diff options
Diffstat (limited to 'xlators/features/marker/src/marker-quota.c')
| -rw-r--r-- | xlators/features/marker/src/marker-quota.c | 4025 |
1 files changed, 1908 insertions, 2117 deletions
diff --git a/xlators/features/marker/src/marker-quota.c b/xlators/features/marker/src/marker-quota.c index 4e820cc2006..3de2ea1c92c 100644 --- a/xlators/features/marker/src/marker-quota.c +++ b/xlators/features/marker/src/marker-quota.c @@ -1,2506 +1,2297 @@ -/*Copyright (c) 2008-2011 Gluster, Inc. <http://www.gluster.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 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 - General Public License for more details. - - You should have received a copy of the GNU General Public License - along with this program. If not, see - <http://www.gnu.org/licenses/>. +/* + Copyright (c) 2008-2012 Red Hat, Inc. <http://www.redhat.com> + This file is part of GlusterFS. + + 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. */ - -#ifndef _CONFIG_H -#define _CONFIG_H -#include "config.h" -#endif - -#include "dict.h" -#include "xlator.h" -#include "defaults.h" +#include <glusterfs/dict.h> +#include <glusterfs/xlator.h> +#include <glusterfs/defaults.h> #include "libxlator.h" -#include "common-utils.h" -#include "byte-order.h" +#include <glusterfs/common-utils.h> +#include <glusterfs/byte-order.h> #include "marker-quota.h" #include "marker-quota-helper.h" +#include <glusterfs/syncop.h> +#include <glusterfs/quota-common-utils.h> int -mq_loc_copy (loc_t *dst, loc_t *src) +mq_loc_copy(loc_t *dst, loc_t *src) { - int ret = -1; + int ret = -1; - GF_VALIDATE_OR_GOTO ("marker", dst, out); - GF_VALIDATE_OR_GOTO ("marker", src, out); + GF_VALIDATE_OR_GOTO("marker", dst, out); + GF_VALIDATE_OR_GOTO("marker", src, out); - if (src->inode == NULL || - src->path == NULL) { - gf_log ("marker", GF_LOG_WARNING, - "src loc is not valid"); - goto out; - } + if (src->inode == NULL || + ((src->parent == NULL) && (gf_uuid_is_null(src->pargfid)) && + !__is_root_gfid(src->inode->gfid))) { + gf_log("marker", GF_LOG_WARNING, "src loc is not valid"); + goto out; + } - ret = loc_copy (dst, src); + ret = loc_copy(dst, src); out: - return ret; + return ret; } -int32_t -mq_get_local_err (quota_local_t *local, - int32_t *val) +static void +mq_set_ctx_status(quota_inode_ctx_t *ctx, gf_boolean_t *flag, + gf_boolean_t status) { - int32_t ret = -1; - - GF_VALIDATE_OR_GOTO ("marker", local, out); - GF_VALIDATE_OR_GOTO ("marker", val, out); - - LOCK (&local->lock); - { - *val = local->err; - } - UNLOCK (&local->lock); - - ret = 0; -out: - return ret; + LOCK(&ctx->lock); + { + *flag = status; + } + UNLOCK(&ctx->lock); } -int32_t -mq_get_ctx_updation_status (quota_inode_ctx_t *ctx, - gf_boolean_t *status) +static void +mq_test_and_set_ctx_status(quota_inode_ctx_t *ctx, gf_boolean_t *flag, + gf_boolean_t *status) { - int32_t ret = -1; - - GF_VALIDATE_OR_GOTO ("marker", ctx, out); - GF_VALIDATE_OR_GOTO ("marker", status, out); - - LOCK (&ctx->lock); - { - *status = ctx->updation_status; - } - UNLOCK (&ctx->lock); - - ret = 0; -out: - return ret; + gf_boolean_t temp = _gf_false; + + LOCK(&ctx->lock); + { + temp = *status; + *status = *flag; + *flag = temp; + } + UNLOCK(&ctx->lock); } +static void +mq_get_ctx_status(quota_inode_ctx_t *ctx, gf_boolean_t *flag, + gf_boolean_t *status) +{ + LOCK(&ctx->lock); + { + *status = *flag; + } + UNLOCK(&ctx->lock); +} int32_t -mq_set_ctx_updation_status (quota_inode_ctx_t *ctx, - gf_boolean_t status) +mq_get_ctx_updation_status(quota_inode_ctx_t *ctx, gf_boolean_t *status) { - int32_t ret = -1; - - if (ctx == NULL) - goto out; - - LOCK (&ctx->lock); - { - ctx->updation_status = status; - } - UNLOCK (&ctx->lock); + GF_VALIDATE_OR_GOTO("marker", ctx, out); + GF_VALIDATE_OR_GOTO("marker", status, out); - ret = 0; + mq_get_ctx_status(ctx, &ctx->updation_status, status); + return 0; out: - return ret; + return -1; } int32_t -mq_test_and_set_ctx_updation_status (quota_inode_ctx_t *ctx, - gf_boolean_t *status) +mq_set_ctx_updation_status(quota_inode_ctx_t *ctx, gf_boolean_t status) { - int32_t ret = -1; - gf_boolean_t temp = _gf_false; - - GF_VALIDATE_OR_GOTO ("marker", ctx, out); - GF_VALIDATE_OR_GOTO ("marker", status, out); - - LOCK (&ctx->lock); - { - temp = *status; - *status = ctx->updation_status; - ctx->updation_status = temp; - } - UNLOCK (&ctx->lock); + GF_VALIDATE_OR_GOTO("marker", ctx, out); - ret = 0; + mq_set_ctx_status(ctx, &ctx->updation_status, status); + return 0; out: - return ret; + return -1; } -void -mq_assign_lk_owner (xlator_t *this, call_frame_t *frame) +int32_t +mq_test_and_set_ctx_updation_status(quota_inode_ctx_t *ctx, + gf_boolean_t *status) { - marker_conf_t *conf = NULL; - uint64_t lk_owner = 0; - - conf = this->private; - - LOCK (&conf->lock); - { - if (++conf->quota_lk_owner == 0) { - ++conf->quota_lk_owner; - } - - lk_owner = conf->quota_lk_owner; - } - UNLOCK (&conf->lock); - - frame->root->lk_owner = lk_owner; + GF_VALIDATE_OR_GOTO("marker", ctx, out); + GF_VALIDATE_OR_GOTO("marker", status, out); - return; + mq_test_and_set_ctx_status(ctx, &ctx->updation_status, status); + return 0; +out: + return -1; } - int32_t -mq_loc_fill_from_name (xlator_t *this, loc_t *newloc, loc_t *oldloc, - uint64_t ino, char *name) +mq_set_ctx_create_status(quota_inode_ctx_t *ctx, gf_boolean_t status) { - int32_t ret = -1; - int32_t len = 0; - char *path = NULL; - - GF_VALIDATE_OR_GOTO ("marker", this, out); - GF_VALIDATE_OR_GOTO ("marker", newloc, out); - GF_VALIDATE_OR_GOTO ("marker", oldloc, out); - GF_VALIDATE_OR_GOTO ("marker", name, out); - - newloc->ino = ino; - - newloc->inode = inode_new (oldloc->inode->table); - - if (!newloc->inode) { - ret = -1; - goto out; - } - - newloc->parent = inode_ref (oldloc->inode); + GF_VALIDATE_OR_GOTO("marker", ctx, out); - len = strlen (oldloc->path); - - if (oldloc->path [len - 1] == '/') - ret = gf_asprintf ((char **) &path, "%s%s", - oldloc->path, name); - else - ret = gf_asprintf ((char **) &path, "%s/%s", - oldloc->path, name); - - if (ret < 0) - goto out; - - newloc->path = path; - - newloc->name = strrchr (newloc->path, '/'); - - if (newloc->name) - newloc->name++; - - gf_log (this->name, GF_LOG_DEBUG, "path = %s name =%s", - newloc->path, newloc->name); + mq_set_ctx_status(ctx, &ctx->create_status, status); + return 0; out: - return ret; + return -1; } int32_t -mq_dirty_inode_updation_done (call_frame_t *frame, void *cookie, xlator_t *this, - int32_t op_ret, int32_t op_errno) +mq_test_and_set_ctx_create_status(quota_inode_ctx_t *ctx, gf_boolean_t *status) { - QUOTA_STACK_DESTROY (frame, this); + GF_VALIDATE_OR_GOTO("marker", ctx, out); + GF_VALIDATE_OR_GOTO("marker", status, out); - return 0; + mq_test_and_set_ctx_status(ctx, &ctx->create_status, status); + return 0; +out: + return -1; } -int32_t -mq_release_lock_on_dirty_inode (call_frame_t *frame, void *cookie, xlator_t *this, - int32_t op_ret, int32_t op_errno) +static void +mq_set_ctx_dirty_status(quota_inode_ctx_t *ctx, gf_boolean_t status) { - struct gf_flock lock = {0, }; - quota_local_t *local = NULL; - loc_t loc = {0, }; - int ret = -1; - - local = frame->local; - - if (op_ret == -1) { - local->err = -1; - - mq_dirty_inode_updation_done (frame, NULL, this, 0, 0); - - return 0; - } - - if (op_ret == 0) - local->ctx->dirty = 0; - - lock.l_type = F_UNLCK; - lock.l_whence = SEEK_SET; - lock.l_start = 0; - lock.l_len = 0; - lock.l_pid = 0; - - ret = loc_copy (&loc, &local->loc); - if (ret == -1) { - local->err = -1; - frame->local = NULL; - mq_dirty_inode_updation_done (frame, NULL, this, 0, 0); - return 0; - } - - if (local->loc.inode == NULL) { - gf_log (this->name, GF_LOG_WARNING, - "Inode is NULL, so can't stackwind."); - goto out; - } - - STACK_WIND (frame, - mq_dirty_inode_updation_done, - FIRST_CHILD(this), - FIRST_CHILD(this)->fops->inodelk, - this->name, &loc, F_SETLKW, &lock); - - loc_wipe (&loc); + GF_VALIDATE_OR_GOTO("marker", ctx, out); - return 0; + mq_set_ctx_status(ctx, &ctx->dirty_status, status); out: - mq_dirty_inode_updation_done (frame, NULL, this, -1, 0); - - return 0; + return; } -int32_t -mq_mark_inode_undirty (call_frame_t *frame, void *cookie, xlator_t *this, - int32_t op_ret, int32_t op_errno, dict_t *dict) +int +mq_build_ancestry(xlator_t *this, loc_t *loc) { - int32_t ret = -1; - int64_t *size = NULL; - dict_t *newdict = NULL; - quota_local_t *local = NULL; - marker_conf_t *priv = NULL; - - local = (quota_local_t *) frame->local; - - if (op_ret == -1) - goto err; - - priv = (marker_conf_t *) this->private; - - if (!dict) - goto wind; - - ret = dict_get_bin (dict, QUOTA_SIZE_KEY, (void **) &size); - if (ret) - goto wind; - - LOCK (&local->ctx->lock); - { - local->ctx->size = ntoh64 (*size); + int32_t ret = -1; + fd_t *fd = NULL; + gf_dirent_t entries; + gf_dirent_t *entry = NULL; + dict_t *xdata = NULL; + inode_t *tmp_parent = NULL; + inode_t *tmp_inode = NULL; + inode_t *linked_inode = NULL; + quota_inode_ctx_t *ctx = NULL; + + INIT_LIST_HEAD(&entries.list); + + xdata = dict_new(); + if (xdata == NULL) { + gf_log(this->name, GF_LOG_ERROR, "dict_new failed"); + ret = -ENOMEM; + goto out; + } + + ret = dict_set_int8(xdata, GET_ANCESTRY_DENTRY_KEY, 1); + if (ret < 0) + goto out; + + fd = fd_anonymous(loc->inode); + if (fd == NULL) { + gf_log(this->name, GF_LOG_ERROR, "fd creation failed"); + ret = -ENOMEM; + goto out; + } + + fd_bind(fd); + + ret = syncop_readdirp(this, fd, 131072, 0, &entries, xdata, NULL); + if (ret < 0) { + gf_log(this->name, + (-ret == ENOENT || -ret == ESTALE) ? GF_LOG_DEBUG : GF_LOG_ERROR, + "readdirp failed " + "for %s: %s", + loc->path, strerror(-ret)); + goto out; + } + + if (list_empty(&entries.list)) { + ret = -1; + goto out; + } + + list_for_each_entry(entry, &entries.list, list) + { + if (__is_root_gfid(entry->inode->gfid)) { + /* The list contains a sub-list for each possible path + * to the target inode. Each sub-list starts with the + * root entry of the tree and is followed by the child + * entries for a particular path to the target entry. + * The root entry is an implied sub-list delimiter, + * as it denotes we have started processing a new path. + * Reset the parent pointer and continue + */ + + tmp_parent = NULL; + } else { + linked_inode = inode_link(entry->inode, tmp_parent, entry->d_name, + &entry->d_stat); + if (linked_inode) { + tmp_inode = entry->inode; + entry->inode = linked_inode; + inode_unref(tmp_inode); + } else { + gf_log(this->name, GF_LOG_ERROR, "inode link failed"); + ret = -EINVAL; + goto out; + } } - UNLOCK (&local->ctx->lock); - -wind: - newdict = dict_new (); - if (!newdict) - goto err; - ret = dict_set_int8 (newdict, QUOTA_DIRTY_KEY, 0); - if (ret) - goto err; - - STACK_WIND (frame, mq_release_lock_on_dirty_inode, - FIRST_CHILD(this), - FIRST_CHILD(this)->fops->setxattr, - &local->loc, newdict, 0); - ret = 0; - -err: - if (op_ret == -1 || ret == -1) { - local->err = -1; - - mq_release_lock_on_dirty_inode (frame, NULL, this, 0, 0); + ctx = mq_inode_ctx_new(entry->inode, this); + if (ctx == NULL) { + gf_log(this->name, GF_LOG_WARNING, + "mq_inode_ctx_new " + "failed for %s", + uuid_utoa(entry->inode->gfid)); + ret = -ENOMEM; + goto out; } - if (newdict) - dict_unref (newdict); - - return 0; -} - -int32_t -mq_update_size_xattr (call_frame_t *frame, void *cookie, xlator_t *this, - int32_t op_ret, int32_t op_errno, inode_t *inode, - struct iatt *buf, dict_t *dict, struct iatt *postparent) -{ - int32_t ret = -1; - dict_t *new_dict = NULL; - int64_t *size = NULL; - int64_t *delta = NULL; - quota_local_t *local = NULL; - marker_conf_t *priv = NULL; - - local = frame->local; + /* For non-directory, posix_get_ancestry_non_directory returns + * all hard-links that are represented by nodes adjacent to + * each other in the dentry-list. + * (Unlike the directory case where adjacent nodes either have + * a parent/child relationship or belong to different paths). + */ + if (entry->inode->ia_type == IA_IFDIR) + tmp_parent = entry->inode; + } - if (op_ret == -1) - goto err; + if (loc->parent) + inode_unref(loc->parent); - priv = this->private; + loc->parent = inode_parent(loc->inode, 0, NULL); + if (loc->parent == NULL) { + ret = -1; + goto out; + } - if (dict == NULL) { - gf_log (this->name, GF_LOG_WARNING, - "Dict is null while updating the size xattr %s", - local->loc.path?local->loc.path:""); - goto err; - } + ret = 0; - ret = dict_get_bin (dict, QUOTA_SIZE_KEY, (void **) &size); - if (!size) { - gf_log (this->name, GF_LOG_WARNING, - "failed to get the size, %s", - local->loc.path?local->loc.path:""); - goto err; - } +out: + gf_dirent_free(&entries); - QUOTA_ALLOC_OR_GOTO (delta, int64_t, ret, err); + if (fd) + fd_unref(fd); - *delta = hton64 (local->sum - ntoh64 (*size)); + if (xdata) + dict_unref(xdata); - gf_log (this->name, GF_LOG_DEBUG, "calculated size = %"PRId64", " - "original size = %"PRIu64 - " path = %s diff = %"PRIu64, local->sum, ntoh64 (*size), - local->loc.path, ntoh64 (*delta)); + return ret; +} - new_dict = dict_new (); - if (!new_dict); +/* This function should be used only in inspect_directory and inspect_file + * function to heal quota xattrs. + * Inode quota feature is introduced in 3.7. + * If gluster setup is upgraded from 3.6 to 3.7, there can be a + * getxattr and setxattr spikes with quota heal as inode quota is missing. + * So this wrapper function is to avoid xattrs spikes during upgrade. + * This function returns success even is inode-quota xattrs are missing and + * hence no healing performed. + */ +static int32_t +_quota_dict_get_meta(xlator_t *this, dict_t *dict, char *key, const int keylen, + quota_meta_t *meta, ia_type_t ia_type, + gf_boolean_t add_delta) +{ + int32_t ret = 0; + marker_conf_t *priv = NULL; - ret = dict_set_bin (new_dict, QUOTA_SIZE_KEY, delta, 8); - if (ret) - goto err; + priv = this->private; - STACK_WIND (frame, mq_mark_inode_undirty, FIRST_CHILD(this), - FIRST_CHILD(this)->fops->xattrop, &local->loc, - GF_XATTROP_ADD_ARRAY64, new_dict); + ret = quota_dict_get_inode_meta(dict, key, keylen, meta); + if (ret == -2 && (priv->feature_enabled & GF_INODE_QUOTA) == 0) { + /* quota_dict_get_inode_meta returns -2 if + * inode quota xattrs are not present. + * if inode quota self heal is turned off, + * then we should skip healing inode quotas + */ + gf_log(this->name, GF_LOG_DEBUG, + "inode quota disabled. " + "inode quota self heal will not be performed"); ret = 0; - -err: - if (op_ret == -1 || ret == -1) { - local->err = -1; - - mq_release_lock_on_dirty_inode (frame, NULL, this, 0, 0); + if (add_delta) { + if (ia_type == IA_IFDIR) + meta->dir_count = 1; + else + meta->file_count = 1; } + } - if (new_dict) - dict_unref (new_dict); - - return 0; + return ret; } int32_t -mq_test_and_set_local_err(quota_local_t *local, - int32_t *val) +quota_dict_set_size_meta(xlator_t *this, dict_t *dict, const quota_meta_t *meta) { - int tmp = 0; - int32_t ret = -1; - - GF_VALIDATE_OR_GOTO ("marker", local, out); - GF_VALIDATE_OR_GOTO ("marker", val, out); - - LOCK (&local->lock); - { - tmp = local->err; - local->err = *val; - *val = tmp; - } - UNLOCK (&local->lock); - - ret = 0; + int32_t ret = -ENOMEM; + quota_meta_t *value = NULL; + char size_key[QUOTA_KEY_MAX] = { + 0, + }; + + value = GF_MALLOC(2 * sizeof(quota_meta_t), gf_common_quota_meta_t); + if (value == NULL) { + goto out; + } + value[0].size = hton64(meta->size); + value[0].file_count = hton64(meta->file_count); + value[0].dir_count = hton64(meta->dir_count); + + value[1].size = 0; + value[1].file_count = 0; + value[1].dir_count = hton64(1); + + GET_SIZE_KEY(this, size_key, ret); + if (ret < 0) + goto out; + ret = dict_set_bin(dict, size_key, value, (sizeof(quota_meta_t) * 2)); + if (ret < 0) { + gf_log_callingfn("quota", GF_LOG_ERROR, "dict set failed"); + GF_FREE(value); + } out: - return ret; + return ret; } -int32_t -mq_get_dirty_inode_size (call_frame_t *frame, xlator_t *this) +void +mq_compute_delta(quota_meta_t *delta, const quota_meta_t *op1, + const quota_meta_t *op2) { - int32_t ret = -1; - dict_t *dict = NULL; - quota_local_t *local = NULL; - marker_conf_t *priv = NULL; - - local = (quota_local_t *) frame->local; - - priv = (marker_conf_t *) this->private; - - dict = dict_new (); - if (!dict) { - ret = -1; - goto err; - } - - ret = dict_set_int64 (dict, QUOTA_SIZE_KEY, 0); - if (ret) - goto err; - - STACK_WIND (frame, mq_update_size_xattr, FIRST_CHILD(this), - FIRST_CHILD(this)->fops->lookup, &local->loc, dict); - ret =0; - -err: - if (ret) { - local->err = -1; - - mq_release_lock_on_dirty_inode (frame, NULL, this, 0, 0); - } - - if (dict) - dict_unref (dict); - - return 0; + delta->size = op1->size - op2->size; + delta->file_count = op1->file_count - op2->file_count; + delta->dir_count = op1->dir_count - op2->dir_count; } -int32_t -mq_get_child_contribution (call_frame_t *frame, - void *cookie, - xlator_t *this, - int32_t op_ret, - int32_t op_errno, - inode_t *inode, - struct iatt *buf, - dict_t *dict, - struct iatt *postparent) +void +mq_add_meta(quota_meta_t *dst, const quota_meta_t *src) { - int32_t ret = -1; - int32_t val = 0; - char contri_key [512] = {0, }; - int64_t *contri = NULL; - quota_local_t *local = NULL; - - local = frame->local; - - frame->local = NULL; - - QUOTA_STACK_DESTROY (frame, this); - - if (op_ret == -1) { - gf_log (this->name, GF_LOG_ERROR, "%s", - strerror (op_errno)); - val = -2; - if (!mq_test_and_set_local_err (local, &val) && - val != -2) - mq_release_lock_on_dirty_inode (local->frame, NULL, this, 0, 0); - - goto exit; - } - - ret = mq_get_local_err (local, &val); - if (!ret && val == -2) - goto exit; - - GET_CONTRI_KEY (contri_key, local->loc.inode->gfid, ret); - if (ret < 0) - goto out; - - if (!dict) - goto out; - - if (dict_get_bin (dict, contri_key, (void **) &contri) == 0) - local->sum += ntoh64 (*contri); - -out: - LOCK (&local->lock); - { - val = --local->dentry_child_count; - } - UNLOCK (&local->lock); - - if (val == 0) { - mq_dirty_inode_readdir (local->frame, NULL, this, - 0, 0, NULL); - } - mq_local_unref (this, local); - - return 0; -exit: - mq_local_unref (this, local); - return 0; + dst->size += src->size; + dst->file_count += src->file_count; + dst->dir_count += src->dir_count; } -int32_t -mq_readdir_cbk (call_frame_t *frame, - void *cookie, - xlator_t *this, - int32_t op_ret, - int32_t op_errno, - gf_dirent_t *entries) +void +mq_sub_meta(quota_meta_t *dst, const quota_meta_t *src) { - char contri_key [512] = {0, }; - int32_t ret = 0; - int32_t val = 0; - off_t offset = 0; - int32_t count = 0; - dict_t *dict = NULL; - quota_local_t *local = NULL; - gf_dirent_t *entry = NULL; - call_frame_t *newframe = NULL; - loc_t loc = {0, }; - - local = mq_local_ref (frame->local); - - if (op_ret == -1) { - gf_log (this->name, GF_LOG_DEBUG, - "readdir failed %s", strerror (op_errno)); - local->err = -1; - - mq_release_lock_on_dirty_inode (frame, NULL, this, 0, 0); - - goto end; - } else if (op_ret == 0) { - mq_get_dirty_inode_size (frame, this); - - goto end; - } - - local->dentry_child_count = 0; - - list_for_each_entry (entry, (&entries->list), list) { - gf_log (this->name, GF_LOG_DEBUG, "entry = %s", entry->d_name); - - if ((!strcmp (entry->d_name, ".")) || (!strcmp (entry->d_name, - ".."))) { - gf_log (this->name, GF_LOG_DEBUG, "entry = %s", - entry->d_name); - continue; - } - - offset = entry->d_off; - count++; - } - - if (count == 0) { - mq_get_dirty_inode_size (frame, this); - goto end; - - } - - local->frame = frame; - - LOCK (&local->lock); - { - local->dentry_child_count = count; - local->d_off = offset; - } - UNLOCK (&local->lock); - - - list_for_each_entry (entry, (&entries->list), list) { - gf_log (this->name, GF_LOG_DEBUG, "entry = %s", entry->d_name); - - if ((!strcmp (entry->d_name, ".")) || (!strcmp (entry->d_name, - ".."))) { - gf_log (this->name, GF_LOG_DEBUG, "entry = %s", - entry->d_name); - continue; - } - - ret = mq_loc_fill_from_name (this, &loc, &local->loc, - entry->d_ino, entry->d_name); - if (ret < 0) - goto out; - - ret = 0; - - LOCK (&local->lock); - { - if (local->err != -2) { - newframe = copy_frame (frame); - if (!newframe) { - ret = -1; - } - } else - ret = -1; - } - UNLOCK (&local->lock); - - if (ret == -1) - goto out; - - newframe->local = mq_local_ref (local); - - dict = dict_new (); - if (!dict) { - ret = -1; - goto out; - } - - GET_CONTRI_KEY (contri_key, local->loc.inode->gfid, ret); - if (ret < 0) - goto out; - - ret = dict_set_int64 (dict, contri_key, 0); - if (ret) - goto out; - - STACK_WIND (newframe, - mq_get_child_contribution, - FIRST_CHILD(this), - FIRST_CHILD(this)->fops->lookup, - &loc, dict); - - offset = entry->d_off; - - loc_wipe (&loc); - - newframe = NULL; - - out: - if (dict) { - dict_unref (dict); - dict = NULL; - } - - if (ret) { - val = -2; - mq_test_and_set_local_err (local, &val); - - if (newframe) { - newframe->local = NULL; - mq_local_unref(this, local); - QUOTA_STACK_DESTROY (newframe, this); - } - - break; - } - } - - if (ret && val != -2) { - mq_release_lock_on_dirty_inode (frame, NULL, this, 0, 0); - } -end: - mq_local_unref (this, local); - - return 0; + if (src == NULL) { + dst->size = -dst->size; + dst->file_count = -dst->file_count; + dst->dir_count = -dst->dir_count; + } else { + dst->size = src->size - dst->size; + dst->file_count = src->file_count - dst->file_count; + dst->dir_count = src->dir_count - dst->dir_count; + } } int32_t -mq_dirty_inode_readdir (call_frame_t *frame, - void *cookie, - xlator_t *this, - int32_t op_ret, - int32_t op_errno, - fd_t *fd) +mq_are_xattrs_set(xlator_t *this, loc_t *loc, gf_boolean_t *contri_set, + gf_boolean_t *size_set) { - quota_local_t *local = NULL; - - local = frame->local; - - if (op_ret == -1) { - local->err = -1; - mq_release_lock_on_dirty_inode (frame, NULL, this, 0, 0); - return 0; - } + int32_t ret = -1; + char contri_key[QUOTA_KEY_MAX] = { + 0, + }; + char size_key[QUOTA_KEY_MAX] = { + 0, + }; + quota_meta_t meta = { + 0, + }; + struct iatt stbuf = { + 0, + }; + dict_t *dict = NULL; + dict_t *rsp_dict = NULL; + + dict = dict_new(); + if (dict == NULL) { + gf_log(this->name, GF_LOG_ERROR, "dict_new failed"); + goto out; + } + + ret = mq_req_xattr(this, loc, dict, contri_key, size_key); + if (ret < 0) + goto out; + + ret = syncop_lookup(FIRST_CHILD(this), loc, &stbuf, NULL, dict, &rsp_dict); + if (ret < 0) { + gf_log_callingfn( + this->name, + (-ret == ENOENT || -ret == ESTALE) ? GF_LOG_DEBUG : GF_LOG_ERROR, + "lookup failed " + "for %s: %s", + loc->path, strerror(-ret)); + goto out; + } + + if (rsp_dict == NULL) + goto out; + + *contri_set = _gf_true; + *size_set = _gf_true; + if (loc->inode->ia_type == IA_IFDIR) { + ret = quota_dict_get_inode_meta(rsp_dict, size_key, strlen(size_key), + &meta); + if (ret < 0 || meta.dir_count == 0) + *size_set = _gf_false; + } + + if (!loc_is_root(loc)) { + ret = quota_dict_get_inode_meta(rsp_dict, contri_key, + strlen(contri_key), &meta); + if (ret < 0) + *contri_set = _gf_false; + } - if (local->fd == NULL) - local->fd = fd_ref (fd); + ret = 0; +out: + if (dict) + dict_unref(dict); - STACK_WIND (frame, - mq_readdir_cbk, - FIRST_CHILD(this), - FIRST_CHILD(this)->fops->readdir, - local->fd, READDIR_BUF, local->d_off); + if (rsp_dict) + dict_unref(rsp_dict); - return 0; + return ret; } int32_t -mq_check_if_still_dirty (call_frame_t *frame, - void *cookie, - xlator_t *this, - int32_t op_ret, - int32_t op_errno, - inode_t *inode, - struct iatt *buf, - dict_t *dict, - struct iatt *postparent) +mq_create_size_xattrs(xlator_t *this, quota_inode_ctx_t *ctx, loc_t *loc) { - int8_t dirty = -1; - int32_t ret = -1; - fd_t *fd = NULL; - quota_local_t *local = NULL; - marker_conf_t *priv = NULL; - - local = frame->local; - - if (op_ret == -1) { - gf_log (this->name, GF_LOG_ERROR, "failed to get " - "the dirty xattr for %s", local->loc.path); - goto err; - } + int32_t ret = -1; + quota_meta_t size = { + 0, + }; + dict_t *dict = NULL; - priv = this->private; - - if (!dict) { - ret = -1; - goto err; - } - - ret = dict_get_int8 (dict, QUOTA_DIRTY_KEY, &dirty); - if (ret) - goto err; - - //the inode is not dirty anymore - if (dirty == 0) { - mq_release_lock_on_dirty_inode (frame, NULL, this, 0, 0); - - return 0; - } - - fd = fd_create (local->loc.inode, frame->root->pid); - - local->d_off = 0; - - STACK_WIND(frame, - mq_dirty_inode_readdir, - FIRST_CHILD(this), - FIRST_CHILD(this)->fops->opendir, - &local->loc, fd); + GF_VALIDATE_OR_GOTO("marker", loc, out); + GF_VALIDATE_OR_GOTO("marker", loc->inode, out); + if (loc->inode->ia_type != IA_IFDIR) { ret = 0; + goto out; + } + + dict = dict_new(); + if (!dict) { + gf_log(this->name, GF_LOG_ERROR, "dict_new failed"); + ret = -1; + goto out; + } + + ret = quota_dict_set_size_meta(this, dict, &size); + if (ret < 0) + goto out; + + ret = syncop_xattrop(FIRST_CHILD(this), loc, + GF_XATTROP_ADD_ARRAY64_WITH_DEFAULT, dict, NULL, NULL, + NULL); + + if (ret < 0) { + gf_log_callingfn( + this->name, + (-ret == ENOENT || -ret == ESTALE) ? GF_LOG_DEBUG : GF_LOG_ERROR, + "xattrop failed " + "for %s: %s", + loc->path, strerror(-ret)); + goto out; + } -err: - if (op_ret == -1 || ret == -1) { - local->err = -1; - mq_release_lock_on_dirty_inode (frame, NULL, this, 0, 0); - } - - if (fd != NULL) { - fd_unref (fd); - } +out: + if (dict) + dict_unref(dict); - return 0; + return ret; } int32_t -mq_get_dirty_xattr (call_frame_t *frame, void *cookie, - xlator_t *this, int32_t op_ret, int32_t op_errno) +mq_lock(xlator_t *this, loc_t *loc, short l_type) { - int32_t ret = -1; - dict_t *xattr_req = NULL; - quota_local_t *local = NULL; - marker_conf_t *priv = NULL; - - if (op_ret == -1) { - mq_dirty_inode_updation_done (frame, NULL, this, 0, 0); - return 0; - } - - priv = (marker_conf_t *) this->private; - - local = frame->local; - - xattr_req = dict_new (); - if (xattr_req == NULL) { - ret = -1; - goto err; - } + struct gf_flock lock = { + 0, + }; + int32_t ret = -1; + + GF_VALIDATE_OR_GOTO("marker", loc, out); + GF_VALIDATE_OR_GOTO("marker", loc->inode, out); + + gf_log(this->name, GF_LOG_DEBUG, "set lock type %d on %s", l_type, + loc->path); + + lock.l_len = 0; + lock.l_start = 0; + lock.l_type = l_type; + lock.l_whence = SEEK_SET; + + ret = syncop_inodelk(FIRST_CHILD(this), this->name, loc, F_SETLKW, &lock, + NULL, NULL); + if (ret < 0) + gf_log_callingfn( + this->name, + (-ret == ENOENT || -ret == ESTALE) ? GF_LOG_DEBUG : GF_LOG_ERROR, + "inodelk failed " + "for %s: %s", + loc->path, strerror(-ret)); - ret = dict_set_int8 (xattr_req, QUOTA_DIRTY_KEY, 0); - if (ret) - goto err; - - STACK_WIND (frame, - mq_check_if_still_dirty, - FIRST_CHILD(this), - FIRST_CHILD(this)->fops->lookup, - &local->loc, - xattr_req); - ret = 0; - -err: - if (ret) { - local->err = -1; - mq_release_lock_on_dirty_inode(frame, NULL, this, 0, 0); - } - - if (xattr_req) - dict_unref (xattr_req); +out: - return 0; + return ret; } -/* return 1 when dirty updation started - * 0 other wise - */ int32_t -mq_update_dirty_inode (xlator_t *this, - loc_t *loc, - quota_inode_ctx_t *ctx, - inode_contribution_t *contribution) +mq_get_dirty(xlator_t *this, loc_t *loc, int32_t *dirty) { - int32_t ret = -1; - quota_local_t *local = NULL; - gf_boolean_t status = _gf_false; - struct gf_flock lock = {0, }; - call_frame_t *frame = NULL; - - ret = mq_get_ctx_updation_status (ctx, &status); - if (ret == -1 || status == _gf_true) { - ret = 0; - goto out; - } - - frame = create_frame (this, this->ctx->pool); - if (frame == NULL) { - ret = -1; - goto out; - } - - mq_assign_lk_owner (this, frame); - - local = mq_local_new (); - if (local == NULL) - goto fr_destroy; - - frame->local = local; - ret = mq_loc_copy (&local->loc, loc); - if (ret < 0) - goto fr_destroy; - - local->ctx = ctx; + int32_t ret = -1; + int8_t value = 0; + dict_t *dict = NULL; + dict_t *rsp_dict = NULL; + struct iatt stbuf = { + 0, + }; + + dict = dict_new(); + if (dict == NULL) { + gf_log(this->name, GF_LOG_ERROR, "dict_new failed"); + goto out; + } + + ret = dict_set_int64(dict, QUOTA_DIRTY_KEY, 0); + if (ret < 0) { + gf_log(this->name, GF_LOG_WARNING, "dict set failed"); + goto out; + } + + ret = syncop_lookup(FIRST_CHILD(this), loc, &stbuf, NULL, dict, &rsp_dict); + if (ret < 0) { + gf_log_callingfn( + this->name, + (-ret == ENOENT || -ret == ESTALE) ? GF_LOG_DEBUG : GF_LOG_ERROR, + "lookup failed " + "for %s: %s", + loc->path, strerror(-ret)); + goto out; + } + + ret = dict_get_int8(rsp_dict, QUOTA_DIRTY_KEY, &value); + if (ret < 0) + goto out; + + *dirty = value; - local->contri = contribution; - - lock.l_type = F_WRLCK; - lock.l_whence = SEEK_SET; - lock.l_start = 0; - lock.l_len = 0; - - if (local->loc.inode == NULL) { - ret = -1; - gf_log (this->name, GF_LOG_WARNING, - "Inode is NULL, so can't stackwind."); - goto fr_destroy; - } - - STACK_WIND (frame, - mq_get_dirty_xattr, - FIRST_CHILD(this), - FIRST_CHILD(this)->fops->inodelk, - this->name, &local->loc, F_SETLKW, &lock); - return 1; - -fr_destroy: - QUOTA_STACK_DESTROY (frame, this); out: + if (dict) + dict_unref(dict); - return 0; -} + if (rsp_dict) + dict_unref(rsp_dict); + return ret; +} int32_t -mq_inode_creation_done (call_frame_t *frame, void *cookie, xlator_t *this, - int32_t op_ret, int32_t op_errno) +mq_get_set_dirty(xlator_t *this, loc_t *loc, int32_t dirty, int32_t *prev_dirty) { - if (frame == NULL) - return 0; + int32_t ret = -1; + int8_t value = 0; + quota_inode_ctx_t *ctx = NULL; + dict_t *dict = NULL; + dict_t *rsp_dict = NULL; + + GF_VALIDATE_OR_GOTO("marker", loc, out); + GF_VALIDATE_OR_GOTO("marker", loc->inode, out); + GF_VALIDATE_OR_GOTO("marker", prev_dirty, out); + + ret = mq_inode_ctx_get(loc->inode, this, &ctx); + if (ret < 0) { + gf_log(this->name, GF_LOG_ERROR, + "failed to get inode ctx for " + "%s", + loc->path); + goto out; + } + + dict = dict_new(); + if (!dict) { + gf_log(this->name, GF_LOG_ERROR, "dict_new failed"); + ret = -1; + goto out; + } + + ret = dict_set_int8(dict, QUOTA_DIRTY_KEY, dirty); + if (ret < 0) { + gf_log(this->name, GF_LOG_ERROR, "dict_set failed"); + goto out; + } + + ret = syncop_xattrop(FIRST_CHILD(this), loc, GF_XATTROP_GET_AND_SET, dict, + NULL, NULL, &rsp_dict); + if (ret < 0) { + gf_log_callingfn( + this->name, + (-ret == ENOENT || -ret == ESTALE) ? GF_LOG_DEBUG : GF_LOG_ERROR, + "xattrop failed " + "for %s: %s", + loc->path, strerror(-ret)); + goto out; + } + + *prev_dirty = 0; + if (rsp_dict) { + ret = dict_get_int8(rsp_dict, QUOTA_DIRTY_KEY, &value); + if (ret == 0) + *prev_dirty = value; + } + + LOCK(&ctx->lock); + { + ctx->dirty = dirty; + } + UNLOCK(&ctx->lock); + ret = 0; +out: + if (dict) + dict_unref(dict); - QUOTA_STACK_DESTROY (frame, this); + if (rsp_dict) + dict_unref(rsp_dict); - return 0; + return ret; } - int32_t -mq_xattr_creation_release_lock (call_frame_t *frame, void *cookie, - xlator_t *this, int32_t op_ret, - int32_t op_errno) +mq_mark_dirty(xlator_t *this, loc_t *loc, int32_t dirty) { - struct gf_flock lock = {0, }; - quota_local_t *local = NULL; - - local = frame->local; - - lock.l_type = F_UNLCK; - lock.l_whence = SEEK_SET; - lock.l_start = 0; - lock.l_len = 0; - lock.l_pid = 0; + int32_t ret = -1; + dict_t *dict = NULL; + quota_inode_ctx_t *ctx = NULL; + + GF_VALIDATE_OR_GOTO("marker", loc, out); + GF_VALIDATE_OR_GOTO("marker", loc->inode, out); + + ret = mq_inode_ctx_get(loc->inode, this, &ctx); + if (ret < 0) { + gf_log(this->name, GF_LOG_ERROR, + "failed to get inode ctx for " + "%s", + loc->path); + ret = 0; + goto out; + } + + dict = dict_new(); + if (!dict) { + ret = -1; + gf_log(this->name, GF_LOG_ERROR, "dict_new failed"); + goto out; + } + + ret = dict_set_int8(dict, QUOTA_DIRTY_KEY, dirty); + if (ret < 0) { + gf_log(this->name, GF_LOG_ERROR, "dict_set failed"); + goto out; + } + + ret = syncop_setxattr(FIRST_CHILD(this), loc, dict, 0, NULL, NULL); + if (ret < 0) { + gf_log_callingfn( + this->name, + (-ret == ENOENT || -ret == ESTALE) ? GF_LOG_DEBUG : GF_LOG_ERROR, + "setxattr dirty = %d " + "failed for %s: %s", + dirty, loc->path, strerror(-ret)); + goto out; + } + + LOCK(&ctx->lock); + { + ctx->dirty = dirty; + } + UNLOCK(&ctx->lock); - STACK_WIND (frame, - mq_inode_creation_done, - FIRST_CHILD(this), - FIRST_CHILD(this)->fops->inodelk, - this->name, &local->loc, - F_SETLKW, &lock); +out: + if (dict) + dict_unref(dict); - return 0; + return ret; } - int32_t -mq_create_dirty_xattr (call_frame_t *frame, void *cookie, xlator_t *this, - int32_t op_ret, int32_t op_errno, dict_t *dict) +_mq_get_metadata(xlator_t *this, loc_t *loc, quota_meta_t *contri, + quota_meta_t *size, uuid_t contri_gfid) { - int32_t ret = -1; - dict_t *newdict = NULL; - quota_local_t *local = NULL; - marker_conf_t *priv = NULL; - - if (op_ret < 0) { - goto err; + int32_t ret = -1; + quota_meta_t meta = { + 0, + }; + char contri_key[QUOTA_KEY_MAX] = { + 0, + }; + char size_key[QUOTA_KEY_MAX] = { + 0, + }; + int keylen = 0; + dict_t *dict = NULL; + dict_t *rsp_dict = NULL; + struct iatt stbuf = { + 0, + }; + + GF_VALIDATE_OR_GOTO("marker", loc, out); + GF_VALIDATE_OR_GOTO("marker", loc->inode, out); + + if (size == NULL && contri == NULL) + goto out; + + dict = dict_new(); + if (dict == NULL) { + gf_log(this->name, GF_LOG_ERROR, "dict_new failed"); + goto out; + } + + if (size && loc->inode->ia_type == IA_IFDIR) { + GET_SIZE_KEY(this, size_key, keylen); + if (keylen < 0) + goto out; + ret = dict_set_int64(dict, size_key, 0); + if (ret < 0) { + gf_log(this->name, GF_LOG_ERROR, "dict_set failed."); + goto out; } + } - local = frame->local; - - priv = (marker_conf_t *) this->private; - - if (local->loc.inode->ia_type == IA_IFDIR) { - newdict = dict_new (); - if (!newdict) { - goto err; - } - - ret = dict_set_int8 (newdict, QUOTA_DIRTY_KEY, 0); - if (ret == -1) { - goto err; - } + if (contri && !loc_is_root(loc)) { + ret = mq_dict_set_contribution(this, dict, loc, contri_gfid, + contri_key); + if (ret < 0) + goto out; + } + + ret = syncop_lookup(FIRST_CHILD(this), loc, &stbuf, NULL, dict, &rsp_dict); + if (ret < 0) { + gf_log_callingfn( + this->name, + (-ret == ENOENT || -ret == ESTALE) ? GF_LOG_DEBUG : GF_LOG_ERROR, + "lookup failed " + "for %s: %s", + loc->path, strerror(-ret)); + goto out; + } + + if (size) { + if (loc->inode->ia_type == IA_IFDIR) { + ret = quota_dict_get_meta(rsp_dict, size_key, keylen, &meta); + if (ret < 0) { + gf_log(this->name, GF_LOG_ERROR, "dict_get failed."); + goto out; + } - STACK_WIND (frame, mq_xattr_creation_release_lock, - FIRST_CHILD(this), - FIRST_CHILD(this)->fops->setxattr, - &local->loc, newdict, 0); + size->size = meta.size; + size->file_count = meta.file_count; + size->dir_count = meta.dir_count; } else { - mq_xattr_creation_release_lock (frame, NULL, this, 0, 0); + size->size = stbuf.ia_blocks * 512; + size->file_count = 1; + size->dir_count = 0; } + } - ret = 0; - -err: + if (contri && !loc_is_root(loc)) { + ret = quota_dict_get_meta(rsp_dict, contri_key, strlen(contri_key), + &meta); if (ret < 0) { - mq_xattr_creation_release_lock (frame, NULL, this, 0, 0); + contri->size = 0; + contri->file_count = 0; + contri->dir_count = 0; + } else { + contri->size = meta.size; + contri->file_count = meta.file_count; + contri->dir_count = meta.dir_count; } + } - if (newdict != NULL) - dict_unref (newdict); + ret = 0; - return 0; -} +out: + if (dict) + dict_unref(dict); + if (rsp_dict) + dict_unref(rsp_dict); + + return ret; +} int32_t -mq_create_xattr (xlator_t *this, call_frame_t *frame) +mq_get_metadata(xlator_t *this, loc_t *loc, quota_meta_t *contri, + quota_meta_t *size, quota_inode_ctx_t *ctx, + inode_contribution_t *contribution) { - int32_t ret = 0; - int64_t *value = NULL; - int64_t *size = NULL; - dict_t *dict = NULL; - char key[512] = {0, }; - quota_local_t *local = NULL; - marker_conf_t *priv = NULL; - quota_inode_ctx_t *ctx = NULL; - inode_contribution_t *contri = NULL; - - if (frame == NULL || this == NULL) - return 0; - - local = frame->local; - - priv = (marker_conf_t *) this->private; - - ret = mq_inode_ctx_get (local->loc.inode, this, &ctx); - if (ret < 0) { - ctx = mq_inode_ctx_new (local->loc.inode, this); - if (ctx == NULL) { - gf_log (this->name, GF_LOG_WARNING, - "mq_inode_ctx_new failed"); - ret = -1; - goto out; - } - } - - dict = dict_new (); - if (!dict) - goto out; - - if (local->loc.inode->ia_type == IA_IFDIR) { - QUOTA_ALLOC_OR_GOTO (size, int64_t, ret, err); - ret = dict_set_bin (dict, QUOTA_SIZE_KEY, size, 8); - if (ret < 0) - goto free_size; - } - - if (strcmp (local->loc.path, "/") != 0) { - contri = mq_add_new_contribution_node (this, ctx, &local->loc); - if (contri == NULL) - goto err; + int32_t ret = -1; - QUOTA_ALLOC_OR_GOTO (value, int64_t, ret, err); - GET_CONTRI_KEY (key, local->loc.parent->gfid, ret); + GF_VALIDATE_OR_GOTO("marker", loc, out); + GF_VALIDATE_OR_GOTO("marker", loc->inode, out); + GF_VALIDATE_OR_GOTO("marker", ctx, out); + GF_VALIDATE_OR_GOTO("marker", contribution, out); - ret = dict_set_bin (dict, key, value, 8); - if (ret < 0) - goto free_value; - } - - STACK_WIND (frame, mq_create_dirty_xattr, FIRST_CHILD(this), - FIRST_CHILD(this)->fops->xattrop, &local->loc, - GF_XATTROP_ADD_ARRAY64, dict); + if (size == NULL && contri == NULL) { ret = 0; + goto out; + } -free_size: - if (ret < 0) { - GF_FREE (size); - } + ret = _mq_get_metadata(this, loc, contri, size, contribution->gfid); + if (ret < 0) + goto out; -free_value: - if (ret < 0) { - GF_FREE (value); + if (size) { + LOCK(&ctx->lock); + { + ctx->size = size->size; + ctx->file_count = size->file_count; + ctx->dir_count = size->dir_count; } + UNLOCK(&ctx->lock); + } -err: - dict_unref (dict); - -out: - if (ret < 0) { - mq_xattr_creation_release_lock (frame, NULL, this, 0, 0); + if (contri) { + LOCK(&contribution->lock); + { + contribution->contribution = contri->size; + contribution->file_count = contri->file_count; + contribution->dir_count = contri->dir_count; } + UNLOCK(&contribution->lock); + } - return 0; +out: + return ret; } - int32_t -mq_check_n_set_inode_xattr (call_frame_t *frame, void *cookie, - xlator_t *this, int32_t op_ret, int32_t op_errno, - inode_t *inode, struct iatt *buf, dict_t *dict, - struct iatt *postparent) +mq_get_delta(xlator_t *this, loc_t *loc, quota_meta_t *delta, + quota_inode_ctx_t *ctx, inode_contribution_t *contribution) { - quota_local_t *local = NULL; - int64_t *size = NULL, *contri = NULL; - int8_t dirty = 0; - marker_conf_t *priv = NULL; - int32_t ret = 0; - char contri_key[512] = {0, }; - - if (op_ret < 0) { - goto out; - } - - local = frame->local; - priv = this->private; + int32_t ret = -1; + quota_meta_t size = { + 0, + }; + quota_meta_t contri = { + 0, + }; - ret = dict_get_bin (dict, QUOTA_SIZE_KEY, (void **) &size); - if (ret < 0) - goto create_xattr; + GF_VALIDATE_OR_GOTO("marker", loc, out); + GF_VALIDATE_OR_GOTO("marker", loc->inode, out); + GF_VALIDATE_OR_GOTO("marker", ctx, out); + GF_VALIDATE_OR_GOTO("marker", contribution, out); - ret = dict_get_int8 (dict, QUOTA_DIRTY_KEY, &dirty); - if (ret < 0) - goto create_xattr; + ret = mq_get_metadata(this, loc, &contri, &size, ctx, contribution); + if (ret < 0) + goto out; - //check contribution xattr if not root - if (strcmp (local->loc.path, "/") != 0) { - GET_CONTRI_KEY (contri_key, local->loc.parent->gfid, ret); - if (ret < 0) - goto out; - - ret = dict_get_bin (dict, contri_key, (void **) &contri); - if (ret < 0) - goto create_xattr; - } + mq_compute_delta(delta, &size, &contri); out: - mq_xattr_creation_release_lock (frame, NULL, this, 0, 0); - return 0; - -create_xattr: - mq_create_xattr (this, frame); - return 0; + return ret; } - int32_t -mq_get_xattr (call_frame_t *frame, void *cookie, xlator_t *this, - int32_t op_ret, int32_t op_errno) +mq_remove_contri(xlator_t *this, loc_t *loc, quota_inode_ctx_t *ctx, + inode_contribution_t *contri, quota_meta_t *delta, + uint32_t nlink) { - dict_t *xattr_req = NULL; - quota_local_t *local = NULL; - int32_t ret = 0; - - if (op_ret < 0) { - goto lock_err; - } - - local = frame->local; + int32_t ret = -1; + char contri_key[QUOTA_KEY_MAX] = { + 0, + }; - xattr_req = dict_new (); - if (xattr_req == NULL) { - goto err; - } - - ret = mq_req_xattr (this, &local->loc, xattr_req); - if (ret < 0) { - gf_log (this->name, GF_LOG_WARNING, "cannot request xattr"); - goto err; + if (nlink == 1) { + /*File was a last link and has been deleted */ + ret = 0; + goto done; + } + + GET_CONTRI_KEY(this, contri_key, contri->gfid, ret); + if (ret < 0) { + gf_log(this->name, GF_LOG_ERROR, + "get contri_key " + "failed for %s", + uuid_utoa(contri->gfid)); + goto out; + } + + ret = syncop_removexattr(FIRST_CHILD(this), loc, contri_key, 0, NULL); + if (ret < 0) { + if (-ret == ENOENT || -ret == ESTALE || -ret == ENODATA || + -ret == ENOATTR) { + /* Remove contri in done when unlink operation is + * performed, so return success on ENOENT/ESTSLE + * rename operation removes xattr earlier, + * so return success on ENODATA + */ + ret = 0; + } else { + gf_log_callingfn(this->name, GF_LOG_ERROR, + "removexattr %s failed for %s: %s", contri_key, + loc->path, strerror(-ret)); + goto out; } + } - STACK_WIND (frame, mq_check_n_set_inode_xattr, FIRST_CHILD(this), - FIRST_CHILD(this)->fops->lookup, &local->loc, xattr_req); +done: + LOCK(&contri->lock); + { + contri->contribution += delta->size; + contri->file_count += delta->file_count; + contri->dir_count += delta->dir_count; + } + UNLOCK(&contri->lock); - dict_unref (xattr_req); + ret = 0; - return 0; - -err: - mq_xattr_creation_release_lock (frame, NULL, this, 0, 0); - - if (xattr_req) - dict_unref (xattr_req); - return 0; +out: + QUOTA_FREE_CONTRIBUTION_NODE(ctx, contri); -lock_err: - mq_inode_creation_done (frame, NULL, this, 0, 0); - return 0; + return ret; } - int32_t -mq_set_inode_xattr (xlator_t *this, loc_t *loc) +mq_update_contri(xlator_t *this, loc_t *loc, inode_contribution_t *contri, + quota_meta_t *delta) { - struct gf_flock lock = {0, }; - quota_local_t *local = NULL; - int32_t ret = 0; - call_frame_t *frame = NULL; - - frame = create_frame (this, this->ctx->pool); - if (!frame) { - ret = -1; - goto err; - } - - local = mq_local_new (); - if (local == NULL) { - goto err; - } - - frame->local = local; + int32_t ret = -1; + char contri_key[QUOTA_KEY_MAX] = { + 0, + }; + dict_t *dict = NULL; + + GF_VALIDATE_OR_GOTO("marker", loc, out); + GF_VALIDATE_OR_GOTO("marker", loc->inode, out); + GF_VALIDATE_OR_GOTO("marker", delta, out); + GF_VALIDATE_OR_GOTO("marker", contri, out); + + if (quota_meta_is_null(delta)) { + ret = 0; + goto out; + } + + dict = dict_new(); + if (!dict) { + gf_log(this->name, GF_LOG_ERROR, "dict_new failed"); + ret = -1; + goto out; + } + + GET_CONTRI_KEY(this, contri_key, contri->gfid, ret); + if (ret < 0) { + gf_log(this->name, GF_LOG_ERROR, + "get contri_key " + "failed for %s", + uuid_utoa(contri->gfid)); + goto out; + } + + ret = quota_dict_set_meta(dict, contri_key, delta, loc->inode->ia_type); + if (ret < 0) + goto out; + + ret = syncop_xattrop(FIRST_CHILD(this), loc, GF_XATTROP_ADD_ARRAY64, dict, + NULL, NULL, NULL); + if (ret < 0) { + gf_log_callingfn( + this->name, + (-ret == ENOENT || -ret == ESTALE) ? GF_LOG_DEBUG : GF_LOG_ERROR, + "xattrop failed " + "for %s: %s", + loc->path, strerror(-ret)); + goto out; + } + + LOCK(&contri->lock); + { + contri->contribution += delta->size; + contri->file_count += delta->file_count; + contri->dir_count += delta->dir_count; + } + UNLOCK(&contri->lock); - ret = loc_copy (&local->loc, loc); - if (ret < 0) { - goto err; - } +out: + if (dict) + dict_unref(dict); - frame->local = local; + return ret; +} - lock.l_len = 0; - lock.l_start = 0; - lock.l_type = F_WRLCK; - lock.l_whence = SEEK_SET; +int32_t +mq_update_size(xlator_t *this, loc_t *loc, quota_meta_t *delta) +{ + int32_t ret = -1; + quota_inode_ctx_t *ctx = NULL; + dict_t *dict = NULL; - STACK_WIND (frame, - mq_get_xattr, - FIRST_CHILD(this), - FIRST_CHILD(this)->fops->inodelk, - this->name, &local->loc, F_SETLKW, &lock); + GF_VALIDATE_OR_GOTO("marker", loc, out); + GF_VALIDATE_OR_GOTO("marker", loc->inode, out); + GF_VALIDATE_OR_GOTO("marker", delta, out); - return 0; + if (quota_meta_is_null(delta)) { + ret = 0; + goto out; + } + + ret = mq_inode_ctx_get(loc->inode, this, &ctx); + if (ret < 0) { + gf_log(this->name, GF_LOG_ERROR, + "failed to get inode ctx for " + "%s", + loc->path); + goto out; + } + + dict = dict_new(); + if (!dict) { + gf_log(this->name, GF_LOG_ERROR, "dict_new failed"); + ret = -1; + goto out; + } + + ret = quota_dict_set_size_meta(this, dict, delta); + if (ret < 0) + goto out; + + ret = syncop_xattrop(FIRST_CHILD(this), loc, + GF_XATTROP_ADD_ARRAY64_WITH_DEFAULT, dict, NULL, NULL, + NULL); + if (ret < 0) { + gf_log_callingfn( + this->name, + (-ret == ENOENT || -ret == ESTALE) ? GF_LOG_DEBUG : GF_LOG_ERROR, + "xattrop failed " + "for %s: %s", + loc->path, strerror(-ret)); + goto out; + } + + LOCK(&ctx->lock); + { + ctx->size += delta->size; + ctx->file_count += delta->file_count; + if (ctx->dir_count == 0) + ctx->dir_count += delta->dir_count + 1; + else + ctx->dir_count += delta->dir_count; + } + UNLOCK(&ctx->lock); -err: - QUOTA_STACK_DESTROY (frame, this); +out: + if (dict) + dict_unref(dict); - return 0; + return ret; } - -int32_t -mq_get_parent_inode_local (xlator_t *this, quota_local_t *local) +int +mq_synctask_cleanup(int ret, call_frame_t *frame, void *opaque) { - int32_t ret = -1; - quota_inode_ctx_t *ctx = NULL; - - GF_VALIDATE_OR_GOTO ("marker", this, out); - GF_VALIDATE_OR_GOTO ("marker", local, out); - - loc_wipe (&local->loc); + quota_synctask_t *args = NULL; - ret = mq_loc_copy (&local->loc, &local->parent_loc); - if (ret < 0) { - gf_log_callingfn (this->name, GF_LOG_WARNING, - "loc copy failed"); - goto out; - } + GF_ASSERT(opaque); - loc_wipe (&local->parent_loc); + args = (quota_synctask_t *)opaque; + loc_wipe(&args->loc); - ret = mq_inode_loc_fill (NULL, local->loc.parent, - &local->parent_loc); - if (ret < 0) { - gf_log_callingfn (this->name, GF_LOG_WARNING, - "failed to build parent loc of %s", - local->loc.path); - goto out; - } + if (args->stub) + call_resume(args->stub); - ret = mq_inode_ctx_get (local->loc.inode, this, &ctx); - if (ret < 0) { - gf_log_callingfn (this->name, GF_LOG_WARNING, - "inode ctx get failed"); - goto out; - } + if (!args->is_static) + GF_FREE(args); - local->ctx = ctx; + return 0; +} - if (list_empty (&ctx->contribution_head)) { - gf_log_callingfn (this->name, GF_LOG_WARNING, - "contribution node list is empty which " - "is an error"); - goto out; +int +mq_synctask1(xlator_t *this, synctask_fn_t task, gf_boolean_t spawn, loc_t *loc, + quota_meta_t *contri, uint32_t nlink, call_stub_t *stub) +{ + int32_t ret = -1; + quota_synctask_t *args = NULL; + quota_synctask_t static_args = { + 0, + }; + + if (spawn) { + QUOTA_ALLOC_OR_GOTO(args, quota_synctask_t, ret, out); + args->is_static = _gf_false; + } else { + args = &static_args; + args->is_static = _gf_true; + } + + args->this = this; + args->stub = stub; + loc_copy(&args->loc, loc); + args->ia_nlink = nlink; + + if (contri) { + args->contri = *contri; + } else { + args->contri.size = -1; + args->contri.file_count = -1; + args->contri.dir_count = -1; + } + + if (spawn) { + ret = synctask_new1(this->ctx->env, 1024 * 16, task, + mq_synctask_cleanup, NULL, args); + if (ret) { + gf_log(this->name, GF_LOG_ERROR, + "Failed to spawn " + "new synctask"); + mq_synctask_cleanup(ret, NULL, args); } + } else { + ret = task(args); + mq_synctask_cleanup(ret, NULL, args); + } - local->contri = (inode_contribution_t *) ctx->contribution_head.next; - - ret = 0; out: - return ret; + return ret; } - -int32_t -mq_xattr_updation_done (call_frame_t *frame, - void *cookie, - xlator_t *this, - int32_t op_ret, - int32_t op_errno, - dict_t *dict) +int +mq_synctask(xlator_t *this, synctask_fn_t task, gf_boolean_t spawn, loc_t *loc) { - QUOTA_STACK_DESTROY (frame, this); - return 0; + return mq_synctask1(this, task, spawn, loc, NULL, -1, NULL); } - int32_t -mq_inodelk_cbk (call_frame_t *frame, void *cookie, - xlator_t *this, int32_t op_ret, int32_t op_errno) +mq_prevalidate_txn(xlator_t *this, loc_t *origin_loc, loc_t *loc, + quota_inode_ctx_t **ctx, struct iatt *buf) { - int32_t ret = 0; - gf_boolean_t status = _gf_false; - quota_local_t *local = NULL; - - local = frame->local; - - if (op_ret == -1 || local->err) { - if (op_ret == -1) { - gf_log (this->name, GF_LOG_WARNING, - "unlocking on path (%s) failed (%s)", - local->parent_loc.path, strerror (op_errno)); - } - - mq_xattr_updation_done (frame, NULL, this, 0, 0, NULL); - - return 0; - } - - gf_log (this->name, GF_LOG_DEBUG, - "inodelk released on %s", local->parent_loc.path); - - if ((strcmp (local->parent_loc.path, "/") == 0) - || (local->delta == 0)) { - mq_xattr_updation_done (frame, NULL, this, 0, 0, NULL); - } else { - ret = mq_get_parent_inode_local (this, local); - if (ret < 0) { - mq_xattr_updation_done (frame, NULL, this, 0, 0, NULL); - goto out; - } - status = _gf_true; - - ret = mq_test_and_set_ctx_updation_status (local->ctx, &status); - if (ret == 0 && status == _gf_false) { - mq_get_lock_on_parent (frame, this); - } else { - mq_xattr_updation_done (frame, NULL, this, 0, 0, NULL); - } - } + int32_t ret = -1; + quota_inode_ctx_t *ctxtmp = NULL; + + if (buf) { + if (buf->ia_type == IA_IFREG && IS_DHT_LINKFILE_MODE(buf)) + goto out; + + if (buf->ia_type != IA_IFREG && buf->ia_type != IA_IFLNK && + buf->ia_type != IA_IFDIR) + goto out; + } + + if (origin_loc == NULL || origin_loc->inode == NULL || + gf_uuid_is_null(origin_loc->inode->gfid)) + goto out; + + loc_copy(loc, origin_loc); + + if (gf_uuid_is_null(loc->gfid)) + gf_uuid_copy(loc->gfid, loc->inode->gfid); + + if (!loc_is_root(loc) && loc->parent == NULL) + loc->parent = inode_parent(loc->inode, 0, NULL); + + ret = mq_inode_ctx_get(loc->inode, this, &ctxtmp); + if (ret < 0) { + gf_log_callingfn(this->name, GF_LOG_WARNING, + "inode ctx for " + "is NULL for %s", + loc->path); + goto out; + } + if (ctx) + *ctx = ctxtmp; + + ret = 0; out: - return 0; + return ret; } - -//now release lock on the parent inode -int32_t -mq_release_parent_lock (call_frame_t *frame, void *cookie, - xlator_t *this, int32_t op_ret, - int32_t op_errno) +int +mq_create_xattrs_task(void *opaque) { - int32_t ret = 0; - quota_local_t *local = NULL; - quota_inode_ctx_t *ctx = NULL; - struct gf_flock lock = {0, }; - - local = frame->local; - - if (local->err != 0) { - gf_log_callingfn (this->name, - (local->err == ENOENT) ? GF_LOG_DEBUG - : GF_LOG_WARNING, - "An operation during quota updation " - "of path (%s) failed (%s)", local->loc.path, - strerror (local->err)); - } - - ret = mq_inode_ctx_get (local->parent_loc.inode, this, &ctx); + int32_t ret = -1; + gf_boolean_t locked = _gf_false; + gf_boolean_t contri_set = _gf_false; + gf_boolean_t size_set = _gf_false; + gf_boolean_t need_txn = _gf_false; + quota_synctask_t *args = NULL; + quota_inode_ctx_t *ctx = NULL; + xlator_t *this = NULL; + loc_t *loc = NULL; + gf_boolean_t status = _gf_false; + + GF_ASSERT(opaque); + + args = (quota_synctask_t *)opaque; + loc = &args->loc; + this = args->this; + THIS = this; + + ret = mq_inode_ctx_get(loc->inode, this, &ctx); + if (ret < 0) { + gf_log(this->name, GF_LOG_WARNING, + "Failed to" + "get inode ctx, aborting quota create txn"); + goto out; + } + + if (loc->inode->ia_type == IA_IFDIR) { + /* lock not required for files */ + ret = mq_lock(this, loc, F_WRLCK); if (ret < 0) - goto wind; - - LOCK (&ctx->lock); - { - ctx->dirty = 0; - } - UNLOCK (&ctx->lock); - - if (local->parent_loc.inode == NULL) { - gf_log (this->name, GF_LOG_WARNING, - "Invalid parent inode."); - goto err; - } - -wind: - lock.l_type = F_UNLCK; - lock.l_whence = SEEK_SET; - lock.l_start = 0; - lock.l_len = 0; - lock.l_pid = 0; - - STACK_WIND (frame, - mq_inodelk_cbk, - FIRST_CHILD(this), - FIRST_CHILD(this)->fops->inodelk, - this->name, &local->parent_loc, - F_SETLKW, &lock); - - return 0; -err: - mq_xattr_updation_done (frame, NULL, this, - 0, 0 , NULL); - return 0; -} + goto out; + locked = _gf_true; + } + ret = mq_are_xattrs_set(this, loc, &contri_set, &size_set); + if (ret < 0 || (contri_set && size_set)) + goto out; -int32_t -mq_mark_undirty (call_frame_t *frame, - void *cookie, - xlator_t *this, - int32_t op_ret, - int32_t op_errno, - dict_t *dict) -{ - int32_t ret = -1; - int64_t *size = NULL; - dict_t *newdict = NULL; - quota_local_t *local = NULL; - quota_inode_ctx_t *ctx = NULL; - marker_conf_t *priv = NULL; - - local = frame->local; - - if (op_ret == -1) { - gf_log (this->name, (op_errno == ENOENT) ? GF_LOG_DEBUG - : GF_LOG_WARNING, "cannot update size of path (%s)(%s)", - local->parent_loc.path, strerror (op_errno)); - local->err = op_errno; - goto err; - } - - priv = this->private; - - //update the size of the parent inode - if (dict != NULL) { - ret = mq_inode_ctx_get (local->parent_loc.inode, this, &ctx); - if (ret < 0) { - op_errno = EINVAL; - goto err; - } - - ret = dict_get_bin (dict, QUOTA_SIZE_KEY, (void **) &size); - if (ret < 0) { - op_errno = EINVAL; - goto err; - } - - LOCK (&ctx->lock); - { - if (size) - ctx->size = ntoh64 (*size); - gf_log (this->name, GF_LOG_DEBUG, "%s %"PRId64, - local->parent_loc.path, ctx->size); - } - UNLOCK (&ctx->lock); - } + mq_set_ctx_create_status(ctx, _gf_false); + status = _gf_true; - newdict = dict_new (); - if (!newdict) { - op_errno = ENOMEM; - goto err; - } - - ret = dict_set_int8 (newdict, QUOTA_DIRTY_KEY, 0); - if (ret < 0) { - op_errno = -ret; - goto err; - } - - STACK_WIND (frame, mq_release_parent_lock, - FIRST_CHILD(this), - FIRST_CHILD(this)->fops->setxattr, - &local->parent_loc, newdict, 0); + if (loc->inode->ia_type == IA_IFDIR && size_set == _gf_false) { + ret = mq_create_size_xattrs(this, ctx, loc); + if (ret < 0) + goto out; + } - ret = 0; -err: - if ((op_ret == -1) || (ret < 0)) { - local->err = op_errno; + need_txn = _gf_true; +out: + if (locked) + ret = mq_lock(this, loc, F_UNLCK); - mq_release_parent_lock (frame, NULL, this, 0, 0); - } + if (status == _gf_false) + mq_set_ctx_create_status(ctx, _gf_false); - if (newdict) - dict_unref (newdict); + if (need_txn) + ret = mq_initiate_quota_blocking_txn(this, loc, NULL); - return 0; + return ret; } - -int32_t -mq_update_parent_size (call_frame_t *frame, - void *cookie, - xlator_t *this, - int32_t op_ret, - int32_t op_errno, - dict_t *dict) +static int +_mq_create_xattrs_txn(xlator_t *this, loc_t *origin_loc, struct iatt *buf, + gf_boolean_t spawn) { - int64_t *size = NULL; - int32_t ret = -1; - dict_t *newdict = NULL; - marker_conf_t *priv = NULL; - quota_local_t *local = NULL; - quota_inode_ctx_t *ctx = NULL; - - local = frame->local; - - if (op_ret == -1) { - gf_log (this->name, ((op_errno == ENOENT) ? GF_LOG_DEBUG : - GF_LOG_WARNING), - "xattrop call on path (%s) failed: %s", - local->loc.path, strerror (op_errno)); - - goto err; - } - - LOCK (&local->contri->lock); - { - local->contri->contribution += local->delta; - } - UNLOCK (&local->contri->lock); - - gf_log (this->name, GF_LOG_DEBUG, "%s %"PRId64 "%"PRId64, - local->loc.path, local->ctx->size, - local->contri->contribution); - - priv = this->private; - - if (dict == NULL) { - op_errno = EINVAL; - goto err; - } - - ret = mq_inode_ctx_get (local->parent_loc.inode, this, &ctx); - if (ret < 0) { - op_errno = EINVAL; - goto err; - } - - newdict = dict_new (); - if (!newdict) { - ret = -1; - op_errno = EINVAL; - goto err; + int32_t ret = -1; + quota_inode_ctx_t *ctx = NULL; + gf_boolean_t status = _gf_true; + loc_t loc = { + 0, + }; + inode_contribution_t *contribution = NULL; + + ret = mq_prevalidate_txn(this, origin_loc, &loc, &ctx, buf); + if (ret < 0) + goto out; + + ret = mq_test_and_set_ctx_create_status(ctx, &status); + if (ret < 0 || status == _gf_true) + goto out; + + if (!loc_is_root(&loc) && loc.parent) { + contribution = mq_add_new_contribution_node(this, ctx, &loc); + if (contribution == NULL) { + gf_log(this->name, GF_LOG_WARNING, + "cannot add a new contribution node " + "(%s)", + uuid_utoa(loc.gfid)); + ret = -1; + goto out; + } else { + GF_REF_PUT(contribution); } + } - QUOTA_ALLOC_OR_GOTO (size, int64_t, ret, err); - - *size = hton64 (local->delta); + ret = mq_synctask(this, mq_create_xattrs_task, spawn, &loc); +out: + if (ret < 0 && status == _gf_false) + mq_set_ctx_create_status(ctx, _gf_false); - ret = dict_set_bin (newdict, QUOTA_SIZE_KEY, size, 8); - if (ret < 0) { - op_errno = -ret; - goto err; - } + loc_wipe(&loc); + return ret; +} - STACK_WIND (frame, - mq_mark_undirty, - FIRST_CHILD(this), - FIRST_CHILD(this)->fops->xattrop, - &local->parent_loc, - GF_XATTROP_ADD_ARRAY64, - newdict); - ret = 0; -err: - if (op_ret == -1 || ret < 0) { - local->err = op_errno; - mq_release_parent_lock (frame, NULL, this, 0, 0); - } +int +mq_create_xattrs_txn(xlator_t *this, loc_t *loc, struct iatt *buf) +{ + int32_t ret = -1; - if (newdict) - dict_unref (newdict); + GF_VALIDATE_OR_GOTO("marker", loc, out); + GF_VALIDATE_OR_GOTO("marker", loc->inode, out); - return 0; + ret = _mq_create_xattrs_txn(this, loc, buf, _gf_true); +out: + return ret; } int32_t -mq_update_inode_contribution (call_frame_t *frame, void *cookie, - xlator_t *this, int32_t op_ret, - int32_t op_errno, inode_t *inode, - struct iatt *buf, dict_t *dict, - struct iatt *postparent) +mq_reduce_parent_size_task(void *opaque) { - int32_t ret = -1; - int64_t *size = NULL, size_int = 0; - int64_t contri_int = 0; - int64_t *contri = NULL; - int64_t *delta = NULL; - char contri_key [512] = {0, }; - dict_t *newdict = NULL; - quota_local_t *local = NULL; - quota_inode_ctx_t *ctx = NULL; - marker_conf_t *priv = NULL; - inode_contribution_t *contribution = NULL; - - local = frame->local; - - if (op_ret == -1) { - gf_log (this->name, ((op_errno == ENOENT) ? GF_LOG_DEBUG : - GF_LOG_WARNING), - "failed to get size and contribution with %s error", - strerror (op_errno)); - goto err; - } - - priv = this->private; - - ctx = local->ctx; - contribution = local->contri; - - //prepare to update size & contribution of the inode - GET_CONTRI_KEY (contri_key, contribution->gfid, ret); - if (ret == -1) { - op_errno = ENOMEM; - goto err; - } - - LOCK (&ctx->lock); - { - if (local->loc.inode->ia_type == IA_IFDIR ) { - ret = dict_get_bin (dict, QUOTA_SIZE_KEY, - (void **) &size); - if (ret < 0) { - op_errno = EINVAL; - goto unlock; - } - - ctx->size = ntoh64 (*size); - } else - ctx->size = buf->ia_blocks * 512; - - size_int = ctx->size; - } -unlock: - UNLOCK (&ctx->lock); - + int32_t ret = -1; + int32_t prev_dirty = 0; + quota_inode_ctx_t *ctx = NULL; + quota_inode_ctx_t *parent_ctx = NULL; + inode_contribution_t *contribution = NULL; + quota_meta_t delta = { + 0, + }; + quota_meta_t contri = { + 0, + }; + loc_t parent_loc = { + 0, + }; + gf_boolean_t locked = _gf_false; + gf_boolean_t dirty = _gf_false; + quota_synctask_t *args = NULL; + xlator_t *this = NULL; + loc_t *loc = NULL; + gf_boolean_t remove_xattr = _gf_true; + uint32_t nlink = 0; + + GF_ASSERT(opaque); + + args = (quota_synctask_t *)opaque; + loc = &args->loc; + contri = args->contri; + nlink = args->ia_nlink; + this = args->this; + THIS = this; + + ret = mq_inode_loc_fill(NULL, loc->parent, &parent_loc); + if (ret < 0) { + gf_log(this->name, GF_LOG_ERROR, + "parent_loc fill failed for " + "child inode %s: ", + uuid_utoa(loc->inode->gfid)); + goto out; + } + + ret = mq_lock(this, &parent_loc, F_WRLCK); + if (ret < 0) + goto out; + locked = _gf_true; + + if (contri.size >= 0) { + /* contri parameter is supplied only for rename operation. + * remove xattr is alreday performed, we need to skip + * removexattr for rename operation + */ + remove_xattr = _gf_false; + delta.size = contri.size; + delta.file_count = contri.file_count; + delta.dir_count = contri.dir_count; + } else { + remove_xattr = _gf_true; + + ret = mq_inode_ctx_get(loc->inode, this, &ctx); if (ret < 0) { - goto err; + gf_log_callingfn(this->name, GF_LOG_WARNING, + "ctx for" + " the node %s is NULL", + loc->path); + goto out; } - ret = dict_get_bin (dict, contri_key, (void **) &contri); + contribution = mq_get_contribution_node(loc->parent, ctx); + if (contribution == NULL) { + ret = -1; + gf_log(this->name, GF_LOG_DEBUG, + "contribution for the node %s is NULL", loc->path); + goto out; + } - LOCK (&contribution->lock); + LOCK(&contribution->lock); { - if (ret < 0) - contribution->contribution = 0; - else - contribution->contribution = ntoh64 (*contri); - - contri_int = contribution->contribution; + delta.size = contribution->contribution; + delta.file_count = contribution->file_count; + delta.dir_count = contribution->dir_count; } - UNLOCK (&contribution->lock); + UNLOCK(&contribution->lock); + } - gf_log (this->name, GF_LOG_DEBUG, "%s %"PRId64 "%"PRId64, - local->loc.path, size_int, contri_int); + ret = mq_get_set_dirty(this, &parent_loc, 1, &prev_dirty); + if (ret < 0) + goto out; + dirty = _gf_true; - local->delta = size_int - contri_int; - - if (local->delta == 0) { - mq_mark_undirty (frame, NULL, this, 0, 0, NULL); - return 0; - } + mq_sub_meta(&delta, NULL); - newdict = dict_new (); - if (newdict == NULL) { - ret = -1; - op_errno = ENOMEM; - goto err; - } + if (remove_xattr) { + ret = mq_remove_contri(this, loc, ctx, contribution, &delta, nlink); + if (ret < 0) + goto out; + } - QUOTA_ALLOC_OR_GOTO (delta, int64_t, ret, err); + if (quota_meta_is_null(&delta)) + goto out; - *delta = hton64 (local->delta); + ret = mq_update_size(this, &parent_loc, &delta); + if (ret < 0) + goto out; - ret = dict_set_bin (newdict, contri_key, delta, 8); - if (ret < 0) { - op_errno = -ret; - ret = -1; - goto err; +out: + if (dirty) { + if (ret < 0 || prev_dirty) { + /* On failure clear dirty status flag. + * In the next lookup inspect_directory_xattr + * can set the status flag and fix the + * dirty directory. + * Do the same if dir was dirty before + * the txn + */ + ret = mq_inode_ctx_get(parent_loc.inode, this, &parent_ctx); + if (ret == 0) + mq_set_ctx_dirty_status(parent_ctx, _gf_false); + } else { + ret = mq_mark_dirty(this, &parent_loc, 0); } + } - STACK_WIND (frame, - mq_update_parent_size, - FIRST_CHILD(this), - FIRST_CHILD(this)->fops->xattrop, - &local->loc, - GF_XATTROP_ADD_ARRAY64, - newdict); - ret = 0; + if (locked) + ret = mq_lock(this, &parent_loc, F_UNLCK); -err: - if (op_ret == -1 || ret < 0) { - local->err = op_errno; + if (ret >= 0) + ret = mq_initiate_quota_blocking_txn(this, &parent_loc, NULL); - mq_release_parent_lock (frame, NULL, this, 0, 0); - } + loc_wipe(&parent_loc); - if (newdict) - dict_unref (newdict); + if (contribution) + GF_REF_PUT(contribution); - return 0; + return ret; } int32_t -mq_fetch_child_size_and_contri (call_frame_t *frame, void *cookie, - xlator_t *this, int32_t op_ret, - int32_t op_errno) +mq_reduce_parent_size_txn(xlator_t *this, loc_t *origin_loc, + quota_meta_t *contri, uint32_t nlink, + call_stub_t *stub) { - int32_t ret = -1; - char contri_key [512] = {0, }; - dict_t *newdict = NULL; - quota_local_t *local = NULL; - marker_conf_t *priv = NULL; - quota_inode_ctx_t *ctx = NULL; - - local = frame->local; - - if (op_ret == -1) { - gf_log (this->name, (op_errno == ENOENT) ? GF_LOG_DEBUG - : GF_LOG_WARNING, - "couldn't mark inode corresponding to path (%s) dirty " - "(%s)", local->parent_loc.path, strerror (op_errno)); - goto err; - } + int32_t ret = -1; + loc_t loc = { + 0, + }; + gf_boolean_t resume_stub = _gf_true; - VALIDATE_OR_GOTO (local->ctx, err); - VALIDATE_OR_GOTO (local->contri, err); + GF_VALIDATE_OR_GOTO("marker", this, out); + GF_VALIDATE_OR_GOTO("marker", origin_loc, out); - gf_log (this->name, GF_LOG_DEBUG, "%s marked dirty", - local->parent_loc.path); + ret = mq_prevalidate_txn(this, origin_loc, &loc, NULL, NULL); + if (ret < 0) + goto out; - priv = this->private; + if (loc_is_root(&loc)) { + ret = 0; + goto out; + } - //update parent ctx - ret = mq_inode_ctx_get (local->parent_loc.inode, this, &ctx); - if (ret == -1) { - op_errno = EINVAL; - goto err; - } + resume_stub = _gf_false; + ret = mq_synctask1(this, mq_reduce_parent_size_task, _gf_true, &loc, contri, + nlink, stub); +out: + loc_wipe(&loc); - LOCK (&ctx->lock); - { - ctx->dirty = 1; - } - UNLOCK (&ctx->lock); + if (resume_stub && stub) + call_resume(stub); - newdict = dict_new (); - if (newdict == NULL) { - op_errno = ENOMEM; - goto err; - } + if (ret) + gf_log_callingfn(this ? this->name : "Marker", GF_LOG_ERROR, + "mq_reduce_parent_size_txn failed"); - if (local->loc.inode->ia_type == IA_IFDIR) { - ret = dict_set_int64 (newdict, QUOTA_SIZE_KEY, 0); - if (ret < 0) { - gf_log (this->name, GF_LOG_WARNING, - "dict_set failed."); - goto err; - } - } + return ret; +} - GET_CONTRI_KEY (contri_key, local->contri->gfid, ret); +int +mq_initiate_quota_task(void *opaque) +{ + int32_t ret = -1; + int32_t prev_dirty = 0; + loc_t child_loc = { + 0, + }; + loc_t parent_loc = { + 0, + }; + gf_boolean_t locked = _gf_false; + gf_boolean_t dirty = _gf_false; + gf_boolean_t status = _gf_false; + quota_meta_t delta = { + 0, + }; + quota_synctask_t *args = NULL; + xlator_t *this = NULL; + loc_t *loc = NULL; + inode_contribution_t *contri = NULL; + quota_inode_ctx_t *ctx = NULL; + quota_inode_ctx_t *parent_ctx = NULL; + inode_t *tmp_parent = NULL; + + GF_VALIDATE_OR_GOTO("marker", opaque, out); + + args = (quota_synctask_t *)opaque; + loc = &args->loc; + this = args->this; + + GF_VALIDATE_OR_GOTO("marker", this, out); + THIS = this; + + GF_VALIDATE_OR_GOTO(this->name, loc, out); + GF_VALIDATE_OR_GOTO(this->name, loc->inode, out); + + ret = mq_loc_copy(&child_loc, loc); + if (ret < 0) { + gf_log(this->name, GF_LOG_ERROR, "loc copy failed"); + goto out; + } + + while (!__is_root_gfid(child_loc.gfid)) { + ret = mq_inode_ctx_get(child_loc.inode, this, &ctx); if (ret < 0) { - op_errno = ENOMEM; - goto err; + gf_log(this->name, GF_LOG_WARNING, + "inode ctx get failed for %s, " + "aborting update txn", + child_loc.path); + goto out; } - ret = dict_set_int64 (newdict, contri_key, 0); - if (ret < 0) { - gf_log (this->name, GF_LOG_WARNING, - "dict_set failed."); - goto err; + /* To improve performance, abort current transaction + * if one is already in progress for same inode + */ + if (status == _gf_true) { + /* status will already set before txn start, + * so it should not be set in first + * loop iteration + */ + ret = mq_test_and_set_ctx_updation_status(ctx, &status); + if (ret < 0 || status == _gf_true) + goto out; } - mq_set_ctx_updation_status (local->ctx, _gf_false); - - STACK_WIND (frame, mq_update_inode_contribution, FIRST_CHILD(this), - FIRST_CHILD(this)->fops->lookup, &local->loc, newdict); - - ret = 0; - -err: - if (op_ret == -1 || ret < 0) { - local->err = op_errno; - - mq_set_ctx_updation_status (local->ctx, _gf_false); - - mq_release_parent_lock (frame, NULL, this, 0, 0); + if (child_loc.parent == NULL) { + ret = mq_build_ancestry(this, &child_loc); + if (ret < 0 || child_loc.parent == NULL) { + /* If application performs parallel remove + * operations on same set of files/directories + * then we may get ENOENT/ESTALE + */ + gf_log(this->name, + (-ret == ENOENT || -ret == ESTALE) ? GF_LOG_DEBUG + : GF_LOG_ERROR, + "build ancestry failed for inode %s", + uuid_utoa(child_loc.inode->gfid)); + ret = -1; + goto out; + } } - if (newdict) - dict_unref (newdict); - - return 0; -} - -int32_t -mq_markdirty (call_frame_t *frame, void *cookie, - xlator_t *this, int32_t op_ret, int32_t op_errno) -{ - int32_t ret = -1; - dict_t *dict = NULL; - quota_local_t *local = NULL; - marker_conf_t *priv = NULL; - - local = frame->local; - - if (op_ret == -1){ - gf_log (this->name, (op_errno == ENOENT) ? GF_LOG_DEBUG - : GF_LOG_WARNING, - "lock setting failed on %s (%s)", - local->parent_loc.path, strerror (op_errno)); - - local->err = op_errno; - - mq_set_ctx_updation_status (local->ctx, _gf_false); - - mq_inodelk_cbk (frame, NULL, this, 0, 0); - - return 0; + ret = mq_inode_loc_fill(NULL, child_loc.parent, &parent_loc); + if (ret < 0) { + gf_log(this->name, GF_LOG_ERROR, + "parent_loc fill " + "failed for child inode %s: ", + uuid_utoa(child_loc.inode->gfid)); + goto out; } - gf_log (this->name, GF_LOG_TRACE, - "inodelk succeeded on %s", local->parent_loc.path); + ret = mq_lock(this, &parent_loc, F_WRLCK); + if (ret < 0) + goto out; + locked = _gf_true; - priv = this->private; + mq_set_ctx_updation_status(ctx, _gf_false); + status = _gf_true; - dict = dict_new (); - if (!dict) { + /* Contribution node can be NULL in below scenarios and + create if needed: + + Scenario 1) + In this case create a new contribution node + Suppose hard link for a file f1 present in a directory d1 is + created in the directory d2 (as f2). Now, since d2's + contribution is not there in f1's inode ctx, d2's + contribution xattr won't be created and will create problems + for quota operations. + + Don't create contribution if parent has been changed after + taking a lock, this can happen when rename is performed + and writes is still in-progress for the same file + + Scenario 2) + When a rename operation is performed, contribution node + for olp path will be removed. + + Create contribution node only if oldparent is same as + newparent. + Consider below example + 1) rename FOP invoked on file 'x' + 2) write is still in progress for file 'x' + 3) rename takes a lock on old-parent + 4) write-update txn blocked on old-parent to acquire lock + 5) in rename_cbk, contri xattrs are removed and contribution + is deleted and lock is released + 6) now write-update txn gets the lock and updates the + wrong parent as it was holding lock on old parent + so validate parent once the lock is acquired + + For more information on this problem, please see + doc for marker_rename in file marker.c + */ + contri = mq_get_contribution_node(child_loc.parent, ctx); + if (contri == NULL) { + tmp_parent = inode_parent(child_loc.inode, 0, NULL); + if (tmp_parent == NULL) { + /* This can happen if application performs + * parallel remove operations on same set + * of files/directories + */ + gf_log(this->name, GF_LOG_WARNING, + "parent is " + "NULL for inode %s", + uuid_utoa(child_loc.inode->gfid)); + ret = -1; + goto out; + } + if (gf_uuid_compare(tmp_parent->gfid, parent_loc.gfid)) { + /* abort txn if parent has changed */ + ret = 0; + goto out; + } + + inode_unref(tmp_parent); + tmp_parent = NULL; + + contri = mq_add_new_contribution_node(this, ctx, &child_loc); + if (contri == NULL) { + gf_log(this->name, GF_LOG_ERROR, + "Failed to " + "create contribution node for %s, " + "abort update txn", + child_loc.path); ret = -1; - goto err; + goto out; + } } - ret = dict_set_int8 (dict, QUOTA_DIRTY_KEY, 1); - if (ret == -1) - goto err; + ret = mq_get_delta(this, &child_loc, &delta, ctx, contri); + if (ret < 0) + goto out; - STACK_WIND (frame, mq_fetch_child_size_and_contri, - FIRST_CHILD(this), - FIRST_CHILD(this)->fops->setxattr, - &local->parent_loc, dict, 0); + if (quota_meta_is_null(&delta)) + goto out; - ret = 0; -err: - if (ret == -1) { - local->err = 1; + ret = mq_get_set_dirty(this, &parent_loc, 1, &prev_dirty); + if (ret < 0) + goto out; + dirty = _gf_true; - mq_set_ctx_updation_status (local->ctx, _gf_false); + ret = mq_update_contri(this, &child_loc, contri, &delta); + if (ret < 0) + goto out; - mq_release_parent_lock (frame, NULL, this, 0, 0); + ret = mq_update_size(this, &parent_loc, &delta); + if (ret < 0) { + gf_log(this->name, GF_LOG_DEBUG, + "rollback " + "contri updation"); + mq_sub_meta(&delta, NULL); + mq_update_contri(this, &child_loc, contri, &delta); + goto out; } - if (dict) - dict_unref (dict); + if (prev_dirty == 0) { + ret = mq_mark_dirty(this, &parent_loc, 0); + } else { + ret = mq_inode_ctx_get(parent_loc.inode, this, &parent_ctx); + if (ret == 0) + mq_set_ctx_dirty_status(parent_ctx, _gf_false); + } + dirty = _gf_false; + prev_dirty = 0; - return 0; -} + ret = mq_lock(this, &parent_loc, F_UNLCK); + locked = _gf_false; + if (__is_root_gfid(parent_loc.gfid)) + break; -int32_t -mq_get_lock_on_parent (call_frame_t *frame, xlator_t *this) -{ - struct gf_flock lock = {0, }; - quota_local_t *local = NULL; + /* Repeate above steps upwards till the root */ + loc_wipe(&child_loc); + ret = mq_loc_copy(&child_loc, &parent_loc); + if (ret < 0) + goto out; - GF_VALIDATE_OR_GOTO ("marker", frame, fr_destroy); + loc_wipe(&parent_loc); + GF_REF_PUT(contri); + contri = NULL; + } - local = frame->local; - gf_log (this->name, GF_LOG_DEBUG, "taking lock on %s", - local->parent_loc.path); +out: + if ((dirty) && (ret < 0)) { + /* On failure clear dirty status flag. + * In the next lookup inspect_directory_xattr + * can set the status flag and fix the + * dirty directory. + * Do the same if the dir was dirty before + * txn + */ + ret = mq_inode_ctx_get(parent_loc.inode, this, &parent_ctx); + if (ret == 0) + mq_set_ctx_dirty_status(parent_ctx, _gf_false); + } - if (local->parent_loc.inode == NULL) { - gf_log (this->name, GF_LOG_DEBUG, - "parent inode is not valid, aborting " - "transaction."); - goto fr_destroy; - } + if (locked) + ret = mq_lock(this, &parent_loc, F_UNLCK); - lock.l_len = 0; - lock.l_start = 0; - lock.l_type = F_WRLCK; - lock.l_whence = SEEK_SET; + if (ctx && status == _gf_false) + mq_set_ctx_updation_status(ctx, _gf_false); - STACK_WIND (frame, - mq_markdirty, - FIRST_CHILD(this), - FIRST_CHILD(this)->fops->inodelk, - this->name, &local->parent_loc, F_SETLKW, &lock); + loc_wipe(&child_loc); + loc_wipe(&parent_loc); - return 0; + if (tmp_parent) + inode_unref(tmp_parent); -fr_destroy: - QUOTA_STACK_DESTROY (frame, this); + if (contri) + GF_REF_PUT(contri); - return -1; + return 0; } - int -mq_start_quota_txn (xlator_t *this, loc_t *loc, - quota_inode_ctx_t *ctx, - inode_contribution_t *contri) +_mq_initiate_quota_txn(xlator_t *this, loc_t *origin_loc, struct iatt *buf, + gf_boolean_t spawn) { - int32_t ret = -1; - call_frame_t *frame = NULL; - quota_local_t *local = NULL; - - frame = create_frame (this, this->ctx->pool); - if (frame == NULL) - goto err; - - mq_assign_lk_owner (this, frame); - - local = mq_local_new (); - if (local == NULL) - goto fr_destroy; - - frame->local = local; - - ret = mq_loc_copy (&local->loc, loc); - if (ret < 0) - goto fr_destroy; - - ret = mq_inode_loc_fill (NULL, local->loc.parent, - &local->parent_loc); - if (ret < 0) - goto fr_destroy; - - local->ctx = ctx; - local->contri = contri; + int32_t ret = -1; + quota_inode_ctx_t *ctx = NULL; + gf_boolean_t status = _gf_true; + loc_t loc = { + 0, + }; + + ret = mq_prevalidate_txn(this, origin_loc, &loc, &ctx, buf); + if (ret < 0) + goto out; + + if (loc_is_root(&loc)) { + ret = 0; + goto out; + } - ret = mq_get_lock_on_parent (frame, this); - if (ret == -1) - goto err; + ret = mq_test_and_set_ctx_updation_status(ctx, &status); + if (ret < 0 || status == _gf_true) + goto out; - return 0; + ret = mq_synctask(this, mq_initiate_quota_task, spawn, &loc); -fr_destroy: - QUOTA_STACK_DESTROY (frame, this); -err: - mq_set_ctx_updation_status (ctx, _gf_false); +out: + if (ret < 0 && status == _gf_false) + mq_set_ctx_updation_status(ctx, _gf_false); - return -1; + loc_wipe(&loc); + return ret; } - int -mq_initiate_quota_txn (xlator_t *this, loc_t *loc) +mq_initiate_quota_txn(xlator_t *this, loc_t *loc, struct iatt *buf) { - int32_t ret = -1; - gf_boolean_t status = _gf_false; - quota_inode_ctx_t *ctx = NULL; - inode_contribution_t *contribution = NULL; - - GF_VALIDATE_OR_GOTO ("marker", this, out); - GF_VALIDATE_OR_GOTO ("marker", loc, out); - GF_VALIDATE_OR_GOTO ("marker", loc->inode, out); - - ret = mq_inode_ctx_get (loc->inode, this, &ctx); - if (ret == -1) { - gf_log (this->name, GF_LOG_WARNING, - "inode ctx get failed, aborting quota txn"); - ret = -1; - goto out; - } + int32_t ret = -1; - contribution = mq_get_contribution_node (loc->parent, ctx); - if (contribution == NULL) - goto out; - - /* To improve performance, donot start another transaction - * if one is already in progress for same inode - */ - status = _gf_true; - - ret = mq_test_and_set_ctx_updation_status (ctx, &status); - if (ret < 0) - goto out; - - if (status == _gf_false) { - mq_start_quota_txn (this, loc, ctx, contribution); - } + GF_VALIDATE_OR_GOTO("marker", this, out); + GF_VALIDATE_OR_GOTO("marker", loc, out); + GF_VALIDATE_OR_GOTO("marker", loc->inode, out); - ret = 0; + ret = _mq_initiate_quota_txn(this, loc, buf, _gf_true); out: - return ret; + return ret; } - -/* int32_t */ -/* validate_inode_size_contribution (xlator_t *this, loc_t *loc, int64_t size, */ -/* int64_t contribution) */ -/* { */ -/* if (size != contribution) { */ -/* mq_initiate_quota_txn (this, loc); */ -/* } */ - -/* return 0; */ -/* } */ - - -int32_t -mq_inspect_directory_xattr (xlator_t *this, - loc_t *loc, - dict_t *dict, - struct iatt buf) +int +mq_initiate_quota_blocking_txn(xlator_t *this, loc_t *loc, struct iatt *buf) { - int32_t ret = 0; - int8_t dirty = -1; - int64_t *size = NULL, size_int = 0; - int64_t *contri = NULL, contri_int = 0; - char contri_key [512] = {0, }; - marker_conf_t *priv = NULL; - gf_boolean_t not_root = _gf_false; - quota_inode_ctx_t *ctx = NULL; - inode_contribution_t *contribution = NULL; - - priv = this->private; - - ret = mq_inode_ctx_get (loc->inode, this, &ctx); - if (ret < 0) { - ctx = mq_inode_ctx_new (loc->inode, this); - if (ctx == NULL) { - gf_log (this->name, GF_LOG_WARNING, - "mq_inode_ctx_new failed"); - ret = -1; - goto out; - } - } - - ret = dict_get_bin (dict, QUOTA_SIZE_KEY, (void **) &size); - if (ret < 0) - goto out; - - ret = dict_get_int8 (dict, QUOTA_DIRTY_KEY, &dirty); - if (ret < 0) - goto out; - - if (strcmp (loc->path, "/") != 0) { - not_root = _gf_true; - - contribution = mq_add_new_contribution_node (this, ctx, loc); - if (contribution == NULL) { - gf_log (this->name, GF_LOG_DEBUG, - "cannot add a new contributio node"); - goto out; - } - - GET_CONTRI_KEY (contri_key, contribution->gfid, ret); - if (ret < 0) - goto out; - - ret = dict_get_bin (dict, contri_key, (void **) &contri); - if (ret < 0) - goto out; - - LOCK (&contribution->lock); - { - contribution->contribution = ntoh64 (*contri); - contri_int = contribution->contribution; - } - UNLOCK (&contribution->lock); - } - - LOCK (&ctx->lock); - { - ctx->size = ntoh64 (*size); - ctx->dirty = dirty; - size_int = ctx->size; - } - UNLOCK (&ctx->lock); + int32_t ret = -1; - gf_log (this->name, GF_LOG_DEBUG, "size=%"PRId64 - " contri=%"PRId64, size_int, contri_int); + GF_VALIDATE_OR_GOTO("marker", this, out); + GF_VALIDATE_OR_GOTO("marker", loc, out); + GF_VALIDATE_OR_GOTO("marker", loc->inode, out); - if (dirty) { - ret = mq_update_dirty_inode (this, loc, ctx, contribution); - } - - if ((!dirty || ret == 0) && (not_root == _gf_true) && - (size_int != contri_int)) { - mq_initiate_quota_txn (this, loc); - } - - ret = 0; + ret = _mq_initiate_quota_txn(this, loc, buf, _gf_false); out: - if (ret) - mq_set_inode_xattr (this, loc); - - return 0; + return ret; } -int32_t -mq_inspect_file_xattr (xlator_t *this, - loc_t *loc, - dict_t *dict, - struct iatt buf) +int +mq_update_dirty_inode_task(void *opaque) { - int32_t ret = -1; - uint64_t contri_int = 0, size = 0; - int64_t *contri_ptr = NULL; - char contri_key [512] = {0, }; - marker_conf_t *priv = NULL; - quota_inode_ctx_t *ctx = NULL; - inode_contribution_t *contribution = NULL; - - priv = this->private; - - ret = mq_inode_ctx_get (loc->inode, this, &ctx); + int32_t ret = -1; + fd_t *fd = NULL; + off_t offset = 0; + gf_dirent_t entries; + gf_dirent_t *entry = NULL; + gf_boolean_t locked = _gf_false; + gf_boolean_t updated = _gf_false; + int32_t dirty = 0; + quota_meta_t contri = { + 0, + }; + quota_meta_t size = { + 0, + }; + quota_meta_t contri_sum = { + 0, + }; + quota_meta_t delta = { + 0, + }; + quota_synctask_t *args = NULL; + xlator_t *this = NULL; + loc_t *loc = NULL; + quota_inode_ctx_t *ctx = NULL; + dict_t *xdata = NULL; + char contri_key[QUOTA_KEY_MAX] = { + 0, + }; + int keylen = 0; + + GF_ASSERT(opaque); + + args = (quota_synctask_t *)opaque; + loc = &args->loc; + this = args->this; + THIS = this; + INIT_LIST_HEAD(&entries.list); + + ret = mq_inode_ctx_get(loc->inode, this, &ctx); + if (ret < 0) + goto out; + + GET_CONTRI_KEY(this, contri_key, loc->gfid, keylen); + if (keylen < 0) { + ret = keylen; + goto out; + } + + xdata = dict_new(); + if (xdata == NULL) { + gf_log(this->name, GF_LOG_ERROR, "dict_new failed"); + ret = -1; + goto out; + } + + ret = dict_set_int64(xdata, contri_key, 0); + if (ret < 0) { + gf_log(this->name, GF_LOG_ERROR, "dict_set failed"); + goto out; + } + + ret = mq_lock(this, loc, F_WRLCK); + if (ret < 0) + goto out; + locked = _gf_true; + + ret = mq_get_dirty(this, loc, &dirty); + if (ret < 0 || dirty == 0) { + ret = 0; + goto out; + } + + fd = fd_create(loc->inode, 0); + if (!fd) { + gf_log(this->name, GF_LOG_ERROR, "Failed to create fd"); + ret = -1; + goto out; + } + + ret = syncop_opendir(this, loc, fd, NULL, NULL); + if (ret < 0) { + gf_log(this->name, + (-ret == ENOENT || -ret == ESTALE) ? GF_LOG_DEBUG : GF_LOG_ERROR, + "opendir failed " + "for %s: %s", + loc->path, strerror(-ret)); + goto out; + } + + fd_bind(fd); + while ((ret = syncop_readdirp(this, fd, 131072, offset, &entries, xdata, + NULL)) != 0) { if (ret < 0) { - ctx = mq_inode_ctx_new (loc->inode, this); - if (ctx == NULL) { - gf_log (this->name, GF_LOG_WARNING, - "mq_inode_ctx_new failed"); - ret = -1; - goto out; - } + gf_log(this->name, + (-ret == ENOENT || -ret == ESTALE) ? GF_LOG_DEBUG + : GF_LOG_ERROR, + "readdirp failed " + "for %s: %s", + loc->path, strerror(-ret)); + goto out; } - contribution = mq_add_new_contribution_node (this, ctx, loc); - if (contribution == NULL) - goto out; + if (list_empty(&entries.list)) + break; - LOCK (&ctx->lock); + list_for_each_entry(entry, &entries.list, list) { - ctx->size = 512 * buf.ia_blocks; - size = ctx->size; - } - UNLOCK (&ctx->lock); - - list_for_each_entry (contribution, &ctx->contribution_head, - contri_list) { - GET_CONTRI_KEY (contri_key, contribution->gfid, ret); - if (ret < 0) - continue; - - ret = dict_get_bin (dict, contri_key, (void **) &contri_int); - if (ret == 0) { - contri_ptr = (int64_t *)(unsigned long)contri_int; - - LOCK (&contribution->lock); - { - contribution->contribution = ntoh64 (*contri_ptr); - contri_int = contribution->contribution; - } - UNLOCK (&contribution->lock); - - gf_log (this->name, GF_LOG_DEBUG, - "size=%"PRId64 " contri=%"PRId64, size, contri_int); - - if (size != contri_int) { - mq_initiate_quota_txn (this, loc); - } - } else - mq_initiate_quota_txn (this, loc); - } - -out: - return ret; -} - -int32_t -mq_xattr_state (xlator_t *this, - loc_t *loc, - dict_t *dict, - struct iatt buf) -{ - if (buf.ia_type == IA_IFREG || - buf.ia_type == IA_IFLNK) { - k ++; - mq_inspect_file_xattr (this, loc, dict, buf); - } else if (buf.ia_type == IA_IFDIR) - mq_inspect_directory_xattr (this, loc, dict, buf); - - return 0; -} - -int32_t -mq_req_xattr (xlator_t *this, - loc_t *loc, - dict_t *dict) -{ - int32_t ret = -1; - marker_conf_t *priv = NULL; - - GF_VALIDATE_OR_GOTO ("marker", this, out); - GF_VALIDATE_OR_GOTO ("marker", loc, out); - GF_VALIDATE_OR_GOTO ("marker", dict, out); + offset = entry->d_off; - priv = this->private; + if (!strcmp(entry->d_name, ".") || !strcmp(entry->d_name, "..")) + continue; - //if not "/" then request contribution - if (strcmp (loc->path, "/") == 0) - goto set_size; + memset(&contri, 0, sizeof(contri)); + quota_dict_get_meta(entry->dict, contri_key, keylen, &contri); + if (quota_meta_is_null(&contri)) + continue; - ret = mq_dict_set_contribution (this, dict, loc); - if (ret == -1) - goto out; - -set_size: - ret = dict_set_uint64 (dict, QUOTA_SIZE_KEY, 0); - if (ret < 0) { - ret = -1; - goto out; + mq_add_meta(&contri_sum, &contri); } - ret = dict_set_int8 (dict, QUOTA_DIRTY_KEY, 0); - if (ret < 0) { - ret = -1; - goto out; - } + gf_dirent_free(&entries); + } + /* Inculde for self */ + contri_sum.dir_count++; - ret = 0; + ret = _mq_get_metadata(this, loc, NULL, &size, 0); + if (ret < 0) + goto out; -out: - return ret; -} + mq_compute_delta(&delta, &contri_sum, &size); + if (quota_meta_is_null(&delta)) + goto out; -int32_t -mq_removexattr_cbk (call_frame_t *frame, void *cookie, xlator_t *this, - int32_t op_ret, int32_t op_errno) -{ - QUOTA_STACK_DESTROY (frame, this); + gf_log(this->name, GF_LOG_INFO, + "calculated size = %" PRId64 ", original size = %" PRIu64 + ", diff = %" PRIu64 ", path = %s ", + contri_sum.size, size.size, delta.size, loc->path); - return 0; -} + gf_log(this->name, GF_LOG_INFO, + "calculated f_count = %" PRId64 ", original f_count = %" PRIu64 + ", diff = %" PRIu64 ", path = %s ", + contri_sum.file_count, size.file_count, delta.file_count, loc->path); -int32_t -_mq_inode_remove_done (call_frame_t *frame, void *cookie, xlator_t *this, - int32_t op_ret, int32_t op_errno) -{ - int32_t ret = 0; - char contri_key [512] = {0, }; - quota_local_t *local = NULL; - - local = (quota_local_t *) frame->local; - - if (op_ret == -1 || local->err == -1) { - mq_removexattr_cbk (frame, NULL, this, -1, 0); - return 0; - } + gf_log(this->name, GF_LOG_INFO, + "calculated d_count = %" PRId64 ", original d_count = %" PRIu64 + ", diff = %" PRIu64 ", path = %s ", + contri_sum.dir_count, size.dir_count, delta.dir_count, loc->path); - frame->local = NULL; - - if (local->hl_count > 1) { - GET_CONTRI_KEY (contri_key, local->contri->gfid, ret); - - STACK_WIND (frame, mq_removexattr_cbk, FIRST_CHILD(this), - FIRST_CHILD(this)->fops->removexattr, - &local->loc, contri_key); - ret = 0; - } else { - mq_removexattr_cbk (frame, NULL, this, 0, 0); - } + ret = mq_update_size(this, loc, &delta); + if (ret < 0) + goto out; - if (strcmp (local->parent_loc.path, "/") != 0) { - ret = mq_get_parent_inode_local (this, local); - if (ret < 0) - goto out; + updated = _gf_true; - mq_start_quota_txn (this, &local->loc, local->ctx, local->contri); - } out: - mq_local_unref (this, local); + gf_dirent_free(&entries); - return 0; -} + if (fd) + fd_unref(fd); -int32_t -mq_inode_remove_done (call_frame_t *frame, void *cookie, xlator_t *this, - int32_t op_ret, int32_t op_errno, dict_t *dict) -{ - int32_t ret = -1; - struct gf_flock lock = {0, }; - quota_inode_ctx_t *ctx = NULL; - quota_local_t *local = NULL; - int64_t contribution = 0; + if (xdata) + dict_unref(xdata); - local = frame->local; - if (op_ret == -1) - local->err = -1; + if (ret < 0) { + /* On failure clear dirty status flag. + * In the next lookup inspect_directory_xattr + * can set the status flag and fix the + * dirty directory + */ + if (ctx) + mq_set_ctx_dirty_status(ctx, _gf_false); + } else if (dirty) { + mq_mark_dirty(this, loc, 0); + } - ret = mq_inode_ctx_get (local->parent_loc.inode, this, &ctx); + if (locked) + mq_lock(this, loc, F_UNLCK); - LOCK (&local->contri->lock); - { - contribution = local->contri->contribution; - } - UNLOCK (&local->contri->lock); - - if (contribution == local->size) { - if (ret == 0) { - LOCK (&ctx->lock); - { - ctx->size -= contribution; - } - UNLOCK (&ctx->lock); - - LOCK (&local->contri->lock); - { - local->contri->contribution = 0; - } - UNLOCK (&local->contri->lock); - } - } + if (updated) + mq_initiate_quota_blocking_txn(this, loc, NULL); - lock.l_type = F_UNLCK; - lock.l_whence = SEEK_SET; - lock.l_start = 0; - lock.l_len = 0; - lock.l_pid = 0; - - STACK_WIND (frame, - _mq_inode_remove_done, - FIRST_CHILD(this), - FIRST_CHILD(this)->fops->inodelk, - this->name, &local->parent_loc, - F_SETLKW, &lock); - return 0; + return ret; } int32_t -mq_reduce_parent_size_xattr (call_frame_t *frame, void *cookie, - xlator_t *this, int32_t op_ret, int32_t op_errno) +mq_update_dirty_inode_txn(xlator_t *this, loc_t *loc, quota_inode_ctx_t *ctx) { - int32_t ret = -1; - int64_t *size = NULL; - dict_t *dict = NULL; - marker_conf_t *priv = NULL; - quota_local_t *local = NULL; - inode_contribution_t *contribution = NULL; - - local = frame->local; - if (op_ret == -1) { - gf_log (this->name, GF_LOG_WARNING, - "inodelk set failed on %s", local->parent_loc.path); - QUOTA_STACK_DESTROY (frame, this); - return 0; - } - - VALIDATE_OR_GOTO (local->contri, err); + int32_t ret = -1; + gf_boolean_t status = _gf_true; - priv = this->private; + GF_VALIDATE_OR_GOTO("marker", loc, out); + GF_VALIDATE_OR_GOTO("marker", loc->inode, out); - contribution = local->contri; + mq_test_and_set_ctx_status(ctx, &ctx->dirty_status, &status); + if (status == _gf_true) + goto out; - dict = dict_new (); - if (dict == NULL) { - ret = -1; - goto err; - } - - QUOTA_ALLOC_OR_GOTO (size, int64_t, ret, err); - - *size = hton64 (-local->size); - - ret = dict_set_bin (dict, QUOTA_SIZE_KEY, size, 8); - if (ret < 0) - goto err; - - - STACK_WIND (frame, mq_inode_remove_done, FIRST_CHILD(this), - FIRST_CHILD(this)->fops->xattrop, &local->parent_loc, - GF_XATTROP_ADD_ARRAY64, dict); - dict_unref (dict); - return 0; + ret = mq_synctask(this, mq_update_dirty_inode_task, _gf_true, loc); +out: + if (ret < 0 && status == _gf_false) + mq_set_ctx_dirty_status(ctx, _gf_false); -err: - local->err = 1; - mq_inode_remove_done (frame, NULL, this, -1, 0, NULL); - if (dict) - dict_unref (dict); - return 0; + return ret; } int32_t -mq_reduce_parent_size (xlator_t *this, loc_t *loc, int64_t contri) +mq_inspect_directory_xattr(xlator_t *this, quota_inode_ctx_t *ctx, + inode_contribution_t *contribution, loc_t *loc, + dict_t *dict) { - int32_t ret = -1; - struct gf_flock lock = {0,}; - call_frame_t *frame = NULL; - marker_conf_t *priv = NULL; - quota_local_t *local = NULL; - quota_inode_ctx_t *ctx = NULL; - inode_contribution_t *contribution = NULL; - - GF_VALIDATE_OR_GOTO ("marker", this, out); - GF_VALIDATE_OR_GOTO ("marker", loc, out); - - priv = this->private; - - ret = mq_inode_ctx_get (loc->inode, this, &ctx); + int32_t ret = -1; + int8_t dirty = -1; + quota_meta_t size = { + 0, + }; + quota_meta_t contri = { + 0, + }; + quota_meta_t delta = { + 0, + }; + char contri_key[QUOTA_KEY_MAX] = { + 0, + }; + char size_key[QUOTA_KEY_MAX] = { + 0, + }; + int keylen = 0; + gf_boolean_t status = _gf_false; + + ret = dict_get_int8(dict, QUOTA_DIRTY_KEY, &dirty); + if (ret < 0) { + /* dirty is set only on the first file write operation + * so ignore this error + */ + ret = 0; + dirty = 0; + } + + GET_SIZE_KEY(this, size_key, keylen); + if (keylen < 0) { + ret = -1; + goto out; + } + ret = _quota_dict_get_meta(this, dict, size_key, keylen, &size, IA_IFDIR, + _gf_false); + if (ret < 0) + goto create_xattr; + + if (!contribution) + goto create_xattr; + + if (!loc_is_root(loc)) { + GET_CONTRI_KEY(this, contri_key, contribution->gfid, keylen); + if (keylen < 0) { + ret = -1; + goto out; + } + ret = _quota_dict_get_meta(this, dict, contri_key, keylen, &contri, + IA_IFDIR, _gf_false); if (ret < 0) - goto out; + goto create_xattr; - contribution = mq_get_contribution_node (loc->parent, ctx); - if (contribution == NULL) - goto out; + LOCK(&contribution->lock); + { + contribution->contribution = contri.size; + contribution->file_count = contri.file_count; + contribution->dir_count = contri.dir_count; + } + UNLOCK(&contribution->lock); + } + + LOCK(&ctx->lock); + { + ctx->size = size.size; + ctx->file_count = size.file_count; + ctx->dir_count = size.dir_count; + ctx->dirty = dirty; + } + UNLOCK(&ctx->lock); + + ret = mq_get_ctx_updation_status(ctx, &status); + if (ret < 0 || status == _gf_true) { + /* If the update txn is in progress abort inspection */ + ret = 0; + goto out; + } - local = mq_local_new (); - if (local == NULL) { - ret = -1; - goto out; - } + mq_compute_delta(&delta, &size, &contri); - if (contri >= 0) { - local->size = contri; - } else { - LOCK (&contribution->lock); - { - local->size = contribution->contribution; - } - UNLOCK (&contribution->lock); - } + if (dirty) { + ret = mq_update_dirty_inode_txn(this, loc, ctx); + goto out; + } - if (local->size == 0) { - ret = 0; - goto out; - } + if (!loc_is_root(loc) && !quota_meta_is_null(&delta)) + mq_initiate_quota_txn(this, loc, NULL); - ret = mq_loc_copy (&local->loc, loc); - if (ret < 0) - goto out; + ret = 0; + goto out; - local->ctx = ctx; - local->contri = contribution; +create_xattr: + if (ret < 0) + ret = mq_create_xattrs_txn(this, loc, NULL); - ret = mq_inode_loc_fill (NULL, loc->parent, &local->parent_loc); - if (ret < 0) - goto out; +out: + return ret; +} - frame = create_frame (this, this->ctx->pool); - if (!frame) { - ret = -1; - goto out; +int32_t +mq_inspect_file_xattr(xlator_t *this, quota_inode_ctx_t *ctx, + inode_contribution_t *contribution, loc_t *loc, + dict_t *dict, struct iatt *buf) +{ + int32_t ret = -1; + quota_meta_t size = { + 0, + }; + quota_meta_t contri = { + 0, + }; + quota_meta_t delta = { + 0, + }; + char contri_key[QUOTA_KEY_MAX] = { + 0, + }; + int keylen = 0; + gf_boolean_t status = _gf_false; + + if (!buf || !contribution || !ctx) + goto out; + + LOCK(&ctx->lock); + { + ctx->size = 512 * buf->ia_blocks; + ctx->file_count = 1; + ctx->dir_count = 0; + + size.size = ctx->size; + size.file_count = ctx->file_count; + size.dir_count = ctx->dir_count; + } + UNLOCK(&ctx->lock); + + GET_CONTRI_KEY(this, contri_key, contribution->gfid, keylen); + if (keylen < 0) { + ret = -1; + goto out; + } + + ret = _quota_dict_get_meta(this, dict, contri_key, keylen, &contri, + IA_IFREG, _gf_true); + if (ret < 0) { + ret = mq_create_xattrs_txn(this, loc, NULL); + } else { + LOCK(&contribution->lock); + { + contribution->contribution = contri.size; + contribution->file_count = contri.file_count; + contribution->dir_count = contri.dir_count; } + UNLOCK(&contribution->lock); - mq_assign_lk_owner (this, frame); - - frame->local = local; - - lock.l_len = 0; - lock.l_start = 0; - lock.l_type = F_WRLCK; - lock.l_whence = SEEK_SET; - - if (local->parent_loc.inode == NULL) { - ret = -1; - gf_log (this->name, GF_LOG_WARNING, - "Inode is NULL, so can't stackwind."); - goto out; + ret = mq_get_ctx_updation_status(ctx, &status); + if (ret < 0 || status == _gf_true) { + /* If the update txn is in progress abort inspection */ + ret = 0; + goto out; } - STACK_WIND (frame, - mq_reduce_parent_size_xattr, - FIRST_CHILD(this), - FIRST_CHILD(this)->fops->inodelk, - this->name, &local->parent_loc, F_SETLKW, &lock); - local = NULL; - ret = 0; + mq_compute_delta(&delta, &size, &contri); + if (!quota_meta_is_null(&delta)) + mq_initiate_quota_txn(this, loc, NULL); + } + /* TODO: revist this code when fixing hardlinks */ out: - if (local != NULL) - mq_local_unref (this, local); - - return ret; + return ret; } - int32_t -init_quota_priv (xlator_t *this) +mq_xattr_state(xlator_t *this, loc_t *origin_loc, dict_t *dict, + struct iatt *buf) { - return 0; -} + int32_t ret = -1; + quota_inode_ctx_t *ctx = NULL; + loc_t loc = { + 0, + }; + inode_contribution_t *contribution = NULL; + + ret = mq_prevalidate_txn(this, origin_loc, &loc, &ctx, buf); + if (ret < 0 || loc.parent == NULL) + goto out; + + if (!loc_is_root(&loc)) { + contribution = mq_add_new_contribution_node(this, ctx, &loc); + if (contribution == NULL) { + if (!gf_uuid_is_null(loc.inode->gfid)) + gf_log(this->name, GF_LOG_WARNING, + "cannot add a new contribution node " + "(%s)", + uuid_utoa(loc.gfid)); + ret = -1; + goto out; + } + if (buf->ia_type == IA_IFDIR) + mq_inspect_directory_xattr(this, ctx, contribution, &loc, dict); + else + mq_inspect_file_xattr(this, ctx, contribution, &loc, dict, buf); + } else { + mq_inspect_directory_xattr(this, ctx, 0, &loc, dict); + } + +out: + loc_wipe(&loc); + + if (contribution) + GF_REF_PUT(contribution); + return ret; +} int32_t -mq_rename_update_newpath (xlator_t *this, loc_t *loc) +mq_req_xattr(xlator_t *this, loc_t *loc, dict_t *dict, char *contri_key, + char *size_key) { - int32_t ret = -1; - quota_inode_ctx_t *ctx = NULL; - inode_contribution_t *contribution = NULL; + int32_t ret = -1; + char key[QUOTA_KEY_MAX] = { + 0, + }; - GF_VALIDATE_OR_GOTO ("marker", this, out); - GF_VALIDATE_OR_GOTO ("marker", loc, out); - GF_VALIDATE_OR_GOTO ("marker", loc->inode, out); + GF_VALIDATE_OR_GOTO("marker", this, out); + GF_VALIDATE_OR_GOTO("marker", loc, out); + GF_VALIDATE_OR_GOTO("marker", dict, out); - ret = mq_inode_ctx_get (loc->inode, this, &ctx); + if (!loc_is_root(loc)) { + ret = mq_dict_set_contribution(this, dict, loc, NULL, contri_key); if (ret < 0) - goto out; + goto out; + } - contribution = mq_add_new_contribution_node (this, ctx, loc); - if (contribution == NULL) { - ret = -1; - goto out; + GET_SIZE_KEY(this, key, ret); + if (ret < 0) + goto out; + if (size_key) + if (snprintf(size_key, QUOTA_KEY_MAX, "%s", key) >= QUOTA_KEY_MAX) { + ret = -1; + goto out; } - mq_initiate_quota_txn (this, loc); + ret = dict_set_uint64(dict, key, 0); + if (ret < 0) + goto out; + + ret = dict_set_int8(dict, QUOTA_DIRTY_KEY, 0); + out: - return ret; + if (ret < 0) + gf_log_callingfn(this ? this->name : "Marker", GF_LOG_ERROR, + "dict set failed"); + return ret; } int32_t -mq_forget (xlator_t *this, quota_inode_ctx_t *ctx) +mq_forget(xlator_t *this, quota_inode_ctx_t *ctx) { - inode_contribution_t *contri = NULL; - inode_contribution_t *next = NULL; + inode_contribution_t *contri = NULL; + inode_contribution_t *next = NULL; - GF_VALIDATE_OR_GOTO ("marker", this, out); - GF_VALIDATE_OR_GOTO ("marker", ctx, out); + GF_VALIDATE_OR_GOTO("marker", this, out); + GF_VALIDATE_OR_GOTO("marker", ctx, out); - list_for_each_entry_safe (contri, next, &ctx->contribution_head, - contri_list) { - list_del (&contri->contri_list); - GF_FREE (contri); - } + list_for_each_entry_safe(contri, next, &ctx->contribution_head, contri_list) + { + list_del_init(&contri->contri_list); + GF_REF_PUT(contri); + } - LOCK_DESTROY (&ctx->lock); - GF_FREE (ctx); + LOCK_DESTROY(&ctx->lock); + GF_FREE(ctx); out: - return 0; + return 0; } |