* 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.
+ *
+ * Copyright (c) 2011 Whamcloud, Inc.
+ *
*/
/*
* This file is part of Lustre, http://www.lustre.org/
#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,
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);
}
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[MAXQUOTAS] = { 0, 0 };
quota_opc = FSFILT_OP_LINK;
mdd_quota_wrapper(la_tmp, qids);
/* get block quota for parent */
- lquota_chkquota(mds_quota_interface_ref, obd,
+ lquota_chkquota(mds_quota_interface_ref, obd, exp,
qids, rec_pending, 1, NULL,
LQUOTA_FLAGS_BLK, data, 1);
}
}
#endif
- mdd_txn_param_build(env, mdd, MDD_TXN_LINK_OP);
+ mdd_txn_param_build(env, mdd, MDD_TXN_LINK_OP, 1);
handle = mdd_trans_start(env, mdd);
if (IS_ERR(handle))
GOTO(out_pending, rc = PTR_ERR(handle));
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:
LASSERT(mdd_write_locked(env, obj) != 0);
- rc = mdd_iattr_get(env, obj, ma);
+ /* 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
LASSERTF(mdd_object_exists(mdd_cobj) > 0, "FID is "DFID"\n",
PFID(mdd_object_fid(mdd_cobj)));
- rc = mdd_log_txn_param_build(env, cobj, ma, MDD_TXN_UNLINK_OP);
+ rc = mdd_log_txn_param_build(env, cobj, ma, MDD_TXN_UNLINK_OP, 1);
if (rc)
RETURN(rc);
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[MAXQUOTAS] = { 0, 0 };
mdd_quota_wrapper(la_tmp, qids);
/* get block quota for parent */
lquota_chkquota(mds_quota_interface_ref, obd,
- qids, rec_pending, 1, NULL,
+ exp, qids, rec_pending, 1, NULL,
LQUOTA_FLAGS_BLK, data, 1);
}
} else {
}
}
#endif
- mdd_txn_param_build(env, mdd, MDD_TXN_INDEX_INSERT_OP);
+ mdd_txn_param_build(env, mdd, MDD_TXN_INDEX_INSERT_OP, 0);
handle = mdd_trans_start(env, mdo2mdd(pobj));
if (IS_ERR(handle))
GOTO(out_pending, rc = PTR_ERR(handle));
}
}
#endif
- mdd_txn_param_build(env, mdd, MDD_TXN_INDEX_DELETE_OP);
+ mdd_txn_param_build(env, mdd, MDD_TXN_INDEX_DELETE_OP, 0);
handle = mdd_trans_start(env, mdd);
if (IS_ERR(handle))
GOTO(out_pending, rc = PTR_ERR(handle));
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 };
quota_popc = FSFILT_OP_LINK;
mdd_quota_wrapper(la_tmp, qpids);
/* get block quota for target parent */
- lquota_chkquota(mds_quota_interface_ref, obd,
+ lquota_chkquota(mds_quota_interface_ref, obd, exp,
qpids, rec_pending, 1, NULL,
LQUOTA_FLAGS_BLK, data, 1);
}
}
#endif
- mdd_txn_param_build(env, mdd, MDD_TXN_RENAME_TGT_OP);
+ if (tobj && mdd_object_exists(mdd_tobj))
+ mdd_log_txn_param_build(env, tobj, ma, MDD_TXN_RENAME_TGT_OP,1);
+ else
+ mdd_txn_param_build(env, mdd, MDD_TXN_RENAME_TGT_OP, 1);
handle = mdd_trans_start(env, mdd);
if (IS_ERR(handle))
GOTO(out_pending, rc = PTR_ERR(handle));
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);
if (rc)
RETURN(rc);
- mdd_txn_param_build(env, mdd, MDD_TXN_CREATE_DATA_OP);
+ mdd_create_txn_param_build(env, mdd, lmm, MDD_TXN_CREATE_DATA_OP, 0);
handle = mdd_trans_start(env, mdd);
if (IS_ERR(handle))
GOTO(out_free, rc = PTR_ERR(handle));
__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 };
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,
- inode_pending, 1, 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, block_pending,
+ exp, qcids, block_pending,
block_count, NULL,
LQUOTA_FLAGS_BLK, NULL, 0);
if (!same)
lquota_chkquota(mds_quota_interface_ref, obd,
- qpids, parent_pending, 1, NULL,
- LQUOTA_FLAGS_BLK, NULL, 0);
+ exp, qpids, parent_pending, 1,
+ NULL, LQUOTA_FLAGS_BLK, NULL,
+ 0);
}
}
#endif
got_def_acl = 1;
}
- mdd_txn_param_build(env, mdd, MDD_TXN_MKDIR_OP);
+ mdd_create_txn_param_build(env, mdd, lmm, MDD_TXN_MKDIR_OP, 1);
handle = mdd_trans_start(env, mdd);
if (IS_ERR(handle))
GOTO(out_free, rc = PTR_ERR(handle));
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
#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 };
mdd_quota_wrapper(la_tmp, qtpids);
/* get block quota for target parent */
lquota_chkquota(mds_quota_interface_ref,
- obd, qtpids,
+ obd, exp, qtpids,
rec_pending, 1, NULL,
LQUOTA_FLAGS_BLK,
data, 1);
}
}
#endif
- mdd_txn_param_build(env, mdd, MDD_TXN_RENAME_OP);
+ if (tobj && mdd_object_exists(mdd_tobj))
+ mdd_log_txn_param_build(env, tobj, ma, MDD_TXN_RENAME_OP, 2);
+ else
+ mdd_txn_param_build(env, mdd, MDD_TXN_RENAME_OP, 2);
handle = mdd_trans_start(env, mdd);
if (IS_ERR(handle))
GOTO(out_pending, rc = PTR_ERR(handle));
GOTO(cleanup, rc);
/* "mv dir1 dir2" needs "dir1/.." link update */
- if (is_dir && mdd_sobj) {
+ 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)
rc = __mdd_index_insert_only(env, mdd_sobj, tpobj_fid, dotdot,
handle, mdd_object_capa(env, mdd_sobj));
- if (rc) {
+ if (rc)
GOTO(fixup_spobj, rc);
- }
}
/* Remove target name from target directory
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. */
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);
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);
static int mdd_lee_pack(struct link_ea_entry *lee, const struct lu_name *lname,
const struct lu_fid *pfid)
{
- int reclen;
+ struct lu_fid tmpfid;
+ int reclen;
- fid_cpu_to_be(&lee->lee_parent_fid, pfid);
- strncpy(lee->lee_name, lname->ln_name, lname->ln_namelen);
+ 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;
lee->lee_reclen[0] = (reclen >> 8) & 0xff;
struct lu_name *lname, struct lu_fid *pfid)
{
*reclen = (lee->lee_reclen[0] << 8) | lee->lee_reclen[1];
- fid_be_to_cpu(pfid, &lee->lee_parent_fid);
+ 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);
}
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,
};