/*
* Copyright (C) 2015, Trustees of Indiana University
*
- * Copyright (c) 2014, Intel Corporation.
+ * Copyright (c) 2017, Intel Corporation.
*
* Author: Joshua Walgenbach <jjw@iu.edu>
* Author: Kit Westneat <cwestnea@iu.edu>
#include <linux/mutex.h>
#include <linux/string.h>
#include <linux/types.h>
-#include <lnet/types.h>
-#include <lustre/lustre_idl.h>
+#include <uapi/linux/lnet/lnet-types.h>
+#include <uapi/linux/lustre/lustre_idl.h>
#include <dt_object.h>
#include <lu_object.h>
#include <lustre_net.h>
NM_FL_ALLOW_ROOT_ACCESS = 0x1,
NM_FL_TRUST_CLIENT_IDS = 0x2,
NM_FL_DENY_UNKNOWN = 0x4,
+ NM_FL_MAP_UID_ONLY = 0x8,
+ NM_FL_MAP_GID_ONLY = 0x10,
+ NM_FL_ENABLE_AUDIT = 0x20,
+ NM_FL_FORBID_ENCRYPT = 0x40,
};
static void nodemap_cluster_key_init(struct nodemap_key *nk, unsigned int nm_id)
static void nodemap_cluster_rec_init(union nodemap_rec *nr,
const struct lu_nodemap *nodemap)
{
- CLASSERT(sizeof(nr->ncr.ncr_name) == sizeof(nodemap->nm_name));
+ BUILD_BUG_ON(sizeof(nr->ncr.ncr_name) != sizeof(nodemap->nm_name));
- strncpy(nr->ncr.ncr_name, nodemap->nm_name, sizeof(nodemap->nm_name));
+ strncpy(nr->ncr.ncr_name, nodemap->nm_name, sizeof(nr->ncr.ncr_name));
nr->ncr.ncr_squash_uid = cpu_to_le32(nodemap->nm_squash_uid);
nr->ncr.ncr_squash_gid = cpu_to_le32(nodemap->nm_squash_gid);
nr->ncr.ncr_flags = cpu_to_le32(
(nodemap->nmf_allow_root_access ?
NM_FL_ALLOW_ROOT_ACCESS : 0) |
(nodemap->nmf_deny_unknown ?
- NM_FL_DENY_UNKNOWN : 0));
+ NM_FL_DENY_UNKNOWN : 0) |
+ (nodemap->nmf_map_uid_only ?
+ NM_FL_MAP_UID_ONLY : 0) |
+ (nodemap->nmf_map_gid_only ?
+ NM_FL_MAP_GID_ONLY : 0) |
+ (nodemap->nmf_enable_audit ?
+ NM_FL_ENABLE_AUDIT : 0) |
+ (nodemap->nmf_forbid_encryption ?
+ NM_FL_FORBID_ENCRYPT : 0));
}
static void nodemap_idmap_key_init(struct nodemap_key *nk, unsigned int nm_id,
dt_version_set(env, nodemap_idx, ver + 1, th);
}
+enum ncfc_find_create {
+ NCFC_CREATE_NEW = 1,
+};
+
static struct dt_object *nodemap_cache_find_create(const struct lu_env *env,
struct dt_device *dev,
struct local_oid_storage *los,
- bool force_create)
+ enum ncfc_find_create create_new)
{
- struct lu_fid root_fid;
+ struct lu_fid tfid;
struct dt_object *root_obj;
struct dt_object *nm_obj;
int rc = 0;
- rc = dt_root_get(env, dev, &root_fid);
+ rc = dt_root_get(env, dev, &tfid);
if (rc < 0)
GOTO(out, nm_obj = ERR_PTR(rc));
- root_obj = dt_locate(env, dev, &root_fid);
+ root_obj = dt_locate(env, dev, &tfid);
if (unlikely(IS_ERR(root_obj)))
GOTO(out, nm_obj = root_obj);
+ rc = dt_lookup_dir(env, root_obj, LUSTRE_NODEMAP_NAME, &tfid);
+ if (rc == -ENOENT) {
+ if (dev->dd_rdonly)
+ GOTO(out_root, nm_obj = ERR_PTR(-EROFS));
+ } else if (rc) {
+ GOTO(out_root, nm_obj = ERR_PTR(rc));
+ } else if (dev->dd_rdonly && create_new == NCFC_CREATE_NEW) {
+ GOTO(out_root, nm_obj = ERR_PTR(-EROFS));
+ }
+
again:
- /* if loading index fails the first time, try again with force_create */
- if (force_create) {
+ /* if loading index fails the first time, create new index */
+ if (create_new == NCFC_CREATE_NEW && rc != -ENOENT) {
CDEBUG(D_INFO, "removing old index, creating new one\n");
rc = local_object_unlink(env, dev, root_obj,
LUSTRE_NODEMAP_NAME);
/* even if loading from tgt fails, connecting to MGS will
* rewrite the config
*/
- if (rc < 0 && !force_create) {
+ if (rc < 0) {
+ dt_object_put(env, nm_obj);
+
+ if (create_new == NCFC_CREATE_NEW)
+ GOTO(out_root, nm_obj = ERR_PTR(rc));
+
CERROR("cannot load nodemap index from disk, creating "
"new index: rc = %d\n", rc);
- lu_object_put(env, &nm_obj->do_lu);
- force_create = true;
+ create_new = NCFC_CREATE_NEW;
goto again;
}
}
- if (rc < 0)
- nm_obj = ERR_PTR(rc);
-
out_root:
- lu_object_put(env, &root_obj->do_lu);
+ dt_object_put(env, root_obj);
out:
return nm_obj;
}
const struct nodemap_key *nk,
const union nodemap_rec *nr)
{
- struct thandle *th;
- struct dt_device *dev = lu2dt_dev(idx->do_lu.lo_dev);
- int rc;
+ struct thandle *th;
+ struct dt_device *dev = lu2dt_dev(idx->do_lu.lo_dev);
+ int rc;
- CLASSERT(sizeof(union nodemap_rec) == 32);
+ BUILD_BUG_ON(sizeof(union nodemap_rec) != 32);
th = dt_trans_create(env, dev);
dt_write_lock(env, idx, 0);
rc = dt_insert(env, idx, (const struct dt_rec *)nr,
- (const struct dt_key *)nk, th, 1);
+ (const struct dt_key *)nk, th);
nodemap_inc_version(env, idx, th);
dt_write_unlock(env, idx);
GOTO(out_lock, rc);
rc = dt_insert(env, idx, (const struct dt_rec *)nr,
- (const struct dt_key *)nk, th, 1);
+ (const struct dt_key *)nk, th);
if (rc != 0)
GOTO(out_lock, rc);
};
static int nodemap_idx_nodemap_add_update(const struct lu_nodemap *nodemap,
+ struct dt_object *idx,
enum nm_add_update update)
{
struct nodemap_key nk;
ENTRY;
- if (nodemap_mgs_ncf == NULL) {
- CERROR("cannot add nodemap config to non-existing MGS.\n");
- return -EINVAL;
- }
-
rc = lu_env_init(&env, LCT_LOCAL);
if (rc)
RETURN(rc);
nodemap_cluster_rec_init(&nr, nodemap);
if (update == NM_UPDATE)
- rc = nodemap_idx_update(&env, nodemap_mgs_ncf->ncf_obj,
- &nk, &nr);
+ rc = nodemap_idx_update(&env, idx, &nk, &nr);
else
- rc = nodemap_idx_insert(&env, nodemap_mgs_ncf->ncf_obj,
- &nk, &nr);
+ rc = nodemap_idx_insert(&env, idx, &nk, &nr);
lu_env_fini(&env);
int nodemap_idx_nodemap_add(const struct lu_nodemap *nodemap)
{
- return nodemap_idx_nodemap_add_update(nodemap, NM_ADD);
+ if (nodemap_mgs_ncf == NULL) {
+ CERROR("cannot add nodemap config to non-existing MGS.\n");
+ return -EINVAL;
+ }
+
+ return nodemap_idx_nodemap_add_update(nodemap, nodemap_mgs_ncf->ncf_obj,
+ NM_ADD);
}
int nodemap_idx_nodemap_update(const struct lu_nodemap *nodemap)
{
- return nodemap_idx_nodemap_add_update(nodemap, NM_UPDATE);
+ if (nodemap_mgs_ncf == NULL) {
+ CERROR("cannot add nodemap config to non-existing MGS.\n");
+ return -EINVAL;
+ }
+
+ return nodemap_idx_nodemap_add_update(nodemap, nodemap_mgs_ncf->ncf_obj,
+ NM_UPDATE);
}
int nodemap_idx_nodemap_del(const struct lu_nodemap *nodemap)
return nodemap_idx_global_add_update(value, NM_UPDATE);
}
+static enum nodemap_idx_type nodemap_get_key_type(const struct nodemap_key *key)
+{
+ u32 nodemap_id;
+
+ nodemap_id = le32_to_cpu(key->nk_nodemap_id);
+ return nm_idx_get_type(nodemap_id);
+}
+
/**
* Process a key/rec pair and modify the new configuration.
*
const union nodemap_rec *rec,
struct lu_nodemap **recent_nodemap)
{
- struct lu_nodemap *nodemap = NULL;
- enum nodemap_idx_type type;
- enum nodemap_id_type id_type;
- u8 flags;
- u32 nodemap_id;
- lnet_nid_t nid[2];
- u32 map[2];
- int rc;
+ struct lu_nodemap *nodemap = NULL;
+ enum nodemap_idx_type type;
+ enum nodemap_id_type id_type;
+ u8 flags;
+ u32 nodemap_id;
+ lnet_nid_t nid[2];
+ u32 map[2];
+ int rc;
+
+ ENTRY;
- CLASSERT(sizeof(union nodemap_rec) == 32);
+ BUILD_BUG_ON(sizeof(union nodemap_rec) != 32);
nodemap_id = le32_to_cpu(key->nk_nodemap_id);
- type = nm_idx_get_type(nodemap_id);
+ type = nodemap_get_key_type(key);
nodemap_id = nm_idx_set_type(nodemap_id, 0);
CDEBUG(D_INFO, "found config entry, nm_id %d type %d\n",
" nodemap_id=%d. nodemap config file corrupt?\n",
nodemap_id);
break;
- case NODEMAP_CLUSTER_IDX:
+ case NODEMAP_CLUSTER_IDX: {
+ struct lu_nodemap *old_nm = NULL;
+
nodemap = cfs_hash_lookup(config->nmc_nodemap_hash,
rec->ncr.ncr_name);
if (nodemap == NULL) {
if (nodemap_id == LUSTRE_NODEMAP_DEFAULT_ID) {
nodemap = nodemap_create(rec->ncr.ncr_name,
config, 1);
- config->nmc_default_nodemap = nodemap;
} else {
nodemap = nodemap_create(rec->ncr.ncr_name,
config, 0);
flags & NM_FL_TRUST_CLIENT_IDS;
nodemap->nmf_deny_unknown =
flags & NM_FL_DENY_UNKNOWN;
+ nodemap->nmf_map_uid_only =
+ flags & NM_FL_MAP_UID_ONLY;
+ nodemap->nmf_map_gid_only =
+ flags & NM_FL_MAP_GID_ONLY;
+ nodemap->nmf_enable_audit =
+ flags & NM_FL_ENABLE_AUDIT;
+ nodemap->nmf_forbid_encryption =
+ flags & NM_FL_FORBID_ENCRYPT;
+
+ /* The fileset should be saved otherwise it will be empty
+ * every time in case of "NODEMAP_CLUSTER_IDX". */
+ mutex_lock(&active_config_lock);
+ old_nm = nodemap_lookup(rec->ncr.ncr_name);
+ if (!IS_ERR(old_nm) && old_nm->nm_fileset[0] != '\0')
+ strlcpy(nodemap->nm_fileset, old_nm->nm_fileset,
+ sizeof(nodemap->nm_fileset));
+ mutex_unlock(&active_config_lock);
+ if (!IS_ERR(old_nm))
+ nodemap_putref(old_nm);
if (*recent_nodemap == NULL) {
*recent_nodemap = nodemap;
}
nodemap_putref(nodemap);
break;
+ }
case NODEMAP_RANGE_IDX:
nid[0] = le64_to_cpu(rec->nrr.nrr_start_nid);
nid[1] = le64_to_cpu(rec->nrr.nrr_end_nid);
rc = type;
+ EXIT;
+
out:
return rc;
}
+enum nm_config_passes {
+ NM_READ_CLUSTERS = 0,
+ NM_READ_ATTRIBUTES = 1,
+};
+
static int nodemap_load_entries(const struct lu_env *env,
struct dt_object *nodemap_idx)
{
- const struct dt_it_ops *iops;
- struct dt_it *it;
- struct lu_nodemap *recent_nodemap = NULL;
- struct nodemap_config *new_config = NULL;
- u64 hash = 0;
- bool activate_nodemap = false;
- bool loaded_global_idx = false;
- int rc = 0;
+ const struct dt_it_ops *iops;
+ struct dt_it *it;
+ struct lu_nodemap *recent_nodemap = NULL;
+ struct nodemap_config *new_config = NULL;
+ u64 hash = 0;
+ bool activate_nodemap = false;
+ bool loaded_global_idx = false;
+ enum nm_config_passes cur_pass = NM_READ_CLUSTERS;
+ int rc = 0;
ENTRY;
while (rc == 0) {
struct nodemap_key *key;
union nodemap_rec rec;
+ enum nodemap_idx_type key_type;
key = (struct nodemap_key *)iops->key(env, it);
- rc = iops->rec(env, it, (struct dt_rec *)&rec, 0);
- if (rc != -ESTALE) {
- if (rc != 0)
- GOTO(out_nodemap_config, rc);
- rc = nodemap_process_keyrec(new_config, key, &rec,
- &recent_nodemap);
- if (rc < 0)
- GOTO(out_nodemap_config, rc);
- if (rc == NODEMAP_GLOBAL_IDX)
- loaded_global_idx = true;
+ key_type = nodemap_get_key_type((struct nodemap_key *)key);
+ if ((cur_pass == NM_READ_CLUSTERS &&
+ key_type == NODEMAP_CLUSTER_IDX) ||
+ (cur_pass == NM_READ_ATTRIBUTES &&
+ key_type != NODEMAP_CLUSTER_IDX &&
+ key_type != NODEMAP_EMPTY_IDX)) {
+ rc = iops->rec(env, it, (struct dt_rec *)&rec, 0);
+ if (rc != -ESTALE) {
+ if (rc != 0)
+ GOTO(out_nodemap_config, rc);
+ rc = nodemap_process_keyrec(new_config, key, &rec,
+ &recent_nodemap);
+ if (rc < 0)
+ GOTO(out_nodemap_config, rc);
+ if (rc == NODEMAP_GLOBAL_IDX)
+ loaded_global_idx = true;
+ }
}
do
rc = iops->next(env, it);
while (rc == -ESTALE);
+
+ /* move to second pass */
+ if (rc > 0 && cur_pass == NM_READ_CLUSTERS) {
+ cur_pass = NM_READ_ATTRIBUTES;
+ rc = iops->load(env, it, 0);
+ if (rc == 0)
+ rc = iops->next(env, it);
+ else if (rc > 0)
+ rc = 0;
+ else
+ GOTO(out, rc);
+ }
}
if (rc > 0)
if (new_config->nmc_default_nodemap == NULL) {
/* new MGS won't have a default nm on disk, so create it here */
- new_config->nmc_default_nodemap =
+ struct lu_nodemap *nodemap =
nodemap_create(DEFAULT_NODEMAP, new_config, 1);
- if (IS_ERR(new_config->nmc_default_nodemap)) {
- rc = PTR_ERR(new_config->nmc_default_nodemap);
+ if (IS_ERR(nodemap)) {
+ rc = PTR_ERR(nodemap);
} else {
rc = nodemap_idx_nodemap_add_update(
new_config->nmc_default_nodemap,
+ nodemap_idx,
NM_ADD);
nodemap_putref(new_config->nmc_default_nodemap);
}
/**
* Step through active config and write to disk.
*/
-int nodemap_save_config_cache(const struct lu_env *env,
- struct nm_config_file *ncf)
+struct dt_object *nodemap_save_config_cache(const struct lu_env *env,
+ struct dt_device *dev,
+ struct local_oid_storage *los)
{
- struct dt_device *dev;
struct dt_object *o;
struct lu_nodemap *nodemap;
struct lu_nodemap *nm_tmp;
ENTRY;
- if (ncf->ncf_los == NULL || ncf->ncf_obj == NULL)
- RETURN(-EIO);
-
- dev = lu2dt_dev(ncf->ncf_obj->do_lu.lo_dev);
-
- /* nodemap_cache_find_create will delete old conf file, so put here */
- lu_object_put_nocache(env, &ncf->ncf_obj->do_lu);
- ncf->ncf_obj = NULL;
-
- /* force create a new index file to fill with active config */
- o = nodemap_cache_find_create(env, dev, ncf->ncf_los, true);
+ /* create a new index file to fill with active config */
+ o = nodemap_cache_find_create(env, dev, los, NCFC_CREATE_NEW);
if (IS_ERR(o))
- GOTO(out, rc = PTR_ERR(o));
-
- ncf->ncf_obj = o;
+ RETURN(o);
mutex_lock(&active_config_lock);
list_for_each_entry_safe(range, range_temp, &nodemap->nm_ranges,
rn_list) {
lnet_nid_t nid[2] = {
- range->rn_node.in_extent.start,
- range->rn_node.in_extent.end
+ range->rn_start,
+ range->rn_end
};
nodemap_range_key_init(&nk, nodemap->nm_id,
range->rn_id);
if (rc2 < 0)
rc = rc2;
-out:
mutex_unlock(&active_config_lock);
- RETURN(rc);
+
+ if (rc < 0) {
+ dt_object_put(env, o);
+ o = ERR_PTR(rc);
+ }
+
+ RETURN(o);
}
static void nodemap_save_all_caches(void)
mutex_lock(&ncf_list_lock);
list_for_each_entry(ncf, &ncf_list_head, ncf_list) {
- rc = nodemap_save_config_cache(&env, ncf);
- if (rc < 0 && ncf->ncf_obj != NULL)
+ struct dt_device *dev = lu2dt_dev(ncf->ncf_obj->do_lu.lo_dev);
+ struct obd_device *obd = ncf->ncf_obj->do_lu.lo_dev->ld_obd;
+ struct dt_object *o;
+
+ /* put current config file so save conf can rewrite it */
+ dt_object_put_nocache(&env, ncf->ncf_obj);
+ ncf->ncf_obj = NULL;
+
+ o = nodemap_save_config_cache(&env, dev, ncf->ncf_los);
+ if (IS_ERR(o))
CWARN("%s: error writing to nodemap config: rc = %d\n",
- ncf->ncf_obj->do_lu.lo_dev->ld_obd->obd_name, rc);
+ obd->obd_name, rc);
+ else
+ ncf->ncf_obj = o;
}
mutex_unlock(&ncf_list_lock);
* \retval -ENOMEM memory allocation failure
* \retval -ENOENT error loading nodemap config
* \retval -EINVAL error loading nodemap config
+ * \retval -EEXIST nodemap config already registered for MGS
*/
-struct nm_config_file *nm_config_file_register(const struct lu_env *env,
- struct dt_object *obj,
- struct local_oid_storage *los,
- enum nm_config_file_type ncf_type)
+struct nm_config_file *nm_config_file_register_mgs(const struct lu_env *env,
+ struct dt_object *obj,
+ struct local_oid_storage *los)
{
struct nm_config_file *ncf;
- bool save_config = false;
int rc = 0;
ENTRY;
+ if (nodemap_mgs_ncf != NULL)
+ GOTO(out, ncf = ERR_PTR(-EEXIST));
+
OBD_ALLOC_PTR(ncf);
if (ncf == NULL)
- RETURN(ERR_PTR(-ENOMEM));
+ GOTO(out, ncf = ERR_PTR(-ENOMEM));
+
+ /* if loading from cache, prevent activation of MGS config until cache
+ * loading is done, so disk config is overwritten by MGS config.
+ */
+ mutex_lock(&nodemap_config_loaded_lock);
+ rc = nodemap_load_entries(env, obj);
+ if (!rc)
+ nodemap_config_loaded = true;
+ mutex_unlock(&nodemap_config_loaded_lock);
+
+ if (rc) {
+ OBD_FREE_PTR(ncf);
+ GOTO(out, ncf = ERR_PTR(rc));
+ }
+
+ lu_object_get(&obj->do_lu);
ncf->ncf_obj = obj;
ncf->ncf_los = los;
- if (ncf_type == NCFT_MGS) {
- nodemap_mgs_ncf = ncf;
- } else {
- mutex_lock(&ncf_list_lock);
- list_add(&ncf->ncf_list, &ncf_list_head);
- mutex_unlock(&ncf_list_lock);
- }
+ nodemap_mgs_ncf = ncf;
- /* prevent activation of config loaded from MGS until disk is loaded
- * so disk config is overwritten by MGS config.
- */
+out:
+ return ncf;
+}
+EXPORT_SYMBOL(nm_config_file_register_mgs);
+
+struct nm_config_file *nm_config_file_register_tgt(const struct lu_env *env,
+ struct dt_device *dev,
+ struct local_oid_storage *los)
+{
+ struct nm_config_file *ncf;
+ struct dt_object *config_obj = NULL;
+ int rc = 0;
+
+ OBD_ALLOC_PTR(ncf);
+ if (ncf == NULL)
+ RETURN(ERR_PTR(-ENOMEM));
+
+ /* don't load from cache if config already loaded */
mutex_lock(&nodemap_config_loaded_lock);
- if (ncf_type == NCFT_MGS || !nodemap_config_loaded)
- rc = nodemap_load_entries(env, obj);
- else
- save_config = true;
- nodemap_config_loaded = true;
+ if (!nodemap_config_loaded) {
+ config_obj = nodemap_cache_find_create(env, dev, los, 0);
+ if (IS_ERR(config_obj))
+ rc = PTR_ERR(config_obj);
+ else
+ rc = nodemap_load_entries(env, config_obj);
+
+ if (!rc)
+ nodemap_config_loaded = true;
+ }
mutex_unlock(&nodemap_config_loaded_lock);
+ if (rc)
+ GOTO(out_ncf, rc);
- /* sync on disk caches with loaded config in memory */
- if (save_config)
- rc = nodemap_save_config_cache(env, ncf);
+ /* sync on disk caches w/ loaded config in memory, ncf_obj may change */
+ if (!config_obj) {
+ config_obj = nodemap_save_config_cache(env, dev, los);
+ if (IS_ERR(config_obj))
+ GOTO(out_ncf, rc = PTR_ERR(config_obj));
+ }
- if (rc < 0) {
- if (ncf_type == NCFT_MGS) {
- nodemap_mgs_ncf = NULL;
- } else {
- mutex_lock(&ncf_list_lock);
- list_del(&ncf->ncf_list);
- mutex_unlock(&ncf_list_lock);
- }
+ ncf->ncf_obj = config_obj;
+ ncf->ncf_los = los;
+
+ mutex_lock(&ncf_list_lock);
+ list_add(&ncf->ncf_list, &ncf_list_head);
+ mutex_unlock(&ncf_list_lock);
+out_ncf:
+ if (rc) {
OBD_FREE_PTR(ncf);
RETURN(ERR_PTR(rc));
}
RETURN(ncf);
}
-EXPORT_SYMBOL(nm_config_file_register);
+EXPORT_SYMBOL(nm_config_file_register_tgt);
/**
* Deregister a nm_config_file. Should be called by targets during cleanup.
*
* \param ncf config file to deregister
*/
-void nm_config_file_deregister(const struct lu_env *env,
- struct nm_config_file *ncf,
- enum nm_config_file_type ncf_type)
+void nm_config_file_deregister_mgs(const struct lu_env *env,
+ struct nm_config_file *ncf)
{
ENTRY;
+ LASSERT(nodemap_mgs_ncf == ncf);
+ nodemap_mgs_ncf = NULL;
if (ncf->ncf_obj)
- lu_object_put(env, &ncf->ncf_obj->do_lu);
+ dt_object_put(env, ncf->ncf_obj);
+
+ OBD_FREE_PTR(ncf);
+
+ EXIT;
+}
+EXPORT_SYMBOL(nm_config_file_deregister_mgs);
+
+void nm_config_file_deregister_tgt(const struct lu_env *env,
+ struct nm_config_file *ncf)
+{
+ ENTRY;
+
+ if (ncf == NULL)
+ return;
+
+ mutex_lock(&ncf_list_lock);
+ list_del(&ncf->ncf_list);
+ mutex_unlock(&ncf_list_lock);
+
+ if (ncf->ncf_obj)
+ dt_object_put(env, ncf->ncf_obj);
- if (ncf_type == NCFT_TGT) {
- mutex_lock(&ncf_list_lock);
- list_del(&ncf->ncf_list);
- mutex_unlock(&ncf_list_lock);
- } else {
- nodemap_mgs_ncf = NULL;
- }
OBD_FREE_PTR(ncf);
EXIT;
}
-EXPORT_SYMBOL(nm_config_file_deregister);
+EXPORT_SYMBOL(nm_config_file_deregister_tgt);
int nodemap_process_idx_pages(struct nodemap_config *config, union lu_page *lip,
struct lu_nodemap **recent_nodemap)
}
EXPORT_SYMBOL(nodemap_process_idx_pages);
+static int nodemap_page_build(const struct lu_env *env, union lu_page *lp,
+ size_t nob, const struct dt_it_ops *iops,
+ struct dt_it *it, __u32 attr, void *arg)
+{
+ struct idx_info *ii = (struct idx_info *)arg;
+ struct lu_idxpage *lip = &lp->lp_idx;
+ char *entry;
+ size_t size = ii->ii_keysize + ii->ii_recsize;
+ int rc;
+ ENTRY;
+
+ if (nob < LIP_HDR_SIZE)
+ return -EINVAL;
+
+ /* initialize the header of the new container */
+ memset(lip, 0, LIP_HDR_SIZE);
+ lip->lip_magic = LIP_MAGIC;
+ nob -= LIP_HDR_SIZE;
+
+ entry = lip->lip_entries;
+ do {
+ char *tmp_entry = entry;
+ struct dt_key *key;
+ __u64 hash;
+ enum nodemap_idx_type key_type;
+
+ /* fetch 64-bit hash value */
+ hash = iops->store(env, it);
+ ii->ii_hash_end = hash;
+
+ if (OBD_FAIL_CHECK(OBD_FAIL_OBD_IDX_READ_BREAK)) {
+ if (lip->lip_nr != 0)
+ GOTO(out, rc = 0);
+ }
+
+ if (nob < size) {
+ if (lip->lip_nr == 0)
+ GOTO(out, rc = -EINVAL);
+ GOTO(out, rc = 0);
+ }
+
+ key = iops->key(env, it);
+ key_type = nodemap_get_key_type((struct nodemap_key *)key);
+
+ /* on the first pass, get only the cluster types. On second
+ * pass, get all the rest */
+ if ((ii->ii_attrs == NM_READ_CLUSTERS &&
+ key_type == NODEMAP_CLUSTER_IDX) ||
+ (ii->ii_attrs == NM_READ_ATTRIBUTES &&
+ key_type != NODEMAP_CLUSTER_IDX &&
+ key_type != NODEMAP_EMPTY_IDX)) {
+ memcpy(tmp_entry, key, ii->ii_keysize);
+ tmp_entry += ii->ii_keysize;
+
+ /* and finally the record */
+ rc = iops->rec(env, it, (struct dt_rec *)tmp_entry,
+ attr);
+ if (rc != -ESTALE) {
+ if (rc != 0)
+ GOTO(out, rc);
+
+ /* hash/key/record successfully copied! */
+ lip->lip_nr++;
+ if (unlikely(lip->lip_nr == 1 &&
+ ii->ii_count == 0))
+ ii->ii_hash_start = hash;
+
+ entry = tmp_entry + ii->ii_recsize;
+ nob -= size;
+ }
+ }
+
+ /* move on to the next record */
+ do {
+ rc = iops->next(env, it);
+ } while (rc == -ESTALE);
+
+ /* move to second pass */
+ if (rc > 0 && ii->ii_attrs == NM_READ_CLUSTERS) {
+ ii->ii_attrs = NM_READ_ATTRIBUTES;
+ rc = iops->load(env, it, 0);
+ if (rc == 0)
+ rc = iops->next(env, it);
+ else if (rc > 0)
+ rc = 0;
+ else
+ GOTO(out, rc);
+ }
+
+ } while (rc == 0);
+
+ GOTO(out, rc);
+out:
+ if (rc >= 0 && lip->lip_nr > 0)
+ /* one more container */
+ ii->ii_count++;
+ if (rc > 0)
+ /* no more entries */
+ ii->ii_hash_end = II_END_OFF;
+ return rc;
+}
+
+
int nodemap_index_read(struct lu_env *env,
struct nm_config_file *ncf,
struct idx_info *ii,
dt_read_lock(env, nodemap_idx, 0);
version = dt_version_get(env, nodemap_idx);
if (rdpg->rp_hash != 0 && ii->ii_version != version) {
- CDEBUG(D_INFO, "nodemap config changed while sending, "
- "old "LPU64", new "LPU64"\n",
+ CDEBUG(D_INFO, "nodemap config changed inflight, old %llu, new %llu\n",
ii->ii_version,
version);
ii->ii_hash_end = 0;
} else {
- rc = dt_index_walk(env, nodemap_idx, rdpg, NULL, ii);
+ rc = dt_index_walk(env, nodemap_idx, rdpg, nodemap_page_build,
+ ii);
CDEBUG(D_INFO, "walked index, hashend %llx\n", ii->ii_hash_end);
}
int nodemap_get_config_req(struct obd_device *mgs_obd,
struct ptlrpc_request *req)
{
+ const struct ptlrpc_bulk_frag_ops *frag_ops = &ptlrpc_bulk_kiov_pin_ops;
struct mgs_config_body *body;
struct mgs_config_res *res;
struct lu_rdpg rdpg;
struct idx_info nodemap_ii;
struct ptlrpc_bulk_desc *desc;
- struct l_wait_info lwi;
struct tg_export_data *rqexp_ted = &req->rq_export->exp_target_data;
int i;
int page_count;
body->mcb_name, rdpg.rp_count);
/* allocate pages to store the containers */
- OBD_ALLOC(rdpg.rp_pages, sizeof(*rdpg.rp_pages) * rdpg.rp_npages);
+ OBD_ALLOC_PTR_ARRAY(rdpg.rp_pages, rdpg.rp_npages);
if (rdpg.rp_pages == NULL)
RETURN(-ENOMEM);
for (i = 0; i < rdpg.rp_npages; i++) {
nodemap_ii.ii_magic = IDX_INFO_MAGIC;
nodemap_ii.ii_flags = II_FL_NOHASH;
nodemap_ii.ii_version = rqexp_ted->ted_nodemap_version;
+ nodemap_ii.ii_attrs = body->mcb_nm_cur_pass;
bytes = nodemap_index_read(req->rq_svc_thread->t_env,
mgs_obd->u.obt.obt_nodemap_config_file,
if (res == NULL)
GOTO(out, rc = -EINVAL);
res->mcr_offset = nodemap_ii.ii_hash_end;
- res->mcr_size = bytes;
+ res->mcr_nm_cur_pass = nodemap_ii.ii_attrs;
page_count = (bytes + PAGE_SIZE - 1) >> PAGE_SHIFT;
LASSERT(page_count <= rdpg.rp_count);
desc = ptlrpc_prep_bulk_exp(req, page_count, 1,
- PTLRPC_BULK_PUT_SOURCE |
- PTLRPC_BULK_BUF_KIOV,
- MGS_BULK_PORTAL,
- &ptlrpc_bulk_kiov_pin_ops);
+ PTLRPC_BULK_PUT_SOURCE,
+ MGS_BULK_PORTAL, frag_ops);
if (desc == NULL)
GOTO(out, rc = -ENOMEM);
for (i = 0; i < page_count && bytes > 0; i++) {
- ptlrpc_prep_bulk_page_pin(desc, rdpg.rp_pages[i], 0,
- min_t(int, bytes, PAGE_SIZE));
+ frag_ops->add_kiov_frag(desc, rdpg.rp_pages[i], 0,
+ min_t(int, bytes, PAGE_SIZE));
bytes -= PAGE_SIZE;
}
- rc = target_bulk_io(req->rq_export, desc, &lwi);
+ rc = target_bulk_io(req->rq_export, desc);
ptlrpc_free_bulk(desc);
out:
for (i = 0; i < rdpg.rp_npages; i++)
if (rdpg.rp_pages[i] != NULL)
__free_page(rdpg.rp_pages[i]);
- OBD_FREE(rdpg.rp_pages,
- rdpg.rp_npages * sizeof(rdpg.rp_pages[0]));
+ OBD_FREE_PTR_ARRAY(rdpg.rp_pages, rdpg.rp_npages);
}
return rc;
}
EXPORT_SYMBOL(nodemap_get_config_req);
-
-int nodemap_fs_init(const struct lu_env *env, struct dt_device *dev,
- struct obd_device *obd, struct local_oid_storage *los)
-{
- struct dt_object *config_obj;
- struct nm_config_file *nm_config_file;
- int rc = 0;
- ENTRY;
-
- CDEBUG(D_INFO, "%s: finding nodemap index\n", obd->obd_name);
- /* load or create the index file from disk (don't force create) */
- config_obj = nodemap_cache_find_create(env, dev, los, false);
- if (IS_ERR(config_obj))
- GOTO(out, rc = PTR_ERR(config_obj));
-
- CDEBUG(D_INFO, "%s: registering nodemap index\n", obd->obd_name);
-
- nm_config_file = nm_config_file_register(env, config_obj, los,
- NCFT_TGT);
- if (IS_ERR(nm_config_file)) {
- CERROR("%s: error loading nodemap config file, file must be "
- "removed via ldiskfs: rc = %ld\n",
- obd->obd_name, PTR_ERR(nm_config_file));
- GOTO(out, rc = PTR_ERR(nm_config_file));
- }
-
- obd->u.obt.obt_nodemap_config_file = nm_config_file;
-
- /* save los in case object needs to be re-created */
- nm_config_file->ncf_los = los;
-
- EXIT;
-
-out:
- return rc;
-}
-EXPORT_SYMBOL(nodemap_fs_init);
-
-void nodemap_fs_fini(const struct lu_env *env, struct obd_device *obd)
-{
- if (obd->u.obt.obt_nodemap_config_file == NULL)
- return;
-
- nm_config_file_deregister(env, obd->u.obt.obt_nodemap_config_file,
- NCFT_TGT);
- obd->u.obt.obt_nodemap_config_file = NULL;
-}
-EXPORT_SYMBOL(nodemap_fs_fini);