/*
* Copyright (c) 2009, 2010, Oracle and/or its affiliates. All rights reserved.
* Use is subject to license terms.
- */
-/*
- * Copyright (c) 2012, 2013, Intel Corporation.
- * Use is subject to license terms.
+ *
+ * Copyright (c) 2012, 2015, Intel Corporation.
*/
/*
* This file is part of Lustre, http://www.lustre.org/
#include <sys/sa_impl.h>
#include <sys/txg.h>
+static inline int osd_object_is_zap(dmu_buf_t *db)
+{
+ dmu_buf_impl_t *dbi = (dmu_buf_impl_t *) db;
+ dnode_t *dn;
+ int rc;
+
+ DB_DNODE_ENTER(dbi);
+ dn = DB_DNODE(dbi);
+ rc = (dn->dn_type == DMU_OT_DIRECTORY_CONTENTS ||
+ dn->dn_type == DMU_OT_USERGROUP_USED);
+ DB_DNODE_EXIT(dbi);
+
+ return rc;
+}
+
+/* We don't actually have direct access to the zap_hashbits() function
+ * so just pretend like we do for now. If this ever breaks we can look at
+ * it at that time. */
+#define zap_hashbits(zc) 48
+/*
+ * ZFS hash format:
+ * | cd (16 bits) | hash (48 bits) |
+ * we need it in other form:
+ * |0| hash (48 bit) | cd (15 bit) |
+ * to be a full 64-bit ordered hash so that Lustre readdir can use it to merge
+ * the readdir hashes from multiple directory stripes uniformly on the client.
+ * Another point is sign bit, the hash range should be in [0, 2^63-1] because
+ * loff_t (for llseek) needs to be a positive value. This means the "cd" field
+ * should only be the low 15 bits.
+ */
+uint64_t osd_zap_cursor_serialize(zap_cursor_t *zc)
+{
+ uint64_t zfs_hash = zap_cursor_serialize(zc) & (~0ULL >> 1);
+
+ return (zfs_hash >> zap_hashbits(zc)) |
+ (zfs_hash << (63 - zap_hashbits(zc)));
+}
+
+void osd_zap_cursor_init_serialized(zap_cursor_t *zc, struct objset *os,
+ uint64_t id, uint64_t dirhash)
+{
+ uint64_t zfs_hash = ((dirhash << zap_hashbits(zc)) & (~0ULL >> 1)) |
+ (dirhash >> (63 - zap_hashbits(zc)));
+
+ zap_cursor_init_serialized(zc, os, id, zfs_hash);
+}
+
+int osd_zap_cursor_init(zap_cursor_t **zc, struct objset *os,
+ uint64_t id, uint64_t dirhash)
+{
+ zap_cursor_t *t;
+
+ OBD_ALLOC_PTR(t);
+ if (unlikely(t == NULL))
+ return -ENOMEM;
+
+ osd_zap_cursor_init_serialized(t, os, id, dirhash);
+ *zc = t;
+
+ return 0;
+}
+
+void osd_zap_cursor_fini(zap_cursor_t *zc)
+{
+ zap_cursor_fini(zc);
+ OBD_FREE_PTR(zc);
+}
+
+static inline void osd_obj_cursor_init_serialized(zap_cursor_t *zc,
+ struct osd_object *o,
+ uint64_t dirhash)
+{
+ struct osd_device *d = osd_obj2dev(o);
+ osd_zap_cursor_init_serialized(zc, d->od_os,
+ o->oo_db->db_object, dirhash);
+}
+
+static inline int osd_obj_cursor_init(zap_cursor_t **zc, struct osd_object *o,
+ uint64_t dirhash)
+{
+ struct osd_device *d = osd_obj2dev(o);
+ return osd_zap_cursor_init(zc, d->od_os, o->oo_db->db_object, dirhash);
+}
+
static struct dt_it *osd_index_it_init(const struct lu_env *env,
struct dt_object *dt,
- __u32 unused,
- struct lustre_capa *capa)
+ __u32 unused)
{
struct osd_thread_info *info = osd_oti_get(env);
struct osd_zap_it *it;
struct osd_object *obj = osd_dt_obj(dt);
- struct osd_device *osd = osd_obj2dev(obj);
struct lu_object *lo = &dt->do_lu;
+ int rc;
ENTRY;
- /* XXX: check capa ? */
+ if (obj->oo_destroyed)
+ RETURN(ERR_PTR(-ENOENT));
LASSERT(lu_object_exists(lo));
LASSERT(obj->oo_db);
- LASSERT(udmu_object_is_zap(obj->oo_db));
+ LASSERT(osd_object_is_zap(obj->oo_db));
LASSERT(info);
- it = &info->oti_it_zap;
-
- if (udmu_zap_cursor_init(&it->ozi_zc, &osd->od_objset,
- obj->oo_db->db_object, 0))
+ OBD_SLAB_ALLOC_PTR_GFP(it, osd_zapit_cachep, GFP_NOFS);
+ if (it == NULL)
RETURN(ERR_PTR(-ENOMEM));
+ rc = osd_obj_cursor_init(&it->ozi_zc, obj, 0);
+ if (rc != 0) {
+ OBD_SLAB_FREE_PTR(it, osd_zapit_cachep);
+ RETURN(ERR_PTR(rc));
+ }
+
it->ozi_obj = obj;
- it->ozi_capa = capa;
it->ozi_reset = 1;
lu_object_get(lo);
static void osd_index_it_fini(const struct lu_env *env, struct dt_it *di)
{
- struct osd_zap_it *it = (struct osd_zap_it *)di;
- struct osd_object *obj;
+ struct osd_zap_it *it = (struct osd_zap_it *)di;
+ struct osd_object *obj;
ENTRY;
LASSERT(it);
obj = it->ozi_obj;
- udmu_zap_cursor_fini(it->ozi_zc);
+ osd_zap_cursor_fini(it->ozi_zc);
lu_object_put(env, &obj->oo_dt.do_lu);
+ OBD_SLAB_FREE_PTR(it, osd_zapit_cachep);
EXIT;
}
* next/finish. */
}
-int udmu_zap_cursor_retrieve_key(const struct lu_env *env,
- zap_cursor_t *zc, char *key, int max)
-{
- zap_attribute_t *za = &osd_oti_get(env)->oti_za;
- int err;
-
- if ((err = zap_cursor_retrieve(zc, za)))
- return err;
-
- if (key)
- strcpy(key, za->za_name);
-
- return 0;
-}
-
-/*
- * zap_cursor_retrieve read from current record.
- * to read bytes we need to call zap_lookup explicitly.
- */
-int udmu_zap_cursor_retrieve_value(const struct lu_env *env,
- zap_cursor_t *zc, char *buf,
- int buf_size, int *bytes_read)
-{
- zap_attribute_t *za = &osd_oti_get(env)->oti_za;
- int err, actual_size;
-
- if ((err = zap_cursor_retrieve(zc, za)))
- return err;
-
- if (za->za_integer_length <= 0)
- return (ERANGE);
-
- actual_size = za->za_integer_length * za->za_num_integers;
-
- if (actual_size > buf_size) {
- actual_size = buf_size;
- buf_size = actual_size / za->za_integer_length;
- } else {
- buf_size = za->za_num_integers;
- }
-
- err = -zap_lookup(zc->zc_objset, zc->zc_zapobj,
- za->za_name, za->za_integer_length,
- buf_size, buf);
-
- if (!err)
- *bytes_read = actual_size;
-
- return err;
-}
-
static inline void osd_it_append_attrs(struct lu_dirent *ent, __u32 attr,
int len, __u16 type)
{
struct dt_object *o,
struct lu_fid *fid)
{
+ struct osd_device *osd = osd_obj2dev(osd_dt_obj(o));
struct lustre_mdt_attrs *lma;
- udmu_objset_t *uos = &osd_obj2dev(osd_dt_obj(o))->od_objset;
struct lu_buf buf;
sa_handle_t *sa_hdl;
nvlist_t *nvbuf = NULL;
/* first of all, get parent dnode from own attributes */
LASSERT(osd_dt_obj(o)->oo_db);
- rc = -sa_handle_get(uos->os, osd_dt_obj(o)->oo_db->db_object,
+ rc = -sa_handle_get(osd->od_os, osd_dt_obj(o)->oo_db->db_object,
NULL, SA_HDL_PRIVATE, &sa_hdl);
if (rc)
RETURN(rc);
dnode = ZFS_NO_OBJECT;
- rc = -sa_lookup(sa_hdl, SA_ZPL_PARENT(uos), &dnode, 8);
+ rc = -sa_lookup(sa_hdl, SA_ZPL_PARENT(osd), &dnode, 8);
sa_handle_destroy(sa_hdl);
if (rc)
RETURN(rc);
/* now get EA buffer */
- rc = __osd_xattr_load(uos, dnode, &nvbuf);
+ rc = __osd_xattr_load(osd, dnode, &nvbuf);
if (rc)
GOTO(regular, rc);
/* no LMA attribute in SA, let's try regular EA */
/* first of all, get parent dnode storing regular EA */
- rc = -sa_handle_get(uos->os, dnode, NULL, SA_HDL_PRIVATE, &sa_hdl);
+ rc = -sa_handle_get(osd->od_os, dnode, NULL, SA_HDL_PRIVATE, &sa_hdl);
if (rc)
GOTO(out, rc);
dnode = ZFS_NO_OBJECT;
- rc = -sa_lookup(sa_hdl, SA_ZPL_XATTR(uos), &dnode, 8);
+ rc = -sa_lookup(sa_hdl, SA_ZPL_XATTR(osd), &dnode, 8);
sa_handle_destroy(sa_hdl);
if (rc)
GOTO(out, rc);
buf.lb_len = sizeof(osd_oti_get(env)->oti_buf);
/* now try to find LMA */
- rc = __osd_xattr_get_large(env, uos, dnode, &buf,
+ rc = __osd_xattr_get_large(env, osd, dnode, &buf,
XATTR_NAME_LMA, &size);
if (rc == 0 && size >= sizeof(*lma)) {
lma = buf.lb_buf;
buf.lb_buf = osd_oti_get(env)->oti_buf;
buf.lb_len = sizeof(osd_oti_get(env)->oti_buf);
- rc = osd_xattr_get(env, o, &buf, XATTR_NAME_LINK, BYPASS_CAPA);
+ rc = osd_xattr_get(env, o, &buf, XATTR_NAME_LINK);
if (rc == -ERANGE) {
- rc = osd_xattr_get(env, o, &LU_BUF_NULL,
- XATTR_NAME_LINK, BYPASS_CAPA);
+ rc = osd_xattr_get(env, o, &LU_BUF_NULL, XATTR_NAME_LINK);
if (rc < 0)
RETURN(rc);
LASSERT(rc > 0);
if (buf.lb_buf == NULL)
RETURN(-ENOMEM);
buf.lb_len = rc;
- rc = osd_xattr_get(env, o, &buf, XATTR_NAME_LINK, BYPASS_CAPA);
+ rc = osd_xattr_get(env, o, &buf, XATTR_NAME_LINK);
}
if (rc < 0)
GOTO(out, rc);
}
static int osd_dir_lookup(const struct lu_env *env, struct dt_object *dt,
- struct dt_rec *rec, const struct dt_key *key,
- struct lustre_capa *capa)
+ struct dt_rec *rec, const struct dt_key *key)
{
struct osd_thread_info *oti = osd_oti_get(env);
struct osd_object *obj = osd_dt_obj(dt);
int rc;
ENTRY;
- LASSERT(udmu_object_is_zap(obj->oo_db));
+ LASSERT(osd_object_is_zap(obj->oo_db));
if (name[0] == '.') {
if (name[1] == 0) {
}
}
- rc = -zap_lookup(osd->od_objset.os, obj->oo_db->db_object,
+ rc = -zap_lookup(osd->od_os, obj->oo_db->db_object,
(char *)key, 8, sizeof(oti->oti_zde) / 8,
(void *)&oti->oti_zde);
memcpy(rec, &oti->oti_zde.lzd_fid, sizeof(struct lu_fid));
{
struct osd_object *obj = osd_dt_obj(dt);
struct osd_thandle *oh;
+ uint64_t object;
ENTRY;
LASSERT(th != NULL);
oh = container_of0(th, struct osd_thandle, ot_super);
- LASSERT(obj->oo_db);
- LASSERT(udmu_object_is_zap(obj->oo_db));
+ /* This is for inserting dot/dotdot for new created dir. */
+ if (obj->oo_db == NULL)
+ object = DMU_NEW_OBJECT;
+ else
+ object = obj->oo_db->db_object;
- dmu_tx_hold_bonus(oh->ot_tx, obj->oo_db->db_object);
- dmu_tx_hold_zap(oh->ot_tx, obj->oo_db->db_object, TRUE, (char *)key);
+ dmu_tx_hold_bonus(oh->ot_tx, object);
+ dmu_tx_hold_zap(oh->ot_tx, object, TRUE, (char *)key);
RETURN(0);
}
lu_object_put(env, &obj->oo_dt.do_lu);
}
-static int osd_mdt_seq_exists(const struct lu_env *env, struct osd_device *osd,
- obd_seq seq)
+static int osd_seq_exists(const struct lu_env *env, struct osd_device *osd,
+ u64 seq)
{
struct lu_seq_range *range = &osd_oti_get(env)->oti_seq_range;
struct seq_server_site *ss = osd_seq_site(osd);
int rc;
ENTRY;
- if (ss == NULL)
- RETURN(1);
+ LASSERT(ss != NULL);
+ LASSERT(ss->ss_server_fld != NULL);
- /* XXX: currently, each MDT only store avaible sequence on disk,
- * and no allocated sequences information on disk, so it has to
- * lookup FLDB. It probably makes more sense also store allocated
- * sequence locally, so we do not need do remote FLDB lookup in OSD */
rc = osd_fld_lookup(env, osd, seq, range);
if (rc != 0) {
- CERROR("%s: Can not lookup fld for "LPX64"\n",
- osd_name(osd), seq);
+ if (rc != -ENOENT)
+ CERROR("%s: Can not lookup fld for "LPX64"\n",
+ osd_name(osd), seq);
RETURN(0);
}
}
static int osd_remote_fid(const struct lu_env *env, struct osd_device *osd,
- struct lu_fid *fid)
+ const struct lu_fid *fid)
{
+ struct seq_server_site *ss = osd_seq_site(osd);
ENTRY;
- if (!fid_is_norm(fid) && !fid_is_root(fid))
+ /* FID seqs not in FLDB, must be local seq */
+ if (unlikely(!fid_seq_in_fldb(fid_seq(fid))))
+ RETURN(0);
+
+ /* If FLD is not being initialized yet, it only happens during the
+ * initialization, likely during mgs initialization, and we assume
+ * this is local FID. */
+ if (ss == NULL || ss->ss_server_fld == NULL)
RETURN(0);
- /* Currently, it only used to check FID on MDT */
- if (osd_mdt_seq_exists(env, osd, fid_seq(fid)))
+ /* Only check the local FLDB here */
+ if (osd_seq_exists(env, osd, fid_seq(fid)))
RETURN(0);
RETURN(1);
* \param key key for index
* \param rec record reference
* \param th transaction handler
- * \param capa capability descriptor
* \param ignore_quota update should not affect quota
*
* \retval 0 success
*/
static int osd_dir_insert(const struct lu_env *env, struct dt_object *dt,
const struct dt_rec *rec, const struct dt_key *key,
- struct thandle *th, struct lustre_capa *capa,
- int ignore_quota)
+ struct thandle *th, int ignore_quota)
{
struct osd_thread_info *oti = osd_oti_get(env);
struct osd_object *parent = osd_dt_obj(dt);
struct osd_device *osd = osd_obj2dev(parent);
- struct lu_fid *fid = (struct lu_fid *)rec;
+ struct dt_insert_rec *rec1 = (struct dt_insert_rec *)rec;
+ const struct lu_fid *fid = rec1->rec_fid;
struct osd_thandle *oh;
struct osd_object *child = NULL;
__u32 attr;
ENTRY;
LASSERT(parent->oo_db);
- LASSERT(udmu_object_is_zap(parent->oo_db));
+ LASSERT(osd_object_is_zap(parent->oo_db));
LASSERT(dt_object_exists(dt));
LASSERT(osd_invariant(parent));
if (unlikely(rc == 1)) {
/* Insert remote entry */
memset(&oti->oti_zde.lzd_reg, 0, sizeof(oti->oti_zde.lzd_reg));
- oti->oti_zde.lzd_reg.zde_type = IFTODT(S_IFDIR & S_IFMT);
+ oti->oti_zde.lzd_reg.zde_type = IFTODT(rec1->rec_type & S_IFMT);
} else {
/*
* To simulate old Orion setups with ./.. stored in the
* during iteration */
GOTO(out, rc = 0);
} else if (name[1] == '.' && name[2] == 0) {
+ if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_PARENT)) {
+ struct lu_fid tfid = *fid;
+
+ osd_object_put(env, child);
+ tfid.f_oid--;
+ child = osd_object_find(env, dt, &tfid);
+ if (IS_ERR(child))
+ RETURN(PTR_ERR(child));
+
+ LASSERT(child->oo_db);
+ }
+
/* update parent dnode in the child.
* later it will be used to generate ".." */
- udmu_objset_t *uos = &osd->od_objset;
rc = osd_object_sa_update(parent,
- SA_ZPL_PARENT(uos),
+ SA_ZPL_PARENT(osd),
&child->oo_db->db_object,
8, oh);
+
GOTO(out, rc);
}
}
oti->oti_zde.lzd_fid = *fid;
/* Insert (key,oid) into ZAP */
- rc = -zap_add(osd->od_objset.os, parent->oo_db->db_object,
+ rc = -zap_add(osd->od_os, parent->oo_db->db_object,
(char *)key, 8, sizeof(oti->oti_zde) / 8,
(void *)&oti->oti_zde, oh->ot_tx);
+ if (unlikely(rc == -EEXIST &&
+ name[0] == '.' && name[1] == '.' && name[2] == 0))
+ /* Update (key,oid) in ZAP */
+ rc = -zap_update(osd->od_os, parent->oo_db->db_object,
+ (char *)key, 8, sizeof(oti->oti_zde) / 8,
+ (void *)&oti->oti_zde, oh->ot_tx);
out:
if (child != NULL)
const struct dt_key *key,
struct thandle *th)
{
- struct osd_object *obj = osd_dt_obj(dt);
+ struct osd_object *obj = osd_dt_obj(dt);
struct osd_thandle *oh;
+ uint64_t dnode;
ENTRY;
LASSERT(dt_object_exists(dt));
LASSERT(th != NULL);
oh = container_of0(th, struct osd_thandle, ot_super);
- LASSERT(obj->oo_db);
- LASSERT(udmu_object_is_zap(obj->oo_db));
-
- dmu_tx_hold_zap(oh->ot_tx, obj->oo_db->db_object, TRUE, (char *)key);
+ if (dt_object_exists(dt)) {
+ LASSERT(obj->oo_db);
+ LASSERT(osd_object_is_zap(obj->oo_db));
+ dnode = obj->oo_db->db_object;
+ } else {
+ dnode = DMU_NEW_OBJECT;
+ }
+ dmu_tx_hold_zap(oh->ot_tx, dnode, TRUE, (char *)key);
RETURN(0);
}
static int osd_dir_delete(const struct lu_env *env, struct dt_object *dt,
- const struct dt_key *key, struct thandle *th,
- struct lustre_capa *capa)
+ const struct dt_key *key, struct thandle *th)
{
struct osd_object *obj = osd_dt_obj(dt);
struct osd_device *osd = osd_obj2dev(obj);
int rc;
ENTRY;
- LASSERT(obj->oo_db);
- LASSERT(udmu_object_is_zap(obj->oo_db));
+ LASSERT(zap_db);
+ LASSERT(osd_object_is_zap(zap_db));
LASSERT(th != NULL);
oh = container_of0(th, struct osd_thandle, ot_super);
}
/* Remove key from the ZAP */
- rc = -zap_remove(osd->od_objset.os, zap_db->db_object,
+ rc = -zap_remove(osd->od_os, zap_db->db_object,
(char *) key, oh->ot_tx);
-#if LUSTRE_VERSION_CODE <= OBD_OCD_VERSION(2, 4, 53, 0)
- if (unlikely(rc == -ENOENT && name[0] == '.' &&
- (name[1] == 0 || (name[1] == '.' && name[2] == 0))))
- rc = 0;
-#endif
if (unlikely(rc && rc != -ENOENT))
CERROR("%s: zap_remove failed: rc = %d\n", osd->od_svname, rc);
static struct dt_it *osd_dir_it_init(const struct lu_env *env,
struct dt_object *dt,
- __u32 unused,
- struct lustre_capa *capa)
+ __u32 unused)
{
struct osd_zap_it *it;
- it = (struct osd_zap_it *)osd_index_it_init(env, dt, unused, capa);
+ it = (struct osd_zap_it *)osd_index_it_init(env, dt, unused);
if (!IS_ERR(it))
it->ozi_pos = 0;
{
struct osd_zap_it *it = (struct osd_zap_it *)di;
struct osd_object *obj = it->ozi_obj;
- struct osd_device *osd = osd_obj2dev(obj);
char *name = (char *)key;
int rc;
ENTRY;
LASSERT(it);
LASSERT(it->ozi_zc);
- udmu_zap_cursor_fini(it->ozi_zc);
-
- if (udmu_zap_cursor_init(&it->ozi_zc, &osd->od_objset,
- obj->oo_db->db_object, 0))
- RETURN(-ENOMEM);
+ /* reset the cursor */
+ zap_cursor_fini(it->ozi_zc);
+ osd_obj_cursor_init_serialized(it->ozi_zc, obj, 0);
/* XXX: implementation of the API is broken at the moment */
LASSERT(((const char *)key)[0] == 0);
zap_attribute_t *za = &osd_oti_get(env)->oti_za;
int rc;
+ ENTRY;
+
/* temp. storage should be enough for any key supported by ZFS */
CLASSERT(sizeof(za->za_name) <= sizeof(it->ozi_name));
it->ozi_pos++;
if (it->ozi_pos <=2)
RETURN(0);
- }
- zap_cursor_advance(it->ozi_zc);
+ } else {
+ zap_cursor_advance(it->ozi_zc);
+ }
/*
* According to current API we need to return error if its last entry.
strcpy(it->ozi_name, za->za_name);
-#if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 3, 90, 0)
- if (za->za_name[0] == '.') {
- if (za->za_name[1] == 0 || (za->za_name[1] == '.' &&
- za->za_name[2] == 0)) {
- /* we should not get onto . and ..
- * stored in the directory. ->next() and
- * other methods should prevent this
- */
- LBUG();
- }
- }
-#endif
-
RETURN((struct dt_key *)it->ozi_name);
}
if ((rc = -zap_cursor_retrieve(it->ozi_zc, za)) == 0)
rc = strlen(za->za_name);
-#if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 3, 90, 0)
- if (rc == 0 && za->za_name[0] == '.') {
- if (za->za_name[1] == 0 || (za->za_name[1] == '.' &&
- za->za_name[2] == 0)) {
- /* we should not get onto . and ..
- * stored in the directory. ->next() and
- * other methods should prevent this
- */
- LBUG();
- }
- }
-#endif
RETURN(rc);
}
osd_it_append_attrs(lde, attr, 2, IFTODT(S_IFDIR));
lde->lde_reclen = cpu_to_le16(lu_dirent_calc_size(2, attr));
rc = osd_find_parent_fid(env, &it->ozi_obj->oo_dt, &lde->lde_fid);
- /*
- * early Orion code was not setting LinkEA, so it's possible
- * some setups still have objects with no LinkEA set.
- * but at that time .. was a real record in the directory
- * so we should try to lookup .. in ZAP
- */
- if (rc != -ENOENT)
- GOTO(out, rc);
+
+ /* ENOENT happens at the root of filesystem so ignore it */
+ if (rc == -ENOENT)
+ rc = 0;
+ GOTO(out, rc);
}
LASSERT(lde);
- lde->lde_hash = cpu_to_le64(udmu_zap_cursor_serialize(it->ozi_zc));
-
- if ((rc = -zap_cursor_retrieve(it->ozi_zc, za)))
+ rc = -zap_cursor_retrieve(it->ozi_zc, za);
+ if (unlikely(rc != 0))
GOTO(out, rc);
+ lde->lde_hash = cpu_to_le64(osd_zap_cursor_serialize(it->ozi_zc));
namelen = strlen(za->za_name);
if (namelen > NAME_MAX)
GOTO(out, rc = -EOVERFLOW);
RETURN(rc);
}
+static int osd_dir_it_rec_size(const struct lu_env *env, const struct dt_it *di,
+ __u32 attr)
+{
+ struct osd_zap_it *it = (struct osd_zap_it *)di;
+ zap_attribute_t *za = &osd_oti_get(env)->oti_za;
+ size_t namelen = 0;
+ int rc;
+ ENTRY;
+
+ if (it->ozi_pos <= 1)
+ namelen = 1;
+ else if (it->ozi_pos == 2)
+ namelen = 2;
+
+ if (namelen > 0) {
+ rc = lu_dirent_calc_size(namelen, attr);
+ RETURN(rc);
+ }
+
+ rc = -zap_cursor_retrieve(it->ozi_zc, za);
+ if (unlikely(rc != 0))
+ RETURN(rc);
+
+ if (za->za_integer_length != 8 || za->za_num_integers < 3) {
+ CERROR("%s: unsupported direntry format: %d %d\n",
+ osd_obj2dev(it->ozi_obj)->od_svname,
+ za->za_integer_length, (int)za->za_num_integers);
+ RETURN(-EIO);
+ }
+
+ namelen = strlen(za->za_name);
+ if (namelen > NAME_MAX)
+ RETURN(-EOVERFLOW);
+
+ rc = lu_dirent_calc_size(namelen, attr);
+
+ RETURN(rc);
+}
+
static __u64 osd_dir_it_store(const struct lu_env *env, const struct dt_it *di)
{
struct osd_zap_it *it = (struct osd_zap_it *)di;
if (it->ozi_pos <= 2)
pos = it->ozi_pos;
else
- pos = udmu_zap_cursor_serialize(it->ozi_zc);
+ pos = osd_zap_cursor_serialize(it->ozi_zc);
RETURN(pos);
}
{
struct osd_zap_it *it = (struct osd_zap_it *)di;
struct osd_object *obj = it->ozi_obj;
- struct osd_device *osd = osd_obj2dev(obj);
zap_attribute_t *za = &osd_oti_get(env)->oti_za;
int rc;
ENTRY;
- udmu_zap_cursor_fini(it->ozi_zc);
- if (udmu_zap_cursor_init(&it->ozi_zc, &osd->od_objset,
- obj->oo_db->db_object, hash))
- RETURN(-ENOMEM);
+ /* reset the cursor */
+ zap_cursor_fini(it->ozi_zc);
+ osd_obj_cursor_init_serialized(it->ozi_zc, obj, hash);
if (hash <= 2) {
it->ozi_pos = hash;
RETURN(rc);
}
-static struct dt_index_operations osd_dir_ops = {
+struct dt_index_operations osd_dir_ops = {
.dio_lookup = osd_dir_lookup,
.dio_declare_insert = osd_declare_dir_insert,
.dio_insert = osd_dir_insert,
.key = osd_dir_it_key,
.key_size = osd_dir_it_key_size,
.rec = osd_dir_it_rec,
+ .rec_size = osd_dir_it_rec_size,
.store = osd_dir_it_store,
.load = osd_dir_it_load
}
* Primitives for index files using binary keys.
*/
-static int osd_prepare_key(struct osd_object *o, __u64 *dst,
- const struct dt_key *src)
+/* key integer_size is 8 */
+static int osd_prepare_key_uint64(struct osd_object *o, __u64 *dst,
+ const struct dt_key *src)
{
int size;
memset(dst + o->oo_keysize, 0, size - o->oo_keysize);
memcpy(dst, (const char *)src, o->oo_keysize);
- return size;
+ return (size/sizeof(__u64));
}
static int osd_index_lookup(const struct lu_env *env, struct dt_object *dt,
- struct dt_rec *rec, const struct dt_key *key,
- struct lustre_capa *capa)
+ struct dt_rec *rec, const struct dt_key *key)
{
struct osd_object *obj = osd_dt_obj(dt);
struct osd_device *osd = osd_obj2dev(obj);
int rc;
ENTRY;
- rc = osd_prepare_key(obj, k, key);
+ rc = osd_prepare_key_uint64(obj, k, key);
- rc = -zap_lookup_uint64(osd->od_objset.os, obj->oo_db->db_object,
+ rc = -zap_lookup_uint64(osd->od_os, obj->oo_db->db_object,
k, rc, obj->oo_recusize, obj->oo_recsize,
(void *)rec);
RETURN(rc == 0 ? 1 : rc);
static int osd_index_insert(const struct lu_env *env, struct dt_object *dt,
const struct dt_rec *rec, const struct dt_key *key,
- struct thandle *th, struct lustre_capa *capa,
- int ignore_quota)
+ struct thandle *th, int ignore_quota)
{
struct osd_object *obj = osd_dt_obj(dt);
struct osd_device *osd = osd_obj2dev(obj);
oh = container_of0(th, struct osd_thandle, ot_super);
- rc = osd_prepare_key(obj, k, key);
+ rc = osd_prepare_key_uint64(obj, k, key);
/* Insert (key,oid) into ZAP */
- rc = -zap_add_uint64(osd->od_objset.os, obj->oo_db->db_object,
+ rc = -zap_add_uint64(osd->od_os, obj->oo_db->db_object,
k, rc, obj->oo_recusize, obj->oo_recsize,
(void *)rec, oh->ot_tx);
RETURN(rc);
}
static int osd_index_delete(const struct lu_env *env, struct dt_object *dt,
- const struct dt_key *key, struct thandle *th,
- struct lustre_capa *capa)
+ const struct dt_key *key, struct thandle *th)
{
struct osd_object *obj = osd_dt_obj(dt);
struct osd_device *osd = osd_obj2dev(obj);
LASSERT(th != NULL);
oh = container_of0(th, struct osd_thandle, ot_super);
- rc = osd_prepare_key(obj, k, key);
+ rc = osd_prepare_key_uint64(obj, k, key);
/* Remove binary key from the ZAP */
- rc = -zap_remove_uint64(osd->od_objset.os, obj->oo_db->db_object,
+ rc = -zap_remove_uint64(osd->od_os, obj->oo_db->db_object,
k, rc, oh->ot_tx);
RETURN(rc);
}
* XXX: we need a binary version of zap_cursor_move_to_key()
* to implement this API */
if (*((const __u64 *)key) != 0)
- CERROR("NOT IMPLEMETED YET (move to %Lx)\n", *((__u64 *)key));
+ CERROR("NOT IMPLEMETED YET (move to "LPX64")\n",
+ *((__u64 *)key));
zap_cursor_fini(it->ozi_zc);
- memset(it->ozi_zc, 0, sizeof(*it->ozi_zc));
- zap_cursor_init(it->ozi_zc, osd->od_objset.os, obj->oo_db->db_object);
+ zap_cursor_init(it->ozi_zc, osd->od_os, obj->oo_db->db_object);
it->ozi_reset = 1;
RETURN(+1);
if (rc)
RETURN(rc);
- rc = osd_prepare_key(obj, k, (const struct dt_key *)za->za_name);
+ rc = osd_prepare_key_uint64(obj, k, (const struct dt_key *)za->za_name);
- rc = -zap_lookup_uint64(osd->od_objset.os, obj->oo_db->db_object,
+ rc = -zap_lookup_uint64(osd->od_os, obj->oo_db->db_object,
k, rc, obj->oo_recusize, obj->oo_recsize,
(void *)rec);
RETURN(rc);
int rc;
ENTRY;
- /* close the current cursor */
+ /* reset the cursor */
zap_cursor_fini(it->ozi_zc);
-
- /* create a new one starting at hash */
- memset(it->ozi_zc, 0, sizeof(*it->ozi_zc));
- zap_cursor_init_serialized(it->ozi_zc, osd->od_objset.os,
+ zap_cursor_init_serialized(it->ozi_zc, osd->od_os,
obj->oo_db->db_object, hash);
it->ozi_reset = 0;
}
};
+struct osd_metadnode_it {
+ struct osd_device *mit_dev;
+ __u64 mit_pos;
+ struct lu_fid mit_fid;
+ int mit_prefetched;
+ __u64 mit_prefetched_dnode;
+};
+
+static struct dt_it *osd_zfs_otable_it_init(const struct lu_env *env,
+ struct dt_object *dt, __u32 attr)
+{
+ struct osd_device *dev = osd_dev(dt->do_lu.lo_dev);
+ struct osd_metadnode_it *it;
+ ENTRY;
+
+ OBD_ALLOC_PTR(it);
+ if (unlikely(it == NULL))
+ RETURN(ERR_PTR(-ENOMEM));
+
+ it->mit_dev = dev;
+
+ /* XXX: dmu_object_next() does NOT find dnodes allocated
+ * in the current non-committed txg, so we force txg
+ * commit to find all existing dnodes ... */
+ txg_wait_synced(dmu_objset_pool(dev->od_os), 0ULL);
+
+ RETURN((struct dt_it *)it);
+}
+
+static void osd_zfs_otable_it_fini(const struct lu_env *env, struct dt_it *di)
+{
+ struct osd_metadnode_it *it = (struct osd_metadnode_it *)di;
+
+ OBD_FREE_PTR(it);
+}
+
+static int osd_zfs_otable_it_get(const struct lu_env *env,
+ struct dt_it *di, const struct dt_key *key)
+{
+ return 0;
+}
+
+static void osd_zfs_otable_it_put(const struct lu_env *env, struct dt_it *di)
+{
+}
+
+#define OTABLE_PREFETCH 256
+
+static void osd_zfs_otable_prefetch(const struct lu_env *env,
+ struct osd_metadnode_it *it)
+{
+ struct osd_device *dev = it->mit_dev;
+ int rc;
+
+ /* can go negative on the very first access to the iterator
+ * or if some non-Lustre objects were found */
+ if (unlikely(it->mit_prefetched < 0))
+ it->mit_prefetched = 0;
+
+ if (it->mit_prefetched >= (OTABLE_PREFETCH >> 1))
+ return;
+
+ if (it->mit_prefetched_dnode == 0)
+ it->mit_prefetched_dnode = it->mit_pos;
+
+ while (it->mit_prefetched < OTABLE_PREFETCH) {
+ rc = -dmu_object_next(dev->od_os, &it->mit_prefetched_dnode,
+ B_FALSE, 0);
+ if (unlikely(rc != 0))
+ break;
+
+ /* dmu_prefetch() was exported in 0.6.2, if you use with
+ * an older release, just comment it out - this is an
+ * optimization */
+ dmu_prefetch(dev->od_os, it->mit_prefetched_dnode, 0, 0);
+
+ it->mit_prefetched++;
+ }
+}
+
+static int osd_zfs_otable_it_next(const struct lu_env *env, struct dt_it *di)
+{
+ struct osd_metadnode_it *it = (struct osd_metadnode_it *)di;
+ struct lustre_mdt_attrs *lma;
+ struct osd_device *dev = it->mit_dev;
+ nvlist_t *nvbuf = NULL;
+ uchar_t *v;
+ __u64 dnode;
+ int rc, s;
+
+ memset(&it->mit_fid, 0, sizeof(it->mit_fid));
+
+ dnode = it->mit_pos;
+ do {
+ rc = -dmu_object_next(dev->od_os, &it->mit_pos, B_FALSE, 0);
+ if (unlikely(rc != 0))
+ GOTO(out, rc = 1);
+ it->mit_prefetched--;
+
+ /* LMA is required for this to be a Lustre object.
+ * If there is no xattr skip it. */
+ rc = __osd_xattr_load(dev, it->mit_pos, &nvbuf);
+ if (unlikely(rc != 0))
+ continue;
+
+ LASSERT(nvbuf != NULL);
+ rc = -nvlist_lookup_byte_array(nvbuf, XATTR_NAME_LMA, &v, &s);
+ if (likely(rc == 0)) {
+ /* Lustre object */
+ lma = (struct lustre_mdt_attrs *)v;
+ lustre_lma_swab(lma);
+ it->mit_fid = lma->lma_self_fid;
+ nvlist_free(nvbuf);
+ break;
+ } else {
+ /* not a Lustre object, try next one */
+ nvlist_free(nvbuf);
+ }
+
+ } while (1);
+
+
+ /* we aren't prefetching in the above loop because the number of
+ * non-Lustre objects is very small and we will be repeating very
+ * rare. in case we want to use this to iterate over non-Lustre
+ * objects (i.e. when we convert regular ZFS in Lustre) it makes
+ * sense to initiate prefetching in the loop */
+
+ /* 0 - there are more items, +1 - the end */
+ if (likely(rc == 0))
+ osd_zfs_otable_prefetch(env, it);
+
+ CDEBUG(D_OTHER, "advance: %llu -> %llu "DFID": %d\n", dnode,
+ it->mit_pos, PFID(&it->mit_fid), rc);
+
+out:
+ return rc;
+}
+
+static struct dt_key *osd_zfs_otable_it_key(const struct lu_env *env,
+ const struct dt_it *di)
+{
+ return NULL;
+}
+
+static int osd_zfs_otable_it_key_size(const struct lu_env *env,
+ const struct dt_it *di)
+{
+ return sizeof(__u64);
+}
+
+static int osd_zfs_otable_it_rec(const struct lu_env *env,
+ const struct dt_it *di,
+ struct dt_rec *rec, __u32 attr)
+{
+ struct osd_metadnode_it *it = (struct osd_metadnode_it *)di;
+ struct lu_fid *fid = (struct lu_fid *)rec;
+ ENTRY;
+
+ *fid = it->mit_fid;
+
+ RETURN(0);
+}
+
+
+static __u64 osd_zfs_otable_it_store(const struct lu_env *env,
+ const struct dt_it *di)
+{
+ struct osd_metadnode_it *it = (struct osd_metadnode_it *)di;
+
+ return it->mit_pos;
+}
+
+static int osd_zfs_otable_it_load(const struct lu_env *env,
+ const struct dt_it *di, __u64 hash)
+{
+ struct osd_metadnode_it *it = (struct osd_metadnode_it *)di;
+
+ it->mit_pos = hash;
+ it->mit_prefetched = 0;
+ it->mit_prefetched_dnode = 0;
+
+ return osd_zfs_otable_it_next(env, (struct dt_it *)di);
+}
+
+static int osd_zfs_otable_it_key_rec(const struct lu_env *env,
+ const struct dt_it *di, void *key_rec)
+{
+ return 0;
+}
+
+const struct dt_index_operations osd_zfs_otable_ops = {
+ .dio_it = {
+ .init = osd_zfs_otable_it_init,
+ .fini = osd_zfs_otable_it_fini,
+ .get = osd_zfs_otable_it_get,
+ .put = osd_zfs_otable_it_put,
+ .next = osd_zfs_otable_it_next,
+ .key = osd_zfs_otable_it_key,
+ .key_size = osd_zfs_otable_it_key_size,
+ .rec = osd_zfs_otable_it_rec,
+ .store = osd_zfs_otable_it_store,
+ .load = osd_zfs_otable_it_load,
+ .key_rec = osd_zfs_otable_it_key_rec,
+ }
+};
+
int osd_index_try(const struct lu_env *env, struct dt_object *dt,
const struct dt_index_features *feat)
{
struct osd_object *obj = osd_dt_obj(dt);
ENTRY;
- LASSERT(dt_object_exists(dt));
-
/*
* XXX: implement support for fixed-size keys sorted with natural
* numerical way (not using internal hash value)
if (feat->dif_flags & DT_IND_RANGE)
RETURN(-ERANGE);
- if (unlikely(feat == &dt_otable_features))
- /* do not support oi scrub yet. */
- RETURN(-ENOTSUPP);
+ if (unlikely(feat == &dt_otable_features)) {
+ dt->do_index_ops = &osd_zfs_otable_ops;
+ RETURN(0);
+ }
- LASSERT(obj->oo_db != NULL);
+ LASSERT(!dt_object_exists(dt) || obj->oo_db != NULL);
if (likely(feat == &dt_directory_features)) {
- if (udmu_object_is_zap(obj->oo_db))
+ if (!dt_object_exists(dt) || osd_object_is_zap(obj->oo_db))
dt->do_index_ops = &osd_dir_ops;
else
RETURN(-ENOTDIR);
} else if (unlikely(feat == &dt_acct_features)) {
LASSERT(fid_is_acct(lu_object_fid(&dt->do_lu)));
dt->do_index_ops = &osd_acct_index_ops;
- } else if (udmu_object_is_zap(obj->oo_db) &&
- dt->do_index_ops == NULL) {
+ } else if (dt->do_index_ops == NULL) {
/* For index file, we don't support variable key & record sizes
* and the key has to be unique */
if ((feat->dif_flags & ~DT_IND_UPDATE) != 0)
RETURN(0);
}
-