* GPL HEADER END
*/
/*
- * Copyright 2008 Sun Microsystems, Inc. All rights reserved
+ * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
* Use is subject to license terms.
*/
/*
#define DEBUG_SUBSYSTEM S_MDS
#include <linux/module.h>
+#ifdef HAVE_EXT4_LDISKFS
+#include <ldiskfs/ldiskfs_jbd2.h>
+#else
#include <linux/jbd.h>
+#endif
#include <obd.h>
#include <obd_class.h>
#include <lustre_ver.h>
#include <obd_support.h>
#include <lprocfs_status.h>
-
+#ifdef HAVE_EXT4_LDISKFS
+#include <ldiskfs/ldiskfs.h>
+#else
#include <linux/ldiskfs_fs.h>
+#endif
#include <lustre_mds.h>
#include <lustre/lustre_idl.h>
#include <lustre_fid.h>
struct mdd_object *mdd_obj,
const struct lu_fid *pfid,
const struct lu_name *lname,
- struct thandle *handle);
+ struct thandle *handle, int first);
static int mdd_links_rename(const struct lu_env *env,
struct mdd_object *mdd_obj,
const struct lu_fid *oldpfid,
*
* returns < 0: if error
*/
-static int mdd_is_subdir(const struct lu_env *env,
- struct md_object *mo, const struct lu_fid *fid,
- struct lu_fid *sfid)
+int mdd_is_subdir(const struct lu_env *env, struct md_object *mo,
+ const struct lu_fid *fid, struct lu_fid *sfid)
{
struct mdd_device *mdd = mdo2mdd(mo);
int rc;
RETURN(-ENOTDIR);
iops = &obj->do_index_ops->dio_it;
- it = iops->init(env, obj, BYPASS_CAPA);
- if (it != NULL) {
+ it = iops->init(env, obj, LUDA_64BITHASH, BYPASS_CAPA);
+ if (!IS_ERR(it)) {
result = iops->get(env, it, (const void *)"");
if (result > 0) {
int i;
iops->put(env, it);
iops->fini(env, it);
} else
- result = -ENOMEM;
+ result = PTR_ERR(it);
RETURN(result);
}
if (!mdd_object_exists(cobj))
RETURN(-ENOENT);
+ if (mdd_is_dead_obj(cobj))
+ RETURN(-ESTALE);
+
if (pobj) {
+ if (!mdd_object_exists(pobj))
+ RETURN(-ENOENT);
+
if (mdd_is_dead_obj(pobj))
RETURN(-ENOENT);
int rc = 0;
ENTRY;
+ if (!mdd_object_exists(src_obj))
+ RETURN(-ENOENT);
+
+ if (mdd_is_dead_obj(src_obj))
+ RETURN(-ESTALE);
+
/* Local ops, no lookup before link, check filename length here. */
if (lname && (lname->ln_namelen > m->mdd_dt_conf.ddp_max_name_len))
RETURN(-ENAMETOOLONG);
mdo_ref_del(env, obj, handle);
}
-/* insert named index, add reference if isdir */
-static int __mdd_index_insert(const struct lu_env *env, struct mdd_object *pobj,
- const struct lu_fid *lf, const char *name, int is_dir,
- struct thandle *handle, struct lustre_capa *capa)
+static int __mdd_index_delete_only(const struct lu_env *env, struct mdd_object *pobj,
+ const char *name, struct thandle *handle,
+ struct lustre_capa *capa)
+{
+ struct dt_object *next = mdd_object_child(pobj);
+ int rc;
+ ENTRY;
+
+ if (dt_try_as_dir(env, next)) {
+ rc = next->do_index_ops->dio_delete(env, next,
+ (struct dt_key *)name,
+ handle, capa);
+ } else
+ rc = -ENOTDIR;
+
+ RETURN(rc);
+}
+
+static int __mdd_index_insert_only(const struct lu_env *env,
+ struct mdd_object *pobj,
+ const struct lu_fid *lf, const char *name,
+ struct thandle *handle,
+ struct lustre_capa *capa)
{
struct dt_object *next = mdd_object_child(pobj);
int rc;
struct md_ucred *uc = md_ucred(env);
rc = next->do_index_ops->dio_insert(env, next,
- __mdd_fid_rec(env, lf),
+ (struct dt_rec*)lf,
(const struct dt_key *)name,
handle, capa, uc->mu_cap &
CFS_CAP_SYS_RESOURCE_MASK);
} else {
rc = -ENOTDIR;
}
-
- if (rc == 0) {
- if (is_dir) {
- mdd_write_lock(env, pobj, MOR_TGT_PARENT);
- __mdd_ref_add(env, pobj, handle);
- mdd_write_unlock(env, pobj);
- }
- }
RETURN(rc);
}
-/* delete named index, drop reference if isdir */
-static int __mdd_index_delete(const struct lu_env *env, struct mdd_object *pobj,
- const char *name, int is_dir, struct thandle *handle,
- struct lustre_capa *capa)
+/* insert named index, add reference if isdir */
+static int __mdd_index_insert(const struct lu_env *env, struct mdd_object *pobj,
+ const struct lu_fid *lf, const char *name, int is_dir,
+ struct thandle *handle, struct lustre_capa *capa)
{
- struct dt_object *next = mdd_object_child(pobj);
int rc;
ENTRY;
- if (dt_try_as_dir(env, next)) {
- rc = next->do_index_ops->dio_delete(env, next,
- (struct dt_key *)name,
- handle, capa);
- if (rc == 0 && is_dir) {
- int is_dot = 0;
-
- if (name != NULL && name[0] == '.' && name[1] == 0)
- is_dot = 1;
- mdd_write_lock(env, pobj, MOR_TGT_PARENT);
- __mdd_ref_del(env, pobj, handle, is_dot);
- mdd_write_unlock(env, pobj);
- }
- } else
- rc = -ENOTDIR;
-
+ rc = __mdd_index_insert_only(env, pobj, lf, name, handle, capa);
+ if (rc == 0 && is_dir) {
+ mdd_write_lock(env, pobj, MOR_TGT_PARENT);
+ __mdd_ref_add(env, pobj, handle);
+ mdd_write_unlock(env, pobj);
+ }
RETURN(rc);
}
-static int
-__mdd_index_insert_only(const struct lu_env *env, struct mdd_object *pobj,
- const struct lu_fid *lf, const char *name,
- struct thandle *handle, struct lustre_capa *capa)
+/* delete named index, drop reference if isdir */
+static int __mdd_index_delete(const struct lu_env *env, struct mdd_object *pobj,
+ const char *name, int is_dir, struct thandle *handle,
+ struct lustre_capa *capa)
{
- struct dt_object *next = mdd_object_child(pobj);
int rc;
ENTRY;
- if (dt_try_as_dir(env, next)) {
- struct md_ucred *uc = md_ucred(env);
+ rc = __mdd_index_delete_only(env, pobj, name, handle, capa);
+ if (rc == 0 && is_dir) {
+ int is_dot = 0;
- rc = next->do_index_ops->dio_insert(env, next,
- __mdd_fid_rec(env, lf),
- (const struct dt_key *)name,
- handle, capa, uc->mu_cap &
- CFS_CAP_SYS_RESOURCE_MASK);
- } else {
- rc = -ENOTDIR;
+ if (name != NULL && name[0] == '.' && name[1] == 0)
+ is_dot = 1;
+ mdd_write_lock(env, pobj, MOR_TGT_PARENT);
+ __mdd_ref_del(env, pobj, handle, is_dot);
+ mdd_write_unlock(env, pobj);
}
+
RETURN(rc);
}
+
/** Store a namespace change changelog record
* If this fails, we must fail the whole transaction; we don't
* want the change to commit without the log entry.
static int mdd_changelog_ns_store(const struct lu_env *env,
struct mdd_device *mdd,
enum changelog_rec_type type,
+ int flags,
struct mdd_object *target,
struct mdd_object *parent,
const struct lu_fid *tf,
int rc;
ENTRY;
+ /* Not recording */
if (!(mdd->mdd_cl.mc_flags & CLM_ON))
RETURN(0);
+ if ((mdd->mdd_cl.mc_mask & (1 << type)) == 0)
+ RETURN(0);
LASSERT(parent != NULL);
LASSERT(tname != NULL);
RETURN(-ENOMEM);
rec = (struct llog_changelog_rec *)buf->lb_buf;
- rec->cr_flags = CLF_VERSION;
- rec->cr_type = (__u32)type;
+ rec->cr.cr_flags = CLF_VERSION | (CLF_FLAGMASK & flags);
+ rec->cr.cr_type = (__u32)type;
tfid = tf ? tf : mdo2fid(target);
- rec->cr_tfid = *tfid;
- rec->cr_pfid = *tpfid;
- rec->cr_namelen = tname->ln_namelen;
- memcpy(rec->cr_name, tname->ln_name, rec->cr_namelen);
+ rec->cr.cr_tfid = *tfid;
+ rec->cr.cr_pfid = *tpfid;
+ rec->cr.cr_namelen = tname->ln_namelen;
+ memcpy(rec->cr.cr_name, tname->ln_name, rec->cr.cr_namelen);
if (likely(target))
target->mod_cltime = cfs_time_current_64();
struct thandle *handle;
#ifdef HAVE_QUOTA_SUPPORT
struct obd_device *obd = mdd->mdd_obd_dev;
+ struct obd_export *exp = md_quota(env)->mq_exp;
struct mds_obd *mds = &obd->u.mds;
unsigned int qids[MAXQUOTAS] = { 0, 0 };
- int quota_opc = 0, rec_pending = 0;
+ int quota_opc = 0, rec_pending[MAXQUOTAS] = { 0, 0 };
#endif
int rc;
ENTRY;
quota_opc = FSFILT_OP_LINK;
mdd_quota_wrapper(la_tmp, qids);
/* get block quota for parent */
- lquota_chkquota(mds_quota_interface_ref, obd,
- qids[USRQUOTA], qids[GRPQUOTA], 1,
- &rec_pending, NULL, LQUOTA_FLAGS_BLK,
- data, 1);
+ lquota_chkquota(mds_quota_interface_ref, obd, exp,
+ qids, rec_pending, 1, NULL,
+ LQUOTA_FLAGS_BLK, data, 1);
}
}
#endif
la->la_valid = LA_CTIME;
rc = mdd_attr_check_set_internal(env, mdd_sobj, la, handle, 0);
- if (rc == 0)
- mdd_links_add(env, mdd_sobj, mdo2fid(mdd_tobj), lname, handle);
+ if (rc == 0) {
+ mdd_links_add(env, mdd_sobj,
+ mdo2fid(mdd_tobj), lname, handle, 0);
+ }
EXIT;
out_unlock:
mdd_pdo_write_unlock(env, mdd_tobj, dlh);
out_trans:
if (rc == 0)
- rc = mdd_changelog_ns_store(env, mdd, CL_HARDLINK, mdd_sobj,
+ rc = mdd_changelog_ns_store(env, mdd, CL_HARDLINK, 0, mdd_sobj,
mdd_tobj, NULL, lname, handle);
mdd_trans_stop(env, mdd, rc, handle);
out_pending:
#ifdef HAVE_QUOTA_SUPPORT
if (quota_opc) {
- if (rec_pending)
- lquota_pending_commit(mds_quota_interface_ref, obd,
- qids[USRQUOTA], qids[GRPQUOTA],
- rec_pending, 1);
+ lquota_pending_commit(mds_quota_interface_ref, obd,
+ qids, rec_pending, 1);
/* Trigger dqacq for the parent owner. If failed,
* the next call for lquota_chkquota will process it. */
lquota_adjust(mds_quota_interface_ref, obd, 0, qids, rc,
int reset = 1;
ENTRY;
- rc = mdd_iattr_get(env, obj, ma);
+ LASSERT(mdd_write_locked(env, obj) != 0);
+
+ /* read HSM flags, needed to set changelogs flags */
+ ma->ma_need = MA_HSM | MA_INODE;
+ rc = mdd_attr_get_internal(env, obj, ma);
if (rc == 0 && ma->ma_attr.la_nlink == 0) {
+ obj->mod_flags |= DEAD_OBJ;
/* add new orphan and the object
* will be deleted during mdd_close() */
if (obj->mod_count) {
rc = __mdd_orphan_add(env, obj, th);
- if (rc == 0) {
- obj->mod_flags |= ORPHAN_OBJ;
- CDEBUG(D_HA, "Object "DFID" is going to be "
- "an orphan, open count = %d\n",
+ if (rc == 0)
+ CDEBUG(D_HA, "Object "DFID" is inserted into "
+ "orphan list, open count = %d\n",
PFID(mdd_object_fid(obj)),
obj->mod_count);
- }
- }
-
- obj->mod_flags |= DEAD_OBJ;
- if (!(obj->mod_flags & ORPHAN_OBJ)) {
+ else
+ CERROR("Object "DFID" fail to be an orphan, "
+ "open count = %d, maybe cause failed "
+ "open replay\n",
+ PFID(mdd_object_fid(obj)),
+ obj->mod_count);
+ } else {
rc = mdd_object_kill(env, obj, ma);
if (rc == 0)
reset = 0;
if (rc)
GOTO(cleanup, rc);
- la->la_valid = LA_CTIME;
- rc = mdd_attr_check_set_internal(env, mdd_cobj, la, handle, 0);
- if (rc)
- GOTO(cleanup, rc);
+ if (ma->ma_attr.la_nlink > 0 || mdd_cobj->mod_count > 0) {
+ /* update ctime of an unlinked file only if it is still
+ * opened or a link still exists */
+ la->la_valid = LA_CTIME;
+ rc = mdd_attr_check_set_internal(env, mdd_cobj, la, handle, 0);
+ if (rc)
+ GOTO(cleanup, rc);
+ }
rc = mdd_finish_unlink(env, mdd_cobj, ma, handle);
#ifdef HAVE_QUOTA_SUPPORT
}
}
#endif
-
- if (rc == 0)
- obd_set_info_async(mdd2obd_dev(mdd)->u.mds.mds_osc_exp,
- sizeof(KEY_UNLINKED), KEY_UNLINKED, 0,
- NULL, NULL);
if (!is_dir)
/* old files may not have link ea; ignore errors */
mdd_links_rename(env, mdd_cobj, mdo2fid(mdd_pobj),
mdd_write_unlock(env, mdd_cobj);
mdd_pdo_write_unlock(env, mdd_pobj, dlh);
out_trans:
- if (rc == 0)
+ if (rc == 0) {
+ int cl_flags;
+
+ cl_flags = (ma->ma_attr.la_nlink == 0) ? CLF_UNLINK_LAST : 0;
+ if ((ma->ma_valid & MA_HSM) &&
+ (ma->ma_hsm.mh_flags & HS_EXISTS))
+ cl_flags |= CLF_UNLINK_HSM_EXISTS;
+
rc = mdd_changelog_ns_store(env, mdd,
- is_dir ? CL_RMDIR : CL_UNLINK,
- mdd_cobj, mdd_pobj, NULL, lname,
- handle);
+ is_dir ? CL_RMDIR : CL_UNLINK, cl_flags,
+ mdd_cobj, mdd_pobj, NULL, lname, handle);
+ }
mdd_trans_stop(env, mdd, rc, handle);
#ifdef HAVE_QUOTA_SUPPORT
#ifdef HAVE_QUOTA_SUPPORT
struct md_ucred *uc = md_ucred(env);
struct obd_device *obd = mdd->mdd_obd_dev;
+ struct obd_export *exp = md_quota(env)->mq_exp;
struct mds_obd *mds = &obd->u.mds;
unsigned int qids[MAXQUOTAS] = { 0, 0 };
- int quota_opc = 0, rec_pending = 0;
+ int quota_opc = 0, rec_pending[MAXQUOTAS] = { 0, 0 };
cfs_cap_t save = uc->mu_cap;
#endif
int rc;
mdd_quota_wrapper(la_tmp, qids);
/* get block quota for parent */
lquota_chkquota(mds_quota_interface_ref, obd,
- qids[USRQUOTA], qids[GRPQUOTA],
- 1, &rec_pending, NULL,
+ exp, qids, rec_pending, 1, NULL,
LQUOTA_FLAGS_BLK, data, 1);
}
} else {
#ifdef HAVE_QUOTA_SUPPORT
if (mds->mds_quota) {
if (quota_opc) {
- if (rec_pending)
- lquota_pending_commit(mds_quota_interface_ref,
- obd, qids[USRQUOTA],
- qids[GRPQUOTA],
- rec_pending, 1);
+ lquota_pending_commit(mds_quota_interface_ref,
+ obd, qids, rec_pending, 1);
/* Trigger dqacq for the parent owner. If failed,
* the next call for lquota_chkquota will process it*/
lquota_adjust(mds_quota_interface_ref, obd, 0, qids,
* processed in cmr_rename_tgt before mdd_rename_tgt and enable
* MDS_PERM_BYPASS.
* So check may_delete, but not check nlink of tgt_pobj. */
- LASSERT(tobj);
+
rc = mdd_may_delete(env, tgt_pobj, tobj, ma, 1, 1);
RETURN(rc);
struct thandle *handle;
#ifdef HAVE_QUOTA_SUPPORT
struct obd_device *obd = mdd->mdd_obd_dev;
+ struct obd_export *exp = md_quota(env)->mq_exp;
struct mds_obd *mds = &obd->u.mds;
unsigned int qcids[MAXQUOTAS] = { 0, 0 };
unsigned int qpids[MAXQUOTAS] = { 0, 0 };
- int quota_opc = 0, rec_pending = 0;
+ int quota_copc = 0, quota_popc = 0;
+ int rec_pending[MAXQUOTAS] = { 0, 0 };
#endif
int rc;
ENTRY;
if (!rc) {
void *data = NULL;
mdd_data_get(env, mdd_tpobj, &data);
- quota_opc = FSFILT_OP_LINK;
+ quota_popc = FSFILT_OP_LINK;
mdd_quota_wrapper(la_tmp, qpids);
/* get block quota for target parent */
- lquota_chkquota(mds_quota_interface_ref, obd,
- qpids[USRQUOTA], qpids[GRPQUOTA], 1,
- &rec_pending, NULL, LQUOTA_FLAGS_BLK,
- data, 1);
+ lquota_chkquota(mds_quota_interface_ref, obd, exp,
+ qpids, rec_pending, 1, NULL,
+ LQUOTA_FLAGS_BLK, data, 1);
}
}
#endif
#ifdef HAVE_QUOTA_SUPPORT
if (mds->mds_quota && ma->ma_valid & MA_INODE &&
ma->ma_attr.la_nlink == 0 && mdd_tobj->mod_count == 0) {
- quota_opc = FSFILT_OP_UNLINK_PARTIAL_CHILD;
+ quota_copc = FSFILT_OP_UNLINK_PARTIAL_CHILD;
mdd_quota_wrapper(&ma->ma_attr, qcids);
}
#endif
if (rc == 0)
/* Bare EXT record with no RENAME in front of it signifies
a partial slave op */
- rc = mdd_changelog_ns_store(env, mdd, CL_EXT, mdd_tobj,
+ rc = mdd_changelog_ns_store(env, mdd, CL_EXT, 0, mdd_tobj,
mdd_tpobj, NULL, lname, handle);
mdd_trans_stop(env, mdd, rc, handle);
out_pending:
#ifdef HAVE_QUOTA_SUPPORT
if (mds->mds_quota) {
- if (rec_pending)
+ if (quota_popc)
lquota_pending_commit(mds_quota_interface_ref, obd,
- qpids[USRQUOTA],
- qpids[GRPQUOTA],
- rec_pending, 1);
- if (quota_opc)
- /* Trigger dqrel/dqacq on the target owner of child and
- * parent. If failed, the next call for lquota_chkquota
+ qpids, rec_pending, 1);
+
+ if (quota_copc)
+ /* Trigger dqrel on the target owner of child.
+ * If failed, the next call for lquota_chkquota
* will process it. */
- lquota_adjust(mds_quota_interface_ref, obd, qcids,
- qpids, rc, quota_opc);
+ lquota_adjust(mds_quota_interface_ref, obd, qcids, qpids,
+ rc, quota_copc);
}
#endif
return rc;
if (!md_should_create(spec->sp_cr_flags))
RETURN(0);
-
+ lmm_size = ma->ma_lmm_size;
rc = mdd_lov_create(env, mdd, mdd_pobj, son, &lmm, &lmm_size,
spec, attr);
if (rc)
struct mdd_object *mdd_obj = md2mdd_obj(pobj);
struct mdd_device *m = mdo2mdd(pobj);
struct dt_object *dir = mdd_object_child(mdd_obj);
- struct lu_fid_pack *pack = &mdd_env_info(env)->mti_pack;
int rc;
ENTRY;
dt_try_as_dir(env, dir))) {
rc = dir->do_index_ops->dio_lookup(env, dir,
- (struct dt_rec *)pack, key,
+ (struct dt_rec *)fid, key,
mdd_object_capa(env, mdd_obj));
if (rc > 0)
- rc = fid_unpack(pack, fid);
+ rc = 0;
else if (rc == 0)
rc = -ENOENT;
} else
__mdd_ref_add(env, child, handle);
rc = __mdd_index_insert_only(env, child, mdo2fid(child),
dot, handle, BYPASS_CAPA);
- if (rc == 0) {
+ if (rc == 0)
rc = __mdd_index_insert_only(env, child, pfid,
dotdot, handle,
BYPASS_CAPA);
- if (rc != 0) {
- int rc2;
-
- rc2 = __mdd_index_delete(env, child, dot, 1,
- handle, BYPASS_CAPA);
- if (rc2 != 0)
- CERROR("Failure to cleanup after dotdot"
- " creation: %d (%d)\n", rc2, rc);
- }
- }
+ if (rc != 0)
+ __mdd_ref_del(env, child, handle, 1);
}
if (rc == 0)
- mdd_links_add(env, child, pfid, lname, handle);
+ mdd_links_add(env, child, pfid, lname, handle, 1);
RETURN(rc);
}
int got_def_acl = 0;
#ifdef HAVE_QUOTA_SUPPORT
struct obd_device *obd = mdd->mdd_obd_dev;
+ struct obd_export *exp = md_quota(env)->mq_exp;
struct mds_obd *mds = &obd->u.mds;
unsigned int qcids[MAXQUOTAS] = { 0, 0 };
unsigned int qpids[MAXQUOTAS] = { 0, 0 };
int quota_opc = 0, block_count = 0;
- int inode_pending = 0, block_pending = 0, parent_pending = 0;
+ int inode_pending[MAXQUOTAS] = { 0, 0 };
+ int block_pending[MAXQUOTAS] = { 0, 0 };
+ int parent_pending[MAXQUOTAS] = { 0, 0 };
#endif
ENTRY;
mdd_quota_wrapper(&ma->ma_attr, qcids);
mdd_quota_wrapper(la_tmp, qpids);
/* get file quota for child */
- lquota_chkquota(mds_quota_interface_ref, obd,
- qcids[USRQUOTA], qcids[GRPQUOTA], 1,
- &inode_pending, NULL, 0, NULL, 0);
+ lquota_chkquota(mds_quota_interface_ref, obd, exp,
+ qcids, inode_pending, 1, NULL, 0, NULL,
+ 0);
switch (ma->ma_attr.la_mode & S_IFMT) {
case S_IFLNK:
case S_IFDIR:
/* get block quota for child and parent */
if (block_count)
lquota_chkquota(mds_quota_interface_ref, obd,
- qcids[USRQUOTA], qcids[GRPQUOTA],
- block_count,
- &block_pending, NULL,
+ exp, qcids, block_pending,
+ block_count, NULL,
LQUOTA_FLAGS_BLK, NULL, 0);
if (!same)
lquota_chkquota(mds_quota_interface_ref, obd,
- qpids[USRQUOTA], qpids[GRPQUOTA], 1,
- &parent_pending, NULL,
- LQUOTA_FLAGS_BLK, NULL, 0);
+ exp, qpids, parent_pending, 1,
+ NULL, LQUOTA_FLAGS_BLK, NULL,
+ 0);
}
}
#endif
* first.
*/
if (S_ISREG(attr->la_mode)) {
+ lmm_size = ma->ma_lmm_size;
rc = mdd_lov_create(env, mdd, mdd_pobj, son, &lmm, &lmm_size,
spec, attr);
if (rc)
S_ISDIR(attr->la_mode) ? CL_MKDIR :
S_ISREG(attr->la_mode) ? CL_CREATE :
S_ISLNK(attr->la_mode) ? CL_SOFTLINK : CL_MKNOD,
- son, mdd_pobj, NULL, lname, handle);
+ 0, son, mdd_pobj, NULL, lname, handle);
mdd_trans_stop(env, mdd, rc, handle);
out_free:
- /* finis lov_create stuff, free all temporary data */
+ /* finish lov_create stuff, free all temporary data */
mdd_lov_create_finish(env, mdd, lmm, lmm_size, spec);
out_pending:
#ifdef HAVE_QUOTA_SUPPORT
if (quota_opc) {
- if (inode_pending)
- lquota_pending_commit(mds_quota_interface_ref, obd,
- qcids[USRQUOTA], qcids[GRPQUOTA],
- inode_pending, 0);
- if (block_pending)
- lquota_pending_commit(mds_quota_interface_ref, obd,
- qcids[USRQUOTA], qcids[GRPQUOTA],
- block_pending, 1);
- if (parent_pending)
- lquota_pending_commit(mds_quota_interface_ref, obd,
- qpids[USRQUOTA], qpids[GRPQUOTA],
- parent_pending, 1);
+ lquota_pending_commit(mds_quota_interface_ref, obd, qcids,
+ inode_pending, 0);
+ lquota_pending_commit(mds_quota_interface_ref, obd, qcids,
+ block_pending, 1);
+ lquota_pending_commit(mds_quota_interface_ref, obd, qpids,
+ parent_pending, 1);
/* Trigger dqacq on the owner of child and parent. If failed,
* the next call for lquota_chkquota will process it. */
lquota_adjust(mds_quota_interface_ref, obd, qcids, qpids, rc,
* the other case has been processed in cml_rename
* before mdd_rename and enable MDS_PERM_BYPASS. */
LASSERT(sobj);
+
rc = mdd_may_delete(env, src_pobj, sobj, ma, 1, 0);
if (rc)
RETURN(rc);
struct dynlock_handle *sdlh, *tdlh;
struct thandle *handle;
const struct lu_fid *tpobj_fid = mdo2fid(mdd_tpobj);
+ const struct lu_fid *spobj_fid = mdo2fid(mdd_spobj);
int is_dir;
- int rc;
+ int rc, rc2;
#ifdef HAVE_QUOTA_SUPPORT
struct obd_device *obd = mdd->mdd_obd_dev;
+ struct obd_export *exp = md_quota(env)->mq_exp;
struct mds_obd *mds = &obd->u.mds;
unsigned int qspids[MAXQUOTAS] = { 0, 0 };
unsigned int qtcids[MAXQUOTAS] = { 0, 0 };
unsigned int qtpids[MAXQUOTAS] = { 0, 0 };
- int quota_opc = 0, rec_pending = 0;
+ int quota_copc = 0, quota_popc = 0;
+ int rec_pending[MAXQUOTAS] = { 0, 0 };
#endif
ENTRY;
if (!rc) {
void *data = NULL;
mdd_data_get(env, mdd_tpobj, &data);
- quota_opc = FSFILT_OP_LINK;
+ quota_popc = FSFILT_OP_LINK;
mdd_quota_wrapper(la_tmp, qtpids);
/* get block quota for target parent */
lquota_chkquota(mds_quota_interface_ref,
- obd, qtpids[USRQUOTA],
- qtpids[GRPQUOTA], 1,
- &rec_pending, NULL,
+ obd, exp, qtpids,
+ rec_pending, 1, NULL,
LQUOTA_FLAGS_BLK,
data, 1);
}
GOTO(cleanup, rc);
/* "mv dir1 dir2" needs "dir1/.." link update */
- if (is_dir && mdd_sobj) {
- rc = __mdd_index_delete(env, mdd_sobj, dotdot, is_dir, handle,
- mdd_object_capa(env, mdd_spobj));
+ if (is_dir && mdd_sobj && !lu_fid_eq(spobj_fid, tpobj_fid)) {
+ rc = __mdd_index_delete_only(env, mdd_sobj, dotdot, handle,
+ mdd_object_capa(env, mdd_sobj));
if (rc)
- GOTO(cleanup, rc);
+ GOTO(fixup_spobj2, rc);
- rc = __mdd_index_insert(env, mdd_sobj, tpobj_fid, dotdot,
- is_dir, handle,
- mdd_object_capa(env, mdd_tpobj));
+ rc = __mdd_index_insert_only(env, mdd_sobj, tpobj_fid, dotdot,
+ handle, mdd_object_capa(env, mdd_sobj));
if (rc)
- GOTO(cleanup, rc);
+ GOTO(fixup_spobj, rc);
}
/* Remove target name from target directory
*/
rc = __mdd_index_delete(env, mdd_tpobj, tname, is_dir, handle,
mdd_object_capa(env, mdd_tpobj));
- if (rc != 0 && rc != -ENOENT)
- GOTO(cleanup, rc);
+ if (rc != 0) {
+ if (mdd_tobj) {
+ /* tname might been renamed to something else */
+ GOTO(fixup_spobj, rc);
+ }
+ if (rc != -ENOENT)
+ GOTO(fixup_spobj, rc);
+ }
/* Insert new fid with target name into target dir */
rc = __mdd_index_insert(env, mdd_tpobj, lf, tname, is_dir, handle,
mdd_object_capa(env, mdd_tpobj));
if (rc)
- GOTO(cleanup, rc);
+ GOTO(fixup_tpobj, rc);
LASSERT(ma->ma_attr.la_valid & LA_CTIME);
la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime;
rc = mdd_attr_check_set_internal_locked(env, mdd_sobj, la,
handle, 0);
if (rc)
- GOTO(cleanup, rc);
+ GOTO(fixup_tpobj, rc);
}
/* Remove old target object
*/
if (tobj && mdd_object_exists(mdd_tobj)) {
mdd_write_lock(env, mdd_tobj, MOR_TGT_CHILD);
+ if (mdd_is_dead_obj(mdd_tobj)) {
+ mdd_write_unlock(env, mdd_tobj);
+ /* shld not be dead, something is wrong */
+ CERROR("tobj is dead, something is wrong\n");
+ rc = -EINVAL;
+ goto cleanup;
+ }
__mdd_ref_del(env, mdd_tobj, handle, 0);
/* Remove dot reference. */
la->la_valid = LA_CTIME;
rc = mdd_attr_check_set_internal(env, mdd_tobj, la, handle, 0);
if (rc)
- GOTO(cleanup, rc);
+ GOTO(fixup_tpobj, rc);
rc = mdd_finish_unlink(env, mdd_tobj, ma, handle);
mdd_write_unlock(env, mdd_tobj);
if (rc)
- GOTO(cleanup, rc);
+ GOTO(fixup_tpobj, rc);
#ifdef HAVE_QUOTA_SUPPORT
if (mds->mds_quota && ma->ma_valid & MA_INODE &&
ma->ma_attr.la_nlink == 0 && mdd_tobj->mod_count == 0) {
- quota_opc = FSFILT_OP_UNLINK_PARTIAL_CHILD;
+ quota_copc = FSFILT_OP_UNLINK_PARTIAL_CHILD;
mdd_quota_wrapper(&ma->ma_attr, qtcids);
}
#endif
la->la_valid = LA_CTIME | LA_MTIME;
rc = mdd_attr_check_set_internal_locked(env, mdd_spobj, la, handle, 0);
if (rc)
- GOTO(cleanup, rc);
+ GOTO(fixup_tpobj, rc);
if (mdd_spobj != mdd_tpobj) {
la->la_valid = LA_CTIME | LA_MTIME;
if (rc == -ENOENT)
/* Old files might not have EA entry */
mdd_links_add(env, mdd_sobj, mdo2fid(mdd_spobj),
- lsname, handle);
+ lsname, handle, 0);
mdd_write_unlock(env, mdd_sobj);
/* We don't fail the transaction if the link ea can't be
updated -- fid2path will use alternate lookup method. */
}
EXIT;
+
+fixup_tpobj:
+ if (rc) {
+ rc2 = __mdd_index_delete(env, mdd_tpobj, tname, is_dir, handle,
+ BYPASS_CAPA);
+ if (rc2)
+ CWARN("tp obj fix error %d\n",rc2);
+
+ if (mdd_tobj && mdd_object_exists(mdd_tobj) &&
+ !mdd_is_dead_obj(mdd_tobj)) {
+ rc2 = __mdd_index_insert(env, mdd_tpobj,
+ mdo2fid(mdd_tobj), tname,
+ is_dir, handle,
+ BYPASS_CAPA);
+
+ if (rc2)
+ CWARN("tp obj fix error %d\n",rc2);
+ }
+ }
+
+fixup_spobj:
+ if (rc && is_dir && mdd_sobj) {
+ rc2 = __mdd_index_delete_only(env, mdd_sobj, dotdot, handle,
+ BYPASS_CAPA);
+
+ if (rc2)
+ CWARN("sp obj dotdot delete error %d\n",rc2);
+
+
+ rc2 = __mdd_index_insert_only(env, mdd_sobj, spobj_fid,
+ dotdot, handle, BYPASS_CAPA);
+ if (rc2)
+ CWARN("sp obj dotdot insert error %d\n",rc2);
+ }
+
+fixup_spobj2:
+ if (rc) {
+ rc2 = __mdd_index_insert(env, mdd_spobj,
+ lf, sname, is_dir, handle, BYPASS_CAPA);
+ if (rc2)
+ CWARN("sp obj fix error %d\n",rc2);
+ }
cleanup:
if (likely(tdlh) && sdlh != tdlh)
mdd_pdo_write_unlock(env, mdd_tpobj, tdlh);
mdd_pdo_write_unlock(env, mdd_spobj, sdlh);
cleanup_unlocked:
if (rc == 0)
- rc = mdd_changelog_ns_store(env, mdd, CL_RENAME, mdd_tobj,
+ rc = mdd_changelog_ns_store(env, mdd, CL_RENAME, 0, mdd_tobj,
mdd_spobj, lf, lsname, handle);
if (rc == 0)
- rc = mdd_changelog_ns_store(env, mdd, CL_EXT, mdd_tobj,
+ rc = mdd_changelog_ns_store(env, mdd, CL_EXT, 0, mdd_tobj,
mdd_tpobj, lf, ltname, handle);
mdd_trans_stop(env, mdd, rc, handle);
out_pending:
#ifdef HAVE_QUOTA_SUPPORT
if (mds->mds_quota) {
- if (rec_pending)
+ if (quota_popc)
lquota_pending_commit(mds_quota_interface_ref, obd,
- qtpids[USRQUOTA],
- qtpids[GRPQUOTA],
- rec_pending, 1);
- /* Trigger dqrel on the source owner of parent.
- * If failed, the next call for lquota_chkquota will
- * process it. */
- lquota_adjust(mds_quota_interface_ref, obd, 0, qspids, rc,
- FSFILT_OP_UNLINK_PARTIAL_PARENT);
- if (quota_opc)
- /* Trigger dqrel/dqacq on the target owner of child and
- * parent. If failed, the next call for lquota_chkquota
+ qtpids, rec_pending, 1);
+
+ if (quota_copc) {
+ /* Trigger dqrel on the source owner of parent.
+ * If failed, the next call for lquota_chkquota will
+ * process it. */
+ lquota_adjust(mds_quota_interface_ref, obd, 0, qspids, rc,
+ FSFILT_OP_UNLINK_PARTIAL_PARENT);
+
+ /* Trigger dqrel on the target owner of child.
+ * If failed, the next call for lquota_chkquota
* will process it. */
lquota_adjust(mds_quota_interface_ref, obd, qtcids,
- qtpids, rc, quota_opc);
+ qtpids, rc, quota_copc);
+ }
}
#endif
return rc;
rc = mdo_xattr_get(env, mdd_obj, buf, XATTR_NAME_LINK, capa);
if (rc == -ERANGE) {
/* Buf was too small, figure out what we need. */
- buf->lb_buf = NULL;
- buf->lb_len = 0;
+ mdd_buf_put(buf);
rc = mdo_xattr_get(env, mdd_obj, buf, XATTR_NAME_LINK, capa);
if (rc < 0)
return ERR_PTR(rc);
/** Pack a link_ea_entry.
* All elements are stored as chars to avoid alignment issues.
* Numbers are always big-endian
- * \param packbuf is a temp fid buffer
* \retval record length
*/
static int mdd_lee_pack(struct link_ea_entry *lee, const struct lu_name *lname,
- const struct lu_fid *pfid, struct lu_fid* packbuf)
+ const struct lu_fid *pfid)
{
- char *ptr;
- int reclen;
+ struct lu_fid tmpfid;
+ int reclen;
+
+ fid_cpu_to_be(&tmpfid, pfid);
+ memcpy(&lee->lee_parent_fid, &tmpfid, sizeof(tmpfid));
+ memcpy(lee->lee_name, lname->ln_name, lname->ln_namelen);
+ reclen = sizeof(struct link_ea_entry) + lname->ln_namelen;
- fid_pack(&lee->lee_parent_fid, pfid, packbuf);
- ptr = (char *)&lee->lee_parent_fid + lee->lee_parent_fid.fp_len;
- strncpy(ptr, lname->ln_name, lname->ln_namelen);
- reclen = lee->lee_parent_fid.fp_len + lname->ln_namelen +
- sizeof(lee->lee_reclen);
lee->lee_reclen[0] = (reclen >> 8) & 0xff;
lee->lee_reclen[1] = reclen & 0xff;
return reclen;
struct lu_name *lname, struct lu_fid *pfid)
{
*reclen = (lee->lee_reclen[0] << 8) | lee->lee_reclen[1];
- fid_unpack(&lee->lee_parent_fid, pfid);
- lname->ln_name = (char *)&lee->lee_parent_fid +
- lee->lee_parent_fid.fp_len;
- lname->ln_namelen = *reclen - lee->lee_parent_fid.fp_len -
- sizeof(lee->lee_reclen);
+ memcpy(pfid, &lee->lee_parent_fid, sizeof(*pfid));
+ fid_be_to_cpu(pfid, pfid);
+ lname->ln_name = lee->lee_name;
+ lname->ln_namelen = *reclen - sizeof(struct link_ea_entry);
}
/** Add a record to the end of link ea buf */
leh = buf->lb_buf;
lee = buf->lb_buf + leh->leh_len;
- reclen = mdd_lee_pack(lee, lname, pfid, &mdd_env_info(env)->mti_fid2);
+ reclen = mdd_lee_pack(lee, lname, pfid);
leh->leh_len += reclen;
leh->leh_reccount++;
return 0;
struct mdd_object *mdd_obj,
const struct lu_fid *pfid,
const struct lu_name *lname,
- struct thandle *handle)
+ struct thandle *handle, int first)
{
struct lu_buf *buf;
struct link_ea_header *leh;
if (!mdd_linkea_enable)
RETURN(0);
- buf = mdd_links_get(env, mdd_obj);
+ buf = first ? ERR_PTR(-ENODATA) : mdd_links_get(env, mdd_obj);
if (IS_ERR(buf)) {
rc = PTR_ERR(buf);
if (rc != -ENODATA) {
rc = __mdd_xattr_set(env, mdd_obj,
mdd_buf_get_const(env, buf->lb_buf, leh->leh_len),
XATTR_NAME_LINK, 0, handle);
- if (rc)
- CERROR("link_ea add failed %d "DFID"\n", rc,
- PFID(mdd_object_fid(mdd_obj)));
+ if (rc) {
+ if (rc == -ENOSPC)
+ CDEBUG(D_INODE, "link_ea add failed %d "DFID"\n", rc,
+ PFID(mdd_object_fid(mdd_obj)));
+ else
+ CERROR("link_ea add failed %d "DFID"\n", rc,
+ PFID(mdd_object_fid(mdd_obj)));
+ }
- if (buf->lb_vmalloc)
+ if (buf->lb_len > OBD_ALLOC_BIG)
/* if we vmalloced a large buffer drop it */
mdd_buf_put(buf);
buf = mdd_links_get(env, mdd_obj);
if (IS_ERR(buf)) {
rc = PTR_ERR(buf);
- CERROR("link_ea read failed %d "DFID"\n",
- rc, PFID(mdd_object_fid(mdd_obj)));
+ if (rc == -ENODATA)
+ CDEBUG(D_INODE, "link_ea read failed %d "DFID"\n",
+ rc, PFID(mdd_object_fid(mdd_obj)));
+ else
+ CERROR("link_ea read failed %d "DFID"\n",
+ rc, PFID(mdd_object_fid(mdd_obj)));
RETURN(rc);
}
leh = buf->lb_buf;
lee = (struct link_ea_entry *)(leh + 1); /* link #0 */
/* Find the old record */
- for(count = 0; count <= leh->leh_reccount; count++) {
+ for(count = 0; count < leh->leh_reccount; count++) {
mdd_lee_unpack(lee, &reclen, tmpname, tmpfid);
if (tmpname->ln_namelen == oldlname->ln_namelen &&
lu_fid_eq(tmpfid, oldpfid) &&
break;
lee = (struct link_ea_entry *)((char *)lee + reclen);
}
- if (count > leh->leh_reccount) {
+ if ((count + 1) > leh->leh_reccount) {
CDEBUG(D_INODE, "Old link_ea name '%.*s' not found\n",
oldlname->ln_namelen, oldlname->ln_name);
GOTO(out, rc = -ENOENT);
oldlname->ln_namelen, oldlname->ln_name, rc,
PFID(mdd_object_fid(mdd_obj)));
- if (buf->lb_vmalloc)
+ if (buf->lb_len > OBD_ALLOC_BIG)
/* if we vmalloced a large buffer drop it */
mdd_buf_put(buf);
.mdo_name_insert = mdd_name_insert,
.mdo_name_remove = mdd_name_remove,
.mdo_rename_tgt = mdd_rename_tgt,
- .mdo_create_data = mdd_create_data
+ .mdo_create_data = mdd_create_data,
};
git://git.whamcloud.com / fs / lustre-release.git / blobdiff