summaryrefslogtreecommitdiffstats
path: root/xlators/cluster/dht/src/nufa.c
diff options
context:
space:
mode:
Diffstat (limited to 'xlators/cluster/dht/src/nufa.c')
-rw-r--r--xlators/cluster/dht/src/nufa.c425
1 files changed, 183 insertions, 242 deletions
diff --git a/xlators/cluster/dht/src/nufa.c b/xlators/cluster/dht/src/nufa.c
index 45fea55ae..e934acdf0 100644
--- a/xlators/cluster/dht/src/nufa.c
+++ b/xlators/cluster/dht/src/nufa.c
@@ -1,20 +1,11 @@
/*
- Copyright (c) 2008-2010 Gluster, Inc. <http://www.gluster.com>
+ Copyright (c) 2008-2012 Red Hat, Inc. <http://www.redhat.com>
This file is part of GlusterFS.
- GlusterFS is free software; you can redistribute it and/or modify
- it under the terms of the GNU Affero General Public License as published
- by the Free Software Foundation; either version 3 of the License,
- or (at your option) any later version.
-
- GlusterFS is distributed in the hope that it will be useful, but
- WITHOUT ANY WARRANTY; without even the implied warranty of
- MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
- Affero General Public License for more details.
-
- You should have received a copy of the GNU Affero General Public License
- along with this program. If not, see
- <http://www.gnu.org/licenses/>.
+ 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.
*/
@@ -23,10 +14,12 @@
#include "config.h"
#endif
-#include "dht-common.c"
+#include "dht-common.h"
/* TODO: all 'TODO's in dht.c holds good */
+extern struct volume_options options[];
+
int
nufa_local_lookup_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
int op_ret, int op_errno,
@@ -44,7 +37,6 @@ nufa_local_lookup_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
int call_cnt = 0;
int ret = 0;
-
conf = this->private;
prev = cookie;
@@ -62,7 +54,8 @@ nufa_local_lookup_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
if (op_ret == -1)
goto out;
- is_linkfile = check_is_linkfile (inode, stbuf, xattr);
+ is_linkfile = check_is_linkfile (inode, stbuf, xattr,
+ conf->link_xattr_name);
is_dir = check_is_dir (inode, stbuf, xattr);
if (!is_dir && !is_linkfile) {
@@ -141,7 +134,7 @@ out:
err:
DHT_STACK_UNWIND (lookup, frame, op_ret, op_errno,
- inode, stbuf, xattr, NULL);
+ inode, stbuf, xattr, postparent);
return 0;
}
@@ -150,7 +143,6 @@ nufa_lookup (call_frame_t *frame, xlator_t *this,
loc_t *loc, dict_t *xattr_req)
{
xlator_t *hashed_subvol = NULL;
- xlator_t *cached_subvol = NULL;
xlator_t *subvol = NULL;
dht_local_t *local = NULL;
dht_conf_t *conf = NULL;
@@ -169,21 +161,12 @@ nufa_lookup (call_frame_t *frame, xlator_t *this,
conf = this->private;
- local = dht_local_init (frame);
+ local = dht_local_init (frame, loc, NULL, GF_FOP_LOOKUP);
if (!local) {
op_errno = ENOMEM;
goto err;
}
- ret = loc_dup (loc, &local->loc);
- if (ret == -1) {
- op_errno = errno;
- gf_log (this->name, GF_LOG_DEBUG,
- "copying location failed for path=%s",
- loc->path);
- goto err;
- }
-
if (xattr_req) {
local->xattr_req = dict_ref (xattr_req);
} else {
@@ -191,14 +174,11 @@ nufa_lookup (call_frame_t *frame, xlator_t *this,
}
hashed_subvol = dht_subvol_get_hashed (this, &local->loc);
- cached_subvol = dht_subvol_get_cached (this, local->loc.inode);
- local->cached_subvol = cached_subvol;
local->hashed_subvol = hashed_subvol;
if (is_revalidate (loc)) {
- local->layout = layout = dht_layout_get (this, loc->inode);
-
+ layout = local->layout;
if (!layout) {
gf_log (this->name, GF_LOG_DEBUG,
"revalidate without cache. path=%s",
@@ -215,7 +195,7 @@ nufa_lookup (call_frame_t *frame, xlator_t *this,
goto do_fresh_lookup;
}
- local->inode = inode_ref (loc->inode);
+ local->inode = inode_ref (loc->inode);
local->call_cnt = layout->cnt;
call_cnt = local->call_cnt;
@@ -224,7 +204,7 @@ nufa_lookup (call_frame_t *frame, xlator_t *this,
* revalidates directly go to the cached-subvolume.
*/
ret = dict_set_uint32 (local->xattr_req,
- "trusted.glusterfs.dht", 4 * 4);
+ conf->xattr_name, 4 * 4);
if (ret < 0) {
gf_log (this->name, GF_LOG_ERROR,
"Failed to set dict value.");
@@ -245,7 +225,7 @@ nufa_lookup (call_frame_t *frame, xlator_t *this,
} else {
do_fresh_lookup:
ret = dict_set_uint32 (local->xattr_req,
- "trusted.glusterfs.dht", 4 * 4);
+ conf->xattr_name, 4 * 4);
if (ret < 0) {
gf_log (this->name, GF_LOG_ERROR,
"Failed to set dict value.");
@@ -254,7 +234,7 @@ nufa_lookup (call_frame_t *frame, xlator_t *this,
}
ret = dict_set_uint32 (local->xattr_req,
- "trusted.glusterfs.dht.linkto", 256);
+ conf->link_xattr_name, 256);
if (ret < 0) {
gf_log (this->name, GF_LOG_ERROR,
"Failed to set dict value.");
@@ -273,7 +253,8 @@ nufa_lookup (call_frame_t *frame, xlator_t *this,
err:
op_errno = (op_errno == -1) ? errno : op_errno;
- DHT_STACK_UNWIND (lookup, frame, -1, op_errno, NULL, NULL, NULL, NULL);
+ DHT_STACK_UNWIND (lookup, frame, -1, op_errno, NULL, NULL, NULL,
+ NULL);
return 0;
}
@@ -282,7 +263,7 @@ nufa_create_linkfile_create_cbk (call_frame_t *frame, void *cookie,
xlator_t *this, int op_ret, int op_errno,
inode_t *inode, struct iatt *stbuf,
struct iatt *preparent,
- struct iatt *postparent)
+ struct iatt *postparent, dict_t *xdata)
{
dht_local_t *local = NULL;
@@ -293,28 +274,27 @@ nufa_create_linkfile_create_cbk (call_frame_t *frame, void *cookie,
STACK_WIND (frame, dht_create_cbk,
local->cached_subvol, local->cached_subvol->fops->create,
- &local->loc, local->flags, local->mode, local->fd,
- local->params);
+ &local->loc, local->flags, local->mode, local->umask,
+ local->fd, local->params);
return 0;
err:
DHT_STACK_UNWIND (create, frame, -1, op_errno,
- NULL, NULL, NULL, NULL, NULL);
+ NULL, NULL, NULL, NULL, NULL, NULL);
return 0;
}
int
nufa_create (call_frame_t *frame, xlator_t *this,
loc_t *loc, int32_t flags, mode_t mode,
- fd_t *fd, dict_t *params)
+ mode_t umask, fd_t *fd, dict_t *params)
{
dht_local_t *local = NULL;
dht_conf_t *conf = NULL;
xlator_t *subvol = NULL;
xlator_t *avail_subvol = NULL;
int op_errno = -1;
- int ret = -1;
VALIDATE_OR_GOTO (frame, err);
VALIDATE_OR_GOTO (this, err);
@@ -324,7 +304,7 @@ nufa_create (call_frame_t *frame, xlator_t *this,
dht_get_du_info (frame, this, loc);
- local = dht_local_init (frame);
+ local = dht_local_init (frame, loc, fd, GF_FOP_CREATE);
if (!local) {
op_errno = ENOMEM;
goto err;
@@ -343,26 +323,19 @@ nufa_create (call_frame_t *frame, xlator_t *this,
if (dht_is_subvol_filled (this, (xlator_t *)conf->private)) {
avail_subvol =
dht_free_disk_available_subvol (this,
- (xlator_t *)conf->private);
+ (xlator_t *)conf->private,
+ local);
}
if (subvol != avail_subvol) {
/* create a link file instead of actual file */
- ret = loc_copy (&local->loc, loc);
- if (ret == -1) {
- op_errno = ENOMEM;
- goto err;
- }
-
- local->fd = fd_ref (fd);
local->params = dict_ref (params);
local->mode = mode;
local->flags = flags;
-
+ local->umask = umask;
local->cached_subvol = avail_subvol;
- dht_linkfile_create (frame,
- nufa_create_linkfile_create_cbk,
- avail_subvol, subvol, loc);
+ dht_linkfile_create (frame, nufa_create_linkfile_create_cbk,
+ this, avail_subvol, subvol, loc);
return 0;
}
@@ -371,14 +344,14 @@ nufa_create (call_frame_t *frame, xlator_t *this,
STACK_WIND (frame, dht_create_cbk,
subvol, subvol->fops->create,
- loc, flags, mode, fd, params);
+ loc, flags, mode, umask, fd, params);
return 0;
err:
op_errno = (op_errno == -1) ? errno : op_errno;
DHT_STACK_UNWIND (create, frame, -1, op_errno,
- NULL, NULL, NULL, NULL, NULL);
+ NULL, NULL, NULL, NULL, NULL, NULL);
return 0;
}
@@ -387,41 +360,45 @@ int
nufa_mknod_linkfile_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
int op_ret, int op_errno, inode_t *inode,
struct iatt *stbuf, struct iatt *preparent,
- struct iatt *postparent)
+ struct iatt *postparent, dict_t *xdata)
{
dht_local_t *local = NULL;
local = frame->local;
+ if (!local || !local->cached_subvol) {
+ op_errno = EINVAL;
+ op_ret = -1;
+ goto err;
+ }
if (op_ret >= 0) {
- STACK_WIND (frame, dht_newfile_cbk,
- local->cached_subvol,
+ STACK_WIND_COOKIE (frame, dht_newfile_cbk,
+ (void *)local->cached_subvol, local->cached_subvol,
local->cached_subvol->fops->mknod,
&local->loc, local->mode, local->rdev,
- local->params);
+ local->umask, local->params);
return 0;
}
-
+err:
WIPE (postparent);
WIPE (preparent);
DHT_STACK_UNWIND (link, frame, op_ret, op_errno,
- inode, stbuf, preparent, postparent);
+ inode, stbuf, preparent, postparent, xdata);
return 0;
}
int
nufa_mknod (call_frame_t *frame, xlator_t *this,
- loc_t *loc, mode_t mode, dev_t rdev, dict_t *params)
+ loc_t *loc, mode_t mode, dev_t rdev, mode_t umask, dict_t *params)
{
dht_local_t *local = NULL;
dht_conf_t *conf = NULL;
xlator_t *subvol = NULL;
xlator_t *avail_subvol = NULL;
int op_errno = -1;
- int ret = -1;
VALIDATE_OR_GOTO (frame, err);
VALIDATE_OR_GOTO (this, err);
@@ -431,7 +408,7 @@ nufa_mknod (call_frame_t *frame, xlator_t *this,
dht_get_du_info (frame, this, loc);
- local = dht_local_init (frame);
+ local = dht_local_init (frame, loc, NULL, GF_FOP_MKNOD);
if (!local) {
op_errno = ENOMEM;
goto err;
@@ -451,23 +428,20 @@ nufa_mknod (call_frame_t *frame, xlator_t *this,
if (dht_is_subvol_filled (this, (xlator_t *)conf->private)) {
avail_subvol =
dht_free_disk_available_subvol (this,
- (xlator_t *)conf->private);
+ (xlator_t *)conf->private,
+ local);
}
if (avail_subvol != subvol) {
/* Create linkfile first */
- ret = loc_copy (&local->loc, loc);
- if (ret == -1) {
- op_errno = ENOMEM;
- goto err;
- }
local->params = dict_ref (params);
local->mode = mode;
+ local->umask = umask;
local->rdev = rdev;
local->cached_subvol = avail_subvol;
- dht_linkfile_create (frame, nufa_mknod_linkfile_cbk,
+ dht_linkfile_create (frame, nufa_mknod_linkfile_cbk, this,
avail_subvol, subvol, loc);
return 0;
}
@@ -475,202 +449,185 @@ nufa_mknod (call_frame_t *frame, xlator_t *this,
gf_log (this->name, GF_LOG_TRACE,
"creating %s on %s", loc->path, subvol->name);
- STACK_WIND (frame, dht_newfile_cbk,
- subvol, subvol->fops->mknod,
- loc, mode, rdev, params);
+ STACK_WIND_COOKIE (frame, dht_newfile_cbk, (void *)subvol, subvol,
+ subvol->fops->mknod, loc, mode, rdev, umask,
+ params);
return 0;
err:
op_errno = (op_errno == -1) ? errno : op_errno;
DHT_STACK_UNWIND (mknod, frame, -1, op_errno,
- NULL, NULL, NULL, NULL);
+ NULL, NULL, NULL, NULL, NULL);
return 0;
}
-int
-notify (xlator_t *this, int event, void *data, ...)
+gf_boolean_t
+same_first_part (char *str1, char term1, char *str2, char term2)
{
- int ret = -1;
+ gf_boolean_t ended1;
+ gf_boolean_t ended2;
+
+ for (;;) {
+ ended1 = ((*str1 == '\0') || (*str1 == term1));
+ ended2 = ((*str2 == '\0') || (*str2 == term2));
+ if (ended1 && ended2) {
+ return _gf_true;
+ }
+ if (ended1 || ended2 || (*str1 != *str2)) {
+ return _gf_false;
+ }
+ ++str1;
+ ++str2;
+ }
+}
- ret = dht_notify (this, event, data);
+typedef struct nufa_args {
+ xlator_t *this;
+ char *volname;
+ gf_boolean_t addr_match;
+} nufa_args_t;
+
+static void
+nufa_find_local_brick (xlator_t *xl, void *data)
+{
+ nufa_args_t *args = data;
+ xlator_t *this = args->this;
+ char *local_volname = args->volname;
+ gf_boolean_t addr_match = args->addr_match;
+ char *brick_host = NULL;
+ dht_conf_t *conf = this->private;
+ int ret = -1;
+
+ /*This means a local subvol was already found. We pick the first brick
+ * that is local*/
+ if (conf->private)
+ return;
+
+ if (strcmp (xl->name, local_volname) == 0) {
+ conf->private = xl;
+ gf_log (this->name, GF_LOG_INFO, "Using specified subvol %s",
+ local_volname);
+ return;
+ }
+
+ if (!addr_match)
+ return;
+
+ ret = dict_get_str (xl->options, "remote-host", &brick_host);
+ if ((ret == 0) &&
+ (gf_is_same_address (local_volname, brick_host) ||
+ gf_is_local_addr (brick_host))) {
+ conf->private = xl;
+ gf_log (this->name, GF_LOG_INFO, "Using the first local "
+ "subvol %s", xl->name);
+ return;
+ }
- return ret;
}
-void
-fini (xlator_t *this)
+static void
+nufa_to_dht (xlator_t *this)
{
- int i = 0;
- dht_conf_t *conf = NULL;
+ GF_ASSERT (this);
+ GF_ASSERT (this->fops);
- conf = this->private;
+ this->fops->lookup = dht_lookup;
+ this->fops->create = dht_create;
+ this->fops->mknod = dht_mknod;
+}
- 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);
- }
+int
+nufa_find_local_subvol (xlator_t *this,
+ void (*fn) (xlator_t *each, void* data), void *data)
+{
+ int ret = -1;
+ dht_conf_t *conf = this->private;
+ xlator_list_t *trav = NULL;
+ xlator_t *parent = NULL;
+ xlator_t *candidate = NULL;
+
+ xlator_foreach_depth_first (this, fn, data);
+ if (!conf->private) {
+ gf_log (this->name, GF_LOG_ERROR, "Couldn't find a local "
+ "brick");
+ return -1;
+ }
- if (conf->subvolumes)
- GF_FREE (conf->subvolumes);
+ candidate = conf->private;
+ trav = candidate->parents;
+ while (trav) {
- if (conf->subvolume_status)
- GF_FREE (conf->subvolume_status);
+ parent = trav->xlator;
+ if (strcmp (parent->type, "cluster/nufa") == 0) {
+ gf_log (this->name, GF_LOG_INFO, "Found local subvol, "
+ "%s", candidate->name);
+ ret = 0;
+ conf->private = candidate;
+ break;
+ }
- GF_FREE (conf);
+ candidate = parent;
+ trav = parent->parents;
}
- return;
+ return ret;
}
int
-init (xlator_t *this)
+nufa_init (xlator_t *this)
{
- dht_conf_t *conf = NULL;
- xlator_list_t *trav = NULL;
data_t *data = NULL;
char *local_volname = NULL;
- char *temp_str = NULL;
int ret = -1;
- int i = 0;
char my_hostname[256];
- uint32_t temp_free_disk = 0;
+ gf_boolean_t addr_match = _gf_false;
+ nufa_args_t args = {0, };
- if (!this->children) {
- gf_log (this->name, GF_LOG_CRITICAL,
- "NUFA needs more than one subvolume");
- return -1;
+ ret = dht_init(this);
+ if (ret) {
+ return ret;
}
- if (!this->parents) {
- gf_log (this->name, GF_LOG_WARNING,
- "dangling volume. check volfile");
- }
+ if ((data = dict_get (this->options, "local-volume-name"))) {
+ local_volname = data->data;
- conf = GF_CALLOC (1, sizeof (*conf),
- gf_dht_mt_dht_conf_t);
- if (!conf) {
- goto err;
- }
+ } else {
+ addr_match = _gf_true;
+ local_volname = "localhost";
+ ret = gethostname (my_hostname, 256);
+ if (ret == 0)
+ local_volname = my_hostname;
- 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;
- }
+ gf_log (this->name, GF_LOG_WARNING,
+ "could not find hostname (%s)",
+ strerror (errno));
- 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";
- ret = gethostname (my_hostname, 256);
- if (ret < 0) {
- gf_log (this->name, GF_LOG_WARNING,
- "could not find hostname (%s)",
- strerror (errno));
- }
-
- if (ret == 0)
- local_volname = my_hostname;
-
- data = dict_get (this->options, "local-volume-name");
- if (data) {
- local_volname = data->data;
}
- trav = this->children;
- while (trav) {
- if (strcmp (trav->xlator->name, local_volname) == 0)
- break;
- trav = trav->next;
+ args.this = this;
+ args.volname = local_volname;
+ args.addr_match = addr_match;
+ ret = nufa_find_local_subvol (this, nufa_find_local_brick, &args);
+ if (ret) {
+ gf_log (this->name, GF_LOG_INFO,
+ "Unable to find local subvolume, switching "
+ "to dht mode");
+ nufa_to_dht (this);
}
-
- if (!trav) {
- gf_log (this->name, GF_LOG_ERROR,
- "Could not find subvolume named '%s'. "
- "Please define volume with the name as the hostname "
- "or override it with 'option local-volume-name'",
- local_volname);
- goto err;
- }
- /* The volume specified exists */
- conf->private = trav->xlator;
-
- 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,
- &conf->min_free_disk);
- conf->disk_unit = 'b';
- } else {
- conf->min_free_disk = (uint64_t)temp_free_disk;
- conf->disk_unit = 'p';
- }
- } else {
- gf_string2bytesize (temp_str,
- &conf->min_free_disk);
- 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->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);
- }
-
- if (conf->subvolumes)
- GF_FREE (conf->subvolumes);
-
- if (conf->subvolume_status)
- GF_FREE (conf->subvolume_status);
-
- if (conf->du_stats)
- GF_FREE (conf->du_stats);
-
- GF_FREE (conf);
- }
- return -1;
-}
+class_methods_t class_methods = {
+ .init = nufa_init,
+ .fini = dht_fini,
+ .reconfigure = dht_reconfigure,
+ .notify = dht_notify
+};
struct xlator_fops fops = {
@@ -717,19 +674,3 @@ struct xlator_fops fops = {
struct xlator_cbks cbks = {
.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} },
-};
generated by cgit (git 2.34.1) at 2026-01-05 01:38:16 +0000