rc); return rc; @@ -745,13 +840,11 @@ static int record_base(const struct lu_env *env, struct llog_handle *llh, return rc; } - static inline int record_add_uuid(const struct lu_env *env, struct llog_handle *llh, uint64_t nid, char *uuid) { return record_base(env, llh, NULL, nid, LCFG_ADD_UUID, uuid, 0, 0, 0); - } static inline int record_add_conn(const struct lu_env *env, @@ -775,6 +868,465 @@ static inline int record_setup(const struct lu_env *env, return record_base(env, llh, devname, 0, LCFG_SETUP, s1, s2, s3, s4); } +/** + * \retval <0 record processing error + * \retval n record is processed. No need copy original one. + * \retval 0 record is not processed. + */ +static int process_command(const struct lu_env *env, struct lustre_cfg *lcfg, + struct mgs_replace_uuid_lookup *mrul) +{ + int nids_added = 0; + lnet_nid_t nid; + char *ptr; + int rc; + + if (lcfg->lcfg_command == LCFG_ADD_UUID) { + /* LCFG_ADD_UUID command found. Let's skip original command + and add passed nids */ + ptr = mrul->target.mti_params; + while (class_parse_nid(ptr, &nid, &ptr) == 0) { + CDEBUG(D_MGS, "add nid %s with uuid %s, " + "device %s\n", libcfs_nid2str(nid), + mrul->target.mti_params, + mrul->target.mti_svname); + rc = record_add_uuid(env, + mrul->temp_llh, nid, + mrul->target.mti_params); + if (!rc) + nids_added++; + } + + if (nids_added == 0) { + CERROR("No new nids were added, nid %s with uuid %s, " + "device %s\n", libcfs_nid2str(nid), + mrul->target.mti_params, + mrul->target.mti_svname); + RETURN(-ENXIO); + } else { + mrul->device_nids_added = 1; + } + + return nids_added; + } + + if (mrul->device_nids_added && lcfg->lcfg_command == LCFG_SETUP) { + /* LCFG_SETUP command found. UUID should be changed */ + rc = record_setup(env, + mrul->temp_llh, + /* devname the same */ + lustre_cfg_string(lcfg, 0), + /* s1 is not changed */ + lustre_cfg_string(lcfg, 1), + /* new uuid should be + the full nidlist */ + mrul->target.mti_params, + /* s3 is not changed */ + lustre_cfg_string(lcfg, 3), + /* s4 is not changed */ + lustre_cfg_string(lcfg, 4)); + return rc ? rc : 1; + } + + /* Another commands in target device block */ + return 0; +} + +/** + * Handler that called for every record in llog. + * Records are processed in order they placed in llog. + * + * \param[in] llh log to be processed + * \param[in] rec current record + * \param[in] data mgs_replace_uuid_lookup structure + * + * \retval 0 success + */ +static int mgs_replace_handler(const struct lu_env *env, + struct llog_handle *llh, + struct llog_rec_hdr *rec, + void *data) +{ + struct llog_rec_hdr local_rec = *rec; + struct mgs_replace_uuid_lookup *mrul; + struct lustre_cfg *lcfg = REC_DATA(rec); + int cfg_len = REC_DATA_LEN(rec); + int rc; + ENTRY; + + mrul = (struct mgs_replace_uuid_lookup *)data; + + if (rec->lrh_type != OBD_CFG_REC) { + CERROR("unhandled lrh_type: %#x, cmd %x %s %s\n", + rec->lrh_type, lcfg->lcfg_command, + lustre_cfg_string(lcfg, 0), + lustre_cfg_string(lcfg, 1)); + RETURN(-EINVAL); + } + + rc = lustre_cfg_sanity_check(lcfg, cfg_len); + if (rc) { + /* Do not copy any invalidated records */ + GOTO(skip_out, rc = 0); + } + + rc = check_markers(lcfg, mrul); + if (rc || mrul->skip_it) + GOTO(skip_out, rc = 0); + + /* Write to new log all commands outside target device block */ + if (!mrul->in_target_device) + GOTO(copy_out, rc = 0); + + /* Skip all other LCFG_ADD_UUID and LCFG_ADD_CONN records + (failover nids) for this target, assuming that if then + primary is changing then so is the failover */ + if (mrul->device_nids_added && + (lcfg->lcfg_command == LCFG_ADD_UUID || + lcfg->lcfg_command == LCFG_ADD_CONN)) + GOTO(skip_out, rc = 0); + + rc = process_command(env, lcfg, mrul); + if (rc < 0) + RETURN(rc); + + if (rc) + RETURN(0); +copy_out: + /* Record is placed in temporary llog as is */ + local_rec.lrh_len -= sizeof(*rec) + sizeof(struct llog_rec_tail); + rc = llog_write(env, mrul->temp_llh, &local_rec, NULL, 0, + (void *)lcfg, -1); + + CDEBUG(D_MGS, "Copied idx=%d, rc=%d, len=%d, cmd %x %s %s\n", + rec->lrh_index, rc, rec->lrh_len, lcfg->lcfg_command, + lustre_cfg_string(lcfg, 0), lustre_cfg_string(lcfg, 1)); + RETURN(rc); + +skip_out: + CDEBUG(D_MGS, "Skipped idx=%d, rc=%d, len=%d, cmd %x %s %s\n", + rec->lrh_index, rc, rec->lrh_len, lcfg->lcfg_command, + lustre_cfg_string(lcfg, 0), lustre_cfg_string(lcfg, 1)); + RETURN(rc); +} + +static int mgs_backup_llog(const struct lu_env *env, + struct obd_device *mgs, + char *fsname, char *backup) +{ + struct obd_uuid *uuid; + struct llog_handle *orig_llh, *bak_llh; + struct llog_ctxt *lctxt; + int rc, rc2; + ENTRY; + + lctxt = llog_get_context(mgs, LLOG_CONFIG_ORIG_CTXT); + if (!lctxt) { + CERROR("%s: missing llog context\n", mgs->obd_name); + GOTO(out, rc = -EINVAL); + } + + /* Make sure there's no old backup log */ + rc = llog_erase(env, lctxt, NULL, backup); + if (rc < 0 && rc != -ENOENT) + GOTO(out_put, rc); + + /* open backup log */ + rc = llog_open_create(env, lctxt, &bak_llh, NULL, backup); + if (rc) { + CERROR("%s: backup logfile open %s: rc = %d\n", + mgs->obd_name, backup, rc); + GOTO(out_put, rc); + } + + /* set the log header uuid */ + OBD_ALLOC_PTR(uuid); + if (uuid == NULL) + GOTO(out_put, rc = -ENOMEM); + obd_str2uuid(uuid, backup); + rc = llog_init_handle(env, bak_llh, LLOG_F_IS_PLAIN, uuid); + OBD_FREE_PTR(uuid); + if (rc) + GOTO(out_close1, rc); + + /* open original log */ + rc = llog_open(env, lctxt, &orig_llh, NULL, fsname, + LLOG_OPEN_EXISTS); + if (rc < 0) { + if (rc == -ENOENT) + rc = 0; + GOTO(out_close1, rc); + } + + rc = llog_init_handle(env, orig_llh, LLOG_F_IS_PLAIN, NULL); + if (rc) + GOTO(out_close2, rc); + + /* Copy remote log */ + rc = llog_process(env, orig_llh, llog_copy_handler, + (void *)bak_llh, NULL); + +out_close2: + rc2 = llog_close(env, orig_llh); + if (!rc) + rc = rc2; +out_close1: + rc2 = llog_close(env, bak_llh); + if (!rc) + rc = rc2; +out_put: + if (lctxt) + llog_ctxt_put(lctxt); +out: + if (rc) + CERROR("%s: Failed to backup log %s: rc = %d\n", + mgs->obd_name, fsname, rc); + RETURN(rc); +} + +static int mgs_log_is_empty(const struct lu_env *env, struct mgs_device *mgs, + char *name); + +static int mgs_replace_nids_log(const struct lu_env *env, + struct obd_device *mgs, struct fs_db *fsdb, + char *logname, char *devname, char *nids) +{ + struct llog_handle *orig_llh, *backup_llh; + struct llog_ctxt *ctxt; + struct mgs_replace_uuid_lookup *mrul; + struct mgs_device *mgs_dev = lu2mgs_dev(mgs->obd_lu_dev); + char *backup; + int rc, rc2; + ENTRY; + + CDEBUG(D_MGS, "Replace nids for %s in %s\n", devname, logname); + + ctxt = llog_get_context(mgs, LLOG_CONFIG_ORIG_CTXT); + LASSERT(ctxt != NULL); + + if (mgs_log_is_empty(env, mgs_dev, logname)) { + /* Log is empty. Nothing to replace */ + GOTO(out_put, rc = 0); + } + + OBD_ALLOC(backup, strlen(logname) + 5); + if (backup == NULL) + GOTO(out_put, rc = -ENOMEM); + + sprintf(backup, "%s.bak", logname); + + rc = mgs_backup_llog(env, mgs, logname, backup); + if (rc < 0) { + CERROR("%s: can't make backup for %s: rc = %d\n", + mgs->obd_name, logname, rc); + GOTO(out_free,rc); + } + + /* Now erase original log file. Connections are not allowed. + Backup is already saved */ + rc = llog_erase(env, ctxt, NULL, logname); + if (rc < 0 && rc != -ENOENT) + GOTO(out_free, rc); + + /* open local log */ + rc = llog_open_create(env, ctxt, &orig_llh, NULL, logname); + if (rc) + GOTO(out_restore, rc); + + rc = llog_init_handle(env, orig_llh, LLOG_F_IS_PLAIN, NULL); + if (rc) + GOTO(out_closel, rc); + + /* open backup llog */ + rc = llog_open(env, ctxt, &backup_llh, NULL, backup, + LLOG_OPEN_EXISTS); + if (rc) + GOTO(out_closel, rc); + + rc = llog_init_handle(env, backup_llh, LLOG_F_IS_PLAIN, NULL); + if (rc) + GOTO(out_close, rc); + + if (llog_get_size(backup_llh) <= 1) + GOTO(out_close, rc = 0); + + OBD_ALLOC_PTR(mrul); + if (!mrul) + GOTO(out_close, rc = -ENOMEM); + /* devname is only needed information to replace UUID records */ + strncpy(mrul->target.mti_svname, devname, MTI_NAME_MAXLEN); + /* parse nids later */ + strncpy(mrul->target.mti_params, nids, MTI_PARAM_MAXLEN); + /* Copy records to this temporary llog */ + mrul->temp_llh = orig_llh; + + rc = llog_process(env, backup_llh, mgs_replace_handler, + (void *)mrul, NULL); + OBD_FREE_PTR(mrul); +out_close: + rc2 = llog_close(NULL, backup_llh); + if (!rc) + rc = rc2; +out_closel: + rc2 = llog_close(NULL, orig_llh); + if (!rc) + rc = rc2; + +out_restore: + if (rc) { + CERROR("%s: llog should be restored: rc = %d\n", + mgs->obd_name, rc); + rc2 = mgs_backup_llog(env, mgs, backup, logname); + if (rc2 < 0) + CERROR("%s: can't restore backup %s: rc = %d\n", + mgs->obd_name, logname, rc2); + } + +out_free: + OBD_FREE(backup, strlen(backup) + 5); + +out_put: + llog_ctxt_put(ctxt); + + if (rc) + CERROR("%s: failed to replace nids in log %s: rc = %d\n", + mgs->obd_name, logname, rc); + + RETURN(rc); +} + +/** + * Parse device name and get file system name and/or device index + * + * \param[in] devname device name (ex. lustre-MDT0000) + * \param[out] fsname file system name(optional) + * \param[out] index device index(optional) + * + * \retval 0 success + */ +static int mgs_parse_devname(char *devname, char *fsname, __u32 *index) +{ + char *ptr; + ENTRY; + + /* Extract fsname */ + ptr = strrchr(devname, '-'); + + if (fsname) { + if (!ptr) { + CDEBUG(D_MGS, "Device name %s without fsname\n", + devname); + RETURN(-EINVAL); + } + memset(fsname, 0, MTI_NAME_MAXLEN); + strncpy(fsname, devname, ptr - devname); + fsname[MTI_NAME_MAXLEN - 1] = 0; + } + + if (index) { + if (server_name2index(ptr, index, NULL) < 0) { + CDEBUG(D_MGS, "Device name with wrong index\n"); + RETURN(-EINVAL); + } + } + + RETURN(0); +} + +static int only_mgs_is_running(struct obd_device *mgs_obd) +{ + /* TDB: Is global variable with devices count exists? */ + int num_devices = get_devices_count(); + /* osd, MGS and MGC + self_export + (wc -l /proc/fs/lustre/devices <= 2) && (num_exports <= 2) */ + return (num_devices <= 3) && (mgs_obd->obd_num_exports <= 2); +} + +static int name_create_mdt(char **logname, char *fsname, int i) +{ + char mdt_index[9]; + + sprintf(mdt_index, "-MDT%04x", i); + return name_create(logname, fsname, mdt_index); +} + +/** + * Replace nids for \a device to \a nids values + * + * \param obd MGS obd device + * \param devname nids need to be replaced for this device + * (ex. lustre-OST0000) + * \param nids nids list (ex. nid1,nid2,nid3) + * + * \retval 0 success + */ +int mgs_replace_nids(const struct lu_env *env, + struct mgs_device *mgs, + char *devname, char *nids) +{ + /* Assume fsname is part of device name */ + char fsname[MTI_NAME_MAXLEN]; + int rc; + __u32 index; + char *logname; + struct fs_db *fsdb; + unsigned int i; + int conn_state; + struct obd_device *mgs_obd = mgs->mgs_obd; + ENTRY; + + /* We can only change NIDs if no other nodes are connected */ + spin_lock(&mgs_obd->obd_dev_lock); + conn_state = mgs_obd->obd_no_conn; + mgs_obd->obd_no_conn = 1; + spin_unlock(&mgs_obd->obd_dev_lock); + + /* We can not change nids if not only MGS is started */ + if (!only_mgs_is_running(mgs_obd)) { + CERROR("Only MGS is allowed to be started\n"); + GOTO(out, rc = -EINPROGRESS); + } + + /* Get fsname and index*/ + rc = mgs_parse_devname(devname, fsname, &index); + if (rc) + GOTO(out, rc); + + rc = mgs_find_or_make_fsdb(env, mgs, fsname, &fsdb); + if (rc) { + CERROR("%s: can't find fsdb: rc = %d\n", fsname, rc); + GOTO(out, rc); + } + + /* Process client llogs */ + name_create(&logname, fsname, "-client"); + rc = mgs_replace_nids_log(env, mgs_obd, fsdb, logname, devname, nids); + name_destroy(&logname); + if (rc) { + CERROR("%s: error while replacing NIDs for %s: rc = %d\n", + fsname, devname, rc); + GOTO(out, rc); + } + + /* Process MDT llogs */ + for (i = 0; i < INDEX_MAP_SIZE * 8; i++) { + if (!test_bit(i, fsdb->fsdb_mdt_index_map)) + continue; + name_create_mdt(&logname, fsname, i); + rc = mgs_replace_nids_log(env, mgs_obd, fsdb, logname, devname, nids); + name_destroy(&logname); + if (rc) + GOTO(out, rc); + } + +out: + spin_lock(&mgs_obd->obd_dev_lock); + mgs_obd->obd_no_conn = conn_state; + spin_unlock(&mgs_obd->obd_dev_lock); + + RETURN(rc); +} + static int record_lov_setup(const struct lu_env *env, struct llog_handle *llh, char *devname, struct lov_desc *desc) { @@ -826,7 +1378,7 @@ static inline int record_lov_add(const struct lu_env *env, char *index, char *gen) { return record_base(env,llh,lov_name,0,LCFG_LOV_ADD_OBD, - ost_uuid,index,gen,0); + ost_uuid, index, gen, 0); } static inline int record_mount_opt(const struct lu_env *env, @@ -846,16 +1398,21 @@ static int record_marker(const struct lu_env *env, struct mgs_thread_info *mgi = mgs_env_info(env); struct lustre_cfg *lcfg; int rc; + int cplen = 0; if (flags & CM_START) fsdb->fsdb_gen++; mgi->mgi_marker.cm_step = fsdb->fsdb_gen; mgi->mgi_marker.cm_flags = flags; mgi->mgi_marker.cm_vers = LUSTRE_VERSION_CODE; - strncpy(mgi->mgi_marker.cm_tgtname, tgtname, - sizeof(mgi->mgi_marker.cm_tgtname)); - strncpy(mgi->mgi_marker.cm_comment, comment, - sizeof(mgi->mgi_marker.cm_comment)); + cplen = strlcpy(mgi->mgi_marker.cm_tgtname, tgtname, + sizeof(mgi->mgi_marker.cm_tgtname)); + if (cplen >= sizeof(mgi->mgi_marker.cm_tgtname)) + return -E2BIG; + cplen = strlcpy(mgi->mgi_marker.cm_comment, comment, + sizeof(mgi->mgi_marker.cm_comment)); + if (cplen >= sizeof(mgi->mgi_marker.cm_comment)) + return -E2BIG; mgi->mgi_marker.cm_createtime = cfs_time_current_sec(); mgi->mgi_marker.cm_canceltime = 0; lustre_cfg_bufs_reset(&mgi->mgi_bufs, NULL); @@ -870,83 +1427,72 @@ static int record_marker(const struct lu_env *env, return rc; } -static int record_start_log(const struct lu_env *env, - struct mgs_device *mgs, - struct llog_handle **llh, char *name) +static int record_start_log(const struct lu_env *env, struct mgs_device *mgs, + struct llog_handle **llh, char *name) { - static struct obd_uuid cfg_uuid = { .uuid = "config_uuid" }; - struct lvfs_run_ctxt saved; - struct llog_ctxt *ctxt; - int rc = 0; + static struct obd_uuid cfg_uuid = { .uuid = "config_uuid" }; + struct llog_ctxt *ctxt; + int rc = 0; - if (*llh) - GOTO(out, rc = -EBUSY); + if (*llh) + GOTO(out, rc = -EBUSY); ctxt = llog_get_context(mgs->mgs_obd, LLOG_CONFIG_ORIG_CTXT); if (!ctxt) GOTO(out, rc = -ENODEV); LASSERT(ctxt->loc_obd == mgs->mgs_obd); - push_ctxt(&saved, &mgs->mgs_obd->obd_lvfs_ctxt, NULL); - rc = llog_open_create(NULL, ctxt, llh, NULL, name); + rc = llog_open_create(env, ctxt, llh, NULL, name); if (rc) GOTO(out_ctxt, rc); - rc = llog_init_handle(NULL, *llh, LLOG_F_IS_PLAIN, &cfg_uuid); - if (rc) { - llog_close(NULL, *llh); - *llh = NULL; - } + rc = llog_init_handle(env, *llh, LLOG_F_IS_PLAIN, &cfg_uuid); + if (rc) + llog_close(env, *llh); out_ctxt: - pop_ctxt(&saved, &mgs->mgs_obd->obd_lvfs_ctxt, NULL); llog_ctxt_put(ctxt); out: - if (rc) - CERROR("Can't start log %s: %d\n", name, rc); + if (rc) { + CERROR("%s: can't start log %s: rc = %d\n", + mgs->mgs_obd->obd_name, name, rc); + *llh = NULL; + } RETURN(rc); } static int record_end_log(const struct lu_env *env, struct llog_handle **llh) { - struct lvfs_run_ctxt saved; - struct obd_device *obd = (*llh)->lgh_ctxt->loc_obd; - int rc = 0; - - push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL); + int rc; - rc = llog_close(NULL, *llh); - *llh = NULL; + rc = llog_close(env, *llh); + *llh = NULL; - pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL); - RETURN(rc); + return rc; } static int mgs_log_is_empty(const struct lu_env *env, struct mgs_device *mgs, char *name) { - struct lvfs_run_ctxt saved; struct llog_handle *llh; struct llog_ctxt *ctxt; int rc = 0; ctxt = llog_get_context(mgs->mgs_obd, LLOG_CONFIG_ORIG_CTXT); LASSERT(ctxt != NULL); - push_ctxt(&saved, &mgs->mgs_obd->obd_lvfs_ctxt, NULL); - rc = llog_open(NULL, ctxt, &llh, NULL, name, LLOG_OPEN_EXISTS); + rc = llog_open(env, ctxt, &llh, NULL, name, LLOG_OPEN_EXISTS); if (rc < 0) { if (rc == -ENOENT) rc = 0; GOTO(out_ctxt, rc); } - llog_init_handle(NULL, llh, LLOG_F_IS_PLAIN, NULL); + rc = llog_init_handle(env, llh, LLOG_F_IS_PLAIN, NULL); if (rc) GOTO(out_close, rc); rc = llog_get_size(llh); out_close: - llog_close(NULL, llh); + llog_close(env, llh); out_ctxt: - pop_ctxt(&saved, &mgs->mgs_obd->obd_lvfs_ctxt, NULL); llog_ctxt_put(ctxt); /* header is record 1 */ return (rc <= 1); @@ -995,8 +1541,8 @@ int mgs_write_log_direct_all(const struct lu_env *env, char *devname, char *comment, int server_only) { - cfs_list_t dentry_list; - struct l_linux_dirent *dirent, *n; + cfs_list_t list; + struct mgs_direntry *dirent, *n; char *fsname = mti->mti_fsname; char *logname; int rc = 0, len = strlen(fsname); @@ -1019,65 +1565,90 @@ int mgs_write_log_direct_all(const struct lu_env *env, RETURN(rc); /* Find all the logs in the CONFIGS directory */ - rc = class_dentry_readdir(env, mgs, &dentry_list); - if (rc) { - CERROR("Can't read %s dir\n", MOUNT_CONFIGS_DIR); + rc = class_dentry_readdir(env, mgs, &list); + if (rc) RETURN(rc); - } /* Could use fsdb index maps instead of directory listing */ - cfs_list_for_each_entry_safe(dirent, n, &dentry_list, lld_list) { - cfs_list_del(&dirent->lld_list); + cfs_list_for_each_entry_safe(dirent, n, &list, list) { + cfs_list_del(&dirent->list); /* don't write to sptlrpc rule log */ - if (strstr(dirent->lld_name, "-sptlrpc") != NULL) + if (strstr(dirent->name, "-sptlrpc") != NULL) goto next; /* caller wants write server logs only */ - if (server_only && strstr(dirent->lld_name, "-client") != NULL) + if (server_only && strstr(dirent->name, "-client") != NULL) goto next; - if (strncmp(fsname, dirent->lld_name, len) == 0) { - CDEBUG(D_MGS, "Changing log %s\n", dirent->lld_name); + if (strncmp(fsname, dirent->name, len) == 0) { + CDEBUG(D_MGS, "Changing log %s\n", dirent->name); /* Erase any old settings of this same parameter */ - mgs_modify(env, mgs, fsdb, mti, dirent->lld_name, - devname, comment, CM_SKIP); + rc = mgs_modify(env, mgs, fsdb, mti, dirent->name, + devname, comment, CM_SKIP); + if (rc < 0) + CERROR("%s: Can't modify llog %s: rc = %d\n", + mgs->mgs_obd->obd_name, dirent->name,rc); /* Write the new one */ if (lcfg) { rc = mgs_write_log_direct(env, mgs, fsdb, - dirent->lld_name, - lcfg, devname, - comment); + dirent->name, + lcfg, devname, + comment); if (rc) - CERROR("err %d writing log %s\n", rc, - dirent->lld_name); + CERROR("%s: writing log %s: rc = %d\n", + mgs->mgs_obd->obd_name, + dirent->name, rc); } } next: - OBD_FREE(dirent, sizeof(*dirent)); + mgs_direntry_free(dirent); } RETURN(rc); } -static int mgs_write_log_mdc_to_mdt(const struct lu_env *env, +static int mgs_write_log_osp_to_mdt(const struct lu_env *env, struct mgs_device *mgs, struct fs_db *fsdb, struct mgs_target_info *mti, - char *logname); + int index, char *logname); static int mgs_write_log_osc_to_lov(const struct lu_env *env, struct mgs_device *mgs, struct fs_db *fsdb, struct mgs_target_info *mti, - char *logname, char *suffix, char *lovname, + char *logname, char *suffix, char *lovname, enum lustre_sec_part sec_part, int flags); static int name_create_mdt_and_lov(char **logname, char **lovname, struct fs_db *fsdb, int i); -static int mgs_steal_llog_handler(const struct lu_env *env, - struct llog_handle *llh, - struct llog_rec_hdr *rec, void *data) +static int add_param(char *params, char *key, char *val) +{ + char *start = params + strlen(params); + char *end = params + sizeof(((struct mgs_target_info *)0)->mti_params); + int keylen = 0; + + if (key != NULL) + keylen = strlen(key); + if (start + 1 + keylen + strlen(val) >= end) { + CERROR("params are too long: %s %s%s\n", + params, key != NULL ? key : "", val); + return -EINVAL; + } + + sprintf(start, " %s%s", key != NULL ? key : "", val); + return 0; +} + +/** + * Walk through client config log record and convert the related records + * into the target. + **/ +static int mgs_steal_client_llog_handler(const struct lu_env *env, + struct llog_handle *llh, + struct llog_rec_hdr *rec, void *data) { struct mgs_device *mgs; + struct obd_device *obd; struct mgs_target_info *mti, *tmti; struct fs_db *fsdb; int cfg_len = rec->lrh_len; @@ -1091,13 +1662,16 @@ static int mgs_steal_llog_handler(const struct lu_env *env, 2: found mdc; */ static int last_step = -1; + int cplen = 0; ENTRY; mti = ((struct temp_comp*)data)->comp_mti; tmti = ((struct temp_comp*)data)->comp_tmti; fsdb = ((struct temp_comp*)data)->comp_fsdb; - mgs = ((struct temp_comp*)data)->comp_mgs; + obd = ((struct temp_comp *)data)->comp_obd; + mgs = lu2mgs_dev(obd->obd_lu_dev); + LASSERT(mgs); if (rec->lrh_type != OBD_CFG_REC) { CERROR("unhandled lrh_type: %#x\n", rec->lrh_type); @@ -1112,68 +1686,84 @@ static int mgs_steal_llog_handler(const struct lu_env *env, lcfg = (struct lustre_cfg *)cfg_buf; - if (lcfg->lcfg_command == LCFG_MARKER) { - struct cfg_marker *marker; - marker = lustre_cfg_buf(lcfg, 1); - if (!strncmp(marker->cm_comment,"add osc",7) && - (marker->cm_flags & CM_START)){ - got_an_osc_or_mdc = 1; - strncpy(tmti->mti_svname, marker->cm_tgtname, - sizeof(tmti->mti_svname)); + if (lcfg->lcfg_command == LCFG_MARKER) { + struct cfg_marker *marker; + marker = lustre_cfg_buf(lcfg, 1); + if (!strncmp(marker->cm_comment, "add osc", 7) && + (marker->cm_flags & CM_START) && + !(marker->cm_flags & CM_SKIP)) { + got_an_osc_or_mdc = 1; + cplen = strlcpy(tmti->mti_svname, marker->cm_tgtname, + sizeof(tmti->mti_svname)); + if (cplen >= sizeof(tmti->mti_svname)) + RETURN(-E2BIG); rc = record_start_log(env, mgs, &mdt_llh, mti->mti_svname); if (rc) RETURN(rc); rc = record_marker(env, mdt_llh, fsdb, CM_START, - mti->mti_svname,"add osc(copied)"); + mti->mti_svname, "add osc(copied)"); record_end_log(env, &mdt_llh); - last_step = marker->cm_step; - RETURN(rc); - } - if (!strncmp(marker->cm_comment,"add osc",7) && - (marker->cm_flags & CM_END)){ - LASSERT(last_step == marker->cm_step); - last_step = -1; - got_an_osc_or_mdc = 0; + last_step = marker->cm_step; + RETURN(rc); + } + if (!strncmp(marker->cm_comment, "add osc", 7) && + (marker->cm_flags & CM_END) && + !(marker->cm_flags & CM_SKIP)) { + LASSERT(last_step == marker->cm_step); + last_step = -1; + got_an_osc_or_mdc = 0; + memset(tmti, 0, sizeof(*tmti)); rc = record_start_log(env, mgs, &mdt_llh, mti->mti_svname); if (rc) RETURN(rc); rc = record_marker(env, mdt_llh, fsdb, CM_END, - mti->mti_svname,"add osc(copied)"); + mti->mti_svname, "add osc(copied)"); record_end_log(env, &mdt_llh); - RETURN(rc); - } - if (!strncmp(marker->cm_comment,"add mdc",7) && - (marker->cm_flags & CM_START)){ - got_an_osc_or_mdc = 2; - last_step = marker->cm_step; - memcpy(tmti->mti_svname, marker->cm_tgtname, - strlen(marker->cm_tgtname)); - - RETURN(rc); - } - if (!strncmp(marker->cm_comment,"add mdc",7) && - (marker->cm_flags & CM_END)){ - LASSERT(last_step == marker->cm_step); - last_step = -1; - got_an_osc_or_mdc = 0; - RETURN(rc); - } - } + RETURN(rc); + } + if (!strncmp(marker->cm_comment, "add mdc", 7) && + (marker->cm_flags & CM_START) && + !(marker->cm_flags & CM_SKIP)) { + got_an_osc_or_mdc = 2; + last_step = marker->cm_step; + memcpy(tmti->mti_svname, marker->cm_tgtname, + strlen(marker->cm_tgtname)); + + RETURN(rc); + } + if (!strncmp(marker->cm_comment, "add mdc", 7) && + (marker->cm_flags & CM_END) && + !(marker->cm_flags & CM_SKIP)) { + LASSERT(last_step == marker->cm_step); + last_step = -1; + got_an_osc_or_mdc = 0; + memset(tmti, 0, sizeof(*tmti)); + RETURN(rc); + } + } if (got_an_osc_or_mdc == 0 || last_step < 0) RETURN(rc); - if (lcfg->lcfg_command == LCFG_ADD_UUID) { - uint64_t nodenid; - nodenid = lcfg->lcfg_nid; + if (lcfg->lcfg_command == LCFG_ADD_UUID) { + uint64_t nodenid = lcfg->lcfg_nid; - tmti->mti_nids[tmti->mti_nid_count] = nodenid; - tmti->mti_nid_count++; + if (strlen(tmti->mti_uuid) == 0) { + /* target uuid not set, this config record is before + * LCFG_SETUP, this nid is one of target node nid. + */ + tmti->mti_nids[tmti->mti_nid_count] = nodenid; + tmti->mti_nid_count++; + } else { + /* failover node nid */ + rc = add_param(tmti->mti_params, PARAM_FAILNODE, + libcfs_nid2str(nodenid)); + } - RETURN(rc); - } + RETURN(rc); + } if (lcfg->lcfg_command == LCFG_SETUP) { char *target; @@ -1197,7 +1787,8 @@ static int mgs_steal_llog_handler(const struct lu_env *env, strlen(mti->mti_fsname)); tmti->mti_stripe_index = index; - rc = mgs_write_log_mdc_to_mdt(env, mgs, fsdb, tmti, + rc = mgs_write_log_osp_to_mdt(env, mgs, fsdb, tmti, + mti->mti_stripe_index, mti->mti_svname); memset(tmti, 0, sizeof(*tmti)); RETURN(rc); @@ -1239,7 +1830,6 @@ static int mgs_steal_llog_for_mdt_from_client(const struct lu_env *env, struct temp_comp* comp) { struct llog_handle *loghandle; - struct lvfs_run_ctxt saved; struct mgs_target_info *tmti; struct llog_ctxt *ctxt; int rc; @@ -1254,11 +1844,9 @@ static int mgs_steal_llog_for_mdt_from_client(const struct lu_env *env, GOTO(out_ctxt, rc = -ENOMEM); comp->comp_tmti = tmti; - comp->comp_mgs = mgs; + comp->comp_obd = mgs->mgs_obd; - push_ctxt(&saved, &mgs->mgs_obd->obd_lvfs_ctxt, NULL); - - rc = llog_open(NULL, ctxt, &loghandle, NULL, client_name, + rc = llog_open(env, ctxt, &loghandle, NULL, client_name, LLOG_OPEN_EXISTS); if (rc < 0) { if (rc == -ENOENT) @@ -1266,17 +1854,16 @@ static int mgs_steal_llog_for_mdt_from_client(const struct lu_env *env, GOTO(out_pop, rc); } - rc = llog_init_handle(NULL, loghandle, LLOG_F_IS_PLAIN, NULL); + rc = llog_init_handle(env, loghandle, LLOG_F_IS_PLAIN, NULL); if (rc) GOTO(out_close, rc); - rc = llog_process_or_fork(env, loghandle, mgs_steal_llog_handler, + rc = llog_process_or_fork(env, loghandle, mgs_steal_client_llog_handler, (void *)comp, NULL, false); CDEBUG(D_MGS, "steal llog re = %d\n", rc); out_close: - llog_close(NULL, loghandle); + llog_close(env, loghandle); out_pop: - pop_ctxt(&saved, &mgs->mgs_obd->obd_lvfs_ctxt, NULL); OBD_FREE_PTR(tmti); out_ctxt: llog_ctxt_put(ctxt); @@ -1514,83 +2101,158 @@ out_free: RETURN(rc); } +static inline int name_create_lov(char **lovname, char *mdtname, + struct fs_db *fsdb, int index) +{ + /* COMPAT_180 */ + if (index == 0 && test_bit(FSDB_OSCNAME18, &fsdb->fsdb_flags)) + return name_create(lovname, fsdb->fsdb_name, "-mdtlov"); + else + return name_create(lovname, mdtname, "-mdtlov"); +} + +static int name_create_mdt_and_lov(char **logname, char **lovname, + struct fs_db *fsdb, int i) +{ + int rc; + + rc = name_create_mdt(logname, fsdb->fsdb_name, i); + if (rc) + return rc; + /* COMPAT_180 */ + if (i == 0 && test_bit(FSDB_OSCNAME18, &fsdb->fsdb_flags)) + rc = name_create(lovname, fsdb->fsdb_name, "-mdtlov"); + else + rc = name_create(lovname, *logname, "-mdtlov"); + if (rc) { + name_destroy(logname); + *logname = NULL; + } + return rc; +} + +static inline int name_create_mdt_osc(char **oscname, char *ostname, + struct fs_db *fsdb, int i) +{ + char suffix[16]; + + if (i == 0 && test_bit(FSDB_OSCNAME18, &fsdb->fsdb_flags)) + sprintf(suffix, "-osc"); + else + sprintf(suffix, "-osc-MDT%04x", i); + return name_create(oscname, ostname, suffix); +} + /* add new mdc to already existent MDS */ -static int mgs_write_log_mdc_to_mdt(const struct lu_env *env, +static int mgs_write_log_osp_to_mdt(const struct lu_env *env, struct mgs_device *mgs, struct fs_db *fsdb, struct mgs_target_info *mti, - char *logname) -{ - struct llog_handle *llh = NULL; - char *nodeuuid = NULL; - char *mdcname = NULL; - char *mdcuuid = NULL; - char *mdtuuid = NULL; - int idx = mti->mti_stripe_index; - char index[9]; - int i, rc; - - ENTRY; - if (mgs_log_is_empty(env, mgs, logname)) { + int mdt_index, char *logname) +{ + struct llog_handle *llh = NULL; + char *nodeuuid = NULL; + char *ospname = NULL; + char *lovuuid = NULL; + char *mdtuuid = NULL; + char *svname = NULL; + char *mdtname = NULL; + char *lovname = NULL; + char index_str[16]; + int i, rc; + + ENTRY; + if (mgs_log_is_empty(env, mgs, mti->mti_svname)) { CERROR("log is empty! Logical error\n"); RETURN (-EINVAL); } - CDEBUG(D_MGS, "adding mdc index %d to %s\n", idx, logname); + CDEBUG(D_MGS, "adding osp index %d to %s\n", mti->mti_stripe_index, + logname); - rc = name_create(&nodeuuid, libcfs_nid2str(mti->mti_nids[0]), ""); + rc = name_create_mdt(&mdtname, fsdb->fsdb_name, mti->mti_stripe_index); if (rc) RETURN(rc); - snprintf(index, sizeof(index), "-mdc%04x", idx); - rc = name_create(&mdcname, logname, index); + + rc = name_create(&nodeuuid, libcfs_nid2str(mti->mti_nids[0]), ""); if (rc) - GOTO(out_free, rc); - rc = name_create(&mdcuuid, mdcname, "_UUID"); + GOTO(out_destory, rc); + + rc = name_create(&svname, mdtname, "-osp"); if (rc) - GOTO(out_free, rc); - rc = name_create(&mdtuuid, logname, "_UUID"); + GOTO(out_destory, rc); + + sprintf(index_str, "-MDT%04x", mdt_index); + rc = name_create(&ospname, svname, index_str); if (rc) - GOTO(out_free, rc); + GOTO(out_destory, rc); + + rc = name_create_lov(&lovname, logname, fsdb, mdt_index); + if (rc) + GOTO(out_destory, rc); + + rc = name_create(&lovuuid, lovname, "_UUID"); + if (rc) + GOTO(out_destory, rc); + + rc = name_create(&mdtuuid, mdtname, "_UUID"); + if (rc) + GOTO(out_destory, rc); rc = record_start_log(env, mgs, &llh, logname); if (rc) - GOTO(out_free, rc); - rc = record_marker(env, llh, fsdb, CM_START, mti->mti_svname, "add mdc"); + GOTO(out_destory, rc); + + rc = record_marker(env, llh, fsdb, CM_START, mti->mti_svname, + "add osp"); if (rc) - GOTO(out_end, rc); - for (i = 0; i < mti->mti_nid_count; i++) { - CDEBUG(D_MGS, "add nid %s for mdt\n", - libcfs_nid2str(mti->mti_nids[i])); + GOTO(out_destory, rc); + + for (i = 0; i < mti->mti_nid_count; i++) { + CDEBUG(D_MGS, "add nid %s for mdt\n", + libcfs_nid2str(mti->mti_nids[i])); rc = record_add_uuid(env, llh, mti->mti_nids[i], nodeuuid); if (rc) GOTO(out_end, rc); } - rc = record_attach(env, llh, mdcname, LUSTRE_MDC_NAME, mdcuuid); + + rc = record_attach(env, llh, ospname, LUSTRE_OSP_NAME, lovuuid); if (rc) GOTO(out_end, rc); - rc = record_setup(env, llh, mdcname, mti->mti_uuid, nodeuuid, 0, 0); + + rc = record_setup(env, llh, ospname, mti->mti_uuid, nodeuuid, + NULL, NULL); if (rc) GOTO(out_end, rc); - rc = mgs_write_log_failnids(env, mti, llh, mdcname); + + rc = mgs_write_log_failnids(env, mti, llh, ospname); if (rc) GOTO(out_end, rc); - snprintf(index, sizeof(index), "%d", idx); - rc = record_mdc_add(env, llh, logname, mdcuuid, mti->mti_uuid, - index, "1"); + /* Add mdc(osp) to lod */ + snprintf(index_str, sizeof(mti->mti_stripe_index), "%d", + mti->mti_stripe_index); + rc = record_base(env, llh, lovname, 0, LCFG_ADD_MDC, mti->mti_uuid, + index_str, "1", NULL); if (rc) GOTO(out_end, rc); - rc = record_marker(env, llh, fsdb, CM_END, mti->mti_svname, "add mdc"); + + rc = record_marker(env, llh, fsdb, CM_END, mti->mti_svname, "add osp"); if (rc) GOTO(out_end, rc); + out_end: record_end_log(env, &llh); -out_free: + +out_destory: name_destroy(&mdtuuid); - name_destroy(&mdcuuid); - name_destroy(&mdcname); + name_destroy(&lovuuid); + name_destroy(&lovname); + name_destroy(&ospname); + name_destroy(&svname); name_destroy(&nodeuuid); - RETURN(rc); + name_destroy(&mdtname); + RETURN(rc); } static int mgs_write_log_mdt0(const struct lu_env *env, @@ -1657,46 +2319,6 @@ out_free: RETURN(rc); } -static inline int name_create_mdt(char **logname, char *fsname, int i) -{ - char mdt_index[9]; - - sprintf(mdt_index, "-MDT%04x", i); - return name_create(logname, fsname, mdt_index); -} - -static int name_create_mdt_and_lov(char **logname, char **lovname, - struct fs_db *fsdb, int i) -{ - int rc; - - rc = name_create_mdt(logname, fsdb->fsdb_name, i); - if (rc) - return rc; - /* COMPAT_180 */ - if (i == 0 && cfs_test_bit(FSDB_OSCNAME18, &fsdb->fsdb_flags)) - rc = name_create(lovname, fsdb->fsdb_name, "-mdtlov"); - else - rc = name_create(lovname, *logname, "-mdtlov"); - if (rc) { - name_destroy(logname); - *logname = NULL; - } - return rc; -} - -static inline int name_create_mdt_osc(char **oscname, char *ostname, - struct fs_db *fsdb, int i) -{ - char suffix[16]; - - if (i == 0 && cfs_test_bit(FSDB_OSCNAME18, &fsdb->fsdb_flags)) - sprintf(suffix, "-osc"); - else - sprintf(suffix, "-osc-MDT%04x", i); - return name_create(oscname, ostname, suffix); -} - /* envelope method for all layers log */ static int mgs_write_log_mdt(const struct lu_env *env, struct mgs_device *mgs, @@ -1778,20 +2400,24 @@ static int mgs_write_log_mdt(const struct lu_env *env, if (rc) GOTO(out_end, rc); + /* for_all_existing_mdt except current one */ - for (i = 0; i < INDEX_MAP_SIZE * 8; i++){ - char *mdtname; - if (i != mti->mti_stripe_index && - cfs_test_bit(i, fsdb->fsdb_mdt_index_map)) { - rc = name_create_mdt(&mdtname, mti->mti_fsname, i); + for (i = 0; i < INDEX_MAP_SIZE * 8; i++) { + if (i != mti->mti_stripe_index && + test_bit(i, fsdb->fsdb_mdt_index_map)) { + char *logname; + + rc = name_create_mdt(&logname, fsdb->fsdb_name, i); if (rc) GOTO(out_end, rc); - rc = mgs_write_log_mdc_to_mdt(env, mgs, fsdb, mti, mdtname); - name_destroy(&mdtname); + + rc = mgs_write_log_osp_to_mdt(env, mgs, fsdb, mti, + i, logname); + name_destroy(&logname); if (rc) GOTO(out_end, rc); - } - } + } + } out_end: record_end_log(env, &llh); out_free: @@ -1830,9 +2456,16 @@ static int mgs_write_log_osc_to_lov(const struct lu_env *env, rc = name_create(&svname, mti->mti_svname, "-osc"); if (rc) GOTO(out_free, rc); - rc = name_create(&oscname, svname, suffix); + + /* for the system upgraded from old 1.8, keep using the old osc naming + * style for mdt, see name_create_mdt_osc(). LU-1257 */ + if (test_bit(FSDB_OSCNAME18, &fsdb->fsdb_flags)) + rc = name_create(&oscname, svname, ""); + else + rc = name_create(&oscname, svname, suffix); if (rc) GOTO(out_free, rc); + rc = name_create(&oscuuid, oscname, "_UUID"); if (rc) GOTO(out_free, rc); @@ -1840,6 +2473,7 @@ static int mgs_write_log_osc_to_lov(const struct lu_env *env, if (rc) GOTO(out_free, rc); + /* #03 L add_uuid nid=uml1@tcp(0x20000c0a80201) 0: 1:uml1_UUID multihomed (#4) @@ -1855,11 +2489,18 @@ static int mgs_write_log_osc_to_lov(const struct lu_env *env, rc = record_start_log(env, mgs, &llh, logname); if (rc) GOTO(out_free, rc); + /* FIXME these should be a single journal transaction */ rc = record_marker(env, llh, fsdb, CM_START | flags, mti->mti_svname, "add osc"); if (rc) GOTO(out_end, rc); + + /* NB: don't change record order, because upon MDT steal OSC config + * from client, it treats all nids before LCFG_SETUP as target nids + * (multiple interfaces), while nids after as failover node nids. + * See mgs_steal_client_llog_handler() LCFG_ADD_UUID. + */ for (i = 0; i < mti->mti_nid_count; i++) { CDEBUG(D_MGS, "add nid %s\n", libcfs_nid2str(mti->mti_nids[i])); rc = record_add_uuid(env, llh, mti->mti_nids[i], nodeuuid); @@ -1875,7 +2516,9 @@ static int mgs_write_log_osc_to_lov(const struct lu_env *env, rc = mgs_write_log_failnids(env, mti, llh, oscname); if (rc) GOTO(out_end, rc); - snprintf(index, sizeof(index), "%d", mti->mti_stripe_index); + + snprintf(index, sizeof(index), "%d", mti->mti_stripe_index); + rc = record_lov_add(env, llh, lovname, mti->mti_uuid, index, "1"); if (rc) GOTO(out_end, rc); @@ -1954,7 +2597,7 @@ out_end: /* We also have to update the other logs where this osc is part of the lov */ - if (cfs_test_bit(FSDB_OLDLOG14, &fsdb->fsdb_flags)) { + if (test_bit(FSDB_OLDLOG14, &fsdb->fsdb_flags)) { /* If we're upgrading, the old mdt log already has our entry. Let's do a fake one for fun. */ /* Note that we can't add any new failnids, since we don't @@ -1972,7 +2615,7 @@ out_end: /* Add ost to all MDT lov defs */ for (i = 0; i < INDEX_MAP_SIZE * 8; i++){ - if (cfs_test_bit(i, fsdb->fsdb_mdt_index_map)) { + if (test_bit(i, fsdb->fsdb_mdt_index_map)) { char mdt_index[9]; rc = name_create_mdt_and_lov(&logname, &lovname, fsdb, @@ -2110,7 +2753,7 @@ static int mgs_write_log_add_failnid(const struct lu_env *env, int i; for (i = 0; i < INDEX_MAP_SIZE * 8; i++) { - if (!cfs_test_bit(i, fsdb->fsdb_mdt_index_map)) + if (!test_bit(i, fsdb->fsdb_mdt_index_map)) continue; rc = name_create_mdt(&logname, mti->mti_fsname, i); if (rc) @@ -2243,12 +2886,11 @@ static int mgs_write_log_quota(const struct lu_env *env, struct mgs_device *mgs, struct fs_db *fsdb, struct mgs_target_info *mti, char *quota, char *ptr) { - struct lustre_cfg_bufs bufs; - struct lustre_cfg *lcfg; - char *tmp; - char sep; - int cmd = LCFG_PARAM; - int rc; + struct mgs_thread_info *mgi = mgs_env_info(env); + struct lustre_cfg *lcfg; + char *tmp; + char sep; + int rc, cmd = LCFG_PARAM; /* support only 'meta' and 'data' pools so far */ if (class_match_param(ptr, QUOTA_METAPOOL_NAME, &tmp) != 0 && @@ -2270,9 +2912,9 @@ static int mgs_write_log_quota(const struct lu_env *env, struct mgs_device *mgs, } } - lustre_cfg_bufs_reset(&bufs, NULL); - lustre_cfg_bufs_set_string(&bufs, 1, quota); - lcfg = lustre_cfg_new(cmd, &bufs); + lustre_cfg_bufs_reset(&mgi->mgi_bufs, mti->mti_fsname); + lustre_cfg_bufs_set_string(&mgi->mgi_bufs, 1, quota); + lcfg = lustre_cfg_new(cmd, &mgi->mgi_bufs); /* truncate the comment to the parameter name */ ptr = tmp - 1; sep = *ptr; @@ -2287,7 +2929,7 @@ static int mgs_write_log_quota(const struct lu_env *env, struct mgs_device *mgs, mti->mti_fsname, quota, 1); *ptr = sep; lustre_cfg_free(lcfg); - return rc; + return rc < 0 ? rc : 0; } static int mgs_srpc_set_param_disk(const struct lu_env *env, @@ -2372,10 +3014,10 @@ static int mgs_srpc_set_param_udesc_mem(struct fs_db *fsdb, goto error_out; if (strcmp(ptr, "yes") == 0) { - cfs_set_bit(FSDB_UDESC, &fsdb->fsdb_flags); + set_bit(FSDB_UDESC, &fsdb->fsdb_flags); CWARN("Enable user descriptor shipping from client to MDT\n"); } else if (strcmp(ptr, "no") == 0) { - cfs_clear_bit(FSDB_UDESC, &fsdb->fsdb_flags); + clear_bit(FSDB_UDESC, &fsdb->fsdb_flags); CWARN("Disable user descriptor shipping from client to MDT\n"); } else { *(ptr - 1) = '='; @@ -2419,7 +3061,7 @@ static int mgs_srpc_set_param_mem(struct fs_db *fsdb, RETURN(rc); /* mgs rules implies must be mgc->mgs */ - if (cfs_test_bit(FSDB_MGS_SELF, &fsdb->fsdb_flags)) { + if (test_bit(FSDB_MGS_SELF, &fsdb->fsdb_flags)) { if ((rule.sr_from != LUSTRE_SP_MGC && rule.sr_from != LUSTRE_SP_ANY) || (rule.sr_to != LUSTRE_SP_MGS && @@ -2503,7 +3145,7 @@ static int mgs_srpc_set_param(const struct lu_env *env, if (rc) goto out_free; - if (cfs_test_bit(FSDB_MGS_SELF, &fsdb->fsdb_flags)) { + if (test_bit(FSDB_MGS_SELF, &fsdb->fsdb_flags)) { /* * for mgs rules, make them effective immediately. */ @@ -2528,7 +3170,7 @@ static int mgs_srpc_read_handler(const struct lu_env *env, { struct mgs_srpc_read_data *msrd = data; struct cfg_marker *marker; - struct lustre_cfg *lcfg = (struct lustre_cfg *)(rec + 1); + struct lustre_cfg *lcfg = REC_DATA(rec); char *svname, *param; int cfg_len, rc; ENTRY; @@ -2591,7 +3233,6 @@ int mgs_get_fsdb_srpc_from_llog(const struct lu_env *env, struct fs_db *fsdb) { struct llog_handle *llh = NULL; - struct lvfs_run_ctxt saved; struct llog_ctxt *ctxt; char *logname; struct mgs_srpc_read_data msrd; @@ -2609,17 +3250,15 @@ int mgs_get_fsdb_srpc_from_llog(const struct lu_env *env, if (mgs_log_is_empty(env, mgs, logname)) GOTO(out, rc = 0); - push_ctxt(&saved, &mgs->mgs_obd->obd_lvfs_ctxt, NULL); - - rc = llog_open(NULL, ctxt, &llh, NULL, logname, + rc = llog_open(env, ctxt, &llh, NULL, logname, LLOG_OPEN_EXISTS); if (rc < 0) { if (rc == -ENOENT) rc = 0; - GOTO(out_pop, rc); + GOTO(out, rc); } - rc = llog_init_handle(NULL, llh, LLOG_F_IS_PLAIN, NULL); + rc = llog_init_handle(env, llh, LLOG_F_IS_PLAIN, NULL); if (rc) GOTO(out_close, rc); @@ -2629,13 +3268,11 @@ int mgs_get_fsdb_srpc_from_llog(const struct lu_env *env, msrd.msrd_fsdb = fsdb; msrd.msrd_skip = 0; - rc = llog_process_or_fork(env, llh, mgs_srpc_read_handler, - (void *)&msrd, NULL, false); + rc = llog_process(env, llh, mgs_srpc_read_handler, (void *)&msrd, + NULL); out_close: - llog_close(NULL, llh); -out_pop: - pop_ctxt(&saved, &mgs->mgs_obd->obd_lvfs_ctxt, NULL); + llog_close(env, llh); out: llog_ctxt_put(ctxt); name_destroy(&logname); @@ -2738,7 +3375,7 @@ static int mgs_write_log_param(const struct lu_env *env, /* Modify mdtlov */ /* Add to all MDT logs for CMD */ for (i = 0; i < INDEX_MAP_SIZE * 8; i++) { - if (!cfs_test_bit(i, fsdb->fsdb_mdt_index_map)) + if (!test_bit(i, fsdb->fsdb_mdt_index_map)) continue; rc = name_create_mdt(&logname, mti->mti_fsname, i); if (rc) @@ -2756,7 +3393,7 @@ static int mgs_write_log_param(const struct lu_env *env, "changes were made to the " "config log.\n", mti->mti_svname, rc); - if (cfs_test_bit(FSDB_OLDLOG14, &fsdb->fsdb_flags)) + if (test_bit(FSDB_OLDLOG14, &fsdb->fsdb_flags)) LCONSOLE_ERROR_MSG(0x146, "This may be" " because the log" "is in the old 1.4" @@ -2814,7 +3451,7 @@ static int mgs_write_log_param(const struct lu_env *env, (class_match_param(ptr, PARAM_LLITE, NULL) == 0)) { char *cname; - if (cfs_test_bit(FSDB_OLDLOG14, &fsdb->fsdb_flags)) { + if (test_bit(FSDB_OLDLOG14, &fsdb->fsdb_flags)) { LCONSOLE_ERROR_MSG(0x148, "Upgraded client logs for %s" " cannot be modified. Consider" " updating the configuration with" @@ -2852,7 +3489,7 @@ static int mgs_write_log_param(const struct lu_env *env, int i; for (i = 0; i < INDEX_MAP_SIZE * 8; i++){ - if (!cfs_test_bit(i, fsdb->fsdb_mdt_index_map)) + if (!test_bit(i, fsdb->fsdb_mdt_index_map)) continue; name_destroy(&cname); rc = name_create_mdt_osc(&cname, mti->mti_svname, @@ -2897,7 +3534,7 @@ static int mgs_write_log_param(const struct lu_env *env, goto active_err; if (rc & LDD_F_SV_ALL) { for (i = 0; i < INDEX_MAP_SIZE * 8; i++) { - if (!cfs_test_bit(i, + if (!test_bit(i, fsdb->fsdb_mdt_index_map)) continue; rc = name_create_mdt(&logname, @@ -2966,9 +3603,9 @@ int mgs_check_failnid(const struct lu_env *env, struct mgs_device *mgs, the failover list. Modify mti->params for rewriting back at server_register_target(). */ - cfs_mutex_lock(&fsdb->fsdb_mutex); + mutex_lock(&fsdb->fsdb_mutex); rc = mgs_write_log_add_failnid(obd, fsdb, mti); - cfs_mutex_unlock(&fsdb->fsdb_mutex); + mutex_unlock(&fsdb->fsdb_mutex); RETURN(rc); #endif @@ -3000,9 +3637,10 @@ int mgs_write_log_target(const struct lu_env *env, get repeat setup instructions for already running osc's. So don't update the client/mdt logs. */ mti->mti_flags &= ~LDD_F_UPDATE; + rc = 0; } - cfs_mutex_lock(&fsdb->fsdb_mutex); + mutex_lock(&fsdb->fsdb_mutex); if (mti->mti_flags & (LDD_F_VIRGIN | LDD_F_UPGRADE14 | LDD_F_WRITECONF)) { @@ -3051,35 +3689,31 @@ int mgs_write_log_target(const struct lu_env *env, OBD_FREE(buf, strlen(mti->mti_params) + 1); out_up: - cfs_mutex_unlock(&fsdb->fsdb_mutex); + mutex_unlock(&fsdb->fsdb_mutex); RETURN(rc); } int mgs_erase_log(const struct lu_env *env, struct mgs_device *mgs, char *name) { - struct lvfs_run_ctxt saved; struct llog_ctxt *ctxt; int rc = 0; - struct obd_device *obd = mgs->mgs_obd; - ctxt = llog_get_context(obd, LLOG_CONFIG_ORIG_CTXT); + ctxt = llog_get_context(mgs->mgs_obd, LLOG_CONFIG_ORIG_CTXT); if (ctxt == NULL) { CERROR("%s: MGS config context doesn't exist\n", - obd->obd_name); + mgs->mgs_obd->obd_name); rc = -ENODEV; } else { - push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL); - rc = llog_erase(NULL, ctxt, NULL, name); + rc = llog_erase(env, ctxt, NULL, name); /* llog may not exist */ if (rc == -ENOENT) rc = 0; - pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL); llog_ctxt_put(ctxt); } if (rc) - CERROR("%s: failed to clear log %s: %d\n", obd->obd_name, - name, rc); + CERROR("%s: failed to clear log %s: %d\n", + mgs->mgs_obd->obd_name, name, rc); return rc; } @@ -3087,42 +3721,40 @@ int mgs_erase_log(const struct lu_env *env, struct mgs_device *mgs, char *name) /* erase all logs for the given fs */ int mgs_erase_logs(const struct lu_env *env, struct mgs_device *mgs, char *fsname) { - struct fs_db *fsdb; - cfs_list_t dentry_list; - struct l_linux_dirent *dirent, *n; - int rc, len = strlen(fsname); - char *suffix; - ENTRY; + struct fs_db *fsdb; + cfs_list_t list; + struct mgs_direntry *dirent, *n; + int rc, len = strlen(fsname); + char *suffix; + ENTRY; - /* Find all the logs in the CONFIGS directory */ - rc = class_dentry_readdir(env, mgs, &dentry_list); - if (rc) { - CERROR("Can't read %s dir\n", MOUNT_CONFIGS_DIR); - RETURN(rc); - } + /* Find all the logs in the CONFIGS directory */ + rc = class_dentry_readdir(env, mgs, &list); + if (rc) + RETURN(rc); - cfs_mutex_lock(&mgs->mgs_mutex); + mutex_lock(&mgs->mgs_mutex); /* Delete the fs db */ fsdb = mgs_find_fsdb(mgs, fsname); if (fsdb) mgs_free_fsdb(mgs, fsdb); - cfs_list_for_each_entry_safe(dirent, n, &dentry_list, lld_list) { - cfs_list_del(&dirent->lld_list); - suffix = strrchr(dirent->lld_name, '-'); - if (suffix != NULL) { - if ((len == suffix - dirent->lld_name) && - (strncmp(fsname, dirent->lld_name, len) == 0)) { - CDEBUG(D_MGS, "Removing log %s\n", - dirent->lld_name); - mgs_erase_log(env, mgs, dirent->lld_name); - } - } - OBD_FREE(dirent, sizeof(*dirent)); - } - - cfs_mutex_unlock(&mgs->mgs_mutex); + mutex_unlock(&mgs->mgs_mutex); + + cfs_list_for_each_entry_safe(dirent, n, &list, list) { + cfs_list_del(&dirent->list); + suffix = strrchr(dirent->name, '-'); + if (suffix != NULL) { + if ((len == suffix - dirent->name) && + (strncmp(fsname, dirent->name, len) == 0)) { + CDEBUG(D_MGS, "Removing log %s\n", + dirent->name); + mgs_erase_log(env, mgs, dirent->name); + } + } + mgs_direntry_free(dirent); + } RETURN(rc); } @@ -3186,27 +3818,25 @@ int mgs_setparam(const struct lu_env *env, struct mgs_device *mgs, RETURN(-ENOSYS); } - /* Extract fsname */ - ptr = strrchr(devname, '-'); - memset(fsname, 0, MTI_NAME_MAXLEN); - if (ptr && (server_name2index(ptr, &index, NULL) >= 0)) { + rc = mgs_parse_devname(devname, fsname, NULL); + if (rc == 0 && !mgs_parse_devname(devname, NULL, &index)) { /* param related to llite isn't allowed to set by OST or MDT */ - if (strncmp(param, PARAM_LLITE, sizeof(PARAM_LLITE)) == 0) + if (rc == 0 && strncmp(param, PARAM_LLITE, + sizeof(PARAM_LLITE)) == 0) RETURN(-EINVAL); - - strncpy(fsname, devname, ptr - devname); } else { /* assume devname is the fsname */ + memset(fsname, 0, MTI_NAME_MAXLEN); strncpy(fsname, devname, MTI_NAME_MAXLEN); + fsname[MTI_NAME_MAXLEN - 1] = 0; } - fsname[MTI_NAME_MAXLEN - 1] = 0; CDEBUG(D_MGS, "setparam fs='%s' device='%s'\n", fsname, devname); rc = mgs_find_or_make_fsdb(env, mgs, fsname, &fsdb); if (rc) RETURN(rc); - if (!cfs_test_bit(FSDB_MGS_SELF, &fsdb->fsdb_flags) && - cfs_test_bit(FSDB_LOG_EMPTY, &fsdb->fsdb_flags)) { + if (!test_bit(FSDB_MGS_SELF, &fsdb->fsdb_flags) && + test_bit(FSDB_LOG_EMPTY, &fsdb->fsdb_flags)) { CERROR("No filesystem targets for %s. cfg_device from lctl " "is '%s'\n", fsname, devname); mgs_free_fsdb(mgs, fsdb); @@ -3217,9 +3847,15 @@ int mgs_setparam(const struct lu_env *env, struct mgs_device *mgs, OBD_ALLOC_PTR(mti); if (!mti) GOTO(out, rc = -ENOMEM); - strncpy(mti->mti_fsname, fsname, MTI_NAME_MAXLEN); - strncpy(mti->mti_svname, devname, MTI_NAME_MAXLEN); - strncpy(mti->mti_params, param, sizeof(mti->mti_params)); + if (strlcpy(mti->mti_fsname, fsname, sizeof(mti->mti_fsname)) + >= sizeof(mti->mti_fsname)) + GOTO(out, rc = -E2BIG); + if (strlcpy(mti->mti_svname, devname, sizeof(mti->mti_svname)) + >= sizeof(mti->mti_svname)) + GOTO(out, rc = -E2BIG); + if (strlcpy(mti->mti_params, param, sizeof(mti->mti_params)) + >= sizeof(mti->mti_params)) + GOTO(out, rc = -E2BIG); rc = server_name2index(mti->mti_svname, &mti->mti_stripe_index, &tmp); if (rc < 0) /* Not a valid server; may be only fsname */ @@ -3232,9 +3868,9 @@ int mgs_setparam(const struct lu_env *env, struct mgs_device *mgs, mti->mti_flags = rc | LDD_F_PARAM; - cfs_mutex_lock(&fsdb->fsdb_mutex); + mutex_lock(&fsdb->fsdb_mutex); rc = mgs_write_log_param(env, mgs, fsdb, mti, mti->mti_params); - cfs_mutex_unlock(&fsdb->fsdb_mutex); + mutex_unlock(&fsdb->fsdb_mutex); /* * Revoke lock so everyone updates. Should be alright if @@ -3291,7 +3927,7 @@ int mgs_pool_cmd(const struct lu_env *env, struct mgs_device *mgs, CERROR("Can't get db for %s\n", fsname); RETURN(rc); } - if (cfs_test_bit(FSDB_LOG_EMPTY, &fsdb->fsdb_flags)) { + if (test_bit(FSDB_LOG_EMPTY, &fsdb->fsdb_flags)) { CERROR("%s is not defined\n", fsname); mgs_free_fsdb(mgs, fsdb); RETURN(-EINVAL); @@ -3345,7 +3981,7 @@ int mgs_pool_cmd(const struct lu_env *env, struct mgs_device *mgs, break; } - cfs_mutex_lock(&fsdb->fsdb_mutex); + mutex_lock(&fsdb->fsdb_mutex); if (canceled_label != NULL) { OBD_ALLOC_PTR(mti); @@ -3355,11 +3991,11 @@ int mgs_pool_cmd(const struct lu_env *env, struct mgs_device *mgs, /* write pool def to all MDT logs */ for (i = 0; i < INDEX_MAP_SIZE * 8; i++) { - if (cfs_test_bit(i, fsdb->fsdb_mdt_index_map)) { + if (test_bit(i, fsdb->fsdb_mdt_index_map)) { rc = name_create_mdt_and_lov(&logname, &lovname, fsdb, i); if (rc) { - cfs_mutex_unlock(&fsdb->fsdb_mutex); + mutex_unlock(&fsdb->fsdb_mutex); GOTO(out_mti, rc); } if (canceled_label != NULL) { @@ -3377,7 +4013,7 @@ int mgs_pool_cmd(const struct lu_env *env, struct mgs_device *mgs, name_destroy(&logname); name_destroy(&lovname); if (rc) { - cfs_mutex_unlock(&fsdb->fsdb_mutex); + mutex_unlock(&fsdb->fsdb_mutex); GOTO(out_mti, rc); } } @@ -3385,14 +4021,14 @@ int mgs_pool_cmd(const struct lu_env *env, struct mgs_device *mgs, rc = name_create(&logname, fsname, "-client"); if (rc) { - cfs_mutex_unlock(&fsdb->fsdb_mutex); + mutex_unlock(&fsdb->fsdb_mutex); GOTO(out_mti, rc); } if (canceled_label != NULL) { rc = mgs_modify(env, mgs, fsdb, mti, logname, fsdb->fsdb_clilov, canceled_label, CM_SKIP); if (rc < 0) { - cfs_mutex_unlock(&fsdb->fsdb_mutex); + mutex_unlock(&fsdb->fsdb_mutex); name_destroy(&logname); GOTO(out_mti, rc); } @@ -3400,7 +4036,7 @@ int mgs_pool_cmd(const struct lu_env *env, struct mgs_device *mgs, rc = mgs_write_log_pool(env, mgs, logname, fsdb, fsdb->fsdb_clilov, cmd, fsname, poolname, ostname, label); - cfs_mutex_unlock(&fsdb->fsdb_mutex); + mutex_unlock(&fsdb->fsdb_mutex); name_destroy(&logname); /* request for update */ mgs_revoke_lock(mgs, fsdb, CONFIG_T_CONFIG); @@ -3417,63 +4053,4 @@ out_label: return rc; } -#if 0 -/******************** unused *********************/ -static int mgs_backup_llog(struct obd_device *obd, char* fsname) -{ - struct file *filp, *bak_filp; - struct lvfs_run_ctxt saved; - char *logname, *buf; - loff_t soff = 0 , doff = 0; - int count = 4096, len; - int rc = 0; - - OBD_ALLOC(logname, PATH_MAX); - if (logname == NULL) - return -ENOMEM; - OBD_ALLOC(buf, count); - if (!buf) - GOTO(out , rc = -ENOMEM); - - len = snprintf(logname, PATH_MAX, "%s/%s.bak", - MOUNT_CONFIGS_DIR, fsname); - - if (len >= PATH_MAX - 1) { - GOTO(out, -ENAMETOOLONG); - } - - push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL); - - bak_filp = l_filp_open(logname, O_RDWR|O_CREAT|O_TRUNC, 0660); - if (IS_ERR(bak_filp)) { - rc = PTR_ERR(bak_filp); - CERROR("backup logfile open %s: %d\n", logname, rc); - GOTO(pop, rc); - } - sprintf(logname, "%s/%s", MOUNT_CONFIGS_DIR, fsname); - filp = l_filp_open(logname, O_RDONLY, 0); - if (IS_ERR(filp)) { - rc = PTR_ERR(filp); - CERROR("logfile open %s: %d\n", logname, rc); - GOTO(close1f, rc); - } - - while ((rc = lustre_fread(filp, buf, count, &soff)) > 0) { - rc = lustre_fwrite(bak_filp, buf, count, &doff); - break; - } - - filp_close(filp, 0); -close1f: - filp_close(bak_filp, 0); -pop: - pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL); -out: - if (buf) - OBD_FREE(buf, count); - OBD_FREE(logname, PATH_MAX); - return rc; -} - -#endif