summaryrefslogtreecommitdiffstats
diff options
context:
space:
mode:
authorKrutika Dhananjay <kdhananj@redhat.com>2015-10-20 11:46:10 +0530
committerPranith Kumar Karampuri <pkarampu@redhat.com>2015-10-29 23:36:57 -0700
commitaf5e4e4ef75c845149aee4a8fefd50115157a6da (patch)
tree475c6c4660d04e4275cb91a6c6f74626932a2736
parent6bbce9b1a48d5d50a2044b4518270e952331f159 (diff)
features/shard: Force cache-refresh when lookup/readdirp/stat detect that xattr value has changed
Backport of: http://review.gluster.org/12400 Change-Id: Ifa51979bc530e31d36781759ca62bcac1de7af24 BUG: 1274600 Signed-off-by: Krutika Dhananjay <kdhananj@redhat.com> Reviewed-on: http://review.gluster.org/12457 Tested-by: Gluster Build System <jenkins@build.gluster.com> Tested-by: NetBSD Build System <jenkins@build.gluster.org> Reviewed-by: Pranith Kumar Karampuri <pkarampu@redhat.com>
-rw-r--r--tests/bugs/shard/bug-1272986.t36
-rw-r--r--xlators/features/shard/src/shard.c130
-rw-r--r--xlators/features/shard/src/shard.h2
3 files changed, 149 insertions, 19 deletions
diff --git a/tests/bugs/shard/bug-1272986.t b/tests/bugs/shard/bug-1272986.t
new file mode 100644
index 00000000000..c041ce49482
--- /dev/null
+++ b/tests/bugs/shard/bug-1272986.t
@@ -0,0 +1,36 @@
+#!/bin/bash
+
+. $(dirname $0)/../../include.rc
+
+cleanup
+
+TEST glusterd
+TEST pidof glusterd
+TEST $CLI volume create $V0 $H0:$B0/${V0}{0,1}
+TEST $CLI volume set $V0 features.shard on
+TEST $CLI volume set $V0 performance.strict-write-ordering on
+TEST $CLI volume start $V0
+
+# $M0 is where the reads will be done and $M1 is where files will be created,
+# written to, etc.
+TEST glusterfs --volfile-id=$V0 --volfile-server=$H0 $M0
+TEST glusterfs --volfile-id=$V0 --volfile-server=$H0 $M1
+
+# Write some data into a file, such that its size crosses the shard block size.
+TEST dd if=/dev/zero of=$M1/file bs=1M count=5 conv=notrunc
+
+md5sum1_reader=$(md5sum $M0/file | awk '{print $1}')
+
+EXPECT "$md5sum1_reader" echo `md5sum $M1/file | awk '{print $1}'`
+
+# Append some more data into the file.
+TEST `echo "abcdefg" >> $M1/file`
+
+md5sum2_reader=$(md5sum $M0/file | awk '{print $1}')
+
+# Test to see if the reader refreshes its cache correctly as part of the reads
+# triggered through md5sum. If it does, then the md5sum on the reader and writer
+# must match.
+EXPECT "$md5sum2_reader" echo `md5sum $M1/file | awk '{print $1}'`
+
+cleanup
diff --git a/xlators/features/shard/src/shard.c b/xlators/features/shard/src/shard.c
index a99eaf92bc1..35c9d1d9d45 100644
--- a/xlators/features/shard/src/shard.c
+++ b/xlators/features/shard/src/shard.c
@@ -155,6 +155,9 @@ __shard_inode_ctx_set (inode_t *inode, xlator_t *this, struct iatt *stbuf,
ctx->stat.ia_blksize = stbuf->ia_blksize;
}
+ if (valid & SHARD_MASK_REFRESH_RESET)
+ ctx->refresh = _gf_false;
+
return 0;
}
@@ -175,6 +178,37 @@ shard_inode_ctx_set (inode_t *inode, xlator_t *this, struct iatt *stbuf,
}
int
+__shard_inode_ctx_invalidate (inode_t *inode, xlator_t *this, struct iatt *stbuf)
+{
+ int ret = -1;
+ shard_inode_ctx_t *ctx = NULL;
+
+ ret = __shard_inode_ctx_get (inode, this, &ctx);
+ if (ret)
+ return ret;
+
+ if ((stbuf->ia_size != ctx->stat.ia_size) ||
+ (stbuf->ia_blocks != ctx->stat.ia_blocks))
+ ctx->refresh = _gf_true;
+
+ return 0;
+}
+
+int
+shard_inode_ctx_invalidate (inode_t *inode, xlator_t *this, struct iatt *stbuf)
+{
+ int ret = -1;
+
+ LOCK (&inode->lock);
+ {
+ ret = __shard_inode_ctx_invalidate (inode, this, stbuf);
+ }
+ UNLOCK (&inode->lock);
+
+ return ret;
+}
+
+int
__shard_inode_ctx_get_block_size (inode_t *inode, xlator_t *this,
uint64_t *block_size)
{
@@ -242,6 +276,46 @@ shard_inode_ctx_get_all (inode_t *inode, xlator_t *this,
return ret;
}
+int
+__shard_inode_ctx_fill_iatt_from_cache (inode_t *inode, xlator_t *this,
+ struct iatt *buf,
+ gf_boolean_t *need_refresh)
+{
+ int ret = -1;
+ uint64_t ctx_uint = 0;
+ shard_inode_ctx_t *ctx = NULL;
+
+ ret = __inode_ctx_get (inode, this, &ctx_uint);
+ if (ret < 0)
+ return ret;
+
+ ctx = (shard_inode_ctx_t *) ctx_uint;
+
+ if (ctx->refresh == _gf_false)
+ *buf = ctx->stat;
+ else
+ *need_refresh = _gf_true;
+
+ return 0;
+}
+
+int
+shard_inode_ctx_fill_iatt_from_cache (inode_t *inode, xlator_t *this,
+ struct iatt *buf,
+ gf_boolean_t *need_refresh)
+{
+ int ret = -1;
+
+ LOCK (&inode->lock);
+ {
+ ret = __shard_inode_ctx_fill_iatt_from_cache (inode, this, buf,
+ need_refresh);
+ }
+ UNLOCK (&inode->lock);
+
+ return ret;
+}
+
void
shard_local_wipe (shard_local_t *local)
{
@@ -733,9 +807,10 @@ shard_inode_ctx_update (inode_t *inode, xlator_t *this, dict_t *xdata,
/* If the file is sharded, also set the remaining attributes,
* except for ia_size and ia_blocks.
*/
- if (size)
+ if (size) {
shard_inode_ctx_set (inode, this, buf, 0, SHARD_LOOKUP_MASK);
-
+ (void) shard_inode_ctx_invalidate (inode, this, buf);
+ }
}
int
@@ -749,6 +824,14 @@ shard_lookup_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
if (IA_ISDIR (buf->ia_type))
goto unwind;
+ /* Also, if the file is sharded, get the file size and block cnt xattr,
+ * and store them in the stbuf appropriately.
+ */
+
+ if (dict_get (xdata, GF_XATTR_SHARD_FILE_SIZE) &&
+ frame->root->pid != GF_CLIENT_PID_GSYNCD)
+ shard_modify_size_and_block_count (buf, xdata);
+
/* If this was a fresh lookup, there are two possibilities:
* 1) If the file is sharded (indicated by the presence of block size
* xattr), store this block size, along with rdev and mode in its
@@ -760,14 +843,6 @@ shard_lookup_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
(void) shard_inode_ctx_update (inode, this, xdata, buf);
- /* Also, if the file is sharded, get the file size and block cnt xattr,
- * and store them in the stbuf appropriately.
- */
-
- if (dict_get (xdata, GF_XATTR_SHARD_FILE_SIZE) &&
- frame->root->pid != GF_CLIENT_PID_GSYNCD)
- shard_modify_size_and_block_count (buf, xdata);
-
unwind:
SHARD_STACK_UNWIND (lookup, frame, op_ret, op_errno, inode, buf,
xdata, postparent);
@@ -871,7 +946,8 @@ shard_lookup_base_file_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
if (shard_inode_ctx_get_all (inode, this, &ctx))
mask = SHARD_ALL_MASK;
- ret = shard_inode_ctx_set (inode, this, &local->prebuf, 0, mask);
+ ret = shard_inode_ctx_set (inode, this, &local->prebuf, 0,
+ (mask | SHARD_MASK_REFRESH_RESET));
if (ret) {
gf_msg (this->name, GF_LOG_ERROR,
SHARD_MSG_INODE_CTX_SET_FAILED, 0, "Failed to set inode"
@@ -891,23 +967,24 @@ int
shard_lookup_base_file (call_frame_t *frame, xlator_t *this, loc_t *loc,
shard_post_fop_handler_t handler)
{
- int ret = -1;
- shard_local_t *local = NULL;
- shard_inode_ctx_t ctx = {0,};
- dict_t *xattr_req = NULL;
+ int ret = -1;
+ shard_local_t *local = NULL;
+ dict_t *xattr_req = NULL;
+ gf_boolean_t need_refresh = _gf_false;
local = frame->local;
local->handler = handler;
- ret = shard_inode_ctx_get_all (loc->inode, this, &ctx);
+ ret = shard_inode_ctx_fill_iatt_from_cache (loc->inode, this,
+ &local->prebuf,
+ &need_refresh);
/* By this time, inode ctx should have been created either in create,
* mknod, readdirp or lookup. If not it is a bug!
*/
- if ((ret == 0) && (ctx.stat.ia_size > 0)) {
+ if ((ret == 0) && (need_refresh == _gf_false)) {
gf_msg_debug (this->name, 0, "Skipping lookup on base file: %s"
"Serving prebuf off the inode ctx cache",
uuid_utoa (loc->gfid));
- local->prebuf = ctx.stat;
goto out;
}
@@ -972,6 +1049,7 @@ shard_common_stat_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
int32_t op_ret, int32_t op_errno, struct iatt *buf,
dict_t *xdata)
{
+ inode_t *inode = NULL;
shard_local_t *local = NULL;
local = frame->local;
@@ -994,6 +1072,13 @@ shard_common_stat_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
}
local->xattr_rsp = dict_ref (xdata);
+ if (local->loc.inode)
+ inode = local->loc.inode;
+ else
+ inode = local->fd->inode;
+
+ shard_inode_ctx_invalidate (inode, this, buf);
+
unwind:
local->handler (frame, this);
return 0;
@@ -3798,7 +3883,14 @@ shard_readdir_past_dot_shard_cbk (call_frame_t *frame, void *cookie,
if (IA_ISDIR (entry->d_stat.ia_type))
continue;
- shard_modify_size_and_block_count (&entry->d_stat, entry->dict);
+ if (dict_get (entry->dict, GF_XATTR_SHARD_FILE_SIZE))
+ shard_modify_size_and_block_count (&entry->d_stat,
+ entry->dict);
+ if (!entry->inode)
+ continue;
+
+ shard_inode_ctx_update (entry->inode, this, entry->dict,
+ &entry->d_stat);
}
local->op_ret += op_ret;
diff --git a/xlators/features/shard/src/shard.h b/xlators/features/shard/src/shard.h
index 601269402e0..13b3476c6e4 100644
--- a/xlators/features/shard/src/shard.h
+++ b/xlators/features/shard/src/shard.h
@@ -42,6 +42,7 @@
#define SHARD_MASK_BLOCKS (1 << 7)
#define SHARD_MASK_TIMES (1 << 8)
#define SHARD_MASK_OTHERS (1 << 9)
+#define SHARD_MASK_REFRESH_RESET (1 << 10)
#define SHARD_INODE_WRITE_MASK (SHARD_MASK_SIZE | SHARD_MASK_BLOCKS \
| SHARD_MASK_TIMES)
@@ -266,6 +267,7 @@ typedef struct shard_inode_ctx {
uint64_t block_size; /* The block size with which this inode is
sharded */
struct iatt stat;
+ gf_boolean_t refresh;
/* The following members of inode ctx will be applicable only to the
* individual shards' ctx and never the base file ctx.
*/