summaryrefslogtreecommitdiffstats
path: root/xlators/cluster/dht/src/dht-common.c
diff options
context:
space:
mode:
Diffstat (limited to 'xlators/cluster/dht/src/dht-common.c')
-rw-r--r--xlators/cluster/dht/src/dht-common.c671
1 files changed, 443 insertions, 228 deletions
diff --git a/xlators/cluster/dht/src/dht-common.c b/xlators/cluster/dht/src/dht-common.c
index 99cf6f787..8f61339e6 100644
--- a/xlators/cluster/dht/src/dht-common.c
+++ b/xlators/cluster/dht/src/dht-common.c
@@ -22,6 +22,7 @@
#include "dht-common.h"
#include "defaults.h"
#include "byte-order.h"
+#include "glusterfs-acl.h"
#include <sys/time.h>
#include <libgen.h>
@@ -62,6 +63,11 @@ dht_aggregate (dict_t *this, char *key, data_t *value, void *data)
}
*size = hton64 (ntoh64 (*size) + ntoh64 (*ptr));
+
+ } else if (fnmatch (GF_XATTR_STIME_PATTERN, key, FNM_NOESCAPE) == 0) {
+ ret = gf_get_min_stime (THIS, dst, key, value);
+ if (ret < 0)
+ return ret;
} else {
/* compare user xattrs only */
if (!strncmp (key, "user.", strlen ("user."))) {
@@ -148,9 +154,11 @@ dht_discover_complete (xlator_t *this, call_frame_t *discover_frame)
int op_errno = 0;
int ret = -1;
dht_layout_t *layout = NULL;
+ dht_conf_t *conf = NULL;
local = discover_frame->local;
layout = local->layout;
+ conf = this->private;
LOCK(&discover_frame->lock);
{
@@ -193,11 +201,14 @@ dht_discover_complete (xlator_t *this, call_frame_t *discover_frame)
"(overlaps/holes present: %s, "
"ENOENT errors: %d)", local->loc.path,
(ret < 0) ? "yes" : "no", (ret > 0) ? ret : 0);
- op_errno = EINVAL;
- goto out;
+ if ((ret > 0) && (ret == conf->subvolume_cnt)) {
+ op_errno = ESTALE;
+ goto out;
+ }
}
- dht_layout_set (this, local->inode, layout);
+ if (local->inode)
+ dht_layout_set (this, local->inode, layout);
}
DHT_STACK_UNWIND (lookup, main_frame, local->op_ret, local->op_errno,
@@ -226,6 +237,7 @@ dht_discover_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
int is_dir = 0;
int is_linkfile = 0;
int attempt_unwind = 0;
+ dht_conf_t *conf = 0;
GF_VALIDATE_OR_GOTO ("dht", frame, out);
GF_VALIDATE_OR_GOTO ("dht", this, out);
@@ -235,6 +247,7 @@ dht_discover_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
local = frame->local;
prev = cookie;
+ conf = this->private;
layout = local->layout;
@@ -269,7 +282,8 @@ dht_discover_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
goto unlock;
}
- 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) {
@@ -328,23 +342,20 @@ dht_discover (call_frame_t *frame, xlator_t *this, loc_t *loc)
int i = 0;
call_frame_t *discover_frame = NULL;
-
conf = this->private;
local = frame->local;
- ret = dict_set_uint32 (local->xattr_req,
- "trusted.glusterfs.dht", 4 * 4);
+ ret = dict_set_uint32 (local->xattr_req, conf->xattr_name, 4 * 4);
if (ret)
gf_log (this->name, GF_LOG_WARNING,
- "%s: failed to set 'trusted.glusterfs.dht' key",
- loc->path);
+ "%s: failed to set '%s' key",
+ loc->path, conf->xattr_name);
- ret = dict_set_uint32 (local->xattr_req,
- "trusted.glusterfs.dht.linkto", 256);
+ ret = dict_set_uint32 (local->xattr_req, conf->link_xattr_name, 256);
if (ret)
gf_log (this->name, GF_LOG_WARNING,
- "%s: failed to set 'trusted.glusterfs.dht.linkto' key",
- loc->path);
+ "%s: failed to set '%s' key",
+ loc->path, conf->link_xattr_name);
call_cnt = conf->subvolume_cnt;
local->call_cnt = call_cnt;
@@ -430,7 +441,7 @@ dht_lookup_dir_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
op_ret, op_errno, xattr);
if (op_ret == -1) {
- local->op_errno = ENOENT;
+ local->op_errno = op_errno;
gf_log (this->name, GF_LOG_DEBUG,
"lookup of %s on %s returned error (%s)",
local->loc.path, prev->this->name,
@@ -585,7 +596,8 @@ dht_revalidate_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
layout = local->layout;
is_dir = check_is_dir (inode, stbuf, xattr);
- is_linkfile = check_is_linkfile (inode, stbuf, xattr);
+ is_linkfile = check_is_linkfile (inode, stbuf, xattr,
+ conf->link_xattr_name);
if (is_linkfile) {
gf_log (this->name, GF_LOG_INFO,
@@ -597,7 +609,7 @@ dht_revalidate_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
}
if (is_dir) {
- ret = dht_dir_has_layout (xattr);
+ ret = dht_dir_has_layout (xattr, conf->xattr_name);
if (ret >= 0) {
if (is_greater_time(local->stbuf.ia_ctime,
local->stbuf.ia_ctime_nsec,
@@ -760,6 +772,9 @@ dht_lookup_linkfile_create_cbk (call_frame_t *frame, void *cookie,
}
unwind:
+ if (local->linked == _gf_true)
+ dht_linkfile_attr_heal (frame, this);
+
DHT_STRIP_PHASE1_FLAGS (&local->stbuf);
DHT_STACK_UNWIND (lookup, frame, local->op_ret, local->op_errno,
local->inode, &local->stbuf, local->xattr,
@@ -883,7 +898,7 @@ dht_lookup_everywhere_done (call_frame_t *frame, xlator_t *this)
hashed_subvol->name);
ret = dht_linkfile_create (frame,
- dht_lookup_linkfile_create_cbk,
+ dht_lookup_linkfile_create_cbk, this,
cached_subvol, hashed_subvol, &local->loc);
return ret;
@@ -921,8 +936,9 @@ dht_lookup_everywhere_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
xlator_t *subvol = NULL;
loc_t *loc = NULL;
xlator_t *link_subvol = NULL;
- int ret = -1;
- int32_t fd_count = 0;
+ int ret = -1;
+ int32_t fd_count = 0;
+ dht_conf_t *conf = NULL;
GF_VALIDATE_OR_GOTO ("dht", frame, out);
GF_VALIDATE_OR_GOTO ("dht", this, out);
@@ -932,6 +948,7 @@ dht_lookup_everywhere_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
local = frame->local;
loc = &local->loc;
+ conf = this->private;
prev = cookie;
subvol = prev->this;
@@ -953,7 +970,8 @@ dht_lookup_everywhere_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
loc->path, prev->this->name);
}
- is_linkfile = check_is_linkfile (inode, buf, xattr);
+ is_linkfile = check_is_linkfile (inode, buf, xattr,
+ conf->link_xattr_name);
is_dir = check_is_dir (inode, buf, xattr);
if (is_linkfile) {
@@ -1114,7 +1132,7 @@ dht_lookup_linkfile_cbk (call_frame_t *frame, void *cookie,
goto err;
}
- if (check_is_linkfile (inode, stbuf, xattr)) {
+ if (check_is_linkfile (inode, stbuf, xattr, conf->link_xattr_name)) {
gf_log (this->name, GF_LOG_INFO,
"lookup of %s on %s (following linkfile) reached link",
local->loc.path, subvol->name);
@@ -1291,7 +1309,8 @@ dht_lookup_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
goto out;
}
- is_linkfile = check_is_linkfile (inode, stbuf, xattr);
+ is_linkfile = check_is_linkfile (inode, stbuf, xattr,
+ conf->link_xattr_name);
if (!is_linkfile) {
/* non-directory and not a linkfile */
@@ -1396,7 +1415,6 @@ dht_lookup (call_frame_t *frame, xlator_t *this,
VALIDATE_OR_GOTO (this, err);
VALIDATE_OR_GOTO (loc, err);
VALIDATE_OR_GOTO (loc->inode, err);
- VALIDATE_OR_GOTO (loc->path, err);
conf = this->private;
if (!conf)
@@ -1471,7 +1489,7 @@ dht_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 (IA_ISDIR (local->inode->ia_type)) {
local->call_cnt = call_cnt = conf->subvolume_cnt;
@@ -1506,10 +1524,10 @@ dht_lookup (call_frame_t *frame, xlator_t *this,
do_fresh_lookup:
/* TODO: remove the hard-coding */
ret = dict_set_uint32 (local->xattr_req,
- "trusted.glusterfs.dht", 4 * 4);
+ conf->xattr_name, 4 * 4);
ret = dict_set_uint32 (local->xattr_req,
- DHT_LINKFILE_KEY, 256);
+ conf->link_xattr_name, 256);
/* need it for self-healing linkfiles which is
'in-migration' state */
@@ -1617,7 +1635,8 @@ dht_unlink_linkfile_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
LOCK (&frame->lock);
{
- if (op_ret == -1) {
+ if ((op_ret == -1) && !((op_errno == ENOENT) ||
+ (op_errno == ENOTCONN))) {
local->op_errno = op_errno;
gf_log (this->name, GF_LOG_DEBUG,
"subvolume %s returned -1 (%s)",
@@ -1630,7 +1649,7 @@ dht_unlink_linkfile_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
unlock:
UNLOCK (&frame->lock);
- if (op_ret == -1)
+ if (local->op_ret == -1)
goto err;
cached_subvol = dht_subvol_get_cached (this, local->loc.inode);
@@ -1654,41 +1673,6 @@ err:
return 0;
}
-static int
-dht_ufo_xattr_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
- int op_ret, int op_errno, dict_t *xdata)
-{
- dht_local_t *local = NULL;
- int this_call_cnt = 0;
- call_frame_t *prev = NULL;
-
- local = frame->local;
- prev = cookie;
-
- LOCK (&frame->lock);
- {
- if (op_ret == -1) {
- local->op_ret = -1;
- local->op_errno = op_errno;
- gf_log (this->name, GF_LOG_DEBUG,
- "subvolume %s returned -1 (%s)",
- prev->this->name, strerror (op_errno));
- goto unlock;
- }
- }
-unlock:
- UNLOCK (&frame->lock);
-
- this_call_cnt = dht_frame_return (frame);
- if (is_last_call (this_call_cnt)) {
- DHT_STACK_UNWIND (setxattr, frame, local->op_ret,
- local->op_errno, NULL);
- }
-
- return 0;
-}
-
-
int
dht_err_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
int op_ret, int op_errno, dict_t *xdata)
@@ -1761,115 +1745,222 @@ dht_fill_pathinfo_xattr (xlator_t *this, dht_local_t *local,
}
int
-dht_vgetxattr_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
- int op_ret, int op_errno, dict_t *xattr, dict_t *xdata)
+dht_vgetxattr_alloc_and_fill (dht_local_t *local, dict_t *xattr, xlator_t *this,
+ int op_errno)
{
- dht_local_t *local = NULL;
- int ret = 0;
- int flag = 0;
- int this_call_cnt = 0;
- char *value_got = NULL;
- char layout_buf[8192] = {0,};
- char *xattr_buf = NULL;
- dict_t *dict = NULL;
- int32_t alloc_len = 0;
- int32_t plen = 0;
- call_frame_t *prev = NULL;
+ int ret = -1;
+ char *value = NULL;
+ int32_t plen = 0;
- local = frame->local;
- prev = cookie;
+ ret = dict_get_str (xattr, local->xsel, &value);
+ if (ret) {
+ gf_log (this->name, GF_LOG_ERROR,
+ "Subvolume %s returned -1 (%s)", this->name,
+ strerror (op_errno));
+ local->op_ret = -1;
+ local->op_errno = op_errno;
+ goto out;
+ }
- if (op_ret >= 0) {
- ret = dict_get_str (xattr, local->xsel, &value_got);
- if (!ret) {
- alloc_len = strlen (value_got);
+ local->alloc_len += strlen(value);
- /**
- * allocate the buffer:- we allocate 10 bytes extra in
- * case we need to append ' Link: ' in the buffer for
- * another STACK_WIND
- */
+ if (!local->xattr_val) {
+ local->alloc_len += (strlen (DHT_PATHINFO_HEADER) + 10);
+ local->xattr_val = GF_CALLOC (local->alloc_len, sizeof (char),
+ gf_common_mt_char);
+ if (!local->xattr_val) {
+ ret = -1;
+ goto out;
+ }
+ }
+
+ if (local->xattr_val) {
+ plen = strlen (local->xattr_val);
+ if (plen) {
+ /* extra byte(s) for \0 to be safe */
+ local->alloc_len += (plen + 2);
+ local->xattr_val = GF_REALLOC (local->xattr_val,
+ local->alloc_len);
if (!local->xattr_val) {
- alloc_len += (strlen (DHT_PATHINFO_HEADER) + 10);
- local->xattr_val =
- GF_CALLOC (alloc_len,
- sizeof (char),
- gf_common_mt_char);
+ ret = -1;
+ goto out;
}
+ }
- if (local->xattr_val) {
- plen = strlen (local->xattr_val);
- if (plen) {
- void *p;
- /* extra byte(s) for \0 to be safe */
- alloc_len += (plen + 2);
- p = GF_REALLOC (local->xattr_val,
- alloc_len);
- if (!p)
- goto out;
- local->xattr_val = p;
- }
+ (void) strcat (local->xattr_val, value);
+ (void) strcat (local->xattr_val, " ");
+ local->op_ret = 0;
+ }
- strcat (local->xattr_val, value_got);
- }
- local->op_ret = 0;
- }
+ ret = 0;
+
+ out:
+ return ret;
+}
+
+int
+dht_vgetxattr_fill_and_set (dht_local_t *local, dict_t **dict, xlator_t *this,
+ gf_boolean_t flag)
+{
+ int ret = -1;
+ char *xattr_buf = NULL;
+ char layout_buf[8192] = {0,};
+
+ if (flag)
+ fill_layout_info (local->layout, layout_buf);
+
+ *dict = dict_new ();
+ if (!*dict)
+ goto out;
+
+ local->xattr_val[strlen (local->xattr_val) - 1] = '\0';
+
+ /* we would need max this many bytes to create xattr string
+ * extra 40 bytes is just an estimated amount of additional
+ * space required as we include translator name and some
+ * spaces, brackets etc. when forming the pathinfo string.
+ *
+ * For node-uuid we just don't have all the pretty formatting,
+ * but since this is a generic routine for pathinfo & node-uuid
+ * we dont have conditional space allocation and try to be
+ * generic
+ */
+ local->alloc_len += (2 * strlen (this->name))
+ + strlen (layout_buf)
+ + 40;
+ xattr_buf = GF_CALLOC (local->alloc_len, sizeof (char),
+ gf_common_mt_char);
+ if (!xattr_buf)
+ goto out;
+
+ if (XATTR_IS_PATHINFO (local->xsel)) {
+ (void) dht_fill_pathinfo_xattr (this, local, xattr_buf,
+ local->alloc_len, flag,
+ layout_buf);
+ } else if (XATTR_IS_NODE_UUID (local->xsel)) {
+ (void) snprintf (xattr_buf, local->alloc_len, "%s",
+ local->xattr_val);
} else {
- local->op_ret = -1;
- local->op_errno = op_errno;
- gf_log (this->name, GF_LOG_ERROR, "Subvolume %s returned -1 "
- "(%s)", prev->this->name, strerror (op_errno));
+ gf_log (this->name, GF_LOG_WARNING,
+ "Unknown local->xsel (%s)", local->xsel);
+ goto out;
}
+ ret = dict_set_dynstr (*dict, local->xsel, xattr_buf);
+ GF_FREE (local->xattr_val);
+
out:
- this_call_cnt = dht_frame_return (frame);
- if (is_last_call (this_call_cnt)) {
+ return ret;
+}
- if (local->op_ret == -1) {
- goto unwind;
- }
+int
+dht_vgetxattr_dir_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
+ int op_ret, int op_errno, dict_t *xattr, dict_t *xdata)
+{
+ int ret = 0;
+ dht_local_t *local = NULL;
+ int this_call_cnt = 0;
+ dict_t *dict = NULL;
- if (local->layout->cnt > 1) {
- /* Set it for directory */
- fill_layout_info (local->layout, layout_buf);
- flag = 1;
- }
+ VALIDATE_OR_GOTO (frame, out);
+ VALIDATE_OR_GOTO (frame->local, out);
- dict = dict_new ();
-
- /* we would need max this many bytes to create xattr string */
- alloc_len += (2 * strlen (this->name))
- + strlen (layout_buf)
- + 40;
- xattr_buf = GF_CALLOC (alloc_len,
- sizeof (char), gf_common_mt_char);
-
- if (XATTR_IS_PATHINFO (local->xsel)) {
- (void) dht_fill_pathinfo_xattr (this, local, xattr_buf,
- alloc_len, flag,
- layout_buf);
- } else if (XATTR_IS_NODE_UUID (local->xsel)) {
- (void) snprintf (xattr_buf, alloc_len, "%s",
- local->xattr_val);
- } else
- gf_log (this->name, GF_LOG_WARNING,
- "Unknown local->xsel (%s)", local->xsel);
+ local = frame->local;
- ret = dict_set_dynstr (dict, local->xsel, xattr_buf);
+ LOCK (&frame->lock);
+ {
+ this_call_cnt = --local->call_cnt;
+ if (op_ret < 0) {
+ if (op_errno != ENOTCONN) {
+ gf_log (this->name, GF_LOG_ERROR,
+ "getxattr err (%s) for dir",
+ strerror (op_errno));
+ local->op_ret = -1;
+ local->op_errno = op_errno;
+ }
- GF_FREE (local->xattr_val);
+ goto unlock;
+ }
- DHT_STACK_UNWIND (getxattr, frame, op_ret, op_errno, dict,
- xdata);
+ ret = dht_vgetxattr_alloc_and_fill (local, xattr, this,
+ op_errno);
+ if (ret)
+ gf_log (this->name, GF_LOG_ERROR,
+ "alloc or fill failure");
+ }
+ unlock:
+ UNLOCK (&frame->lock);
- if (dict)
- dict_unref (dict);
+ if (!is_last_call (this_call_cnt))
+ goto out;
- return 0;
+ /* -- last call: do patch ups -- */
+
+ if (local->op_ret == -1) {
+ goto unwind;
}
-unwind:
+ ret = dht_vgetxattr_fill_and_set (local, &dict, this, _gf_true);
+ if (ret)
+ goto unwind;
+
+ DHT_STACK_UNWIND (getxattr, frame, 0, 0, dict, xdata);
+ goto cleanup;
+
+ unwind:
DHT_STACK_UNWIND (getxattr, frame, -1, local->op_errno, NULL, NULL);
+ cleanup:
+ if (dict)
+ dict_unref (dict);
+ out:
+ return 0;
+}
+
+int
+dht_vgetxattr_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
+ int op_ret, int op_errno, dict_t *xattr, dict_t *xdata)
+{
+ dht_local_t *local = NULL;
+ int ret = 0;
+ dict_t *dict = NULL;
+ call_frame_t *prev = NULL;
+ gf_boolean_t flag = _gf_true;
+
+ local = frame->local;
+ prev = cookie;
+
+ if (op_ret < 0) {
+ local->op_ret = -1;
+ local->op_errno = op_errno;
+ gf_log (this->name, GF_LOG_ERROR, "Subvolume %s returned -1 "
+ "(%s)", prev->this->name, strerror (op_errno));
+ goto unwind;
+ }
+
+ ret = dht_vgetxattr_alloc_and_fill (local, xattr, this,
+ op_errno);
+ if (ret) {
+ gf_log (this->name, GF_LOG_ERROR,
+ "alloc or fill failure");
+ goto unwind;
+ }
+
+ flag = (local->layout->cnt > 1) ? _gf_true : _gf_false;
+
+ ret = dht_vgetxattr_fill_and_set (local, &dict, this, flag);
+ if (ret)
+ goto unwind;
+
+ DHT_STACK_UNWIND (getxattr, frame, 0, 0, dict, xdata);
+ goto cleanup;
+
+ unwind:
+ DHT_STACK_UNWIND (getxattr, frame, -1, local->op_errno,
+ NULL, NULL);
+ cleanup:
+ if (dict)
+ dict_unref (dict);
+
return 0;
}
@@ -1902,10 +1993,13 @@ dht_getxattr_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
{
int this_call_cnt = 0;
dht_local_t *local = NULL;
+ dht_conf_t *conf = NULL;
VALIDATE_OR_GOTO (frame, out);
VALIDATE_OR_GOTO (frame->local, out);
+ VALIDATE_OR_GOTO (this->private, out);
+ conf = this->private;
local = frame->local;
this_call_cnt = dht_frame_return (frame);
@@ -1913,8 +2007,8 @@ dht_getxattr_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
if (!xattr || (op_ret == -1))
goto out;
- if (dict_get (xattr, "trusted.glusterfs.dht")) {
- dict_del (xattr, "trusted.glusterfs.dht");
+ if (dict_get (xattr, conf->xattr_name)) {
+ dict_del (xattr, conf->xattr_name);
}
local->op_ret = 0;
@@ -1947,9 +2041,72 @@ dht_getxattr_unwind (call_frame_t *frame,
int
+dht_getxattr_get_real_filename_cbk (call_frame_t *frame, void *cookie,
+ xlator_t *this, int op_ret, int op_errno,
+ dict_t *xattr, dict_t *xdata)
+{
+ int this_call_cnt = 0;
+ dht_local_t *local = NULL;
+
+
+ local = frame->local;
+
+ if (op_ret != -1) {
+ if (local->xattr)
+ dict_unref (local->xattr);
+ local->xattr = dict_ref (xattr);
+
+ if (local->xattr_req)
+ dict_unref (local->xattr_req);
+ local->xattr_req = dict_ref (xdata);
+ }
+
+ this_call_cnt = dht_frame_return (frame);
+ if (is_last_call (this_call_cnt)) {
+ DHT_STACK_UNWIND (getxattr, frame, local->op_ret, op_errno,
+ local->xattr, local->xattr_req);
+ }
+
+ return 0;
+}
+
+
+int
+dht_getxattr_get_real_filename (call_frame_t *frame, xlator_t *this,
+ loc_t *loc, const char *key, dict_t *xdata)
+{
+ dht_local_t *local = NULL;
+ int i = 0;
+ dht_layout_t *layout = NULL;
+ int cnt = 0;
+ xlator_t *subvol = NULL;
+
+
+ local = frame->local;
+ layout = local->layout;
+
+ cnt = local->call_cnt = layout->cnt;
+
+ local->op_ret = -1;
+ local->op_errno = ENODATA;
+
+ for (i = 0; i < cnt; i++) {
+ subvol = layout->list[i].xlator;
+ STACK_WIND (frame, dht_getxattr_get_real_filename_cbk,
+ subvol, subvol->fops->getxattr,
+ loc, key, xdata);
+ }
+
+ return 0;
+}
+
+
+int
dht_getxattr (call_frame_t *frame, xlator_t *this,
loc_t *loc, const char *key, dict_t *xdata)
+#define DHT_IS_DIR(layout) (layout->cnt > 1)
{
+
xlator_t *subvol = NULL;
xlator_t *hashed_subvol = NULL;
xlator_t *cached_subvol = NULL;
@@ -1965,7 +2122,6 @@ dht_getxattr (call_frame_t *frame, xlator_t *this,
VALIDATE_OR_GOTO (this, err);
VALIDATE_OR_GOTO (loc, err);
VALIDATE_OR_GOTO (loc->inode, err);
- VALIDATE_OR_GOTO (loc->path, err);
VALIDATE_OR_GOTO (this->private, err);
conf = this->private;
@@ -1993,8 +2149,40 @@ dht_getxattr (call_frame_t *frame, xlator_t *this,
}
}
- if (key && ((strcmp (key, GF_XATTR_PATHINFO_KEY) == 0)
- || strcmp (key, GF_XATTR_NODE_UUID_KEY) == 0)) {
+ if (key &&
+ (strncmp (key, GF_XATTR_GET_REAL_FILENAME_KEY,
+ strlen (GF_XATTR_GET_REAL_FILENAME_KEY)) == 0)
+ && DHT_IS_DIR(layout)) {
+ dht_getxattr_get_real_filename (frame, this, loc, key, xdata);
+ return 0;
+ }
+
+ /* for file use cached subvolume (obviously!): see if {}
+ * below
+ * for directory:
+ * wind to all subvolumes and exclude subvolumes which
+ * return ENOTCONN (in callback)
+ *
+ * NOTE: Don't trust inode here, as that may not be valid
+ * (until inode_link() happens)
+ */
+ if (key && DHT_IS_DIR(layout) &&
+ ((strcmp (key, GF_XATTR_PATHINFO_KEY) == 0)
+ || (strcmp (key, GF_XATTR_NODE_UUID_KEY) == 0))) {
+ (void) strncpy (local->xsel, key, 256);
+ cnt = local->call_cnt = layout->cnt;
+ for (i = 0; i < cnt; i++) {
+ subvol = layout->list[i].xlator;
+ STACK_WIND (frame, dht_vgetxattr_dir_cbk,
+ subvol, subvol->fops->getxattr,
+ loc, key, NULL);
+ }
+ return 0;
+ }
+
+ /* node-uuid or pathinfo for files */
+ if (key && ((strcmp (key, GF_XATTR_NODE_UUID_KEY) == 0)
+ || (strcmp (key, GF_XATTR_PATHINFO_KEY) == 0))) {
cached_subvol = local->cached_subvol;
(void) strncpy (local->xsel, key, 256);
@@ -2004,6 +2192,7 @@ dht_getxattr (call_frame_t *frame, xlator_t *this,
return 0;
}
+
if (key && (strcmp (key, GF_XATTR_LINKINFO_KEY) == 0)) {
hashed_subvol = dht_subvol_get_hashed (this, loc);
if (!hashed_subvol) {
@@ -2036,13 +2225,13 @@ dht_getxattr (call_frame_t *frame, xlator_t *this,
}
if (key && (!strcmp (GF_XATTR_MARKER_KEY, key))
- && (-1 == frame->root->pid)) {
-
- if (loc->inode-> ia_type == IA_IFDIR) {
+ && (GF_CLIENT_PID_GSYNCD == frame->root->pid)) {
+ if (DHT_IS_DIR(layout)) {
cnt = layout->cnt;
} else {
cnt = 1;
}
+
sub_volumes = alloca ( cnt * sizeof (xlator_t *));
for (i = 0; i < cnt; i++)
*(sub_volumes + i) = layout->list[i].xlator;
@@ -2050,7 +2239,8 @@ dht_getxattr (call_frame_t *frame, xlator_t *this,
if (cluster_getmarkerattr (frame, this, loc, key,
local, dht_getxattr_unwind,
sub_volumes, cnt,
- MARKER_UUID_TYPE, conf->vol_uuid)) {
+ MARKER_UUID_TYPE, marker_uuid_default_gauge,
+ conf->vol_uuid)) {
op_errno = EINVAL;
goto err;
}
@@ -2060,8 +2250,8 @@ dht_getxattr (call_frame_t *frame, xlator_t *this,
if (key && *conf->vol_uuid) {
if ((match_uuid_local (key, conf->vol_uuid) == 0) &&
- (-1 == frame->root->pid)) {
- if (loc->inode-> ia_type == IA_IFDIR) {
+ (GF_CLIENT_PID_GSYNCD == frame->root->pid)) {
+ if (DHT_IS_DIR(layout)) {
cnt = layout->cnt;
} else {
cnt = 1;
@@ -2074,6 +2264,7 @@ dht_getxattr (call_frame_t *frame, xlator_t *this,
local, dht_getxattr_unwind,
sub_volumes, cnt,
MARKER_XTIME_TYPE,
+ marker_xtime_default_gauge,
conf->vol_uuid)) {
op_errno = EINVAL;
goto err;
@@ -2083,7 +2274,7 @@ dht_getxattr (call_frame_t *frame, xlator_t *this,
}
}
- if (loc->inode-> ia_type == IA_IFDIR) {
+ if (DHT_IS_DIR(layout)) {
cnt = local->call_cnt = layout->cnt;
} else {
cnt = local->call_cnt = 1;
@@ -2103,6 +2294,7 @@ err:
return 0;
}
+#undef DHT_IS_DIR
int
dht_fgetxattr (call_frame_t *frame, xlator_t *this,
@@ -2174,13 +2366,17 @@ dht_fsetxattr (call_frame_t *frame, xlator_t *this,
xlator_t *subvol = NULL;
dht_local_t *local = NULL;
int op_errno = EINVAL;
+ dht_conf_t *conf = NULL;
VALIDATE_OR_GOTO (frame, err);
VALIDATE_OR_GOTO (this, err);
VALIDATE_OR_GOTO (fd, err);
VALIDATE_OR_GOTO (fd->inode, err);
+ VALIDATE_OR_GOTO (this->private, err);
+
+ conf = this->private;
- GF_IF_INTERNAL_XATTR_GOTO ("trusted.glusterfs.dht*", xattr,
+ GF_IF_INTERNAL_XATTR_GOTO (conf->wild_xattr_name, xattr,
op_errno, err);
local = dht_local_init (frame, NULL, fd, GF_FOP_FSETXATTR);
@@ -2284,12 +2480,12 @@ dht_setxattr (call_frame_t *frame, xlator_t *this,
VALIDATE_OR_GOTO (this, err);
VALIDATE_OR_GOTO (loc, err);
VALIDATE_OR_GOTO (loc->inode, err);
- VALIDATE_OR_GOTO (loc->path, err);
- GF_IF_INTERNAL_XATTR_GOTO ("trusted.glusterfs.dht*", xattr,
+ conf = this->private;
+
+ GF_IF_INTERNAL_XATTR_GOTO (conf->wild_xattr_name, xattr,
op_errno, err);
- conf = this->private;
local = dht_local_init (frame, loc, NULL, GF_FOP_SETXATTR);
if (!local) {
op_errno = ENOMEM;
@@ -2314,25 +2510,6 @@ dht_setxattr (call_frame_t *frame, xlator_t *this,
local->call_cnt = call_cnt = layout->cnt;
- /* This key is sent by Unified File and Object storage
- * to test xattr support in backend.
- */
- tmp = dict_get (xattr, "user.ufo-test");
- if (tmp) {
- if (IA_ISREG (loc->inode->ia_type)) {
- op_errno = ENOTSUP;
- goto err;
- }
- local->op_ret = 0;
- for (i = 0; i < call_cnt; i++) {
- STACK_WIND (frame, dht_ufo_xattr_cbk,
- layout->list[i].xlator,
- layout->list[i].xlator->fops->setxattr,
- loc, xattr, flags, NULL);
- }
- return 0;
- }
-
tmp = dict_get (xattr, "distribute.migrate-data");
if (tmp) {
if (IA_ISDIR (loc->inode->ia_type)) {
@@ -2502,18 +2679,20 @@ dht_removexattr (call_frame_t *frame, xlator_t *this,
dht_local_t *local = NULL;
dht_layout_t *layout = NULL;
int call_cnt = 0;
+ dht_conf_t *conf = NULL;
int i;
VALIDATE_OR_GOTO (this, err);
+ VALIDATE_OR_GOTO (this->private, err);
+
+ conf = this->private;
- GF_IF_NATIVE_XATTR_GOTO ("trusted.glusterfs.dht*",
- key, op_errno, err);
+ GF_IF_NATIVE_XATTR_GOTO (conf->wild_xattr_name, key, op_errno, err);
VALIDATE_OR_GOTO (frame, err);
VALIDATE_OR_GOTO (loc, err);
VALIDATE_OR_GOTO (loc->inode, err);
- VALIDATE_OR_GOTO (loc->path, err);
local = dht_local_init (frame, loc, NULL, GF_FOP_REMOVEXATTR);
if (!local) {
@@ -2565,13 +2744,16 @@ dht_fremovexattr (call_frame_t *frame, xlator_t *this,
dht_local_t *local = NULL;
dht_layout_t *layout = NULL;
int call_cnt = 0;
+ dht_conf_t *conf = 0;
int i;
VALIDATE_OR_GOTO (this, err);
+ VALIDATE_OR_GOTO (this->private, err);
- GF_IF_NATIVE_XATTR_GOTO ("trusted.glusterfs.dht*",
- key, op_errno, err);
+ conf = this->private;
+
+ GF_IF_NATIVE_XATTR_GOTO (conf->wild_xattr_name, key, op_errno, err);
VALIDATE_OR_GOTO (frame, err);
@@ -2742,7 +2924,6 @@ dht_statfs (call_frame_t *frame, xlator_t *this, loc_t *loc, dict_t *xdata)
VALIDATE_OR_GOTO (this, err);
VALIDATE_OR_GOTO (loc, err);
VALIDATE_OR_GOTO (loc->inode, err);
- VALIDATE_OR_GOTO (loc->path, err);
VALIDATE_OR_GOTO (this->private, err);
conf = this->private;
@@ -2862,10 +3043,13 @@ dht_readdirp_cbk (call_frame_t *frame, void *cookie, xlator_t *this, int op_ret,
list_for_each_entry (orig_entry, (&orig_entries->list), list) {
next_offset = orig_entry->d_off;
- if ((check_is_dir (NULL, (&orig_entry->d_stat), NULL) &&
- (prev->this != dht_first_up_subvol (this))) ||
- check_is_linkfile (NULL, (&orig_entry->d_stat),
- orig_entry->dict)) {
+ if (check_is_dir (NULL, (&orig_entry->d_stat), NULL) &&
+ (prev->this != local->first_up_subvol)) {
+ continue;
+ }
+ if (check_is_linkfile (NULL, (&orig_entry->d_stat),
+ orig_entry->dict,
+ conf->link_xattr_name)) {
continue;
}
@@ -2942,13 +3126,16 @@ done:
}
if (conf->readdir_optimize == _gf_true) {
- if (next_subvol != dht_first_up_subvol (this)) {
+ if (next_subvol != local->first_up_subvol) {
ret = dict_set_int32 (local->xattr,
GF_READDIR_SKIP_DIRS, 1);
if (ret)
gf_log (this->name, GF_LOG_ERROR,
"dict set failed");
- }
+ } else {
+ dict_del (local->xattr,
+ GF_READDIR_SKIP_DIRS);
+ }
}
STACK_WIND (frame, dht_readdirp_cbk,
@@ -3081,6 +3268,7 @@ dht_do_readdir (call_frame_t *frame, xlator_t *this, fd_t *fd, size_t size,
VALIDATE_OR_GOTO (frame, err);
VALIDATE_OR_GOTO (this, err);
VALIDATE_OR_GOTO (fd, err);
+ VALIDATE_OR_GOTO (this->private, err);
conf = this->private;
@@ -3093,6 +3281,7 @@ dht_do_readdir (call_frame_t *frame, xlator_t *this, fd_t *fd, size_t size,
local->fd = fd_ref (fd);
local->size = size;
local->xattr_req = (dict)? dict_ref (dict) : NULL;
+ local->first_up_subvol = dht_first_up_subvol (this);
dht_deitransform (this, yoff, &xvol, (uint64_t *)&xoff);
@@ -3105,20 +3294,22 @@ dht_do_readdir (call_frame_t *frame, xlator_t *this, fd_t *fd, size_t size,
if (local->xattr) {
ret = dict_set_uint32 (local->xattr,
- "trusted.glusterfs.dht.linkto",
- 256);
+ conf->link_xattr_name, 256);
if (ret)
gf_log (this->name, GF_LOG_WARNING,
- "failed to set 'glusterfs.dht.linkto'"
- " key");
+ "failed to set '%s' key",
+ conf->link_xattr_name);
if (conf->readdir_optimize == _gf_true) {
- if (xvol != dht_first_up_subvol (this)) {
+ if (xvol != local->first_up_subvol) {
ret = dict_set_int32 (local->xattr,
GF_READDIR_SKIP_DIRS, 1);
if (ret)
gf_log (this->name,
GF_LOG_ERROR,
"Dict set failed");
+ } else {
+ dict_del (local->xattr,
+ GF_READDIR_SKIP_DIRS);
}
}
}
@@ -3288,6 +3479,8 @@ dht_newfile_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
op_errno = EINVAL;
goto out;
}
+ if (local->linked == _gf_true)
+ dht_linkfile_attr_heal (frame, this);
out:
/*
* FIXME: ia_size and st_blocks of preparent and postparent do not have
@@ -3296,7 +3489,6 @@ out:
* corresponding values from each of the subvolume.
* See dht_iatt_merge for reference.
*/
-
DHT_STRIP_PHASE1_FLAGS (stbuf);
DHT_STACK_UNWIND (mknod, frame, op_ret, op_errno, inode, stbuf,
preparent, postparent, xdata);
@@ -3375,7 +3567,9 @@ dht_mknod (call_frame_t *frame, xlator_t *this,
subvol, subvol->fops->mknod, loc, mode,
rdev, umask, params);
} else {
- avail_subvol = dht_free_disk_available_subvol (this, subvol);
+
+ avail_subvol = dht_free_disk_available_subvol (this, subvol,
+ local);
if (avail_subvol != subvol) {
/* Choose the minimum filled volume, and create the
files there */
@@ -3387,7 +3581,7 @@ dht_mknod (call_frame_t *frame, xlator_t *this,
local->umask = umask;
dht_linkfile_create (frame,
dht_mknod_linkfile_create_cbk,
- avail_subvol, subvol, loc);
+ this, avail_subvol, subvol, loc);
} else {
gf_log (this->name, GF_LOG_TRACE,
"creating %s on %s", loc->path, subvol->name);
@@ -3556,7 +3750,10 @@ dht_link_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
dht_inode_ctx_time_update (local->loc.parent, this,
postparent, 1);
}
-
+ if (local->linked == _gf_true) {
+ local->stbuf = *stbuf;
+ dht_linkfile_attr_heal (frame, this);
+ }
out:
DHT_STRIP_PHASE1_FLAGS (stbuf);
DHT_STACK_UNWIND (link, frame, op_ret, op_errno, inode, stbuf, preparent,
@@ -3643,7 +3840,7 @@ dht_link (call_frame_t *frame, xlator_t *this,
if (hashed_subvol != cached_subvol) {
uuid_copy (local->gfid, oldloc->inode->gfid);
- dht_linkfile_create (frame, dht_link_linkfile_cbk,
+ dht_linkfile_create (frame, dht_link_linkfile_cbk, this,
cached_subvol, hashed_subvol, newloc);
} else {
STACK_WIND (frame, dht_link_cbk,
@@ -3700,7 +3897,10 @@ dht_create_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
op_errno = EINVAL;
goto out;
}
-
+ if (local->linked == _gf_true) {
+ local->stbuf = *stbuf;
+ dht_linkfile_attr_heal (frame, this);
+ }
out:
DHT_STRIP_PHASE1_FLAGS (stbuf);
DHT_STACK_UNWIND (create, frame, op_ret, op_errno, fd, inode, stbuf, preparent,
@@ -3790,7 +3990,7 @@ dht_create (call_frame_t *frame, xlator_t *this,
}
/* Choose the minimum filled volume, and create the
files there */
- avail_subvol = dht_free_disk_available_subvol (this, subvol);
+ avail_subvol = dht_free_disk_available_subvol (this, subvol, local);
if (avail_subvol != subvol) {
local->params = dict_ref (params);
local->flags = flags;
@@ -3801,9 +4001,8 @@ dht_create (call_frame_t *frame, xlator_t *this,
gf_log (this->name, GF_LOG_TRACE,
"creating %s on %s (link at %s)", loc->path,
avail_subvol->name, subvol->name);
- dht_linkfile_create (frame,
- dht_create_linkfile_create_cbk,
- avail_subvol, subvol, loc);
+ dht_linkfile_create (frame, dht_create_linkfile_create_cbk,
+ this, avail_subvol, subvol, loc);
goto done;
}
gf_log (this->name, GF_LOG_TRACE,
@@ -3876,6 +4075,15 @@ dht_mkdir_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
ret = dht_layout_merge (this, layout, prev->this,
-1, ENOSPC, NULL);
} else {
+ if (op_ret == -1 && op_errno == EEXIST)
+ /* Very likely just a race between mkdir and
+ self-heal (from lookup of a concurrent mkdir
+ attempt).
+ Ignore error for now. layout setting will
+ anyways fail if this was a different (old)
+ pre-existing different directory.
+ */
+ op_ret = 0;
ret = dht_layout_merge (this, layout, prev->this,
op_ret, op_errno, NULL);
}
@@ -4333,6 +4541,7 @@ dht_rmdir_lookup_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
call_frame_t *main_frame = NULL;
dht_local_t *main_local = NULL;
int this_call_cnt = 0;
+ dht_conf_t *conf = this->private;
local = frame->local;
prev = cookie;
@@ -4344,7 +4553,7 @@ dht_rmdir_lookup_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
if (op_ret != 0)
goto err;
- if (check_is_linkfile (inode, stbuf, xattr) == 0) {
+ if (!check_is_linkfile (inode, stbuf, xattr, conf->link_xattr_name)) {
main_local->op_ret = -1;
main_local->op_errno = ENOTEMPTY;
@@ -4379,6 +4588,7 @@ dht_rmdir_is_subvol_empty (call_frame_t *frame, xlator_t *this,
dht_local_t *lookup_local = NULL;
dht_local_t *local = NULL;
dict_t *xattrs = NULL;
+ dht_conf_t *conf = this->private;
local = frame->local;
@@ -4387,7 +4597,8 @@ dht_rmdir_is_subvol_empty (call_frame_t *frame, xlator_t *this,
continue;
if (strcmp (trav->d_name, "..") == 0)
continue;
- if (check_is_linkfile (NULL, (&trav->d_stat), trav->dict)) {
+ if (check_is_linkfile (NULL, (&trav->d_stat), trav->dict,
+ conf->link_xattr_name)) {
ret++;
continue;
}
@@ -4405,7 +4616,7 @@ dht_rmdir_is_subvol_empty (call_frame_t *frame, xlator_t *this,
return -1;
}
- ret = dict_set_uint32 (xattrs, DHT_LINKFILE_KEY, 256);
+ ret = dict_set_uint32 (xattrs, conf->link_xattr_name, 256);
if (ret) {
gf_log (this->name, GF_LOG_ERROR, "failed to set linkto key"
" in dict");
@@ -4528,6 +4739,7 @@ dht_rmdir_opendir_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
call_frame_t *prev = NULL;
dict_t *dict = NULL;
int ret = 0;
+ dht_conf_t *conf = this->private;
local = frame->local;
prev = cookie;
@@ -4551,12 +4763,11 @@ dht_rmdir_opendir_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
goto err;
}
- ret = dict_set_uint32 (dict,
- "trusted.glusterfs.dht.linkto", 256);
+ ret = dict_set_uint32 (dict, conf->link_xattr_name, 256);
if (ret)
gf_log (this->name, GF_LOG_WARNING,
- "%s: failed to set 'trusted.glusterfs.dht.linkto' key",
- local->loc.path);
+ "%s: failed to set '%s' key",
+ local->loc.path, conf->link_xattr_name);
STACK_WIND (frame, dht_rmdir_readdirp_cbk,
prev->this, prev->this->fops->readdirp,
@@ -4655,7 +4866,6 @@ dht_entrylk (call_frame_t *frame, xlator_t *this,
VALIDATE_OR_GOTO (this, err);
VALIDATE_OR_GOTO (loc, err);
VALIDATE_OR_GOTO (loc->inode, err);
- VALIDATE_OR_GOTO (loc->path, err);
local = dht_local_init (frame, loc, NULL, GF_FOP_ENTRYLK);
if (!local) {
@@ -4936,15 +5146,7 @@ unlock:
or wait for anything else. Just propagate blindly */
if (have_heard_from_all) {
propagate = 1;
- if (conf->defrag) {
- ret = pthread_create (&conf->defrag->th, NULL,
- gf_defrag_start, this);
- if (ret) {
- conf->defrag = NULL;
- GF_FREE (conf->defrag);
- kill (getpid(), SIGTERM);
- }
- }
+
}
@@ -4966,6 +5168,19 @@ unlock:
/* continue to check other events for CHILD_UP */
}
}
+
+ /* rebalance is started with assert_no_child_down. So we do
+ * not need to handle CHILD_DOWN event here.
+ */
+ if (conf->defrag) {
+ ret = gf_thread_create (&conf->defrag->th, NULL,
+ gf_defrag_start, this);
+ if (ret) {
+ conf->defrag = NULL;
+ GF_FREE (conf->defrag);
+ kill (getpid(), SIGTERM);
+ }
+ }
}
ret = 0;
generated by cgit (git 2.34.1) at 2026-01-04 09:56:39 +0000