*
* You should have received a copy of the GNU General Public License
* version 2 along with this program; If not, see
- * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
- *
- * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
- * CA 95054 USA or visit www.sun.com if you need additional information or
- * have any questions.
+ * http://www.gnu.org/licenses/gpl-2.0.html
*
* GPL HEADER END
*/
* Copyright (c) 2010, Oracle and/or its affiliates. All rights reserved.
* Use is subject to license terms.
*
- * Copyright (c) 2012, 2013, Intel Corporation.
+ * Copyright (c) 2012, 2015, Intel Corporation.
*/
/*
* This file is part of Lustre, http://www.lustre.org/
newctxt->pwdmnt = dev->od_mnt;
newctxt->pwd = dev->od_mnt->mnt_root;
newctxt->fs = get_ds();
+ newctxt->umask = current_umask();
+ newctxt->dt = NULL;
push_ctxt(save, newctxt);
}
return dchild;
}
-int osd_last_rcvd_subdir_count(struct osd_device *osd)
+static int osd_last_rcvd_subdir_count(struct osd_device *osd)
{
struct lr_server_data lsd;
struct dentry *dlast;
struct osd_thread_info *oti = osd_oti_get(env);
struct lustre_mdt_attrs *lma = &oti->oti_mdt_attrs;
char *name = oti->oti_name;
+ struct osd_thread_info *info = osd_oti_get(env);
struct dentry *dentry;
struct dentry *parent;
int rc;
dentry = osd_child_dentry_by_inode(env, parent->d_inode,
name, strlen(name));
mutex_lock(&parent->d_inode->i_mutex);
- rc = osd_ldiskfs_add_entry(oh->ot_handle, dentry, obj->oo_inode,
- NULL);
+ rc = osd_ldiskfs_add_entry(info, osd, oh->ot_handle, dentry,
+ obj->oo_inode, NULL);
CDEBUG(D_INODE, "%s: add %s:%lu to remote parent %lu.\n", osd_name(osd),
name, obj->oo_inode->i_ino, parent->d_inode->i_ino);
ldiskfs_inc_count(oh->ot_handle, parent->d_inode);
/* Check lma to see whether it is remote object */
rc = osd_get_lma(oti, obj->oo_inode, &oti->oti_obj_dentry, lma);
- if (rc != 0)
+ if (rc != 0) {
+ /* No LMA if the directory is created before 2.0 */
+ if (rc == -ENODATA)
+ rc = 0;
RETURN(rc);
+ }
if (likely(!(lma->lma_incompat & LMAI_REMOTE_PARENT)))
RETURN(0);
mutex_lock(&parent->d_inode->i_mutex);
bh = osd_ldiskfs_find_entry(parent->d_inode, &dentry->d_name, &de,
NULL, NULL);
- if (bh == NULL) {
+ if (IS_ERR(bh)) {
mutex_unlock(&parent->d_inode->i_mutex);
- RETURN(-ENOENT);
+ RETURN(PTR_ERR(bh));
}
CDEBUG(D_INODE, "%s: el %s:%lu to remote parent %lu.\n", osd_name(osd),
name, obj->oo_inode->i_ino, parent->d_inode->i_ino);
mutex_lock(&parent->d_inode->i_mutex);
bh = osd_ldiskfs_find_entry(parent->d_inode, &dentry->d_name, &de,
NULL, NULL);
- if (bh == NULL) {
- rc = -ENOENT;
+ if (IS_ERR(bh)) {
+ rc = PTR_ERR(bh);
} else {
- rc = 0;
+ struct inode *inode;
+
osd_id_gen(id, le32_to_cpu(de->inode), OSD_OII_NOGEN);
brelse(bh);
+ inode = osd_iget(oti, osd, id);
+ if (IS_ERR(inode)) {
+ rc = PTR_ERR(inode);
+ if (rc == -ESTALE)
+ rc = -ENOENT;
+ } else {
+ iput(inode);
+ rc = 0;
+ }
}
mutex_unlock(&parent->d_inode->i_mutex);
if (rc == 0)
/* to get subdir count from last_rcvd */
rc = osd_last_rcvd_subdir_count(dev);
- if (rc < 0) {
- OBD_FREE_PTR(dev->od_ost_map);
- RETURN(rc);
- }
+ if (rc < 0)
+ GOTO(cleanup_alloc, rc);
dev->od_ost_map->om_subdir_count = rc;
rc = 0;
- CFS_INIT_LIST_HEAD(&dev->od_ost_map->om_seq_list);
+ INIT_LIST_HEAD(&dev->od_ost_map->om_seq_list);
rwlock_init(&dev->od_ost_map->om_seq_list_lock);
mutex_init(&dev->od_ost_map->om_dir_init_mutex);
d = ll_lookup_one_len("O", rootd, strlen("O"));
if (IS_ERR(d))
- GOTO(cleanup, rc = PTR_ERR(d));
+ GOTO(cleanup_ctxt, rc = PTR_ERR(d));
if (d->d_inode == NULL) {
dput(d);
/* The lookup() may be called again inside simple_mkdir().
* mount time, it will not affect the whole performance. */
d = simple_mkdir(rootd, dev->od_mnt, "O", 0755, 1);
if (IS_ERR(d))
- GOTO(cleanup, rc = PTR_ERR(d));
+ GOTO(cleanup_ctxt, rc = PTR_ERR(d));
/* It is quite probably that the device is new formatted. */
dev->od_maybe_new = 1;
* has no OI mapping, and only is visible inside the OSD.*/
lu_igif_build(fid, inode->i_ino, inode->i_generation);
rc = osd_ea_fid_set(info, inode, fid,
- LMAC_NOT_IN_OI | LMAC_FID_ON_OST, 0);
+ LMAC_NOT_IN_OI | LMAC_FID_ON_OST, 0);
+ if (rc)
+ GOTO(cleanup_dentry, rc);
- GOTO(cleanup, rc);
+ pop_ctxt(&save, &new);
+ RETURN(0);
-cleanup:
+cleanup_dentry:
+ dput(d);
+cleanup_ctxt:
pop_ctxt(&save, &new);
- if (IS_ERR(d)) {
- OBD_FREE_PTR(dev->od_ost_map);
- RETURN(PTR_ERR(d));
- }
+cleanup_alloc:
+ OBD_FREE_PTR(dev->od_ost_map);
return rc;
}
-static void osd_seq_free(struct osd_obj_map *map,
- struct osd_obj_seq *osd_seq)
+static void osd_seq_free(struct osd_obj_seq *osd_seq)
{
int j;
- cfs_list_del_init(&osd_seq->oos_seq_list);
-
if (osd_seq->oos_dirs) {
for (j = 0; j < osd_seq->oos_subdir_count; j++) {
if (osd_seq->oos_dirs[j])
return;
write_lock(&map->om_seq_list_lock);
- cfs_list_for_each_entry_safe(osd_seq, tmp,
- &map->om_seq_list,
- oos_seq_list) {
- osd_seq_free(map, osd_seq);
+ list_for_each_entry_safe(osd_seq, tmp, &map->om_seq_list,
+ oos_seq_list) {
+ list_del_init(&osd_seq->oos_seq_list);
+ write_unlock(&map->om_seq_list_lock);
+ osd_seq_free(osd_seq);
+ write_lock(&map->om_seq_list_lock);
}
write_unlock(&map->om_seq_list_lock);
if (map->om_root)
/* prepare structures for MDS */
rc = osd_mdt_init(env, dev);
+ if (rc)
+ osd_ost_fini(dev);
RETURN(rc);
}
-struct osd_obj_seq *osd_seq_find_locked(struct osd_obj_map *map, obd_seq seq)
+static struct osd_obj_seq *osd_seq_find_locked(struct osd_obj_map *map, u64 seq)
{
struct osd_obj_seq *osd_seq;
- cfs_list_for_each_entry(osd_seq, &map->om_seq_list, oos_seq_list) {
+ list_for_each_entry(osd_seq, &map->om_seq_list, oos_seq_list) {
if (osd_seq->oos_seq == seq)
return osd_seq;
}
return NULL;
}
-struct osd_obj_seq *osd_seq_find(struct osd_obj_map *map, obd_seq seq)
+static struct osd_obj_seq *osd_seq_find(struct osd_obj_map *map, u64 seq)
{
struct osd_obj_seq *osd_seq;
ll_vfs_dq_init(parent);
mutex_lock(&parent->i_mutex);
bh = osd_ldiskfs_find_entry(parent, &child->d_name, &de, NULL, NULL);
- if (bh == NULL)
- GOTO(out, rc = -ENOENT);
+ if (IS_ERR(bh))
+ GOTO(out, rc = PTR_ERR(bh));
if (le32_to_cpu(de->inode) == id->oii_ino)
GOTO(out, rc = 1);
GOTO(out, rc);
}
+ /* The EA inode should NOT be in OI, old OI scrub may added
+ * such OI mapping by wrong, replace it. */
+ if (unlikely(osd_is_ea_inode(inode))) {
+ iput(inode);
+ goto update;
+ }
+
rc = osd_get_lma(info, inode, dentry, lma);
if (rc == -ENODATA) {
rc = osd_get_idif(info, inode, dentry, oi_fid);
- if (rc > 0) {
+ if (rc > 0 || rc == -ENODATA) {
oi_fid = NULL;
rc = 0;
}
}
iput(inode);
- /* If the OST-object has neither FID-in-LMA nor FID-in-ff, it is
- * either a crashed object or a uninitialized one. Replace it. */
- if (rc == -ENODATA || oi_fid == NULL)
- goto update;
-
if (rc != 0)
GOTO(out, rc);
- if (lu_fid_eq(fid, oi_fid)) {
+ /* If the OST-object has neither FID-in-LMA nor FID-in-ff, it is
+ * either a crashed object or a uninitialized one. Replace it. */
+ if (oi_fid != NULL && lu_fid_eq(fid, oi_fid)) {
CERROR("%s: the FID "DFID" is used by two objects: "
"%u/%u %u/%u\n", osd_name(osd), PFID(fid),
oi_id->oii_ino, oi_id->oii_gen,
GOTO(out, rc = -EEXIST);
}
+ if (fid_is_idif(fid) && oi_fid != NULL && fid_is_idif(oi_fid)) {
+ __u32 idx1 = fid_idif_ost_idx(fid);
+ __u32 idx2 = fid_idif_ost_idx(oi_fid);
+ struct ost_id *ostid = &info->oti_ostid;
+ struct lu_fid *tfid = &info->oti_fid3;
+
+ LASSERTF(idx1 == 0 || idx1 == osd->od_index,
+ "invalid given FID "DFID", not match the "
+ "device index %u\n", PFID(fid), osd->od_index);
+
+ if (idx1 != idx2) {
+ if (idx1 == 0 && idx2 == osd->od_index) {
+ fid_to_ostid(fid, ostid);
+ ostid_to_fid(tfid, ostid, idx2);
+ if (lu_fid_eq(tfid, oi_fid)) {
+ CERROR("%s: the FID "DFID" is used by "
+ "two objects(2): %u/%u %u/%u\n",
+ osd_name(osd), PFID(fid),
+ oi_id->oii_ino, oi_id->oii_gen,
+ id->oii_ino, id->oii_gen);
+
+ GOTO(out, rc = -EEXIST);
+ }
+ } else if (idx2 == 0 && idx1 == osd->od_index) {
+ fid_to_ostid(oi_fid, ostid);
+ ostid_to_fid(tfid, ostid, idx1);
+ if (lu_fid_eq(tfid, fid)) {
+ CERROR("%s: the FID "DFID" is used by "
+ "two objects(2): %u/%u %u/%u\n",
+ osd_name(osd), PFID(fid),
+ oi_id->oii_ino, oi_id->oii_gen,
+ id->oii_ino, id->oii_gen);
+
+ GOTO(out, rc = -EEXIST);
+ }
+ }
+ }
+ }
+
update:
/* There may be temporary inconsistency: On one hand, the new
* object may be referenced by multiple entries, which is out
GOTO(out, rc);
de->inode = cpu_to_le32(id->oii_ino);
- rc = ldiskfs_journal_dirty_metadata(th, bh);
+ rc = ldiskfs_handle_dirty_metadata(th, NULL, bh);
GOTO(out, rc);
out:
- brelse(bh);
+ if (!IS_ERR(bh))
+ brelse(bh);
mutex_unlock(&parent->i_mutex);
return rc;
}
ll_vfs_dq_init(dir);
mutex_lock(&dir->i_mutex);
- rc = -ENOENT;
bh = osd_ldiskfs_find_entry(dir, &child->d_name, &de, NULL, NULL);
- if (bh) {
+ if (IS_ERR(bh)) {
+ rc = PTR_ERR(bh);
+ } else {
rc = ldiskfs_delete_entry(th, dir, de, bh);
brelse(bh);
}
RETURN(rc);
}
-int osd_obj_add_entry(struct osd_thread_info *info,
- struct osd_device *osd,
- struct dentry *dir, char *name,
- const struct osd_inode_id *id,
- handle_t *th)
+static int osd_obj_add_entry(struct osd_thread_info *info,
+ struct osd_device *osd,
+ struct dentry *dir, char *name,
+ const struct osd_inode_id *id,
+ handle_t *th)
{
struct dentry *child;
struct inode *inode;
LASSERT(th != NULL);
LASSERT(th->h_transaction != NULL);
- inode = &info->oti_inode;
+ inode = info->oti_inode;
+ if (unlikely(inode == NULL)) {
+ struct ldiskfs_inode_info *lii;
+ OBD_ALLOC_PTR(lii);
+ if (lii == NULL)
+ RETURN(-ENOMEM);
+ inode = info->oti_inode = &lii->vfs_inode;
+ }
+
inode->i_sb = osd_sb(osd);
osd_id_to_inode(inode, id);
inode->i_mode = S_IFREG; /* for type in ldiskfs dir entry */
ll_vfs_dq_init(dir->d_inode);
mutex_lock(&dir->d_inode->i_mutex);
- rc = osd_ldiskfs_add_entry(th, child, inode, NULL);
+ rc = osd_ldiskfs_add_entry(info, osd, th, child, inode, NULL);
mutex_unlock(&dir->d_inode->i_mutex);
RETURN(rc);
}
/**
- * Use LPU64 for legacy OST sequences, but use LPX64i for new
+ * Use %llu for legacy OST sequences, but use %llx for new
* sequences names, so that the O/{seq}/dN/{oid} more closely
* follows the DFID/PFID format. This makes it easier to map from
* debug messages to objects in the future, and the legacy space
* of FID_SEQ_OST_MDT0 will be unused in the future.
**/
-static inline void osd_seq_name(char *seq_name, size_t name_size, obd_seq seq)
+static inline void osd_seq_name(char *seq_name, size_t name_size, u64 seq)
{
snprintf(seq_name, name_size,
(fid_seq_is_rsvd(seq) ||
- fid_seq_is_mdt0(seq)) ? LPU64 : LPX64i,
+ fid_seq_is_mdt0(seq)) ? "%llu" : "%llx",
fid_seq_is_idif(seq) ? 0 : seq);
}
static inline void osd_oid_name(char *name, size_t name_size,
- const struct lu_fid *fid, obd_id id)
+ const struct lu_fid *fid, u64 id)
{
snprintf(name, name_size,
(fid_seq_is_rsvd(fid_seq(fid)) ||
fid_seq_is_mdt0(fid_seq(fid)) ||
- fid_seq_is_idif(fid_seq(fid))) ? LPU64 : LPX64i, id);
+ fid_seq_is_idif(fid_seq(fid))) ? "%llu" : "%llx", id);
}
/* external locking is required */
}
static struct osd_obj_seq *osd_seq_load(struct osd_thread_info *info,
- struct osd_device *osd, obd_seq seq)
+ struct osd_device *osd, u64 seq)
{
struct osd_obj_map *map;
struct osd_obj_seq *osd_seq;
if (osd_seq == NULL)
GOTO(cleanup, rc = -ENOMEM);
- CFS_INIT_LIST_HEAD(&osd_seq->oos_seq_list);
+ INIT_LIST_HEAD(&osd_seq->oos_seq_list);
osd_seq->oos_seq = seq;
/* Init subdir count to be 32, but each seq can have
* different subdir count */
GOTO(cleanup, rc);
write_lock(&map->om_seq_list_lock);
- cfs_list_add(&osd_seq->oos_seq_list, &map->om_seq_list);
+ list_add(&osd_seq->oos_seq_list, &map->om_seq_list);
write_unlock(&map->om_seq_list_lock);
cleanup:
bh = osd_ldiskfs_find_entry(dir, &child->d_name, &de, NULL, NULL);
mutex_unlock(&dir->i_mutex);
- if (bh == NULL)
- RETURN(-ENOENT);
+ if (IS_ERR(bh))
+ RETURN(PTR_ERR(bh));
osd_id_gen(id, le32_to_cpu(de->inode), OSD_OII_NOGEN);
brelse(bh);
inode = osd_iget(info, dev, id);
- if (IS_ERR(inode))
- RETURN(PTR_ERR(inode));
+ if (IS_ERR(inode)) {
+ int rc = PTR_ERR(inode);
+
+ RETURN(rc == -ENOENT ? -ESTALE : rc);
+ }
iput(inode);
RETURN(0);
struct osd_obj_seq *osd_seq;
struct dentry *d;
struct ost_id *ostid = &info->oti_ostid;
- obd_id oid;
+ u64 oid;
int dirn, rc = 0;
char name[32];
ENTRY;
mutex_lock(&src_parent->i_mutex);
mutex_lock(&dir->i_mutex);
bh = osd_ldiskfs_find_entry(dir, &tgt_child->d_name, &de, NULL, NULL);
- if (bh != NULL) {
+ if (!IS_ERR(bh)) {
/* XXX: If some other object occupied the same slot. And If such
* inode is zero-sized and with SUID+SGID, then means it is
* a new created one. Maybe we can remove it and insert the
bh = osd_ldiskfs_find_entry(src_parent, &src_child->d_name, &de,
NULL, NULL);
- if (unlikely(bh == NULL))
- GOTO(unlock, rc = -ENOENT);
+ if (unlikely(IS_ERR(bh)))
+ GOTO(unlock, rc = PTR_ERR(bh));
rc = ldiskfs_delete_entry(jh, src_parent, de, bh);
brelse(bh);
if (rc != 0)
GOTO(unlock, rc);
- rc = osd_ldiskfs_add_entry(jh, tgt_child, inode, NULL);
+ rc = osd_ldiskfs_add_entry(info, osd, jh, tgt_child, inode, NULL);
GOTO(unlock, rc);