andy/glusterfs
1
0
Fork 0
Code Issues Pull Requests Releases Wiki Activity

dht: make nufa/switch call dht's init/fini

Browse Source 
These functions keep changing as new functionality is added, so copying
and pasting the code is not a good solution.  This way ensures that all
fields get initialized properly no matter how much new stuff we throw in.

Change-Id: I9e9b043d2d305d31e80cf5689465555b70312756
BUG: 924488
Signed-off-by: Jeff Darcy <jdarcy@redhat.com>
Reviewed-on: http://review.gluster.org/4710
Tested-by: Gluster Build System <jenkins@build.gluster.com>
Reviewed-by: Anand Avati <avati@redhat.com>
This commit is contained in:
Jeff Darcy 2013-03-21 19:11:25 -04:00 committed by Anand Avati
parent 42a9d608d5
commit 0106fce7fe
8 changed files with 806 additions and 1046 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

27
libglusterfs/src/xlator.c
View File

@ -225,8 +225,10 @@ xlator_dynload (xlator_t *xl)
*/ */
vtbl = dlsym(handle,"class_methods"); vtbl = dlsym(handle,"class_methods");
if (vtbl) { if (vtbl) {
xl->init = vtbl->init; xl->init = vtbl->init;
xl->fini = vtbl->fini; xl->fini = vtbl->fini;
xl->reconfigure = vtbl->reconfigure;
xl->notify = vtbl->notify;
} }
else { else {
if (!(*VOID(&xl->init) = dlsym (handle, "init"))) { if (!(*VOID(&xl->init) = dlsym (handle, "init"))) {
@ -240,11 +242,18 @@ xlator_dynload (xlator_t *xl)
dlerror ()); dlerror ());
goto out; goto out;
} }
} if (!(*VOID(&(xl->reconfigure)) = dlsym (handle,
"reconfigure"))) {
gf_log ("xlator", GF_LOG_TRACE,
"dlsym(reconfigure) on %s -- neglecting",
dlerror());
}
if (!(*VOID(&(xl->notify)) = dlsym (handle, "notify"))) {
gf_log ("xlator", GF_LOG_TRACE,
"dlsym(notify) on %s -- neglecting",
dlerror ());
}
if (!(*VOID(&(xl->notify)) = dlsym (handle, "notify"))) {
gf_log ("xlator", GF_LOG_TRACE,
"dlsym(notify) on %s -- neglecting", dlerror ());
} }
if (!(xl->dumpops = dlsym (handle, "dumpops"))) { if (!(xl->dumpops = dlsym (handle, "dumpops"))) {
@ -258,12 +267,6 @@ xlator_dynload (xlator_t *xl)
dlerror ()); dlerror ());
} }
if (!(*VOID(&(xl->reconfigure)) = dlsym (handle, "reconfigure"))) {
gf_log ("xlator", GF_LOG_TRACE,
"dlsym(reconfigure) on %s -- neglecting",
dlerror());
}
vol_opt = GF_CALLOC (1, sizeof (volume_opt_list_t), vol_opt = GF_CALLOC (1, sizeof (volume_opt_list_t),
gf_common_mt_volume_opt_list_t); gf_common_mt_volume_opt_list_t);

7
libglusterfs/src/xlator.h
View File

@ -828,8 +828,11 @@ struct _xlator {
}; };
typedef struct { typedef struct {
int32_t (*init) (xlator_t *this); int32_t (*init) (xlator_t *this);
void (*fini) (xlator_t *this); void (*fini) (xlator_t *this);
int32_t (*reconfigure) (xlator_t *this,
dict_t *options);
event_notify_fn_t notify;
} class_methods_t; } class_methods_t;
#define xlator_has_parent(xl) (xl->parents != NULL) #define xlator_has_parent(xl) (xl->parents != NULL)

2
xlators/cluster/dht/src/Makefile.am
View File

@ -4,7 +4,7 @@ xlatordir = $(libdir)/glusterfs/$(PACKAGE_VERSION)/xlator/cluster
dht_common_source = dht-layout.c dht-helper.c dht-linkfile.c dht-rebalance.c \ dht_common_source = dht-layout.c dht-helper.c dht-linkfile.c dht-rebalance.c \
dht-selfheal.c dht-rename.c dht-hashfn.c dht-diskusage.c \ dht-selfheal.c dht-rename.c dht-hashfn.c dht-diskusage.c \
dht-common.c dht-inode-write.c dht-inode-read.c \ dht-common.c dht-inode-write.c dht-inode-read.c dht-shared.c \
$(top_builddir)/xlators/lib/src/libxlator.c $(top_builddir)/xlators/lib/src/libxlator.c
dht_la_SOURCES = $(dht_common_source) dht.c dht_la_SOURCES = $(dht_common_source) dht.c

11
xlators/cluster/dht/src/dht-common.h
View File

@ -684,6 +684,9 @@ int32_t dht_setattr (call_frame_t *frame, xlator_t *this, loc_t *loc,
int32_t dht_fsetattr (call_frame_t *frame, xlator_t *this, fd_t *fd, int32_t dht_fsetattr (call_frame_t *frame, xlator_t *this, fd_t *fd,
struct iatt *stbuf, int32_t valid, dict_t *xdata); struct iatt *stbuf, int32_t valid, dict_t *xdata);
int32_t dht_init (xlator_t *this);
void dht_fini (xlator_t *this);
int dht_reconfigure (xlator_t *this, dict_t *options);
int32_t dht_notify (xlator_t *this, int32_t event, void *data, ...); int32_t dht_notify (xlator_t *this, int32_t event, void *data, ...);
/* definitions for nufa/switch */ /* definitions for nufa/switch */
@ -754,4 +757,12 @@ xlator_t *
dht_subvol_maxspace_nonzeroinode (xlator_t *this, xlator_t *subvol); dht_subvol_maxspace_nonzeroinode (xlator_t *this, xlator_t *subvol);
int int
dht_linkfile_attr_heal (call_frame_t *frame, xlator_t *this); dht_linkfile_attr_heal (call_frame_t *frame, xlator_t *this);
void
dht_layout_dump (dht_layout_t *layout, const char *prefix);
int32_t
dht_priv_dump (xlator_t *this);
int32_t
dht_inodectx_dump (xlator_t *this, inode_t *inode);
#endif/* _DHT_H */ #endif/* _DHT_H */

732
xlators/cluster/dht/src/dht-shared.c Normal file
View File

@ -0,0 +1,732 @@
/*
Copyright (c) 2008-2012 Red Hat, Inc. <http://www.redhat.com>
This file is part of GlusterFS.
This file is licensed to you under your choice of the GNU Lesser
General Public License, version 3 or any later version (LGPLv3 or
later), or the GNU General Public License, version 2 (GPLv2), in all
cases as published by the Free Software Foundation.
*/
#ifndef _CONFIG_H
#define _CONFIG_H
#include "config.h"
#endif
/* TODO: add NS locking */
#include "statedump.h"
#include "dht-common.h"
/* TODO:
- use volumename in xattr instead of "dht"
- use NS locks
- handle all cases in self heal layout reconstruction
- complete linkfile selfheal
*/
struct volume_options options[];
void
dht_layout_dump (dht_layout_t *layout, const char *prefix)
{
char key[GF_DUMP_MAX_BUF_LEN];
int i = 0;
if (!layout)
goto out;
if (!prefix)
goto out;
gf_proc_dump_build_key(key, prefix, "cnt");
gf_proc_dump_write(key, "%d", layout->cnt);
gf_proc_dump_build_key(key, prefix, "preset");
gf_proc_dump_write(key, "%d", layout->preset);
gf_proc_dump_build_key(key, prefix, "gen");
gf_proc_dump_write(key, "%d", layout->gen);
if (layout->type != IA_INVAL) {
gf_proc_dump_build_key(key, prefix, "inode type");
gf_proc_dump_write(key, "%d", layout->type);
}
if (!IA_ISDIR (layout->type))
goto out;
for (i = 0; i < layout->cnt; i++) {
gf_proc_dump_build_key(key, prefix,"list[%d].err", i);
gf_proc_dump_write(key, "%d", layout->list[i].err);
gf_proc_dump_build_key(key, prefix,"list[%d].start", i);
gf_proc_dump_write(key, "%u", layout->list[i].start);
gf_proc_dump_build_key(key, prefix,"list[%d].stop", i);
gf_proc_dump_write(key, "%u", layout->list[i].stop);
if (layout->list[i].xlator) {
gf_proc_dump_build_key(key, prefix,
"list[%d].xlator.type", i);
gf_proc_dump_write(key, "%s",
layout->list[i].xlator->type);
gf_proc_dump_build_key(key, prefix,
"list[%d].xlator.name", i);
gf_proc_dump_write(key, "%s",
layout->list[i].xlator->name);
}
}
out:
return;
}
int32_t
dht_priv_dump (xlator_t *this)
{
char key_prefix[GF_DUMP_MAX_BUF_LEN];
char key[GF_DUMP_MAX_BUF_LEN];
int i = 0;
dht_conf_t *conf = NULL;
int ret = -1;
if (!this)
goto out;
conf = this->private;
if (!conf)
goto out;
ret = TRY_LOCK(&conf->subvolume_lock);
if (ret != 0) {
return ret;
}
gf_proc_dump_add_section("xlator.cluster.dht.%s.priv", this->name);
gf_proc_dump_build_key(key_prefix,"xlator.cluster.dht","%s.priv",
this->name);
gf_proc_dump_write("subvol_cnt","%d", conf->subvolume_cnt);
for (i = 0; i < conf->subvolume_cnt; i++) {
sprintf (key, "subvolumes[%d]", i);
gf_proc_dump_write(key, "%s.%s", conf->subvolumes[i]->type,
conf->subvolumes[i]->name);
if (conf->file_layouts && conf->file_layouts[i]){
sprintf (key, "file_layouts[%d]", i);
dht_layout_dump(conf->file_layouts[i], key);
}
if (conf->dir_layouts && conf->dir_layouts[i]) {
sprintf (key, "dir_layouts[%d]", i);
dht_layout_dump(conf->dir_layouts[i], key);
}
if (conf->subvolume_status) {
sprintf (key, "subvolume_status[%d]", i);
gf_proc_dump_write(key, "%d",
(int)conf->subvolume_status[i]);
}
}
gf_proc_dump_write("search_unhashed", "%d", conf->search_unhashed);
gf_proc_dump_write("gen", "%d", conf->gen);
gf_proc_dump_write("min_free_disk", "%lf", conf->min_free_disk);
gf_proc_dump_write("min_free_inodes", "%lf", conf->min_free_inodes);
gf_proc_dump_write("disk_unit", "%c", conf->disk_unit);
gf_proc_dump_write("refresh_interval", "%d", conf->refresh_interval);
gf_proc_dump_write("unhashed_sticky_bit", "%d", conf->unhashed_sticky_bit);
if (conf ->du_stats) {
gf_proc_dump_write("du_stats.avail_percent", "%lf",
conf->du_stats->avail_percent);
gf_proc_dump_write("du_stats.avail_space", "%lu",
conf->du_stats->avail_space);
gf_proc_dump_write("du_stats.avail_inodes", "%lf",
conf->du_stats->avail_inodes);
gf_proc_dump_write("du_stats.log", "%lu", conf->du_stats->log);
}
if (conf->last_stat_fetch.tv_sec)
gf_proc_dump_write("last_stat_fetch", "%s",
ctime(&conf->last_stat_fetch.tv_sec));
UNLOCK(&conf->subvolume_lock);
out:
return ret;
}
int32_t
dht_inodectx_dump (xlator_t *this, inode_t *inode)
{
int ret = -1;
dht_layout_t *layout = NULL;
if (!this)
goto out;
if (!inode)
goto out;
ret = dht_inode_ctx_layout_get (inode, this, &layout);
if ((ret != 0) || !layout)
return ret;
gf_proc_dump_add_section("xlator.cluster.dht.%s.inode", this->name);
dht_layout_dump(layout, "layout");
out:
return ret;
}
void
dht_fini (xlator_t *this)
{
int i = 0;
dht_conf_t *conf = NULL;
GF_VALIDATE_OR_GOTO ("dht", this, out);
conf = this->private;
this->private = NULL;
if (conf) {
if (conf->file_layouts) {
for (i = 0; i < conf->subvolume_cnt; i++) {
GF_FREE (conf->file_layouts[i]);
}
GF_FREE (conf->file_layouts);
}
GF_FREE (conf->subvolumes);
GF_FREE (conf->subvolume_status);
GF_FREE (conf);
}
out:
return;
}
int32_t
mem_acct_init (xlator_t *this)
{
int ret = -1;
GF_VALIDATE_OR_GOTO ("dht", this, out);
ret = xlator_mem_acct_init (this, gf_dht_mt_end + 1);
if (ret != 0) {
gf_log (this->name, GF_LOG_ERROR, "Memory accounting init"
"failed");
return ret;
}
out:
return ret;
}
int
dht_parse_decommissioned_bricks (xlator_t *this, dht_conf_t *conf,
const char *bricks)
{
int i = 0;
int ret = -1;
char *tmpstr = NULL;
char *dup_brick = NULL;
char *node = NULL;
if (!conf || !bricks)
goto out;
dup_brick = gf_strdup (bricks);
node = strtok_r (dup_brick, ",", &tmpstr);
while (node) {
for (i = 0; i < conf->subvolume_cnt; i++) {
if (!strcmp (conf->subvolumes[i]->name, node)) {
conf->decommissioned_bricks[i] =
conf->subvolumes[i];
conf->decommission_subvols_cnt++;
gf_log (this->name, GF_LOG_INFO,
"decommissioning subvolume %s",
conf->subvolumes[i]->name);
break;
}
}
if (i == conf->subvolume_cnt) {
/* Wrong node given. */
goto out;
}
node = strtok_r (NULL, ",", &tmpstr);
}
ret = 0;
conf->decommission_in_progress = 1;
out:
GF_FREE (dup_brick);
return ret;
}
void
dht_init_regex (xlator_t *this, dict_t *odict, char *name,
regex_t *re, gf_boolean_t *re_valid)
{
char *temp_str;
if (dict_get_str (odict, name, &temp_str) != 0) {
if (strcmp(name,"rsync-hash-regex")) {
return;
}
temp_str = "^\\.(.+)\\.[^.]+$";
}
if (*re_valid) {
regfree(re);
*re_valid = _gf_false;
}
if (!strcmp(temp_str,"none")) {
return;
}
if (regcomp(re,temp_str,REG_EXTENDED) == 0) {
gf_log (this->name, GF_LOG_INFO,
"using regex %s = %s", name, temp_str);
*re_valid = _gf_true;
}
else {
gf_log (this->name, GF_LOG_WARNING,
"compiling regex %s failed", temp_str);
}
}
int
dht_reconfigure (xlator_t *this, dict_t *options)
{
dht_conf_t *conf = NULL;
char *temp_str = NULL;
gf_boolean_t search_unhashed;
int ret = -1;
GF_VALIDATE_OR_GOTO ("dht", this, out);
GF_VALIDATE_OR_GOTO ("dht", options, out);
conf = this->private;
if (!conf)
return 0;
if (dict_get_str (options, "lookup-unhashed", &temp_str) == 0) {
/* If option is not "auto", other options _should_ be boolean*/
if (strcasecmp (temp_str, "auto")) {
if (!gf_string2boolean (temp_str, &search_unhashed)) {
gf_log(this->name, GF_LOG_DEBUG, "Reconfigure:"
" lookup-unhashed reconfigured (%s)",
temp_str);
conf->search_unhashed = search_unhashed;
} else {
gf_log(this->name, GF_LOG_ERROR, "Reconfigure:"
" lookup-unhashed should be boolean,"
" not (%s), defaulting to (%d)",
temp_str, conf->search_unhashed);
//return -1;
ret = -1;
goto out;
}
} else {
gf_log(this->name, GF_LOG_DEBUG, "Reconfigure:"
" lookup-unhashed reconfigured auto ");
conf->search_unhashed = GF_DHT_LOOKUP_UNHASHED_AUTO;
}
}
GF_OPTION_RECONF ("min-free-disk", conf->min_free_disk, options,
percent_or_size, out);
/* option can be any one of percent or bytes */
conf->disk_unit = 0;
if (conf->min_free_disk < 100.0)
conf->disk_unit = 'p';
GF_OPTION_RECONF ("min-free-inodes", conf->min_free_inodes, options,
percent, out);
GF_OPTION_RECONF ("directory-layout-spread", conf->dir_spread_cnt,
options, uint32, out);
GF_OPTION_RECONF ("readdir-optimize", conf->readdir_optimize, options,
bool, out);
if (conf->defrag) {
GF_OPTION_RECONF ("rebalance-stats", conf->defrag->stats,
options, bool, out);
}
if (dict_get_str (options, "decommissioned-bricks", &temp_str) == 0) {
ret = dht_parse_decommissioned_bricks (this, conf, temp_str);
if (ret == -1)
goto out;
}
dht_init_regex (this, options, "rsync-hash-regex",
&conf->rsync_regex, &conf->rsync_regex_valid);
dht_init_regex (this, options, "extra-hash-regex",
&conf->extra_regex, &conf->extra_regex_valid);
ret = 0;
out:
return ret;
}
static int
gf_defrag_pattern_list_fill (xlator_t *this, gf_defrag_info_t *defrag, char *data)
{
int ret = -1;
char *tmp_str = NULL;
char *tmp_str1 = NULL;
char *dup_str = NULL;
char *num = NULL;
char *pattern_str = NULL;
char *pattern = NULL;
gf_defrag_pattern_list_t *temp_list = NULL;
gf_defrag_pattern_list_t *pattern_list = NULL;
if (!this || !defrag || !data)
goto out;
/* Get the pattern for pattern list. "pattern:<optional-size>"
* eg: *avi, *pdf:10MB, *:1TB
*/
pattern_str = strtok_r (data, ",", &tmp_str);
while (pattern_str) {
dup_str = gf_strdup (pattern_str);
pattern_list = GF_CALLOC (1, sizeof (gf_defrag_pattern_list_t),
1);
if (!pattern_list) {
goto out;
}
pattern = strtok_r (dup_str, ":", &tmp_str1);
num = strtok_r (NULL, ":", &tmp_str1);
if (!pattern)
goto out;
if (!num) {
if (gf_string2bytesize(pattern, &pattern_list->size)
== 0) {
pattern = "*";
}
} else if (gf_string2bytesize (num, &pattern_list->size) != 0) {
gf_log (this->name, GF_LOG_ERROR,
"invalid number format \"%s\"", num);
goto out;
}
memcpy (pattern_list->path_pattern, pattern, strlen (dup_str));
if (!defrag->defrag_pattern)
temp_list = NULL;
else
temp_list = defrag->defrag_pattern;
pattern_list->next = temp_list;
defrag->defrag_pattern = pattern_list;
pattern_list = NULL;
GF_FREE (dup_str);
dup_str = NULL;
pattern_str = strtok_r (NULL, ",", &tmp_str);
}
ret = 0;
out:
if (ret)
GF_FREE (pattern_list);
GF_FREE (dup_str);
return ret;
}
int
dht_init (xlator_t *this)
{
dht_conf_t *conf = NULL;
char *temp_str = NULL;
int ret = -1;
int i = 0;
gf_defrag_info_t *defrag = NULL;
int cmd = 0;
char *node_uuid = NULL;
GF_VALIDATE_OR_GOTO ("dht", this, err);
if (!this->children) {
gf_log (this->name, GF_LOG_CRITICAL,
"Distribute needs more than one subvolume");
return -1;
}
if (!this->parents) {
gf_log (this->name, GF_LOG_WARNING,
"dangling volume. check volfile");
}
conf = GF_CALLOC (1, sizeof (*conf), gf_dht_mt_dht_conf_t);
if (!conf) {
goto err;
}
ret = dict_get_int32 (this->options, "rebalance-cmd", &cmd);
if (cmd) {
defrag = GF_CALLOC (1, sizeof (gf_defrag_info_t),
gf_defrag_info_mt);
GF_VALIDATE_OR_GOTO (this->name, defrag, err);
LOCK_INIT (&defrag->lock);
defrag->is_exiting = 0;
conf->defrag = defrag;
ret = dict_get_str (this->options, "node-uuid", &node_uuid);
if (ret) {
gf_log (this->name, GF_LOG_ERROR, "node-uuid not "
"specified");
goto err;
}
if (uuid_parse (node_uuid, defrag->node_uuid)) {
gf_log (this->name, GF_LOG_ERROR, "Cannot parse "
"glusterd node uuid");
goto err;
}
defrag->cmd = cmd;
defrag->stats = _gf_false;
}
conf->search_unhashed = GF_DHT_LOOKUP_UNHASHED_ON;
if (dict_get_str (this->options, "lookup-unhashed", &temp_str) == 0) {
/* If option is not "auto", other options _should_ be boolean */
if (strcasecmp (temp_str, "auto"))
gf_string2boolean (temp_str, &conf->search_unhashed);
else
conf->search_unhashed = GF_DHT_LOOKUP_UNHASHED_AUTO;
}
GF_OPTION_INIT ("unhashed-sticky-bit", conf->unhashed_sticky_bit, bool,
err);
GF_OPTION_INIT ("use-readdirp", conf->use_readdirp, bool, err);
GF_OPTION_INIT ("min-free-disk", conf->min_free_disk, percent_or_size,
err);
GF_OPTION_INIT ("min-free-inodes", conf->min_free_inodes, percent,
err);
conf->dir_spread_cnt = conf->subvolume_cnt;
GF_OPTION_INIT ("directory-layout-spread", conf->dir_spread_cnt,
uint32, err);
GF_OPTION_INIT ("assert-no-child-down", conf->assert_no_child_down,
bool, err);
GF_OPTION_INIT ("readdir-optimize", conf->readdir_optimize, bool, err);
if (defrag) {
GF_OPTION_INIT ("rebalance-stats", defrag->stats, bool, err);
if (dict_get_str (this->options, "rebalance-filter", &temp_str)
== 0) {
if (gf_defrag_pattern_list_fill (this, defrag, temp_str)
== -1) {
gf_log (this->name, GF_LOG_ERROR, "Cannot parse"
" rebalance-filter (%s)", temp_str);
goto err;
}
}
}
/* option can be any one of percent or bytes */
conf->disk_unit = 0;
if (conf->min_free_disk < 100)
conf->disk_unit = 'p';
ret = dht_init_subvolumes (this, conf);
if (ret == -1) {
goto err;
}
if (dict_get_str (this->options, "decommissioned-bricks", &temp_str) == 0) {
ret = dht_parse_decommissioned_bricks (this, conf, temp_str);
if (ret == -1)
goto err;
}
dht_init_regex (this, this->options, "rsync-hash-regex",
&conf->rsync_regex, &conf->rsync_regex_valid);
dht_init_regex (this, this->options, "extra-hash-regex",
&conf->extra_regex, &conf->extra_regex_valid);
ret = dht_layouts_init (this, conf);
if (ret == -1) {
goto err;
}
LOCK_INIT (&conf->subvolume_lock);
LOCK_INIT (&conf->layout_lock);
conf->gen = 1;
this->local_pool = mem_pool_new (dht_local_t, 512);
if (!this->local_pool) {
gf_log (this->name, GF_LOG_ERROR,
"failed to create local_t's memory pool");
goto err;
}
GF_OPTION_INIT ("xattr-name", conf->xattr_name, str, err);
gf_asprintf (&conf->link_xattr_name, "%s.linkto", conf->xattr_name);
gf_asprintf (&conf->wild_xattr_name, "%s*", conf->xattr_name);
if (!conf->link_xattr_name || !conf->wild_xattr_name) {
goto err;
}
this->private = conf;
return 0;
err:
if (conf) {
if (conf->file_layouts) {
for (i = 0; i < conf->subvolume_cnt; i++) {
GF_FREE (conf->file_layouts[i]);
}
GF_FREE (conf->file_layouts);
}
GF_FREE (conf->subvolumes);
GF_FREE (conf->subvolume_status);
GF_FREE (conf->du_stats);
GF_FREE (conf->defrag);
GF_FREE (conf->xattr_name);
GF_FREE (conf->link_xattr_name);
GF_FREE (conf->wild_xattr_name);
GF_FREE (conf);
}
return -1;
}
struct volume_options options[] = {
{ .key = {"lookup-unhashed"},
.value = {"auto", "yes", "no", "enable", "disable", "1", "0",
"on", "off"},
.type = GF_OPTION_TYPE_STR,
.default_value = "on",
.description = "This option if set to ON, does a lookup through "
"all the sub-volumes, in case a lookup didn't return any result "
"from the hash subvolume. If set to OFF, it does not do a lookup "
"on the remaining subvolumes."
},
{ .key = {"min-free-disk"},
.type = GF_OPTION_TYPE_PERCENT_OR_SIZET,
.default_value = "10%",
.description = "Percentage/Size of disk space, after which the "
"process starts balancing out the cluster, and logs will appear "
"in log files",
},
{ .key = {"min-free-inodes"},
.type = GF_OPTION_TYPE_PERCENT,
.default_value = "5%",
.description = "after system has only N% of inodes, warnings "
"starts to appear in log files",
},
{ .key = {"unhashed-sticky-bit"},
.type = GF_OPTION_TYPE_BOOL,
.default_value = "off",
},
{ .key = {"use-readdirp"},
.type = GF_OPTION_TYPE_BOOL,
.default_value = "on",
.description = "This option if set to ON, forces the use of "
"readdirp, and hence also displays the stats of the files."
},
{ .key = {"assert-no-child-down"},
.type = GF_OPTION_TYPE_BOOL,
.default_value = "off",
.description = "This option if set to ON, in the event of "
"CHILD_DOWN, will call exit."
},
{ .key = {"directory-layout-spread"},
.type = GF_OPTION_TYPE_INT,
.min = 1,
.validate = GF_OPT_VALIDATE_MIN,
.description = "Specifies the directory layout spread."
},
{ .key = {"decommissioned-bricks"},
.type = GF_OPTION_TYPE_ANY,
.description = "This option if set to ON, decommissions "
"the brick, so that no new data is allowed to be created "
"on that brick."
},
{ .key = {"rebalance-cmd"},
.type = GF_OPTION_TYPE_INT,
},
{ .key = {"node-uuid"},
.type = GF_OPTION_TYPE_STR,
},
{ .key = {"rebalance-stats"},
.type = GF_OPTION_TYPE_BOOL,
.default_value = "off",
.description = "This option if set to ON displays and logs the "
" time taken for migration of each file, during the rebalance "
"process. If set to OFF, the rebalance logs will only display the "
"time spent in each directory."
},
{ .key = {"readdir-optimize"},
.type = GF_OPTION_TYPE_BOOL,
.default_value = "off",
.description = "This option if set to ON enables the optimization "
"that allows DHT to requests non-first subvolumes to filter out "
"directory entries."
},
{ .key = {"rsync-hash-regex"},
.type = GF_OPTION_TYPE_STR,
/* Setting a default here doesn't work. See dht_init_regex. */
.description = "Regular expression for stripping temporary-file "
"suffix and prefix used by rsync, to prevent relocation when the "
"file is renamed."
},
{ .key = {"extra-hash-regex"},
.type = GF_OPTION_TYPE_STR,
/* Setting a default here doesn't work. See dht_init_regex. */
.description = "Regular expression for stripping temporary-file "
"suffix and prefix used by an application, to prevent relocation when "
"the file is renamed."
},
{ .key = {"rebalance-filter"},
.type = GF_OPTION_TYPE_STR,
},
{ .key = {"xattr-name"},
.type = GF_OPTION_TYPE_STR,
.default_value = "trusted.glusterfs.dht",
.description = "Base for extended attributes used by this "
"translator instance, to avoid conflicts with others above or "
"below it."
},
/* NUFA option */
{ .key = {"local-volume-name"},
.type = GF_OPTION_TYPE_XLATOR
},
/* switch option */
{ .key = {"pattern.switch.case"},
.type = GF_OPTION_TYPE_ANY
},
{ .key = {NULL} },
};

733
xlators/cluster/dht/src/dht.c
View File

@ -14,633 +14,15 @@
#include "config.h" #include "config.h"
#endif #endif
/* TODO: add NS locking */
#include "statedump.h" #include "statedump.h"
#include "dht-common.h" #include "dht-common.h"
/* TODO: class_methods_t class_methods = {
- use volumename in xattr instead of "dht" .init = dht_init,
- use NS locks .fini = dht_fini,
- handle all cases in self heal layout reconstruction .reconfigure = dht_reconfigure,
- complete linkfile selfheal .notify = dht_notify
*/ };
struct volume_options options[];
void
dht_layout_dump (dht_layout_t *layout, const char *prefix)
{
char key[GF_DUMP_MAX_BUF_LEN];
int i = 0;
if (!layout)
goto out;
if (!prefix)
goto out;
gf_proc_dump_build_key(key, prefix, "cnt");
gf_proc_dump_write(key, "%d", layout->cnt);
gf_proc_dump_build_key(key, prefix, "preset");
gf_proc_dump_write(key, "%d", layout->preset);
gf_proc_dump_build_key(key, prefix, "gen");
gf_proc_dump_write(key, "%d", layout->gen);
if (layout->type != IA_INVAL) {
gf_proc_dump_build_key(key, prefix, "inode type");
gf_proc_dump_write(key, "%d", layout->type);
}
if (!IA_ISDIR (layout->type))
goto out;
for (i = 0; i < layout->cnt; i++) {
gf_proc_dump_build_key(key, prefix,"list[%d].err", i);
gf_proc_dump_write(key, "%d", layout->list[i].err);
gf_proc_dump_build_key(key, prefix,"list[%d].start", i);
gf_proc_dump_write(key, "%u", layout->list[i].start);
gf_proc_dump_build_key(key, prefix,"list[%d].stop", i);
gf_proc_dump_write(key, "%u", layout->list[i].stop);
if (layout->list[i].xlator) {
gf_proc_dump_build_key(key, prefix,
"list[%d].xlator.type", i);
gf_proc_dump_write(key, "%s",
layout->list[i].xlator->type);
gf_proc_dump_build_key(key, prefix,
"list[%d].xlator.name", i);
gf_proc_dump_write(key, "%s",
layout->list[i].xlator->name);
}
}
out:
return;
}
int32_t
dht_priv_dump (xlator_t *this)
{
char key_prefix[GF_DUMP_MAX_BUF_LEN];
char key[GF_DUMP_MAX_BUF_LEN];
int i = 0;
dht_conf_t *conf = NULL;
int ret = -1;
if (!this)
goto out;
conf = this->private;
if (!conf)
goto out;
ret = TRY_LOCK(&conf->subvolume_lock);
if (ret != 0) {
return ret;
}
gf_proc_dump_add_section("xlator.cluster.dht.%s.priv", this->name);
gf_proc_dump_build_key(key_prefix,"xlator.cluster.dht","%s.priv",
this->name);
gf_proc_dump_write("subvol_cnt","%d", conf->subvolume_cnt);
for (i = 0; i < conf->subvolume_cnt; i++) {
sprintf (key, "subvolumes[%d]", i);
gf_proc_dump_write(key, "%s.%s", conf->subvolumes[i]->type,
conf->subvolumes[i]->name);
if (conf->file_layouts && conf->file_layouts[i]){
sprintf (key, "file_layouts[%d]", i);
dht_layout_dump(conf->file_layouts[i], key);
}
if (conf->dir_layouts && conf->dir_layouts[i]) {
sprintf (key, "dir_layouts[%d]", i);
dht_layout_dump(conf->dir_layouts[i], key);
}
if (conf->subvolume_status) {
sprintf (key, "subvolume_status[%d]", i);
gf_proc_dump_write(key, "%d",
(int)conf->subvolume_status[i]);
}
}
gf_proc_dump_write("search_unhashed", "%d", conf->search_unhashed);
gf_proc_dump_write("gen", "%d", conf->gen);
gf_proc_dump_write("min_free_disk", "%lf", conf->min_free_disk);
gf_proc_dump_write("min_free_inodes", "%lf", conf->min_free_inodes);
gf_proc_dump_write("disk_unit", "%c", conf->disk_unit);
gf_proc_dump_write("refresh_interval", "%d", conf->refresh_interval);
gf_proc_dump_write("unhashed_sticky_bit", "%d", conf->unhashed_sticky_bit);
if (conf ->du_stats) {
gf_proc_dump_write("du_stats.avail_percent", "%lf",
conf->du_stats->avail_percent);
gf_proc_dump_write("du_stats.avail_space", "%lu",
conf->du_stats->avail_space);
gf_proc_dump_write("du_stats.avail_inodes", "%lf",
conf->du_stats->avail_inodes);
gf_proc_dump_write("du_stats.log", "%lu", conf->du_stats->log);
}
if (conf->last_stat_fetch.tv_sec)
gf_proc_dump_write("last_stat_fetch", "%s",
ctime(&conf->last_stat_fetch.tv_sec));
UNLOCK(&conf->subvolume_lock);
out:
return ret;
}
int32_t
dht_inodectx_dump (xlator_t *this, inode_t *inode)
{
int ret = -1;
dht_layout_t *layout = NULL;
if (!this)
goto out;
if (!inode)
goto out;
ret = dht_inode_ctx_layout_get (inode, this, &layout);
if ((ret != 0) || !layout)
return ret;
gf_proc_dump_add_section("xlator.cluster.dht.%s.inode", this->name);
dht_layout_dump(layout, "layout");
out:
return ret;
}
int
notify (xlator_t *this, int event, void *data, ...)
{
int ret = -1;
va_list ap;
dict_t *output = NULL;
GF_VALIDATE_OR_GOTO ("dht", this, out);
if (!data)
goto out;
va_start (ap, data);
output = va_arg (ap, dict_t*);
ret = dht_notify (this, event, data, output);
out:
return ret;
}
void
fini (xlator_t *this)
{
int i = 0;
dht_conf_t *conf = NULL;
GF_VALIDATE_OR_GOTO ("dht", this, out);
conf = this->private;
this->private = NULL;
if (conf) {
if (conf->file_layouts) {
for (i = 0; i < conf->subvolume_cnt; i++) {
GF_FREE (conf->file_layouts[i]);
}
GF_FREE (conf->file_layouts);
}
GF_FREE (conf->subvolumes);
GF_FREE (conf->subvolume_status);
GF_FREE (conf);
}
out:
return;
}
int32_t
mem_acct_init (xlator_t *this)
{
int ret = -1;
GF_VALIDATE_OR_GOTO ("dht", this, out);
ret = xlator_mem_acct_init (this, gf_dht_mt_end + 1);
if (ret != 0) {
gf_log (this->name, GF_LOG_ERROR, "Memory accounting init"
"failed");
return ret;
}
out:
return ret;
}
int
dht_parse_decommissioned_bricks (xlator_t *this, dht_conf_t *conf,
const char *bricks)
{
int i = 0;
int ret = -1;
char *tmpstr = NULL;
char *dup_brick = NULL;
char *node = NULL;
if (!conf || !bricks)
goto out;
dup_brick = gf_strdup (bricks);
node = strtok_r (dup_brick, ",", &tmpstr);
while (node) {
for (i = 0; i < conf->subvolume_cnt; i++) {
if (!strcmp (conf->subvolumes[i]->name, node)) {
conf->decommissioned_bricks[i] =
conf->subvolumes[i];
conf->decommission_subvols_cnt++;
gf_log (this->name, GF_LOG_INFO,
"decommissioning subvolume %s",
conf->subvolumes[i]->name);
break;
}
}
if (i == conf->subvolume_cnt) {
/* Wrong node given. */
goto out;
}
node = strtok_r (NULL, ",", &tmpstr);
}
ret = 0;
conf->decommission_in_progress = 1;
out:
GF_FREE (dup_brick);
return ret;
}
void
dht_init_regex (xlator_t *this, dict_t *odict, char *name,
regex_t *re, gf_boolean_t *re_valid)
{
char *temp_str;
if (dict_get_str (odict, name, &temp_str) != 0) {
if (strcmp(name,"rsync-hash-regex")) {
return;
}
temp_str = "^\\.(.+)\\.[^.]+$";
}
if (*re_valid) {
regfree(re);
*re_valid = _gf_false;
}
if (!strcmp(temp_str,"none")) {
return;
}
if (regcomp(re,temp_str,REG_EXTENDED) == 0) {
gf_log (this->name, GF_LOG_INFO,
"using regex %s = %s", name, temp_str);
*re_valid = _gf_true;
}
else {
gf_log (this->name, GF_LOG_WARNING,
"compiling regex %s failed", temp_str);
}
}
int
reconfigure (xlator_t *this, dict_t *options)
{
dht_conf_t *conf = NULL;
char *temp_str = NULL;
gf_boolean_t search_unhashed;
int ret = -1;
GF_VALIDATE_OR_GOTO ("dht", this, out);
GF_VALIDATE_OR_GOTO ("dht", options, out);
conf = this->private;
if (!conf)
return 0;
if (dict_get_str (options, "lookup-unhashed", &temp_str) == 0) {
/* If option is not "auto", other options _should_ be boolean*/
if (strcasecmp (temp_str, "auto")) {
if (!gf_string2boolean (temp_str, &search_unhashed)) {
gf_log(this->name, GF_LOG_DEBUG, "Reconfigure:"
" lookup-unhashed reconfigured (%s)",
temp_str);
conf->search_unhashed = search_unhashed;
} else {
gf_log(this->name, GF_LOG_ERROR, "Reconfigure:"
" lookup-unhashed should be boolean,"
" not (%s), defaulting to (%d)",
temp_str, conf->search_unhashed);
//return -1;
ret = -1;
goto out;
}
} else {
gf_log(this->name, GF_LOG_DEBUG, "Reconfigure:"
" lookup-unhashed reconfigured auto ");
conf->search_unhashed = GF_DHT_LOOKUP_UNHASHED_AUTO;
}
}
GF_OPTION_RECONF ("min-free-disk", conf->min_free_disk, options,
percent_or_size, out);
/* option can be any one of percent or bytes */
conf->disk_unit = 0;
if (conf->min_free_disk < 100.0)
conf->disk_unit = 'p';
GF_OPTION_RECONF ("min-free-inodes", conf->min_free_inodes, options,
percent, out);
GF_OPTION_RECONF ("directory-layout-spread", conf->dir_spread_cnt,
options, uint32, out);
GF_OPTION_RECONF ("readdir-optimize", conf->readdir_optimize, options,
bool, out);
if (conf->defrag) {
GF_OPTION_RECONF ("rebalance-stats", conf->defrag->stats,
options, bool, out);
}
if (dict_get_str (options, "decommissioned-bricks", &temp_str) == 0) {
ret = dht_parse_decommissioned_bricks (this, conf, temp_str);
if (ret == -1)
goto out;
}
dht_init_regex (this, options, "rsync-hash-regex",
&conf->rsync_regex, &conf->rsync_regex_valid);
dht_init_regex (this, options, "extra-hash-regex",
&conf->extra_regex, &conf->extra_regex_valid);
ret = 0;
out:
return ret;
}
static int
gf_defrag_pattern_list_fill (xlator_t *this, gf_defrag_info_t *defrag, char *data)
{
int ret = -1;
char *tmp_str = NULL;
char *tmp_str1 = NULL;
char *dup_str = NULL;
char *num = NULL;
char *pattern_str = NULL;
char *pattern = NULL;
gf_defrag_pattern_list_t *temp_list = NULL;
gf_defrag_pattern_list_t *pattern_list = NULL;
if (!this || !defrag || !data)
goto out;
/* Get the pattern for pattern list. "pattern:<optional-size>"
* eg: *avi, *pdf:10MB, *:1TB
*/
pattern_str = strtok_r (data, ",", &tmp_str);
while (pattern_str) {
dup_str = gf_strdup (pattern_str);
pattern_list = GF_CALLOC (1, sizeof (gf_defrag_pattern_list_t),
1);
if (!pattern_list) {
goto out;
}
pattern = strtok_r (dup_str, ":", &tmp_str1);
num = strtok_r (NULL, ":", &tmp_str1);
if (!pattern)
goto out;
if (!num) {
if (gf_string2bytesize(pattern, &pattern_list->size)
== 0) {
pattern = "*";
}
} else if (gf_string2bytesize (num, &pattern_list->size) != 0) {
gf_log (this->name, GF_LOG_ERROR,
"invalid number format \"%s\"", num);
goto out;
}
memcpy (pattern_list->path_pattern, pattern, strlen (dup_str));
if (!defrag->defrag_pattern)
temp_list = NULL;
else
temp_list = defrag->defrag_pattern;
pattern_list->next = temp_list;
defrag->defrag_pattern = pattern_list;
pattern_list = NULL;
GF_FREE (dup_str);
dup_str = NULL;
pattern_str = strtok_r (NULL, ",", &tmp_str);
}
ret = 0;
out:
if (ret)
GF_FREE (pattern_list);
GF_FREE (dup_str);
return ret;
}
int
init (xlator_t *this)
{
dht_conf_t *conf = NULL;
char *temp_str = NULL;
int ret = -1;
int i = 0;
gf_defrag_info_t *defrag = NULL;
int cmd = 0;
char *node_uuid = NULL;
GF_VALIDATE_OR_GOTO ("dht", this, err);
if (!this->children) {
gf_log (this->name, GF_LOG_CRITICAL,
"Distribute needs more than one subvolume");
return -1;
}
if (!this->parents) {
gf_log (this->name, GF_LOG_WARNING,
"dangling volume. check volfile");
}
conf = GF_CALLOC (1, sizeof (*conf), gf_dht_mt_dht_conf_t);
if (!conf) {
goto err;
}
memset (conf, 0, sizeof(*conf));
ret = dict_get_int32 (this->options, "rebalance-cmd", &cmd);
if (cmd) {
defrag = GF_CALLOC (1, sizeof (gf_defrag_info_t),
gf_defrag_info_mt);
GF_VALIDATE_OR_GOTO (this->name, defrag, err);
LOCK_INIT (&defrag->lock);
defrag->is_exiting = 0;
conf->defrag = defrag;
ret = dict_get_str (this->options, "node-uuid", &node_uuid);
if (ret) {
gf_log (this->name, GF_LOG_ERROR, "node-uuid not "
"specified");
goto err;
}
if (uuid_parse (node_uuid, defrag->node_uuid)) {
gf_log (this->name, GF_LOG_ERROR, "Cannot parse "
"glusterd node uuid");
goto err;
}
defrag->cmd = cmd;
defrag->stats = _gf_false;
}
conf->search_unhashed = GF_DHT_LOOKUP_UNHASHED_ON;
if (dict_get_str (this->options, "lookup-unhashed", &temp_str) == 0) {
/* If option is not "auto", other options _should_ be boolean */
if (strcasecmp (temp_str, "auto"))
gf_string2boolean (temp_str, &conf->search_unhashed);
else
conf->search_unhashed = GF_DHT_LOOKUP_UNHASHED_AUTO;
}
GF_OPTION_INIT ("unhashed-sticky-bit", conf->unhashed_sticky_bit, bool,
err);
GF_OPTION_INIT ("use-readdirp", conf->use_readdirp, bool, err);
GF_OPTION_INIT ("min-free-disk", conf->min_free_disk, percent_or_size,
err);
GF_OPTION_INIT ("min-free-inodes", conf->min_free_inodes, percent,
err);
conf->dir_spread_cnt = conf->subvolume_cnt;
GF_OPTION_INIT ("directory-layout-spread", conf->dir_spread_cnt,
uint32, err);
GF_OPTION_INIT ("assert-no-child-down", conf->assert_no_child_down,
bool, err);
GF_OPTION_INIT ("readdir-optimize", conf->readdir_optimize, bool, err);
if (defrag) {
GF_OPTION_INIT ("rebalance-stats", defrag->stats, bool, err);
if (dict_get_str (this->options, "rebalance-filter", &temp_str)
== 0) {
if (gf_defrag_pattern_list_fill (this, defrag, temp_str)
== -1) {
gf_log (this->name, GF_LOG_ERROR, "Cannot parse"
" rebalance-filter (%s)", temp_str);
goto err;
}
}
}
/* option can be any one of percent or bytes */
conf->disk_unit = 0;
if (conf->min_free_disk < 100)
conf->disk_unit = 'p';
ret = dht_init_subvolumes (this, conf);
if (ret == -1) {
goto err;
}
if (dict_get_str (this->options, "decommissioned-bricks", &temp_str) == 0) {
ret = dht_parse_decommissioned_bricks (this, conf, temp_str);
if (ret == -1)
goto err;
}
dht_init_regex (this, this->options, "rsync-hash-regex",
&conf->rsync_regex, &conf->rsync_regex_valid);
dht_init_regex (this, this->options, "extra-hash-regex",
&conf->extra_regex, &conf->extra_regex_valid);
ret = dht_layouts_init (this, conf);
if (ret == -1) {
goto err;
}
LOCK_INIT (&conf->subvolume_lock);
LOCK_INIT (&conf->layout_lock);
conf->gen = 1;
this->local_pool = mem_pool_new (dht_local_t, 512);
if (!this->local_pool) {
gf_log (this->name, GF_LOG_ERROR,
"failed to create local_t's memory pool");
goto err;
}
GF_OPTION_INIT ("xattr-name", conf->xattr_name, str, err);
gf_asprintf (&conf->link_xattr_name, "%s.linkto", conf->xattr_name);
gf_asprintf (&conf->wild_xattr_name, "%s*", conf->xattr_name);
if (!conf->link_xattr_name || !conf->wild_xattr_name) {
goto err;
}
this->private = conf;
return 0;
err:
if (conf) {
if (conf->file_layouts) {
for (i = 0; i < conf->subvolume_cnt; i++) {
GF_FREE (conf->file_layouts[i]);
}
GF_FREE (conf->file_layouts);
}
GF_FREE (conf->subvolumes);
GF_FREE (conf->subvolume_status);
GF_FREE (conf->du_stats);
GF_FREE (conf->defrag);
GF_FREE (conf->xattr_name);
GF_FREE (conf->link_xattr_name);
GF_FREE (conf->wild_xattr_name);
GF_FREE (conf);
}
return -1;
}
struct xlator_fops fops = { struct xlator_fops fops = {
.lookup = dht_lookup, .lookup = dht_lookup,
@ -701,105 +83,4 @@ struct xlator_cbks cbks = {
// .releasedir = dht_releasedir, // .releasedir = dht_releasedir,
.forget = dht_forget .forget = dht_forget
}; };
;
struct volume_options options[] = {
{ .key = {"lookup-unhashed"},
.value = {"auto", "yes", "no", "enable", "disable", "1", "0",
"on", "off"},
.type = GF_OPTION_TYPE_STR,
.default_value = "on",
.description = "This option if set to ON, does a lookup through "
"all the sub-volumes, in case a lookup didn't return any result "
"from the hash subvolume. If set to OFF, it does not do a lookup "
"on the remaining subvolumes."
},
{ .key = {"min-free-disk"},
.type = GF_OPTION_TYPE_PERCENT_OR_SIZET,
.default_value = "10%",
.description = "Percentage/Size of disk space, after which the "
"process starts balancing out the cluster, and logs will appear "
"in log files",
},
{ .key = {"min-free-inodes"},
.type = GF_OPTION_TYPE_PERCENT,
.default_value = "5%",
.description = "after system has only N% of inodes, warnings "
"starts to appear in log files",
},
{ .key = {"unhashed-sticky-bit"},
.type = GF_OPTION_TYPE_BOOL,
.default_value = "off",
},
{ .key = {"use-readdirp"},
.type = GF_OPTION_TYPE_BOOL,
.default_value = "on",
.description = "This option if set to ON, forces the use of "
"readdirp, and hence also displays the stats of the files."
},
{ .key = {"assert-no-child-down"},
.type = GF_OPTION_TYPE_BOOL,
.default_value = "off",
.description = "This option if set to ON, in the event of "
"CHILD_DOWN, will call exit."
},
{ .key = {"directory-layout-spread"},
.type = GF_OPTION_TYPE_INT,
.min = 1,
.validate = GF_OPT_VALIDATE_MIN,
.description = "Specifies the directory layout spread."
},
{ .key = {"decommissioned-bricks"},
.type = GF_OPTION_TYPE_ANY,
.description = "This option if set to ON, decommissions "
"the brick, so that no new data is allowed to be created "
"on that brick."
},
{ .key = {"rebalance-cmd"},
.type = GF_OPTION_TYPE_INT,
},
{ .key = {"node-uuid"},
.type = GF_OPTION_TYPE_STR,
},
{ .key = {"rebalance-stats"},
.type = GF_OPTION_TYPE_BOOL,
.default_value = "off",
.description = "This option if set to ON displays and logs the "
" time taken for migration of each file, during the rebalance "
"process. If set to OFF, the rebalance logs will only display the "
"time spent in each directory."
},
{ .key = {"readdir-optimize"},
.type = GF_OPTION_TYPE_BOOL,
.default_value = "off",
.description = "This option if set to ON enables the optimization "
"that allows DHT to requests non-first subvolumes to filter out "
"directory entries."
},
{ .key = {"rsync-hash-regex"},
.type = GF_OPTION_TYPE_STR,
/* Setting a default here doesn't work. See dht_init_regex. */
.description = "Regular expression for stripping temporary-file "
"suffix and prefix used by rsync, to prevent relocation when the "
"file is renamed."
},
{ .key = {"extra-hash-regex"},
.type = GF_OPTION_TYPE_STR,
/* Setting a default here doesn't work. See dht_init_regex. */
.description = "Regular expression for stripping temporary-file "
"suffix and prefix used by an application, to prevent relocation when "
"the file is renamed."
},
{ .key = {"rebalance-filter"},
.type = GF_OPTION_TYPE_STR,
},
{ .key = {"xattr-name"},
.type = GF_OPTION_TYPE_STR,
.default_value = "trusted.glusterfs.dht",
.description = "Base for extended attributes used by this "
"translator instance, to avoid conflicts with others above or "
"below it."
},
{ .key = {NULL} },
};

177
xlators/cluster/dht/src/nufa.c
View File

@ -18,6 +18,8 @@
/* TODO: all 'TODO's in dht.c holds good */ /* TODO: all 'TODO's in dht.c holds good */
extern struct volume_options options[];
int int
nufa_local_lookup_cbk (call_frame_t *frame, void *cookie, xlator_t *this, nufa_local_lookup_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
int op_ret, int op_errno, int op_ret, int op_errno,
@ -460,42 +462,6 @@ err:
} }
int
notify (xlator_t *this, int event, void *data, ...)
{
int ret = -1;
ret = dht_notify (this, event, data);
return ret;
}
void
fini (xlator_t *this)
{
int i = 0;
dht_conf_t *conf = NULL;
conf = this->private;
if (conf) {
if (conf->file_layouts) {
for (i = 0; i < conf->subvolume_cnt; i++) {
GF_FREE (conf->file_layouts[i]);
}
GF_FREE (conf->file_layouts);
}
GF_FREE (conf->subvolumes);
GF_FREE (conf->subvolume_status);
GF_FREE (conf);
}
return;
}
gf_boolean_t gf_boolean_t
same_first_part (char *str1, char term1, char *str2, char term2) same_first_part (char *str1, char term1, char *str2, char term2)
{ {
@ -516,72 +482,24 @@ same_first_part (char *str1, char term1, char *str2, char term2)
} }
} }
int32_t
mem_acct_init (xlator_t *this)
{
int ret = -1;
ret = xlator_mem_acct_init (this, gf_dht_mt_end + 1);
return ret;
}
int int
init (xlator_t *this) nufa_init (xlator_t *this)
{ {
dht_conf_t *conf = NULL; dht_conf_t *conf = NULL;
xlator_list_t *trav = NULL; xlator_list_t *trav = NULL;
data_t *data = NULL; data_t *data = NULL;
char *local_volname = NULL; char *local_volname = NULL;
char *temp_str = NULL;
int ret = -1; int ret = -1;
int i = 0;
char my_hostname[256]; char my_hostname[256];
double temp_free_disk = 0;
uint64_t size = 0;
xlator_t *local_subvol = NULL; xlator_t *local_subvol = NULL;
char *brick_host = NULL; char *brick_host = NULL;
xlator_t *kid = NULL; xlator_t *kid = NULL;
if (!this->children) { ret = dht_init(this);
gf_log (this->name, GF_LOG_CRITICAL, if (ret) {
"NUFA needs more than one subvolume"); return ret;
return -1;
} }
conf = this->private;
if (!this->parents) {
gf_log (this->name, GF_LOG_WARNING,
"dangling volume. check volfile");
}
conf = GF_CALLOC (1, sizeof (*conf),
gf_dht_mt_dht_conf_t);
if (!conf) {
goto err;
}
conf->search_unhashed = GF_DHT_LOOKUP_UNHASHED_ON;
if (dict_get_str (this->options, "lookup-unhashed", &temp_str) == 0) {
/* If option is not "auto", other options _should_ be boolean */
if (strcasecmp (temp_str, "auto"))
gf_string2boolean (temp_str, &conf->search_unhashed);
else
conf->search_unhashed = GF_DHT_LOOKUP_UNHASHED_AUTO;
}
ret = dht_init_subvolumes (this, conf);
if (ret == -1) {
goto err;
}
ret = dht_layouts_init (this, conf);
if (ret == -1) {
goto err;
}
LOCK_INIT (&conf->subvolume_lock);
LOCK_INIT (&conf->layout_lock);
conf->gen = 1;
local_volname = "localhost"; local_volname = "localhost";
ret = gethostname (my_hostname, 256); ret = gethostname (my_hostname, 256);
@ -650,69 +568,22 @@ init (xlator_t *this)
} }
/* The volume specified exists */
conf->min_free_disk = 10;
conf->disk_unit = 'p';
if (dict_get_str (this->options, "min-free-disk",
&temp_str) == 0) {
if (gf_string2percent (temp_str,
&temp_free_disk) == 0) {
if (temp_free_disk > 100) {
gf_string2bytesize (temp_str, &size);
conf->min_free_disk = size;
conf->disk_unit = 'b';
} else {
conf->min_free_disk = temp_free_disk;
conf->disk_unit = 'p';
}
} else {
gf_string2bytesize (temp_str, &size);
conf->min_free_disk = size;
conf->disk_unit = 'b';
}
}
conf->du_stats = GF_CALLOC (conf->subvolume_cnt, sizeof (dht_du_t),
gf_dht_mt_dht_du_t);
if (!conf->du_stats) {
goto err;
}
this->local_pool = mem_pool_new (dht_local_t, 128);
if (!this->local_pool) {
gf_log (this->name, GF_LOG_ERROR,
"failed to create local_t's memory pool");
goto err;
}
this->private = conf;
return 0; return 0;
err: err:
if (conf) { dht_fini(this);
if (conf->file_layouts) {
for (i = 0; i < conf->subvolume_cnt; i++) {
GF_FREE (conf->file_layouts[i]);
}
GF_FREE (conf->file_layouts);
}
GF_FREE (conf->subvolumes);
GF_FREE (conf->subvolume_status);
GF_FREE (conf->du_stats);
GF_FREE (conf);
}
return -1; return -1;
} }
class_methods_t class_methods = {
.init = nufa_init,
.fini = dht_fini,
.reconfigure = dht_reconfigure,
.notify = dht_notify
};
struct xlator_fops fops = { struct xlator_fops fops = {
.lookup = nufa_lookup, .lookup = nufa_lookup,
.create = nufa_create, .create = nufa_create,
@ -757,19 +628,3 @@ struct xlator_fops fops = {
struct xlator_cbks cbks = { struct xlator_cbks cbks = {
.forget = dht_forget .forget = dht_forget
}; };
struct volume_options options[] = {
{ .key = {"lookup-unhashed"},
.value = {"auto", "yes", "no", "enable", "disable", "1", "0",
"on", "off"},
.type = GF_OPTION_TYPE_STR
},
{ .key = {"local-volume-name"},
.type = GF_OPTION_TYPE_XLATOR
},
{ .key = {"min-free-disk"},
.type = GF_OPTION_TYPE_PERCENT_OR_SIZET,
},
{ .key = {NULL} },
};

163
xlators/cluster/dht/src/switch.c
View File

@ -22,6 +22,8 @@
#include <fnmatch.h> #include <fnmatch.h>
#include <string.h> #include <string.h>
extern struct volume_options options[];
struct switch_sched_array { struct switch_sched_array {
xlator_t *xl; xlator_t *xl;
int32_t eligible; int32_t eligible;
@ -569,20 +571,9 @@ err:
} }
int
notify (xlator_t *this, int event, void *data, ...)
{
int ret = -1;
ret = dht_notify (this, event, data);
return ret;
}
void void
fini (xlator_t *this) switch_fini (xlator_t *this)
{ {
int i = 0;
dht_conf_t *conf = NULL; dht_conf_t *conf = NULL;
struct switch_struct *trav = NULL; struct switch_struct *trav = NULL;
struct switch_struct *prev = NULL; struct switch_struct *prev = NULL;
@ -598,22 +589,9 @@ fini (xlator_t *this)
trav = trav->next; trav = trav->next;
GF_FREE (prev); GF_FREE (prev);
} }
if (conf->file_layouts) {
for (i = 0; i < conf->subvolume_cnt; i++) {
GF_FREE (conf->file_layouts[i]);
}
GF_FREE (conf->file_layouts);
}
GF_FREE (conf->subvolumes);
GF_FREE (conf->subvolume_status);
GF_FREE (conf);
} }
return; dht_fini(this);
} }
int int
@ -839,68 +817,18 @@ err:
} }
int int32_t
init (xlator_t *this) switch_init (xlator_t *this)
{ {
dht_conf_t *conf = NULL; dht_conf_t *conf = NULL;
data_t *data = NULL; data_t *data = NULL;
char *temp_str = NULL;
int ret = -1; int ret = -1;
int i = 0;
double temp_free_disk = 0;
uint64_t size = 0;
if (!this->children) { ret = dht_init(this);
gf_log (this->name, GF_LOG_CRITICAL, if (ret) {
"SWITCH needs more than one subvolume"); return ret;
return -1;
}
if (!this->parents) {
gf_log (this->name, GF_LOG_WARNING,
"dangling volume. check volfile");
}
conf = GF_CALLOC (1, sizeof (*conf), gf_switch_mt_dht_conf_t);
if (!conf) {
goto err;
}
conf->search_unhashed = GF_DHT_LOOKUP_UNHASHED_ON;
if (dict_get_str (this->options, "lookup-unhashed", &temp_str) == 0) {
/* If option is not "auto", other options _should_ be boolean */
if (strcasecmp (temp_str, "auto"))
gf_string2boolean (temp_str, &conf->search_unhashed);
else
conf->search_unhashed = GF_DHT_LOOKUP_UNHASHED_AUTO;
}
conf->unhashed_sticky_bit = 0;
if (dict_get_str (this->options, "unhashed-sticky-bit",
&temp_str) == 0) {
gf_string2boolean (temp_str, &conf->unhashed_sticky_bit);
}
conf->min_free_disk = 10.0;
conf->disk_unit = 'p';
if (dict_get_str (this->options, "min-free-disk",
&temp_str) == 0) {
if (gf_string2percent (temp_str, &temp_free_disk) == 0) {
if (temp_free_disk > 100) {
gf_string2bytesize (temp_str, &size);
conf->min_free_disk = size;
conf->disk_unit = 'b';
} else {
conf->min_free_disk = temp_free_disk;
conf->disk_unit = 'p';
}
} else {
gf_string2bytesize (temp_str, &size);
conf->min_free_disk = size;
conf->disk_unit = 'b';
}
} }
conf = this->private;
data = dict_get (this->options, "pattern.switch.case"); data = dict_get (this->options, "pattern.switch.case");
if (data) { if (data) {
@ -911,60 +839,23 @@ init (xlator_t *this)
} }
} }
ret = dht_init_subvolumes (this, conf);
if (ret == -1) {
goto err;
}
ret = dht_layouts_init (this, conf);
if (ret == -1) {
goto err;
}
LOCK_INIT (&conf->subvolume_lock);
LOCK_INIT (&conf->layout_lock);
conf->gen = 1;
conf->du_stats = GF_CALLOC (conf->subvolume_cnt, sizeof (dht_du_t),
gf_switch_mt_dht_du_t);
if (!conf->du_stats) {
goto err;
}
this->local_pool = mem_pool_new (dht_local_t, 128);
if (!this->local_pool) {
gf_log (this->name, GF_LOG_ERROR,
"failed to create local_t's memory pool");
goto err;
}
this->private = conf; this->private = conf;
return 0; return 0;
err: err:
if (conf) { dht_fini(this);
if (conf->file_layouts) {
for (i = 0; i < conf->subvolume_cnt; i++) {
GF_FREE (conf->file_layouts[i]);
}
GF_FREE (conf->file_layouts);
}
GF_FREE (conf->subvolumes);
GF_FREE (conf->subvolume_status);
GF_FREE (conf->du_stats);
GF_FREE (conf);
}
return -1; return -1;
} }
class_methods_t class_methods = {
.init = switch_init,
.fini = switch_fini,
.reconfigure = dht_reconfigure,
.notify = dht_notify
};
struct xlator_fops fops = { struct xlator_fops fops = {
.lookup = switch_lookup, .lookup = switch_lookup,
.create = switch_create, .create = switch_create,
@ -1009,19 +900,3 @@ struct xlator_fops fops = {
struct xlator_cbks cbks = { struct xlator_cbks cbks = {
.forget = dht_forget .forget = dht_forget
}; };
struct volume_options options[] = {
{ .key = {"lookup-unhashed"},
.value = {"auto", "yes", "no", "enable", "disable", "1", "0",
"on", "off"},
.type = GF_OPTION_TYPE_STR
},
{ .key = {"pattern.switch.case"},
.type = GF_OPTION_TYPE_ANY
},
{ .key = {"min-free-disk"},
.type = GF_OPTION_TYPE_PERCENT_OR_SIZET,
},
{ .key = {NULL} },
};