summaryrefslogtreecommitdiffstats
path: root/xlators/cluster/dht/src
diff options
context:
space:
mode:
Diffstat (limited to 'xlators/cluster/dht/src')
-rw-r--r--xlators/cluster/dht/src/dht-common.c32
-rw-r--r--xlators/cluster/dht/src/dht-common.h5
-rw-r--r--xlators/cluster/dht/src/dht-selfheal.c319
3 files changed, 297 insertions, 59 deletions
diff --git a/xlators/cluster/dht/src/dht-common.c b/xlators/cluster/dht/src/dht-common.c
index 4bc87fc2c43..237a0789426 100644
--- a/xlators/cluster/dht/src/dht-common.c
+++ b/xlators/cluster/dht/src/dht-common.c
@@ -2080,7 +2080,6 @@ dht_setxattr (call_frame_t *frame, xlator_t *this,
dht_layout_t *layout = NULL;
int i = 0;
int op_errno = EINVAL;
- int flag = 0;
int ret = -1;
data_t *tmp = NULL;
@@ -2115,29 +2114,18 @@ dht_setxattr (call_frame_t *frame, xlator_t *this,
tmp = dict_get (xattr, GF_XATTR_FIX_LAYOUT_KEY);
if (tmp) {
- for (i = 0; i < layout->cnt; i++) {
- if (layout->list[i].start == layout->list[i].stop) {
- flag = 1;
- break;
- }
- }
- if ((layout->cnt < conf->subvolume_cnt) || flag) {
- gf_log (this->name, GF_LOG_INFO,
- "expanding layout of %s from %d to %d",
- loc->path, layout->cnt, conf->subvolume_cnt);
-
- ret = loc_dup (loc, &local->loc);
- if (ret == -1) {
- op_errno = ENOMEM;
- goto err;
- }
+ gf_log (this->name, GF_LOG_INFO,
+ "fixing the layout of %s", loc->path);
- dht_selfheal_new_directory (frame, dht_fix_layout_cbk,
- layout);
- return 0;
+ ret = loc_dup (loc, &local->loc);
+ if (ret == -1) {
+ op_errno = ENOMEM;
+ goto err;
}
- op_errno = ENOTSUP;
- goto err;
+
+ dht_fix_directory_layout (frame, dht_fix_layout_cbk,
+ layout);
+ return 0;
}
local->call_cnt = layout->cnt;
diff --git a/xlators/cluster/dht/src/dht-common.h b/xlators/cluster/dht/src/dht-common.h
index 461add16897..c85ba9cfcc2 100644
--- a/xlators/cluster/dht/src/dht-common.h
+++ b/xlators/cluster/dht/src/dht-common.h
@@ -318,4 +318,9 @@ int dht_rename_links_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
int dht_linkfile_recreate(call_frame_t *frame, fop_mknod_cbk_t linkfile_cbk,
xlator_t *tovol, xlator_t *fromvol, loc_t *loc);
+
+int dht_fix_directory_layout (call_frame_t *frame,
+ dht_selfheal_dir_cbk_t dir_cbk,
+ dht_layout_t *layout);
+
#endif /* _DHT_H */
diff --git a/xlators/cluster/dht/src/dht-selfheal.c b/xlators/cluster/dht/src/dht-selfheal.c
index e05894a5c2d..a20526067b4 100644
--- a/xlators/cluster/dht/src/dht-selfheal.c
+++ b/xlators/cluster/dht/src/dht-selfheal.c
@@ -28,6 +28,53 @@
#include "dht-common.h"
+#define DHT_SET_LAYOUT_RANGE(layout,i,srt,chunk,cnt,path) do { \
+ layout->list[i].start = srt; \
+ layout->list[i].stop = srt + chunk - 1; \
+ \
+ gf_log (this->name, GF_LOG_TRACE, \
+ "gave fix: %u - %u on %s for %s", \
+ layout->list[i].start, layout->list[i].stop, \
+ layout->list[i].xlator->name, path); \
+ } while (0)
+
+
+static inline uint32_t
+dht_find_overlap (int idx, int cnk_idx, uint32_t start, uint32_t stop,
+ uint32_t chunk_size)
+{
+ uint32_t overlap = 0;
+ uint32_t chunk_begin = 0;
+
+ chunk_begin = cnk_idx * chunk_size;
+
+ /* There is no chance of overlap */
+ if ((chunk_begin > stop) ||
+ ((chunk_begin + chunk_size) < start))
+ goto out;
+
+ if ((chunk_begin <= start) &&
+ ((chunk_begin + chunk_size) <= stop)) {
+ overlap = ((chunk_begin + chunk_size) - start);
+ goto out;
+ }
+
+ if ((chunk_begin <= start) &&
+ ((chunk_begin + chunk_size) >= stop)) {
+ overlap = (stop - start);
+ goto out;
+ }
+
+ if ((chunk_begin < stop) &&
+ ((chunk_begin + chunk_size) >= stop)) {
+ overlap = (stop - chunk_begin);
+ goto out;
+ }
+
+out:
+ return overlap;
+}
+
int
dht_selfheal_dir_finish (call_frame_t *frame, xlator_t *this, int ret)
{
@@ -144,6 +191,30 @@ err:
return 0;
}
+int
+dht_fix_dir_xattr (call_frame_t *frame, loc_t *loc, dht_layout_t *layout)
+{
+ dht_local_t *local = NULL;
+ int i = 0;
+ int count = 0;
+ xlator_t *this = NULL;
+
+ local = frame->local;
+ this = frame->this;
+
+ gf_log (this->name, GF_LOG_DEBUG,
+ "writing the new range for all subvolumes");
+
+ local->call_cnt = count = layout->cnt;
+
+ for (i = 0; i < layout->cnt; i++) {
+ dht_selfheal_dir_xattr_persubvol (frame, loc, layout, i);
+
+ if (--count == 0)
+ break;
+ }
+ return 0;
+}
int
dht_selfheal_dir_xattr (call_frame_t *frame, loc_t *loc, dht_layout_t *layout)
@@ -385,40 +456,204 @@ dht_selfheal_layout_alloc_start (xlator_t *this, loc_t *loc,
return start;
}
-
-void
-dht_selfheal_layout_new_directory (call_frame_t *frame, loc_t *loc,
- dht_layout_t *layout)
+static inline int
+dht_get_layout_count (xlator_t *this, dht_layout_t *layout)
{
- xlator_t *this = NULL;
- uint32_t chunk = 0;
- int i = 0;
- uint32_t start = 0;
- int cnt = 0;
- int err = 0;
- int start_subvol = 0;
-
- this = frame->this;
+ int i = 0;
+ int err = 0;
+ int count = 0;
for (i = 0; i < layout->cnt; i++) {
err = layout->list[i].err;
if (err == -1 || err == 0) {
layout->list[i].err = -1;
- cnt++;
+ count++;
}
}
/* no subvolume has enough space, but can't stop directory creation */
- if (!cnt) {
+ if (!count) {
for (i = 0; i < layout->cnt; i++) {
err = layout->list[i].err;
if (err == ENOSPC) {
layout->list[i].err = -1;
- cnt++;
+ count++;
+ }
+ }
+ }
+
+ return count;
+}
+
+
+dht_layout_t *
+dht_fix_layout_of_directory (call_frame_t *frame, loc_t *loc,
+ dht_layout_t *layout)
+{
+ uint32_t chunk = 0;
+ uint32_t start = 0;
+ uint32_t stop = 0;
+ uint32_t overlap = 0;
+ uint32_t max_overlap = 0;
+ uint32_t chunk_begin = 0;
+ int count = 0;
+ int cnt = 0;
+ int i = 0;
+ int j = 0;
+ int k = 0;
+ int loop_cnt = 0;
+ int start_subvol = 0;
+ int *fix_array = NULL;
+ xlator_t *this = NULL;
+ dht_layout_t *new_layout = NULL;
+ dht_conf_t *priv = NULL;
+ dht_local_t *local = NULL;
+
+ this = frame->this;
+ priv = this->private;
+ local = frame->local;
+
+ count = cnt = dht_get_layout_count (this, layout);
+
+ chunk = ((unsigned long) 0xffffffff) / ((cnt) ? cnt : 1);
+
+ start_subvol = dht_selfheal_layout_alloc_start (this, loc, layout);
+
+ fix_array = GF_CALLOC (sizeof (int), layout->cnt, gf_common_mt_char);
+ if (!fix_array) {
+ /* No fix, use the existing layout itself */
+ goto done;
+ }
+
+ new_layout = dht_layout_new (this, priv->subvolume_cnt);
+ if (!new_layout)
+ goto done;
+
+ for (i = 0; i < new_layout->cnt; i++) {
+ /* TODO: fix this in layout_alloc() itself */
+ new_layout->list[i].err = -ENOENT;
+ if (i < layout->cnt)
+ new_layout->list[i].xlator = layout->list[i].xlator;
+ }
+
+ /* Check if there are any overlap in layout, and give the proper fix */
+ for (i = 0; i < layout->cnt; i++) {
+ /* No need to fix if 'err' is not '-1' */
+ if (layout->list[i].err != -1)
+ continue;
+
+ /* If already existing layout is having no range, skip it */
+ start = layout->list[i].start;
+ stop = layout->list[i].stop;
+ if ((stop - start) == 0)
+ continue;
+
+ max_overlap = 0;
+
+ /* 'j' is used as starting point of each chunk */
+ for (j = 1; j <= count; j++) {
+ /* if chunk is already used, don't use it again */
+ for (k = 0; k < i; k++)
+ if (j == fix_array[k])
+ break;
+ if (k < i)
+ continue;
+
+ overlap = dht_find_overlap (i, (j-1), start, stop, chunk);
+ if (max_overlap < overlap) {
+ max_overlap = overlap;
+ fix_array[i] = j;
+ }
+ }
+
+ /* If we have any overlap, then use that itself as new
+ layout for the subvolume */
+ if (fix_array[i]) {
+ chunk_begin = chunk * (fix_array[i] - 1);
+ new_layout->list[i].err = -1;
+ DHT_SET_LAYOUT_RANGE (new_layout, i, chunk_begin,
+ chunk, cnt, loc->path);
+ /* make sure to give (max - 1) as 'stop' range,
+ if it is last chunk */
+ if (fix_array[i] == count)
+ new_layout->list[i].stop = 0xffffffff;
+ if (--cnt == 0)
+ goto done;
+
+ }
+ }
+
+ /* Now, look for layouts which are not having any overlaps
+ and give it a fix */
+ for (loop_cnt = 0, i = start_subvol; loop_cnt < new_layout->cnt;
+ i++, loop_cnt++) {
+ if (i == new_layout->cnt)
+ i = 0;
+
+ /* If 'fix_array[i]' is set, the layout is already fixed. */
+ if (fix_array[i])
+ continue;
+
+ if (layout->list[i].err != -1) {
+ new_layout->list[i].err = layout->list[i].err;
+ continue;
+ }
+
+ for (k = 1; k <= count; k++) {
+ for (j = 0; j < new_layout->cnt; j++) {
+ if (k == fix_array[j])
+ break;
}
+ /* Didn't find any of the list begining with 'k' */
+ if (j == new_layout->cnt)
+ break;
}
+
+ fix_array[i] = k;
+ chunk_begin = (k - 1) * chunk;
+ new_layout->list[i].err = -1;
+ DHT_SET_LAYOUT_RANGE (new_layout, i, chunk_begin, chunk, cnt,
+ loc->path);
+ /* make sure to give (max - 1) as 'stop' range,
+ if it is last chunk */
+ if (k == count)
+ new_layout->list[i].stop = 0xffffffff;
+ if (--cnt == 0)
+ goto done;
}
+done:
+ if (new_layout) {
+ /* Now that the new layout has all the proper layout, change the
+ inode context */
+ dht_layout_set (this, loc->inode, new_layout);
+
+ /* Make sure the extra 'ref' for existing layout is removed */
+ dht_layout_unref (this, local->layout);
+
+ local->layout = new_layout;
+ }
+
+ return new_layout;
+}
+
+
+void
+dht_selfheal_layout_new_directory (call_frame_t *frame, loc_t *loc,
+ dht_layout_t *layout)
+{
+ xlator_t *this = NULL;
+ uint32_t chunk = 0;
+ int i = 0;
+ uint32_t start = 0;
+ int cnt = 0;
+ int err = 0;
+ int start_subvol = 0;
+
+ this = frame->this;
+
+ cnt = dht_get_layout_count (this, layout);
+
chunk = ((unsigned long) 0xffffffff) / ((cnt) ? cnt : 1);
start_subvol = dht_selfheal_layout_alloc_start (this, loc, layout);
@@ -426,42 +661,32 @@ dht_selfheal_layout_new_directory (call_frame_t *frame, loc_t *loc,
for (i = start_subvol; i < layout->cnt; i++) {
err = layout->list[i].err;
if (err == -1) {
- layout->list[i].start = start;
- layout->list[i].stop = start + chunk - 1;
-
- start = start + chunk;
-
- gf_log (this->name, GF_LOG_TRACE,
- "gave fix: %u - %u on %s for %s",
- layout->list[i].start, layout->list[i].stop,
- layout->list[i].xlator->name, loc->path);
+ DHT_SET_LAYOUT_RANGE(layout, i, start, chunk,
+ cnt, loc->path);
if (--cnt == 0) {
layout->list[i].stop = 0xffffffff;
- break;
+ goto done;
}
+ start += chunk;
}
}
for (i = 0; i < start_subvol; i++) {
err = layout->list[i].err;
if (err == -1) {
- layout->list[i].start = start;
- layout->list[i].stop = start + chunk - 1;
-
- start = start + chunk;
-
- gf_log (this->name, GF_LOG_TRACE,
- "gave fix: %u - %u on %s for %s",
- layout->list[i].start, layout->list[i].stop,
- layout->list[i].xlator->name, loc->path);
+ DHT_SET_LAYOUT_RANGE(layout, i, start, chunk,
+ cnt, loc->path);
if (--cnt == 0) {
layout->list[i].stop = 0xffffffff;
- break;
+ goto done;
}
+ start += chunk;
}
}
-}
+done:
+ return;
+}
int
dht_selfheal_dir_getafix (call_frame_t *frame, loc_t *loc,
@@ -532,6 +757,26 @@ dht_selfheal_new_directory (call_frame_t *frame,
return 0;
}
+int
+dht_fix_directory_layout (call_frame_t *frame,
+ dht_selfheal_dir_cbk_t dir_cbk,
+ dht_layout_t *layout)
+{
+ dht_local_t *local = NULL;
+ dht_layout_t *tmp_layout = NULL;
+
+ local = frame->local;
+
+ local->selfheal.dir_cbk = dir_cbk;
+ local->selfheal.layout = dht_layout_ref (frame->this, layout);
+
+ /* No layout sorting required here */
+ tmp_layout = dht_fix_layout_of_directory (frame, &local->loc, layout);
+ dht_fix_dir_xattr (frame, &local->loc, tmp_layout);
+
+ return 0;
+}
+
int
dht_selfheal_directory (call_frame_t *frame, dht_selfheal_dir_cbk_t dir_cbk,