summaryrefslogtreecommitdiffstats
path: root/xlators/cluster/afr/src/afr-dir-read.c
diff options
context:
space:
mode:
Diffstat (limited to 'xlators/cluster/afr/src/afr-dir-read.c')
-rw-r--r--xlators/cluster/afr/src/afr-dir-read.c899
1 files changed, 252 insertions, 647 deletions
diff --git a/xlators/cluster/afr/src/afr-dir-read.c b/xlators/cluster/afr/src/afr-dir-read.c
index b661cfa9c85..f8bf8340dab 100644
--- a/xlators/cluster/afr/src/afr-dir-read.c
+++ b/xlators/cluster/afr/src/afr-dir-read.c
@@ -8,734 +8,339 @@
cases as published by the Free Software Foundation.
*/
-
#include <libgen.h>
#include <unistd.h>
-#include <fnmatch.h>
#include <sys/time.h>
#include <stdlib.h>
#include <signal.h>
#include <string.h>
-#ifndef _CONFIG_H
-#define _CONFIG_H
-#include "config.h"
-#endif
-
-#include "glusterfs.h"
-#include "dict.h"
-#include "xlator.h"
-#include "hashfn.h"
-#include "logging.h"
-#include "stack.h"
-#include "list.h"
-#include "call-stub.h"
-#include "defaults.h"
-#include "common-utils.h"
-#include "compat-errno.h"
-#include "compat.h"
-#include "checksum.h"
+#include <glusterfs/glusterfs.h>
+#include <glusterfs/dict.h>
+#include <glusterfs/list.h>
+#include <glusterfs/common-utils.h>
+#include <glusterfs/compat-errno.h>
+#include <glusterfs/compat.h>
#include "afr.h"
-#include "afr-self-heal.h"
-#include "afr-self-heal-common.h"
-
-int
-afr_examine_dir_sh_unwind (call_frame_t *frame, xlator_t *this, int32_t op_ret,
- int32_t op_errno, int32_t sh_failed)
-{
- afr_local_t *local = NULL;
-
- local = frame->local;
-
- afr_set_opendir_done (this, local->fd->inode);
-
- AFR_STACK_UNWIND (opendir, frame, local->op_ret,
- local->op_errno, local->fd, NULL);
-
- return 0;
-}
-
-
-gf_boolean_t
-__checksums_differ (uint32_t *checksum, int child_count,
- unsigned char *child_up)
-{
- int ret = _gf_false;
- int i = 0;
- uint32_t cksum = 0;
- gf_boolean_t activate_check = _gf_false;
-
- for (i = 0; i < child_count; i++) {
- if (!child_up[i])
- continue;
- if (_gf_false == activate_check) {
- cksum = checksum[i];
- activate_check = _gf_true;
- continue;
- }
-
- if (cksum != checksum[i]) {
- ret = _gf_true;
- break;
- }
-
- cksum = checksum[i];
- }
-
- return ret;
-}
-
+#include "afr-transaction.h"
int32_t
-afr_examine_dir_readdir_cbk (call_frame_t *frame, void *cookie,
- xlator_t *this, int32_t op_ret, int32_t op_errno,
- gf_dirent_t *entries, dict_t *xdata)
+afr_opendir_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+ int32_t op_ret, int32_t op_errno, fd_t *fd, dict_t *xdata)
{
- afr_private_t * priv = NULL;
- afr_local_t * local = NULL;
- afr_self_heal_t * sh = NULL;
- gf_dirent_t * entry = NULL;
- gf_dirent_t * tmp = NULL;
- char *reason = NULL;
- int child_index = 0;
- uint32_t entry_cksum = 0;
- int call_count = 0;
- off_t last_offset = 0;
- inode_t *inode = NULL;
-
- priv = this->private;
- local = frame->local;
- sh = &local->self_heal;
- inode = local->fd->inode;
-
- child_index = (long) cookie;
-
- if (op_ret == -1) {
- gf_log (this->name, GF_LOG_INFO,
- "%s: failed to do opendir on %s",
- local->loc.path, priv->children[child_index]->name);
- local->op_ret = -1;
- local->op_ret = op_errno;
- goto out;
- }
+ afr_local_t *local = NULL;
+ int call_count = -1;
+ int32_t child_index = 0;
+ afr_fd_ctx_t *fd_ctx = NULL;
- if (op_ret == 0) {
- gf_log (this->name, GF_LOG_DEBUG,
- "%s: no entries found in %s",
- local->loc.path, priv->children[child_index]->name);
- goto out;
- }
+ local = frame->local;
+ fd_ctx = local->fd_ctx;
+ child_index = (long)cookie;
- list_for_each_entry_safe (entry, tmp, &entries->list, list) {
- entry_cksum = gf_rsync_weak_checksum ((unsigned char *)entry->d_name,
- strlen (entry->d_name));
- local->cont.opendir.checksum[child_index] ^= entry_cksum;
- }
+ local->replies[child_index].valid = 1;
+ local->replies[child_index].op_ret = op_ret;
+ local->replies[child_index].op_errno = op_errno;
- list_for_each_entry (entry, &entries->list, list) {
- last_offset = entry->d_off;
+ LOCK(&frame->lock);
+ {
+ if (op_ret == -1) {
+ local->op_errno = op_errno;
+ fd_ctx->opened_on[child_index] = AFR_FD_NOT_OPENED;
+ } else {
+ local->op_ret = op_ret;
+ fd_ctx->opened_on[child_index] = AFR_FD_OPENED;
+ if (!local->xdata_rsp && xdata)
+ local->xdata_rsp = dict_ref(xdata);
}
+ call_count = --local->call_count;
+ }
+ UNLOCK(&frame->lock);
- /* read more entries */
+ if (call_count == 0) {
+ afr_handle_replies_quorum(frame, this);
+ AFR_STACK_UNWIND(opendir, frame, local->op_ret, local->op_errno,
+ local->fd, NULL);
+ }
- STACK_WIND_COOKIE (frame, afr_examine_dir_readdir_cbk,
- (void *) (long) child_index,
- priv->children[child_index],
- priv->children[child_index]->fops->readdir,
- local->fd, 131072, last_offset, NULL);
-
- return 0;
-
-out:
- call_count = afr_frame_return (frame);
-
- if (call_count == 0) {
- if (__checksums_differ (local->cont.opendir.checksum,
- priv->child_count,
- local->child_up)) {
-
- sh->do_entry_self_heal = _gf_true;
- sh->forced_merge = _gf_true;
-
- reason = "checksums of directory differ";
- afr_launch_self_heal (frame, this, inode, _gf_false,
- inode->ia_type, reason, NULL,
- afr_examine_dir_sh_unwind);
- } else {
- afr_set_opendir_done (this, inode);
-
- AFR_STACK_UNWIND (opendir, frame, local->op_ret,
- local->op_errno, local->fd, NULL);
- }
- }
-
- return 0;
+ return 0;
}
-
int
-afr_examine_dir (call_frame_t *frame, xlator_t *this)
-{
- afr_private_t * priv = NULL;
- afr_local_t * local = NULL;
- int i = 0;
- int call_count = 0;
-
- local = frame->local;
- priv = this->private;
-
- local->cont.opendir.checksum = GF_CALLOC (priv->child_count,
- sizeof (*local->cont.opendir.checksum),
- gf_afr_mt_int32_t);
-
- call_count = afr_up_children_count (local->child_up, priv->child_count);
-
- local->call_count = call_count;
-
- for (i = 0; i < priv->child_count; i++) {
- if (local->child_up[i]) {
- STACK_WIND_COOKIE (frame, afr_examine_dir_readdir_cbk,
- (void *) (long) i,
- priv->children[i],
- priv->children[i]->fops->readdir,
- local->fd, 131072, 0, NULL);
-
- if (!--call_count)
- break;
- }
- }
-
- return 0;
-}
-
-
-int32_t
-afr_opendir_cbk (call_frame_t *frame, void *cookie,
- xlator_t *this, int32_t op_ret, int32_t op_errno,
- fd_t *fd, dict_t *xdata)
+afr_opendir(call_frame_t *frame, xlator_t *this, loc_t *loc, fd_t *fd,
+ dict_t *xdata)
{
- afr_private_t *priv = NULL;
- afr_local_t *local = NULL;
- int32_t up_children_count = 0;
- int ret = -1;
- int call_count = -1;
- int32_t child_index = 0;
-
- priv = this->private;
- local = frame->local;
- child_index = (long) cookie;
-
- up_children_count = afr_up_children_count (local->child_up,
- priv->child_count);
-
- LOCK (&frame->lock);
- {
- if (op_ret >= 0) {
- local->op_ret = op_ret;
- ret = afr_child_fd_ctx_set (this, fd, child_index, 0);
- if (ret) {
- local->op_ret = -1;
- local->op_errno = -ret;
- goto unlock;
- }
- }
-
- local->op_errno = op_errno;
- }
-unlock:
- UNLOCK (&frame->lock);
-
- call_count = afr_frame_return (frame);
-
- if (call_count == 0) {
- if (local->op_ret != 0)
- goto out;
-
- if (!afr_is_opendir_done (this, local->fd->inode) &&
- up_children_count > 1) {
-
- /*
- * This is the first opendir on this inode. We need
- * to check if the directory's entries are the same
- * on all subvolumes. This is needed in addition
- * to regular entry self-heal because the readdir
- * call is sent only to the first subvolume, and
- * thus files that exist only there will never be healed
- * otherwise (assuming changelog shows no anomalies).
- */
-
- gf_log (this->name, GF_LOG_TRACE,
- "reading contents of directory %s looking for mismatch",
- local->loc.path);
-
- afr_examine_dir (frame, this);
-
- } else {
- /* do the unwind */
- goto out;
- }
- }
+ afr_private_t *priv = NULL;
+ afr_local_t *local = NULL;
+ int i = 0;
+ int call_count = -1;
+ int32_t op_errno = ENOMEM;
+ afr_fd_ctx_t *fd_ctx = NULL;
- return 0;
+ priv = this->private;
-out:
- AFR_STACK_UNWIND (opendir, frame, local->op_ret,
- local->op_errno, local->fd, NULL);
+ local = AFR_FRAME_INIT(frame, op_errno);
+ if (!local)
+ goto out;
- return 0;
-}
+ local->op = GF_FOP_OPENDIR;
+ if (priv->quorum_count && !afr_has_quorum(local->child_up, this, NULL)) {
+ op_errno = afr_quorum_errno(priv);
+ goto out;
+ }
-int32_t
-afr_opendir (call_frame_t *frame, xlator_t *this,
- loc_t *loc, fd_t *fd)
-{
- afr_private_t * priv = NULL;
- afr_local_t * local = NULL;
- int child_count = 0;
- int i = 0;
- int ret = -1;
- int call_count = -1;
- int32_t op_errno = 0;
+ if (!afr_is_consistent_io_possible(local, priv, &op_errno))
+ goto out;
- VALIDATE_OR_GOTO (frame, out);
- VALIDATE_OR_GOTO (this, out);
- VALIDATE_OR_GOTO (this->private, out);
+ fd_ctx = afr_fd_ctx_get(fd, this);
+ if (!fd_ctx)
+ goto out;
- priv = this->private;
+ loc_copy(&local->loc, loc);
- child_count = priv->child_count;
+ local->fd = fd_ref(fd);
+ local->fd_ctx = fd_ctx;
- AFR_LOCAL_ALLOC_OR_GOTO (frame->local, out);
- local = frame->local;
+ call_count = local->call_count;
- ret = afr_local_init (local, priv, &op_errno);
- if (ret < 0)
- goto out;
+ for (i = 0; i < priv->child_count; i++) {
+ if (local->child_up[i]) {
+ STACK_WIND_COOKIE(frame, afr_opendir_cbk, (void *)(long)i,
+ priv->children[i],
+ priv->children[i]->fops->opendir, loc, fd, NULL);
- loc_copy (&local->loc, loc);
-
- local->fd = fd_ref (fd);
-
- call_count = local->call_count;
-
- for (i = 0; i < child_count; i++) {
- if (local->child_up[i]) {
- STACK_WIND_COOKIE (frame, afr_opendir_cbk,
- (void*) (long) i,
- priv->children[i],
- priv->children[i]->fops->opendir,
- loc, fd, NULL);
-
- if (!--call_count)
- break;
- }
+ if (!--call_count)
+ break;
}
+ }
- ret = 0;
+ return 0;
out:
- if (ret < 0)
- AFR_STACK_UNWIND (opendir, frame, -1, op_errno, fd, NULL);
-
- return 0;
+ AFR_STACK_UNWIND(opendir, frame, -1, op_errno, fd, NULL);
+ return 0;
}
-
-/**
- * Common algorithm for directory read calls:
- *
- * - Try the fop on the first child that is up
- * - if we have failed due to ENOTCONN:
- * try the next child
- *
- * Applicable to: readdir
- */
-
-
-struct entry_name {
- char *name;
- struct list_head list;
-};
-
-
-static gf_boolean_t
-remembered_name (const char *name, struct list_head *entries)
+static int
+afr_validate_read_subvol(inode_t *inode, xlator_t *this, int par_read_subvol)
{
- struct entry_name *e = NULL;
- gf_boolean_t ret = _gf_false;
-
- list_for_each_entry (e, entries, list) {
- if (!strcmp (name, e->name)) {
- ret = _gf_true;
- goto out;
- }
- }
+ int gen = 0;
+ int entry_read_subvol = 0;
+ unsigned char *data_readable = NULL;
+ unsigned char *metadata_readable = NULL;
+ afr_private_t *priv = NULL;
+
+ priv = this->private;
+ data_readable = alloca0(priv->child_count);
+ metadata_readable = alloca0(priv->child_count);
+
+ afr_inode_read_subvol_get(inode, this, data_readable, metadata_readable,
+ &gen);
+
+ if (gen != priv->event_generation || !data_readable[par_read_subvol] ||
+ !metadata_readable[par_read_subvol])
+ return -1;
+
+ /* Once the control reaches the following statement, it means that the
+ * parent's read subvol is perfectly readable. So calling
+ * either afr_data_subvol_get() or afr_metadata_subvol_get() would
+ * yield the same result. Hence, choosing afr_data_subvol_get() below.
+ */
+
+ if (!priv->consistent_metadata)
+ return 0;
-out:
- return ret;
+ /* For an inode fetched through readdirp which is yet to be linked,
+ * inode ctx would not be initialised (yet). So this function returns
+ * -1 above due to gen being 0, which is why it is OK to pass NULL for
+ * read_subvol_args here.
+ */
+ entry_read_subvol = afr_data_subvol_get(inode, this, NULL, NULL, NULL,
+ NULL);
+ if (entry_read_subvol != par_read_subvol)
+ return -1;
+
+ return 0;
}
-
static void
-afr_remember_entries (gf_dirent_t *entries, fd_t *fd)
+afr_readdir_transform_entries(call_frame_t *frame, gf_dirent_t *subvol_entries,
+ int subvol, gf_dirent_t *entries, fd_t *fd)
{
- struct entry_name *n = NULL;
- gf_dirent_t *entry = NULL;
- int ret = 0;
- uint64_t ctx = 0;
- afr_fd_ctx_t *fd_ctx = NULL;
-
- ret = fd_ctx_get (fd, THIS, &ctx);
- if (ret < 0) {
- gf_log (THIS->name, GF_LOG_INFO,
- "could not get fd ctx for fd=%p", fd);
- return;
+ int ret = -1;
+ gf_dirent_t *entry = NULL;
+ gf_dirent_t *tmp = NULL;
+ xlator_t *this = NULL;
+ afr_private_t *priv = NULL;
+ gf_boolean_t need_heal = _gf_false;
+ gf_boolean_t validate_subvol = _gf_false;
+
+ this = THIS;
+ priv = this->private;
+
+ need_heal = afr_get_need_heal(this);
+ validate_subvol = need_heal | priv->consistent_metadata;
+
+ list_for_each_entry_safe(entry, tmp, &subvol_entries->list, list)
+ {
+ if (afr_is_private_directory(priv, fd->inode->gfid, entry->d_name,
+ frame->root->pid)) {
+ continue;
}
- fd_ctx = (afr_fd_ctx_t *)(long) ctx;
+ list_del_init(&entry->list);
+ list_add_tail(&entry->list, &entries->list);
- list_for_each_entry (entry, &entries->list, list) {
- n = GF_CALLOC (1, sizeof (*n), gf_afr_mt_entry_name);
- n->name = gf_strdup (entry->d_name);
- INIT_LIST_HEAD (&n->list);
+ if (!validate_subvol)
+ continue;
- list_add (&n->list, &fd_ctx->entries);
+ if (entry->inode) {
+ ret = afr_validate_read_subvol(entry->inode, this, subvol);
+ if (ret == -1) {
+ inode_unref(entry->inode);
+ entry->inode = NULL;
+ continue;
+ }
}
+ }
}
-
-static off_t
-afr_filter_entries (gf_dirent_t *entries, fd_t *fd)
+int32_t
+afr_readdir_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+ int32_t op_ret, int32_t op_errno, gf_dirent_t *subvol_entries,
+ dict_t *xdata)
{
- gf_dirent_t *entry = NULL;
- gf_dirent_t *tmp = NULL;
- int ret = 0;
- uint64_t ctx = 0;
- afr_fd_ctx_t *fd_ctx = NULL;
- off_t offset = 0;
-
- ret = fd_ctx_get (fd, THIS, &ctx);
- if (ret < 0) {
- gf_log (THIS->name, GF_LOG_INFO,
- "could not get fd ctx for fd=%p", fd);
- return -1;
- }
+ afr_local_t *local = NULL;
+ gf_dirent_t entries;
- fd_ctx = (afr_fd_ctx_t *)(long) ctx;
+ INIT_LIST_HEAD(&entries.list);
- list_for_each_entry_safe (entry, tmp, &entries->list, list) {
- offset = entry->d_off;
+ local = frame->local;
- if (remembered_name (entry->d_name, &fd_ctx->entries)) {
- list_del (&entry->list);
- GF_FREE (entry);
- }
- }
+ if (op_ret < 0 && !local->cont.readdir.offset) {
+ /* failover only if this was first readdir, detected
+ by offset == 0 */
+ local->op_ret = op_ret;
+ local->op_errno = op_errno;
- return offset;
-}
+ afr_read_txn_continue(frame, this, (long)cookie);
+ return 0;
+ }
+ if (op_ret >= 0)
+ afr_readdir_transform_entries(frame, subvol_entries, (long)cookie,
+ &entries, local->fd);
-static void
-afr_forget_entries (fd_t *fd)
-{
- struct entry_name *entry = NULL;
- struct entry_name *tmp = NULL;
- int ret = 0;
- uint64_t ctx = 0;
- afr_fd_ctx_t *fd_ctx = NULL;
-
- ret = fd_ctx_get (fd, THIS, &ctx);
- if (ret < 0) {
- gf_log (THIS->name, GF_LOG_INFO,
- "could not get fd ctx for fd=%p", fd);
- return;
- }
+ AFR_STACK_UNWIND(readdir, frame, op_ret, op_errno, &entries, xdata);
- fd_ctx = (afr_fd_ctx_t *)(long) ctx;
+ gf_dirent_free(&entries);
- list_for_each_entry_safe (entry, tmp, &fd_ctx->entries, list) {
- GF_FREE (entry->name);
- list_del (&entry->list);
- GF_FREE (entry);
- }
+ return 0;
}
-
-int32_t
-afr_readdir_cbk (call_frame_t *frame, void *cookie,
- xlator_t *this, int32_t op_ret, int32_t op_errno,
- gf_dirent_t *entries, dict_t *xdata)
+int
+afr_readdir_wind(call_frame_t *frame, xlator_t *this, int subvol)
{
- afr_local_t * local = NULL;
- gf_dirent_t * entry = NULL;
- gf_dirent_t * tmp = NULL;
-
- local = frame->local;
-
- if (op_ret == -1)
- goto out;
-
- list_for_each_entry_safe (entry, tmp, &entries->list, list) {
- if ((local->fd->inode == local->fd->inode->table->root)
- && !strcmp (entry->d_name, GF_REPLICATE_TRASH_DIR)) {
- list_del_init (&entry->list);
- GF_FREE (entry);
- }
- }
-
-out:
- AFR_STACK_UNWIND (readdir, frame, op_ret, op_errno, entries, NULL);
-
+ afr_local_t *local = NULL;
+ afr_private_t *priv = NULL;
+ afr_fd_ctx_t *fd_ctx = NULL;
+
+ priv = this->private;
+ local = frame->local;
+ fd_ctx = afr_fd_ctx_get(local->fd, this);
+ if (!fd_ctx) {
+ local->op_errno = EINVAL;
+ local->op_ret = -1;
+ }
+
+ if (subvol == -1 || !fd_ctx) {
+ AFR_STACK_UNWIND(readdir, frame, local->op_ret, local->op_errno, 0, 0);
return 0;
+ }
+
+ fd_ctx->readdir_subvol = subvol;
+
+ if (local->op == GF_FOP_READDIR)
+ STACK_WIND_COOKIE(frame, afr_readdir_cbk, (void *)(long)subvol,
+ priv->children[subvol],
+ priv->children[subvol]->fops->readdir, local->fd,
+ local->cont.readdir.size, local->cont.readdir.offset,
+ local->xdata_req);
+ else
+ STACK_WIND_COOKIE(frame, afr_readdir_cbk, (void *)(long)subvol,
+ priv->children[subvol],
+ priv->children[subvol]->fops->readdirp, local->fd,
+ local->cont.readdir.size, local->cont.readdir.offset,
+ local->xdata_req);
+ return 0;
}
-
-int32_t
-afr_readdirp_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
- int32_t op_ret, int32_t op_errno, gf_dirent_t *entries,
- dict_t *xdata)
+int
+afr_do_readdir(call_frame_t *frame, xlator_t *this, fd_t *fd, size_t size,
+ off_t offset, int whichop, dict_t *dict)
{
- afr_private_t * priv = NULL;
- afr_local_t * local = NULL;
- xlator_t ** children = NULL;
- int32_t next_call_child = -1;
- int ret = 0;
- gf_dirent_t * entry = NULL;
- gf_dirent_t * tmp = NULL;
- int32_t *last_index = NULL;
- int32_t read_child = -1;
- int32_t *fresh_children = NULL;
- uint64_t ctx = 0;
- afr_fd_ctx_t *fd_ctx = NULL;
- off_t offset = 0;
- int32_t call_child = -1;
-
- priv = this->private;
- children = priv->children;
-
- local = frame->local;
-
- read_child = (long) cookie;
- last_index = &local->cont.readdir.last_index;
- fresh_children = local->fresh_children;
-
- /* the value of the last_index changes if afr_next_call_child is
- * called. So to find the call_child of this callback use last_index
- * before the next_call_child call.
- */
- if (*last_index == -1)
- call_child = read_child;
- else
- call_child = fresh_children[*last_index];
-
- if (priv->strict_readdir) {
- ret = fd_ctx_get (local->fd, this, &ctx);
- if (ret < 0) {
- gf_log (this->name, GF_LOG_INFO,
- "could not get fd ctx for fd=%p", local->fd);
- op_ret = -1;
- op_errno = -ret;
- goto out;
- }
-
- fd_ctx = (afr_fd_ctx_t *)(long) ctx;
-
- if (op_ret == -1) {
- next_call_child = afr_next_call_child (fresh_children,
- local->child_up,
- priv->child_count,
- last_index,
- read_child);
- if (next_call_child < 0)
- goto out;
- gf_log (this->name, GF_LOG_TRACE,
- "starting readdir afresh on child %d, offset %"PRId64,
- next_call_child, (uint64_t) 0);
-
- fd_ctx->failed_over = _gf_true;
-
- STACK_WIND_COOKIE (frame, afr_readdirp_cbk,
- (void *) (long) read_child,
- children[next_call_child],
- children[next_call_child]->fops->readdirp,
- local->fd,
- local->cont.readdir.size, 0,
- local->cont.readdir.dict);
- return 0;
- }
- }
-
- if (op_ret != -1) {
- list_for_each_entry_safe (entry, tmp, &entries->list, list) {
- if ((local->fd->inode == local->fd->inode->table->root)
- && !strcmp (entry->d_name, GF_REPLICATE_TRASH_DIR)) {
- list_del_init (&entry->list);
- GF_FREE (entry);
- }
- }
- }
-
- if (priv->strict_readdir) {
- if (fd_ctx->failed_over) {
- if (list_empty (&entries->list)) {
- gf_log (this->name, GF_LOG_DEBUG,
- "no entries found");
- goto out;
- }
-
- offset = afr_filter_entries (entries, local->fd);
-
- afr_remember_entries (entries, local->fd);
-
- if (list_empty (&entries->list)) {
- /* All the entries we got were duplicate. We
- shouldn't send an empty list now, because
- that will make the application stop reading. So
- try to get more entries */
-
- gf_log (this->name, GF_LOG_TRACE,
- "trying to fetch non-duplicate entries "
- "from offset %"PRId64", child %s",
- offset, children[call_child]->name);
-
- STACK_WIND_COOKIE (frame, afr_readdirp_cbk,
- (void *) (long) read_child,
- children[call_child],
- children[call_child]->fops->readdirp,
- local->fd, local->cont.readdir.size, offset,
- local->cont.readdir.dict);
- return 0;
- }
- } else {
- afr_remember_entries (entries, local->fd);
- }
- }
-
+ afr_local_t *local = NULL;
+ int32_t op_errno = 0;
+ int subvol = -1;
+ afr_fd_ctx_t *fd_ctx = NULL;
+
+ local = AFR_FRAME_INIT(frame, op_errno);
+ if (!local)
+ goto out;
+
+ fd_ctx = afr_fd_ctx_get(fd, this);
+ if (!fd_ctx) {
+ op_errno = EINVAL;
+ goto out;
+ }
+
+ local->op = whichop;
+ local->fd = fd_ref(fd);
+ local->cont.readdir.size = size;
+ local->cont.readdir.offset = offset;
+ local->xdata_req = (dict) ? dict_ref(dict) : NULL;
+
+ subvol = fd_ctx->readdir_subvol;
+
+ if (offset == 0 || subvol == -1) {
+ /* First readdir has option of failing over and selecting
+ an appropriate read subvolume */
+ afr_read_txn(frame, this, fd->inode, afr_readdir_wind,
+ AFR_DATA_TRANSACTION);
+ } else {
+ /* But continued readdirs MUST stick to the same subvolume
+ without an option to failover */
+ afr_readdir_wind(frame, this, subvol);
+ }
+
+ return 0;
out:
- AFR_STACK_UNWIND (readdirp, frame, op_ret, op_errno, entries, NULL);
-
- return 0;
+ AFR_STACK_UNWIND(readdir, frame, -1, op_errno, NULL, NULL);
+ return 0;
}
int32_t
-afr_do_readdir (call_frame_t *frame, xlator_t *this,
- fd_t *fd, size_t size, off_t offset, int whichop, dict_t *dict)
+afr_readdir(call_frame_t *frame, xlator_t *this, fd_t *fd, size_t size,
+ off_t offset, dict_t *xdata)
{
- afr_private_t * priv = NULL;
- xlator_t ** children = NULL;
- int call_child = 0;
- afr_local_t *local = NULL;
- uint64_t ctx = 0;
- afr_fd_ctx_t *fd_ctx = NULL;
- int ret = -1;
- int32_t op_errno = 0;
- uint64_t read_child = 0;
-
- VALIDATE_OR_GOTO (frame, out);
- VALIDATE_OR_GOTO (this, out);
- VALIDATE_OR_GOTO (this->private, out);
-
- priv = this->private;
- children = priv->children;
-
- AFR_LOCAL_ALLOC_OR_GOTO (frame->local, out);
- local = frame->local;
-
- ret = afr_local_init (local, priv, &op_errno);
- if (ret < 0)
- goto out;
-
- local->fresh_children = afr_children_create (priv->child_count);
- if (!local->fresh_children) {
- op_errno = ENOMEM;
- goto out;
- }
-
- read_child = afr_inode_get_read_ctx (this, fd->inode,
- local->fresh_children);
- ret = afr_get_call_child (this, local->child_up, read_child,
- local->fresh_children,
- &call_child,
- &local->cont.readdir.last_index);
- if (ret < 0) {
- op_errno = -ret;
- goto out;
- }
-
- local->fd = fd_ref (fd);
- local->cont.readdir.size = size;
- local->cont.readdir.dict = (dict)? dict_ref (dict) : NULL;
-
- if (priv->strict_readdir) {
- ret = fd_ctx_get (fd, this, &ctx);
- if (ret < 0) {
- gf_log (this->name, GF_LOG_INFO,
- "could not get fd ctx for fd=%p", fd);
- op_errno = -ret;
- goto out;
- }
-
- fd_ctx = (afr_fd_ctx_t *)(long) ctx;
+ afr_do_readdir(frame, this, fd, size, offset, GF_FOP_READDIR, xdata);
- if (fd_ctx->last_tried != call_child) {
- gf_log (this->name, GF_LOG_TRACE,
- "first up child has changed from %d to %d, "
- "restarting readdir from offset 0",
- fd_ctx->last_tried, call_child);
-
- fd_ctx->failed_over = _gf_true;
- offset = 0;
- }
-
- fd_ctx->last_tried = call_child;
- }
-
- if (whichop == GF_FOP_READDIR)
- STACK_WIND_COOKIE (frame, afr_readdir_cbk,
- (void *) (long) call_child,
- children[call_child],
- children[call_child]->fops->readdir, fd,
- size, offset, dict);
- else
- STACK_WIND_COOKIE (frame, afr_readdirp_cbk,
- (void *) (long) call_child,
- children[call_child],
- children[call_child]->fops->readdirp, fd,
- size, offset, dict);
-
- ret = 0;
-out:
- if (ret < 0)
- AFR_STACK_UNWIND (readdir, frame, -1, op_errno, NULL, NULL);
- return 0;
+ return 0;
}
-
int32_t
-afr_readdir (call_frame_t *frame, xlator_t *this, fd_t *fd, size_t size,
- off_t offset, dict_t *xdata)
+afr_readdirp(call_frame_t *frame, xlator_t *this, fd_t *fd, size_t size,
+ off_t offset, dict_t *dict)
{
- afr_do_readdir (frame, this, fd, size, offset, GF_FOP_READDIR, xdata);
- return 0;
-}
-
+ afr_do_readdir(frame, this, fd, size, offset, GF_FOP_READDIRP, dict);
-int32_t
-afr_readdirp (call_frame_t *frame, xlator_t *this, fd_t *fd, size_t size,
- off_t offset, dict_t *dict)
-{
- afr_do_readdir (frame, this, fd, size, offset, GF_FOP_READDIRP, dict);
- return 0;
+ return 0;
}
-
int32_t
-afr_releasedir (xlator_t *this, fd_t *fd)
+afr_releasedir(xlator_t *this, fd_t *fd)
{
- afr_forget_entries (fd);
- afr_cleanup_fd_ctx (this, fd);
+ afr_cleanup_fd_ctx(this, fd);
- return 0;
+ return 0;
}