int osp_init_precreate(struct osp_device *d)
{
struct l_wait_info lwi = { 0 };
struct task_struct *task;
ENTRY;
OBD_ALLOC_PTR(d->opd_pre);
if (d->opd_pre == NULL)
RETURN(-ENOMEM);
/* initially precreation isn't ready */
d->opd_pre_status = -EAGAIN;
fid_zero(&d->opd_pre_used_fid);
d->opd_pre_used_fid.f_oid = 1;
fid_zero(&d->opd_pre_last_created_fid);
d->opd_pre_last_created_fid.f_oid = 1;
d->opd_pre_reserved = 0;
d->opd_got_disconnected = 1;
d->opd_pre_grow_slow = 0;
d->opd_pre_grow_count = OST_MIN_PRECREATE;
d->opd_pre_min_grow_count = OST_MIN_PRECREATE;
d->opd_pre_max_grow_count = OST_MAX_PRECREATE;
spin_lock_init(&d->opd_pre_lock);
init_waitqueue_head(&d->opd_pre_waitq);
init_waitqueue_head(&d->opd_pre_user_waitq);
init_waitqueue_head(&d->opd_pre_thread.t_ctl_waitq);
/*
* Initialize statfs-related things
*/
d->opd_statfs_maxage = 5; /* default update interval */
d->opd_statfs_fresh_till = cfs_time_shift(-1000);
CDEBUG(D_OTHER, "current %llu, fresh till %llu\n",
(unsigned long long)cfs_time_current(),
(unsigned long long)d->opd_statfs_fresh_till);
cfs_timer_init(&d->opd_statfs_timer, osp_statfs_timer_cb, d);
/*
* start thread handling precreation and statfs updates
*/
task = kthread_run(osp_precreate_thread, d,
"osp-pre-%u-%u", d->opd_index, d->opd_group);
if (IS_ERR(task)) {
CERROR("can't start precreate thread %ld\n", PTR_ERR(task));
RETURN(PTR_ERR(task));
}
l_wait_event(d->opd_pre_thread.t_ctl_waitq,
osp_precreate_running(d) || osp_precreate_stopped(d),
&lwi);
RETURN(0);
}
int osp_init_precreate(struct osp_device *d)
{
struct l_wait_info lwi = { 0 };
int rc;
ENTRY;
/* initially precreation isn't ready */
d->opd_pre_status = -EAGAIN;
fid_zero(&d->opd_pre_used_fid);
d->opd_pre_used_fid.f_oid = 1;
fid_zero(&d->opd_pre_last_created_fid);
d->opd_pre_last_created_fid.f_oid = 1;
d->opd_pre_reserved = 0;
d->opd_got_disconnected = 1;
d->opd_pre_grow_slow = 0;
d->opd_pre_grow_count = OST_MIN_PRECREATE;
d->opd_pre_min_grow_count = OST_MIN_PRECREATE;
d->opd_pre_max_grow_count = OST_MAX_PRECREATE;
spin_lock_init(&d->opd_pre_lock);
cfs_waitq_init(&d->opd_pre_waitq);
cfs_waitq_init(&d->opd_pre_user_waitq);
cfs_waitq_init(&d->opd_pre_thread.t_ctl_waitq);
/*
* Initialize statfs-related things
*/
d->opd_statfs_maxage = 5; /* default update interval */
d->opd_statfs_fresh_till = cfs_time_shift(-1000);
CDEBUG(D_OTHER, "current %llu, fresh till %llu\n",
(unsigned long long)cfs_time_current(),
(unsigned long long)d->opd_statfs_fresh_till);
cfs_timer_init(&d->opd_statfs_timer, osp_statfs_timer_cb, d);
/*
* start thread handling precreation and statfs updates
*/
rc = cfs_create_thread(osp_precreate_thread, d, 0);
if (rc < 0) {
CERROR("can't start precreate thread %d\n", rc);
RETURN(rc);
}
l_wait_event(d->opd_pre_thread.t_ctl_waitq,
osp_precreate_running(d) || osp_precreate_stopped(d),
&lwi);
RETURN(0);
}
/*
* Finish the current sequence due to disconnect.
* See mdc_import_event()
*/
void seq_client_flush(struct lu_client_seq *seq)
{
wait_queue_t link;
LASSERT(seq != NULL);
init_waitqueue_entry(&link, current);
mutex_lock(&seq->lcs_mutex);
while (seq->lcs_update) {
add_wait_queue(&seq->lcs_waitq, &link);
set_current_state(TASK_UNINTERRUPTIBLE);
mutex_unlock(&seq->lcs_mutex);
schedule();
mutex_lock(&seq->lcs_mutex);
remove_wait_queue(&seq->lcs_waitq, &link);
set_current_state(TASK_RUNNING);
}
fid_zero(&seq->lcs_fid);
/**
* this id shld not be used for seq range allocation.
* set to -1 for dgb check.
*/
seq->lcs_space.lsr_index = -1;
lu_seq_range_init(&seq->lcs_space);
mutex_unlock(&seq->lcs_mutex);
}
int ofd_object_ff_load(const struct lu_env *env, struct ofd_object *fo)
{
struct ofd_thread_info *info = ofd_info(env);
struct filter_fid_old *ff = &info->fti_mds_fid_old;
struct lu_buf *buf = &info->fti_buf;
struct lu_fid *pfid = &fo->ofo_pfid;
int rc = 0;
if (fid_is_sane(pfid))
return 0;
buf->lb_buf = ff;
buf->lb_len = sizeof(*ff);
rc = dt_xattr_get(env, ofd_object_child(fo), buf, XATTR_NAME_FID,
BYPASS_CAPA);
if (rc < 0)
return rc;
if (rc < sizeof(struct lu_fid)) {
fid_zero(pfid);
return -ENODATA;
}
pfid->f_seq = le64_to_cpu(ff->ff_parent.f_seq);
pfid->f_oid = le32_to_cpu(ff->ff_parent.f_oid);
/* Currently, the filter_fid::ff_parent::f_ver is not the real parent
* MDT-object's FID::f_ver, instead it is the OST-object index in its
* parent MDT-object's layout EA. */
pfid->f_stripe_idx = le32_to_cpu(ff->ff_parent.f_stripe_idx);
return 0;
}
/**
* open the PENDING directory for device \a mdd
*
* The PENDING directory persistently tracks files and directories that were
* unlinked from the namespace (nlink == 0) but are still held open by clients.
* Those inodes shouldn't be deleted if the MDS crashes, because the clients
* would not be able to recover and reopen those files. Instead, these inodes
* are linked into the PENDING directory on disk, and only deleted if all
* clients close them, or the MDS finishes client recovery without any client
* reopening them (i.e. former clients didn't join recovery).
* \param d mdd device being started.
*
* \retval 0 success
* \retval -ve index operation error.
*
*/
int orph_index_init(const struct lu_env *env, struct mdd_device *mdd)
{
struct lu_fid fid;
struct dt_object *d;
int rc = 0;
ENTRY;
/* create PENDING dir */
fid_zero(&fid);
rc = mdd_local_file_create(env, mdd, &mdd->mdd_local_root_fid,
orph_index_name, S_IFDIR | S_IRUGO |
S_IWUSR | S_IXUGO, &fid);
if (rc < 0)
RETURN(rc);
d = dt_locate(env, mdd->mdd_child, &fid);
if (IS_ERR(d))
RETURN(PTR_ERR(d));
LASSERT(lu_object_exists(&d->do_lu));
if (!dt_try_as_dir(env, d)) {
CERROR("%s: \"%s\" is not an index: rc = %d\n",
mdd2obd_dev(mdd)->obd_name, orph_index_name, rc);
lu_object_put(env, &d->do_lu);
RETURN(-ENOTDIR);
}
mdd->mdd_orphans = d;
RETURN(0);
}
void llu_prep_md_op_data(struct md_op_data *op_data, struct inode *i1,
struct inode *i2, const char *name, int namelen,
int mode, __u32 opc)
{
LASSERT(i1 != NULL || i2 != NULL);
LASSERT(op_data);
if (i1) {
ll_i2gids(op_data->op_suppgids, i1, i2);
op_data->op_fid1 = *ll_inode2fid(i1);
}else {
ll_i2gids(op_data->op_suppgids, i2, i1);
op_data->op_fid1 = *ll_inode2fid(i2);
}
if (i2)
op_data->op_fid2 = *ll_inode2fid(i2);
else
fid_zero(&op_data->op_fid2);
op_data->op_opc = opc;
op_data->op_name = name;
op_data->op_mode = mode;
op_data->op_namelen = namelen;
op_data->op_mod_time = CFS_CURRENT_TIME;
op_data->op_data = NULL;
}
/*
* Finish the current sequence due to disconnect.
* See mdc_import_event()
*/
void seq_client_flush(struct lu_client_seq *seq)
{
cfs_waitlink_t link;
LASSERT(seq != NULL);
cfs_waitlink_init(&link);
mutex_lock(&seq->lcs_mutex);
while (seq->lcs_update) {
cfs_waitq_add(&seq->lcs_waitq, &link);
cfs_set_current_state(CFS_TASK_UNINT);
mutex_unlock(&seq->lcs_mutex);
cfs_waitq_wait(&link, CFS_TASK_UNINT);
mutex_lock(&seq->lcs_mutex);
cfs_waitq_del(&seq->lcs_waitq, &link);
cfs_set_current_state(CFS_TASK_RUNNING);
}
fid_zero(&seq->lcs_fid);
/**
* this id shld not be used for seq range allocation.
* set to -1 for dgb check.
*/
seq->lcs_space.lsr_index = -1;
range_init(&seq->lcs_space);
mutex_unlock(&seq->lcs_mutex);
}
int osp_init_pre_fid(struct osp_device *osp)
{
struct lu_env env;
struct osp_thread_info *osi;
struct lu_client_seq *cli_seq;
struct lu_fid *last_fid;
int rc;
ENTRY;
LASSERT(osp->opd_pre != NULL);
/* Return if last_used fid has been initialized */
if (!fid_is_zero(&osp->opd_last_used_fid))
RETURN(0);
rc = lu_env_init(&env, osp->opd_dt_dev.dd_lu_dev.ld_type->ldt_ctx_tags);
if (rc) {
CERROR("%s: init env error: rc = %d\n",
osp->opd_obd->obd_name, rc);
RETURN(rc);
}
osi = osp_env_info(&env);
last_fid = &osi->osi_fid;
fid_zero(last_fid);
/* For a freshed fs, it will allocate a new sequence first */
if (osp_is_fid_client(osp) && osp->opd_group != 0) {
cli_seq = osp->opd_obd->u.cli.cl_seq;
rc = seq_client_get_seq(&env, cli_seq, &last_fid->f_seq);
if (rc != 0) {
CERROR("%s: alloc fid error: rc = %d\n",
osp->opd_obd->obd_name, rc);
GOTO(out, rc);
}
} else {
last_fid->f_seq = fid_idif_seq(0, osp->opd_index);
}
last_fid->f_oid = 1;
last_fid->f_ver = 0;
spin_lock(&osp->opd_pre_lock);
osp->opd_last_used_fid = *last_fid;
osp->opd_pre_used_fid = *last_fid;
osp->opd_pre_last_created_fid = *last_fid;
spin_unlock(&osp->opd_pre_lock);
rc = osp_write_last_oid_seq_files(&env, osp, last_fid, 1);
if (rc != 0) {
CERROR("%s: write fid error: rc = %d\n",
osp->opd_obd->obd_name, rc);
GOTO(out, rc);
}
out:
lu_env_fini(&env);
RETURN(rc);
}
/*
* Handler for: getattr, lookup and revalidate cases.
*/
int lmv_intent_lookup(struct obd_export *exp, struct md_op_data *op_data,
void *lmm, int lmmsize, struct lookup_intent *it,
int flags, struct ptlrpc_request **reqp,
ldlm_blocking_callback cb_blocking,
__u64 extra_lock_flags)
{
struct obd_device *obd = exp->exp_obd;
struct lmv_obd *lmv = &obd->u.lmv;
struct lmv_tgt_desc *tgt = NULL;
struct mdt_body *body;
int rc = 0;
tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid1);
if (IS_ERR(tgt))
return PTR_ERR(tgt);
if (!fid_is_sane(&op_data->op_fid2))
fid_zero(&op_data->op_fid2);
CDEBUG(D_INODE, "LOOKUP_INTENT with fid1="DFID", fid2="DFID
", name='%s' -> mds #%d\n", PFID(&op_data->op_fid1),
PFID(&op_data->op_fid2),
op_data->op_name ? op_data->op_name : "<NULL>",
tgt->ltd_idx);
op_data->op_bias &= ~MDS_CROSS_REF;
rc = md_intent_lock(tgt->ltd_exp, op_data, lmm, lmmsize, it,
flags, reqp, cb_blocking, extra_lock_flags);
if (rc < 0 || *reqp == NULL)
return rc;
/*
* MDS has returned success. Probably name has been resolved in
* remote inode. Let's check this.
*/
body = req_capsule_server_get(&(*reqp)->rq_pill, &RMF_MDT_BODY);
if (body == NULL)
return -EPROTO;
/* Not cross-ref case, just get out of here. */
if (likely(!(body->valid & OBD_MD_MDS)))
return 0;
rc = lmv_intent_remote(exp, lmm, lmmsize, it, NULL, flags, reqp,
cb_blocking, extra_lock_flags);
return rc;
}
int osd_oi_delete(struct osd_thread_info *info,
struct osd_device *osd, const struct lu_fid *fid,
handle_t *th, enum oi_check_flags flags)
{
struct lu_fid *oi_fid = &info->oti_fid2;
/* clear idmap cache */
if (lu_fid_eq(fid, &info->oti_cache.oic_fid))
fid_zero(&info->oti_cache.oic_fid);
if (fid_is_last_id(fid))
return 0;
if (fid_is_on_ost(info, osd, fid, flags) || fid_is_llog(fid))
return osd_obj_map_delete(info, osd, fid, th);
fid_cpu_to_be(oi_fid, fid);
return osd_oi_iam_delete(info, osd_fid2oi(osd, fid),
(const struct dt_key *)oi_fid, th);
}
static int
lfsck_namespace_dump(const struct lu_env *env, struct lfsck_component *com,
char *buf, int len)
{
struct lfsck_instance *lfsck = com->lc_lfsck;
struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
struct lfsck_namespace *ns = com->lc_file_ram;
int save = len;
int ret = -ENOSPC;
int rc;
down_read(&com->lc_sem);
rc = snprintf(buf, len,
"name: lfsck_namespace\n"
"magic: %#x\n"
"version: %d\n"
"status: %s\n",
ns->ln_magic,
bk->lb_version,
lfsck_status2names(ns->ln_status));
if (rc <= 0)
goto out;
buf += rc;
len -= rc;
rc = lfsck_bits_dump(&buf, &len, ns->ln_flags, lfsck_flags_names,
"flags");
if (rc < 0)
goto out;
rc = lfsck_bits_dump(&buf, &len, bk->lb_param, lfsck_param_names,
"param");
if (rc < 0)
goto out;
rc = lfsck_time_dump(&buf, &len, ns->ln_time_last_complete,
"time_since_last_completed");
if (rc < 0)
goto out;
rc = lfsck_time_dump(&buf, &len, ns->ln_time_latest_start,
"time_since_latest_start");
if (rc < 0)
goto out;
rc = lfsck_time_dump(&buf, &len, ns->ln_time_last_checkpoint,
"time_since_last_checkpoint");
if (rc < 0)
goto out;
rc = lfsck_pos_dump(&buf, &len, &ns->ln_pos_latest_start,
"latest_start_position");
if (rc < 0)
goto out;
rc = lfsck_pos_dump(&buf, &len, &ns->ln_pos_last_checkpoint,
"last_checkpoint_position");
if (rc < 0)
goto out;
rc = lfsck_pos_dump(&buf, &len, &ns->ln_pos_first_inconsistent,
"first_failure_position");
if (rc < 0)
goto out;
if (ns->ln_status == LS_SCANNING_PHASE1) {
struct lfsck_position pos;
const struct dt_it_ops *iops;
cfs_duration_t duration = cfs_time_current() -
lfsck->li_time_last_checkpoint;
__u64 checked = ns->ln_items_checked + com->lc_new_checked;
__u64 speed = checked;
__u64 new_checked = com->lc_new_checked * HZ;
__u32 rtime = ns->ln_run_time_phase1 +
cfs_duration_sec(duration + HALF_SEC);
if (duration != 0)
do_div(new_checked, duration);
if (rtime != 0)
do_div(speed, rtime);
rc = snprintf(buf, len,
"checked_phase1: "LPU64"\n"
"checked_phase2: "LPU64"\n"
"updated_phase1: "LPU64"\n"
"updated_phase2: "LPU64"\n"
"failed_phase1: "LPU64"\n"
"failed_phase2: "LPU64"\n"
"dirs: "LPU64"\n"
"M-linked: "LPU64"\n"
"nlinks_repaired: "LPU64"\n"
"lost_found: "LPU64"\n"
"success_count: %u\n"
"run_time_phase1: %u seconds\n"
"run_time_phase2: %u seconds\n"
"average_speed_phase1: "LPU64" items/sec\n"
"average_speed_phase2: N/A\n"
"real-time_speed_phase1: "LPU64" items/sec\n"
"real-time_speed_phase2: N/A\n",
checked,
ns->ln_objs_checked_phase2,
ns->ln_items_repaired,
ns->ln_objs_repaired_phase2,
ns->ln_items_failed,
ns->ln_objs_failed_phase2,
ns->ln_dirs_checked,
ns->ln_mlinked_checked,
ns->ln_objs_nlink_repaired,
ns->ln_objs_lost_found,
ns->ln_success_count,
rtime,
ns->ln_run_time_phase2,
speed,
new_checked);
if (rc <= 0)
goto out;
buf += rc;
len -= rc;
LASSERT(lfsck->li_di_oit != NULL);
iops = &lfsck->li_obj_oit->do_index_ops->dio_it;
/* The low layer otable-based iteration position may NOT
* exactly match the namespace-based directory traversal
* cookie. Generally, it is not a serious issue. But the
* caller should NOT make assumption on that. */
pos.lp_oit_cookie = iops->store(env, lfsck->li_di_oit);
if (!lfsck->li_current_oit_processed)
pos.lp_oit_cookie--;
spin_lock(&lfsck->li_lock);
if (lfsck->li_di_dir != NULL) {
pos.lp_dir_cookie = lfsck->li_cookie_dir;
if (pos.lp_dir_cookie >= MDS_DIR_END_OFF) {
fid_zero(&pos.lp_dir_parent);
pos.lp_dir_cookie = 0;
} else {
pos.lp_dir_parent =
*lfsck_dto2fid(lfsck->li_obj_dir);
}
} else {
fid_zero(&pos.lp_dir_parent);
pos.lp_dir_cookie = 0;
}
spin_unlock(&lfsck->li_lock);
rc = lfsck_pos_dump(&buf, &len, &pos, "current_position");
if (rc <= 0)
goto out;
} else if (ns->ln_status == LS_SCANNING_PHASE2) {
cfs_duration_t duration = cfs_time_current() -
lfsck->li_time_last_checkpoint;
__u64 checked = ns->ln_objs_checked_phase2 +
com->lc_new_checked;
__u64 speed1 = ns->ln_items_checked;
__u64 speed2 = checked;
__u64 new_checked = com->lc_new_checked * HZ;
__u32 rtime = ns->ln_run_time_phase2 +
cfs_duration_sec(duration + HALF_SEC);
if (duration != 0)
do_div(new_checked, duration);
if (ns->ln_run_time_phase1 != 0)
do_div(speed1, ns->ln_run_time_phase1);
if (rtime != 0)
do_div(speed2, rtime);
rc = snprintf(buf, len,
"checked_phase1: "LPU64"\n"
"checked_phase2: "LPU64"\n"
"updated_phase1: "LPU64"\n"
"updated_phase2: "LPU64"\n"
"failed_phase1: "LPU64"\n"
"failed_phase2: "LPU64"\n"
"dirs: "LPU64"\n"
"M-linked: "LPU64"\n"
"nlinks_repaired: "LPU64"\n"
"lost_found: "LPU64"\n"
"success_count: %u\n"
"run_time_phase1: %u seconds\n"
"run_time_phase2: %u seconds\n"
"average_speed_phase1: "LPU64" items/sec\n"
"average_speed_phase2: "LPU64" objs/sec\n"
"real-time_speed_phase1: N/A\n"
"real-time_speed_phase2: "LPU64" objs/sec\n"
"current_position: "DFID"\n",
ns->ln_items_checked,
checked,
ns->ln_items_repaired,
ns->ln_objs_repaired_phase2,
ns->ln_items_failed,
ns->ln_objs_failed_phase2,
ns->ln_dirs_checked,
ns->ln_mlinked_checked,
ns->ln_objs_nlink_repaired,
ns->ln_objs_lost_found,
ns->ln_success_count,
ns->ln_run_time_phase1,
rtime,
speed1,
speed2,
new_checked,
PFID(&ns->ln_fid_latest_scanned_phase2));
if (rc <= 0)
goto out;
buf += rc;
len -= rc;
} else {
__u64 speed1 = ns->ln_items_checked;
__u64 speed2 = ns->ln_objs_checked_phase2;
if (ns->ln_run_time_phase1 != 0)
do_div(speed1, ns->ln_run_time_phase1);
if (ns->ln_run_time_phase2 != 0)
do_div(speed2, ns->ln_run_time_phase2);
rc = snprintf(buf, len,
"checked_phase1: "LPU64"\n"
"checked_phase2: "LPU64"\n"
"updated_phase1: "LPU64"\n"
"updated_phase2: "LPU64"\n"
"failed_phase1: "LPU64"\n"
"failed_phase2: "LPU64"\n"
"dirs: "LPU64"\n"
"M-linked: "LPU64"\n"
"nlinks_repaired: "LPU64"\n"
"lost_found: "LPU64"\n"
"success_count: %u\n"
"run_time_phase1: %u seconds\n"
"run_time_phase2: %u seconds\n"
"average_speed_phase1: "LPU64" items/sec\n"
"average_speed_phase2: "LPU64" objs/sec\n"
"real-time_speed_phase1: N/A\n"
"real-time_speed_phase2: N/A\n"
"current_position: N/A\n",
ns->ln_items_checked,
ns->ln_objs_checked_phase2,
ns->ln_items_repaired,
ns->ln_objs_repaired_phase2,
ns->ln_items_failed,
ns->ln_objs_failed_phase2,
ns->ln_dirs_checked,
ns->ln_mlinked_checked,
ns->ln_objs_nlink_repaired,
ns->ln_objs_lost_found,
ns->ln_success_count,
ns->ln_run_time_phase1,
ns->ln_run_time_phase2,
speed1,
speed2);
if (rc <= 0)
goto out;
buf += rc;
len -= rc;
}
ret = save - len;
out:
up_read(&com->lc_sem);
return ret;
}
/*
* Handler for: getattr, lookup and revalidate cases.
*/
static int
lmv_intent_lookup(struct obd_export *exp, struct md_op_data *op_data,
struct lookup_intent *it, struct ptlrpc_request **reqp,
ldlm_blocking_callback cb_blocking,
__u64 extra_lock_flags)
{
struct obd_device *obd = exp->exp_obd;
struct lmv_obd *lmv = &obd->u.lmv;
struct lmv_tgt_desc *tgt = NULL;
struct mdt_body *body;
struct lmv_stripe_md *lsm = op_data->op_mea1;
int rc = 0;
ENTRY;
/* If it returns ERR_PTR(-EBADFD) then it is an unknown hash type
* it will try all stripes to locate the object */
tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid1);
if (IS_ERR(tgt) && (PTR_ERR(tgt) != -EBADFD))
RETURN(PTR_ERR(tgt));
/* Both migrating dir and unknown hash dir need to try
* all of sub-stripes */
if (lsm != NULL && !lmv_is_known_hash_type(lsm->lsm_md_hash_type)) {
struct lmv_oinfo *oinfo;
oinfo = &lsm->lsm_md_oinfo[0];
op_data->op_fid1 = oinfo->lmo_fid;
op_data->op_mds = oinfo->lmo_mds;
tgt = lmv_get_target(lmv, oinfo->lmo_mds, NULL);
if (IS_ERR(tgt))
RETURN(PTR_ERR(tgt));
}
if (!fid_is_sane(&op_data->op_fid2))
fid_zero(&op_data->op_fid2);
CDEBUG(D_INODE, "LOOKUP_INTENT with fid1="DFID", fid2="DFID
", name='%s' -> mds #%u lsm=%p lsm_magic=%x\n",
PFID(&op_data->op_fid1), PFID(&op_data->op_fid2),
op_data->op_name ? op_data->op_name : "<NULL>",
tgt->ltd_idx, lsm, lsm == NULL ? -1 : lsm->lsm_md_magic);
op_data->op_bias &= ~MDS_CROSS_REF;
rc = md_intent_lock(tgt->ltd_exp, op_data, it, reqp, cb_blocking,
extra_lock_flags);
if (rc < 0)
RETURN(rc);
if (*reqp == NULL) {
/* If RPC happens, lsm information will be revalidated
* during update_inode process (see ll_update_lsm_md) */
if (op_data->op_mea2 != NULL) {
rc = lmv_revalidate_slaves(exp, op_data->op_mea2,
cb_blocking,
extra_lock_flags);
if (rc != 0)
RETURN(rc);
}
RETURN(rc);
} else if (it_disposition(it, DISP_LOOKUP_NEG) && lsm != NULL &&
lmv_need_try_all_stripes(lsm)) {
/* For migrating and unknown hash type directory, it will
* try to target the entry on other stripes */
int stripe_index;
for (stripe_index = 1;
stripe_index < lsm->lsm_md_stripe_count &&
it_disposition(it, DISP_LOOKUP_NEG); stripe_index++) {
struct lmv_oinfo *oinfo;
/* release the previous request */
ptlrpc_req_finished(*reqp);
it->d.lustre.it_data = NULL;
*reqp = NULL;
oinfo = &lsm->lsm_md_oinfo[stripe_index];
tgt = lmv_find_target(lmv, &oinfo->lmo_fid);
if (IS_ERR(tgt))
RETURN(PTR_ERR(tgt));
CDEBUG(D_INODE, "Try other stripes " DFID"\n",
PFID(&oinfo->lmo_fid));
op_data->op_fid1 = oinfo->lmo_fid;
it->d.lustre.it_disposition &= ~DISP_ENQ_COMPLETE;
rc = md_intent_lock(tgt->ltd_exp, op_data, it, reqp,
cb_blocking, extra_lock_flags);
if (rc != 0)
RETURN(rc);
}
}
/*
* MDS has returned success. Probably name has been resolved in
* remote inode. Let's check this.
*/
body = req_capsule_server_get(&(*reqp)->rq_pill, &RMF_MDT_BODY);
if (body == NULL)
RETURN(-EPROTO);
/* Not cross-ref case, just get out of here. */
if (unlikely((body->mbo_valid & OBD_MD_MDS))) {
rc = lmv_intent_remote(exp, it, NULL, reqp, cb_blocking,
extra_lock_flags);
if (rc != 0)
RETURN(rc);
body = req_capsule_server_get(&(*reqp)->rq_pill, &RMF_MDT_BODY);
if (body == NULL)
RETURN(-EPROTO);
}
RETURN(rc);
}
static int lfsck_namespace_prep(const struct lu_env *env,
struct lfsck_component *com,
struct lfsck_start_param *lsp)
{
struct lfsck_instance *lfsck = com->lc_lfsck;
struct lfsck_namespace *ns = com->lc_file_ram;
struct lfsck_position *pos = &com->lc_pos_start;
if (ns->ln_status == LS_COMPLETED) {
int rc;
rc = lfsck_namespace_reset(env, com, false);
if (rc != 0)
return rc;
}
down_write(&com->lc_sem);
ns->ln_time_latest_start = cfs_time_current_sec();
spin_lock(&lfsck->li_lock);
if (ns->ln_flags & LF_SCANNED_ONCE) {
if (!lfsck->li_drop_dryrun ||
lfsck_pos_is_zero(&ns->ln_pos_first_inconsistent)) {
ns->ln_status = LS_SCANNING_PHASE2;
cfs_list_del_init(&com->lc_link);
cfs_list_add_tail(&com->lc_link,
&lfsck->li_list_double_scan);
if (!cfs_list_empty(&com->lc_link_dir))
cfs_list_del_init(&com->lc_link_dir);
lfsck_pos_set_zero(pos);
} else {
ns->ln_status = LS_SCANNING_PHASE1;
ns->ln_run_time_phase1 = 0;
ns->ln_run_time_phase2 = 0;
ns->ln_items_checked = 0;
ns->ln_items_repaired = 0;
ns->ln_items_failed = 0;
ns->ln_dirs_checked = 0;
ns->ln_mlinked_checked = 0;
ns->ln_objs_checked_phase2 = 0;
ns->ln_objs_repaired_phase2 = 0;
ns->ln_objs_failed_phase2 = 0;
ns->ln_objs_nlink_repaired = 0;
ns->ln_objs_lost_found = 0;
fid_zero(&ns->ln_fid_latest_scanned_phase2);
if (cfs_list_empty(&com->lc_link_dir))
cfs_list_add_tail(&com->lc_link_dir,
&lfsck->li_list_dir);
*pos = ns->ln_pos_first_inconsistent;
}
} else {
ns->ln_status = LS_SCANNING_PHASE1;
if (cfs_list_empty(&com->lc_link_dir))
cfs_list_add_tail(&com->lc_link_dir,
&lfsck->li_list_dir);
if (!lfsck->li_drop_dryrun ||
lfsck_pos_is_zero(&ns->ln_pos_first_inconsistent)) {
*pos = ns->ln_pos_last_checkpoint;
pos->lp_oit_cookie++;
} else {
*pos = ns->ln_pos_first_inconsistent;
}
}
spin_unlock(&lfsck->li_lock);
up_write(&com->lc_sem);
return 0;
}
/*
* VBR: save parent version in reply and child version getting by its name.
* Version of child is getting and checking during its lookup. If
*/
static int mdt_reint_unlink(struct mdt_thread_info *info,
struct mdt_lock_handle *lhc)
{
struct mdt_reint_record *rr = &info->mti_rr;
struct ptlrpc_request *req = mdt_info_req(info);
struct md_attr *ma = &info->mti_attr;
struct lu_fid *child_fid = &info->mti_tmp_fid1;
struct mdt_object *mp;
struct mdt_object *mc;
struct mdt_lock_handle *parent_lh;
struct mdt_lock_handle *child_lh;
struct lu_name *lname;
int rc;
ENTRY;
DEBUG_REQ(D_INODE, req, "unlink "DFID"/%s", PFID(rr->rr_fid1),
rr->rr_name);
if (info->mti_dlm_req)
ldlm_request_cancel(req, info->mti_dlm_req, 0);
if (OBD_FAIL_CHECK(OBD_FAIL_MDS_REINT_UNLINK))
RETURN(err_serious(-ENOENT));
/*
* step 1: Found the parent.
*/
mp = mdt_object_find(info->mti_env, info->mti_mdt, rr->rr_fid1);
if (IS_ERR(mp)) {
rc = PTR_ERR(mp);
GOTO(out, rc);
}
if (mdt_object_obf(mp))
GOTO(put_parent, rc = -EPERM);
parent_lh = &info->mti_lh[MDT_LH_PARENT];
lname = mdt_name(info->mti_env, (char *)rr->rr_name, rr->rr_namelen);
if (mdt_object_remote(mp)) {
mdt_lock_reg_init(parent_lh, LCK_EX);
rc = mdt_remote_object_lock(info, mp, &parent_lh->mlh_rreg_lh,
parent_lh->mlh_rreg_mode,
MDS_INODELOCK_UPDATE);
if (rc != ELDLM_OK)
GOTO(put_parent, rc);
} else {
mdt_lock_pdo_init(parent_lh, LCK_PW, rr->rr_name,
rr->rr_namelen);
rc = mdt_object_lock(info, mp, parent_lh, MDS_INODELOCK_UPDATE,
MDT_LOCAL_LOCK);
if (rc)
GOTO(put_parent, rc);
rc = mdt_version_get_check_save(info, mp, 0);
if (rc)
GOTO(unlock_parent, rc);
}
/* step 2: find & lock the child */
/* lookup child object along with version checking */
fid_zero(child_fid);
rc = mdt_lookup_version_check(info, mp, lname, child_fid, 1);
if (rc != 0)
GOTO(unlock_parent, rc);
mdt_reint_init_ma(info, ma);
/* We will lock the child regardless it is local or remote. No harm. */
mc = mdt_object_find(info->mti_env, info->mti_mdt, child_fid);
if (IS_ERR(mc))
GOTO(unlock_parent, rc = PTR_ERR(mc));
child_lh = &info->mti_lh[MDT_LH_CHILD];
mdt_lock_reg_init(child_lh, LCK_EX);
if (mdt_object_remote(mc)) {
struct mdt_body *repbody;
if (!fid_is_zero(rr->rr_fid2)) {
CDEBUG(D_INFO, "%s: name %s can not find "DFID"\n",
mdt2obd_dev(info->mti_mdt)->obd_name,
(char *)rr->rr_name, PFID(mdt_object_fid(mc)));
GOTO(unlock_parent, rc = -ENOENT);
}
CDEBUG(D_INFO, "%s: name %s: "DFID" is another MDT\n",
mdt2obd_dev(info->mti_mdt)->obd_name,
(char *)rr->rr_name, PFID(mdt_object_fid(mc)));
if (info->mti_spec.sp_rm_entry) {
struct lu_ucred *uc = mdt_ucred(info);
if (!md_capable(uc, CFS_CAP_SYS_ADMIN)) {
CERROR("%s: unlink remote entry is only "
"permitted for administrator: rc = %d\n",
mdt2obd_dev(info->mti_mdt)->obd_name,
-EPERM);
GOTO(unlock_parent, rc = -EPERM);
}
ma->ma_need = MA_INODE;
ma->ma_valid = 0;
mdt_set_capainfo(info, 1, child_fid, BYPASS_CAPA);
rc = mdo_unlink(info->mti_env, mdt_object_child(mp),
NULL, lname, ma);
mdt_object_put(info->mti_env, mc);
GOTO(unlock_parent, rc);
}
/* Revoke the LOOKUP lock of the remote object granted by
* this MDT. Since the unlink will happen on another MDT,
* it will release the LOOKUP lock right away. Then What
* would happen if another client try to grab the LOOKUP
* lock at the same time with unlink XXX */
mdt_object_lock(info, mc, child_lh, MDS_INODELOCK_LOOKUP,
MDT_CROSS_LOCK);
repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
LASSERT(repbody != NULL);
repbody->fid1 = *mdt_object_fid(mc);
repbody->valid |= (OBD_MD_FLID | OBD_MD_MDS);
mdt_object_unlock_put(info, mc, child_lh, rc);
GOTO(unlock_parent, rc = -EREMOTE);
} else if (info->mti_spec.sp_rm_entry) {
CERROR("%s: lfs rmdir should not be used on local dir %s\n",
mdt2obd_dev(info->mti_mdt)->obd_name,
(char *)rr->rr_name);
mdt_object_put(info->mti_env, mc);
GOTO(unlock_parent, rc = -EPERM);
}
rc = mdt_object_lock(info, mc, child_lh, MDS_INODELOCK_FULL,
MDT_CROSS_LOCK);
if (rc != 0) {
mdt_object_put(info->mti_env, mc);
GOTO(unlock_parent, rc);
}
mdt_fail_write(info->mti_env, info->mti_mdt->mdt_bottom,
OBD_FAIL_MDS_REINT_UNLINK_WRITE);
/* save version when object is locked */
mdt_version_get_save(info, mc, 1);
/*
* Now we can only make sure we need MA_INODE, in mdd layer, will check
* whether need MA_LOV and MA_COOKIE.
*/
ma->ma_need = MA_INODE;
ma->ma_valid = 0;
mdt_set_capainfo(info, 1, child_fid, BYPASS_CAPA);
rc = mdo_unlink(info->mti_env, mdt_object_child(mp),
mdt_object_child(mc), lname, ma);
if (rc == 0 && !lu_object_is_dying(&mc->mot_header))
rc = mdt_attr_get_complex(info, mc, ma);
if (rc == 0)
mdt_handle_last_unlink(info, mc, ma);
if (ma->ma_valid & MA_INODE) {
switch (ma->ma_attr.la_mode & S_IFMT) {
case S_IFDIR:
mdt_counter_incr(req, LPROC_MDT_RMDIR);
break;
case S_IFREG:
case S_IFLNK:
case S_IFCHR:
case S_IFBLK:
case S_IFIFO:
case S_IFSOCK:
mdt_counter_incr(req, LPROC_MDT_UNLINK);
break;
default:
LASSERTF(0, "bad file type %o unlinking\n",
ma->ma_attr.la_mode);
}
}
EXIT;
mdt_object_unlock_put(info, mc, child_lh, rc);
unlock_parent:
mdt_object_unlock(info, mp, parent_lh, rc);
put_parent:
mdt_object_put(info->mti_env, mp);
out:
return rc;
}
/*
* VBR: rename versions in reply: 0 - src parent; 1 - tgt parent;
* 2 - src child; 3 - tgt child.
* Update on disk version of src child.
*/
static int mdt_reint_rename(struct mdt_thread_info *info,
struct mdt_lock_handle *lhc)
{
struct mdt_reint_record *rr = &info->mti_rr;
struct md_attr *ma = &info->mti_attr;
struct ptlrpc_request *req = mdt_info_req(info);
struct mdt_object *msrcdir;
struct mdt_object *mtgtdir;
struct mdt_object *mold;
struct mdt_object *mnew = NULL;
struct mdt_lock_handle *lh_srcdirp;
struct mdt_lock_handle *lh_tgtdirp;
struct mdt_lock_handle *lh_oldp;
struct mdt_lock_handle *lh_newp;
struct lu_fid *old_fid = &info->mti_tmp_fid1;
struct lu_fid *new_fid = &info->mti_tmp_fid2;
struct lustre_handle rename_lh = { 0 };
struct lu_name slname = { 0 };
struct lu_name *lname;
int rc;
ENTRY;
if (info->mti_dlm_req)
ldlm_request_cancel(req, info->mti_dlm_req, 0);
DEBUG_REQ(D_INODE, req, "rename "DFID"/%s to "DFID"/%s",
PFID(rr->rr_fid1), rr->rr_name,
PFID(rr->rr_fid2), rr->rr_tgt);
rc = mdt_rename_lock(info, &rename_lh);
if (rc) {
CERROR("Can't lock FS for rename, rc %d\n", rc);
RETURN(rc);
}
lh_newp = &info->mti_lh[MDT_LH_NEW];
/* step 1: lock the source dir. */
lh_srcdirp = &info->mti_lh[MDT_LH_PARENT];
mdt_lock_pdo_init(lh_srcdirp, LCK_PW, rr->rr_name,
rr->rr_namelen);
msrcdir = mdt_object_find_lock(info, rr->rr_fid1, lh_srcdirp,
MDS_INODELOCK_UPDATE);
if (IS_ERR(msrcdir))
GOTO(out_rename_lock, rc = PTR_ERR(msrcdir));
if (mdt_object_obf(msrcdir))
GOTO(out_unlock_source, rc = -EPERM);
rc = mdt_version_get_check_save(info, msrcdir, 0);
if (rc)
GOTO(out_unlock_source, rc);
/* step 2: find & lock the target dir. */
lh_tgtdirp = &info->mti_lh[MDT_LH_CHILD];
mdt_lock_pdo_init(lh_tgtdirp, LCK_PW, rr->rr_tgt,
rr->rr_tgtlen);
if (lu_fid_eq(rr->rr_fid1, rr->rr_fid2)) {
mdt_object_get(info->mti_env, msrcdir);
mtgtdir = msrcdir;
if (lh_tgtdirp->mlh_pdo_hash != lh_srcdirp->mlh_pdo_hash) {
rc = mdt_pdir_hash_lock(info, lh_tgtdirp, mtgtdir,
MDS_INODELOCK_UPDATE);
if (rc)
GOTO(out_unlock_source, rc);
OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_PDO_LOCK2, 10);
}
} else {
mtgtdir = mdt_object_find(info->mti_env, info->mti_mdt,
rr->rr_fid2);
if (IS_ERR(mtgtdir))
GOTO(out_unlock_source, rc = PTR_ERR(mtgtdir));
if (mdt_object_obf(mtgtdir))
GOTO(out_put_target, rc = -EPERM);
/* check early, the real version will be saved after locking */
rc = mdt_version_get_check(info, mtgtdir, 1);
if (rc)
GOTO(out_put_target, rc);
if (unlikely(mdt_object_remote(mtgtdir))) {
CDEBUG(D_INFO, "Source dir "DFID" target dir "DFID
"on different MDTs\n", PFID(rr->rr_fid1),
PFID(rr->rr_fid2));
GOTO(out_put_target, rc = -EXDEV);
} else {
if (likely(mdt_object_exists(mtgtdir))) {
/* we lock the target dir if it is local */
rc = mdt_object_lock(info, mtgtdir, lh_tgtdirp,
MDS_INODELOCK_UPDATE,
MDT_LOCAL_LOCK);
if (rc != 0)
GOTO(out_put_target, rc);
/* get and save correct version after locking */
mdt_version_get_save(info, mtgtdir, 1);
} else {
GOTO(out_put_target, rc = -ESTALE);
}
}
}
/* step 3: find & lock the old object. */
lname = mdt_name(info->mti_env, (char *)rr->rr_name, rr->rr_namelen);
mdt_name_copy(&slname, lname);
fid_zero(old_fid);
rc = mdt_lookup_version_check(info, msrcdir, &slname, old_fid, 2);
if (rc != 0)
GOTO(out_unlock_target, rc);
if (lu_fid_eq(old_fid, rr->rr_fid1) || lu_fid_eq(old_fid, rr->rr_fid2))
GOTO(out_unlock_target, rc = -EINVAL);
mold = mdt_object_find(info->mti_env, info->mti_mdt, old_fid);
if (IS_ERR(mold))
GOTO(out_unlock_target, rc = PTR_ERR(mold));
if (mdt_object_remote(mold)) {
mdt_object_put(info->mti_env, mold);
CDEBUG(D_INFO, "Source child "DFID" is on another MDT\n",
PFID(old_fid));
GOTO(out_unlock_target, rc = -EXDEV);
}
if (mdt_object_obf(mold)) {
mdt_object_put(info->mti_env, mold);
GOTO(out_unlock_target, rc = -EPERM);
}
lh_oldp = &info->mti_lh[MDT_LH_OLD];
mdt_lock_reg_init(lh_oldp, LCK_EX);
rc = mdt_object_lock(info, mold, lh_oldp, MDS_INODELOCK_LOOKUP,
MDT_CROSS_LOCK);
if (rc != 0) {
mdt_object_put(info->mti_env, mold);
GOTO(out_unlock_target, rc);
}
info->mti_mos = mold;
/* save version after locking */
mdt_version_get_save(info, mold, 2);
mdt_set_capainfo(info, 2, old_fid, BYPASS_CAPA);
/* step 4: find & lock the new object. */
/* new target object may not exist now */
lname = mdt_name(info->mti_env, (char *)rr->rr_tgt, rr->rr_tgtlen);
/* lookup with version checking */
fid_zero(new_fid);
rc = mdt_lookup_version_check(info, mtgtdir, lname, new_fid, 3);
if (rc == 0) {
/* the new_fid should have been filled at this moment */
if (lu_fid_eq(old_fid, new_fid))
GOTO(out_unlock_old, rc);
if (lu_fid_eq(new_fid, rr->rr_fid1) ||
lu_fid_eq(new_fid, rr->rr_fid2))
GOTO(out_unlock_old, rc = -EINVAL);
mdt_lock_reg_init(lh_newp, LCK_EX);
mnew = mdt_object_find(info->mti_env, info->mti_mdt, new_fid);
if (IS_ERR(mnew))
GOTO(out_unlock_old, rc = PTR_ERR(mnew));
if (mdt_object_obf(mnew)) {
mdt_object_put(info->mti_env, mnew);
GOTO(out_unlock_old, rc = -EPERM);
}
if (mdt_object_remote(mnew)) {
mdt_object_put(info->mti_env, mnew);
CDEBUG(D_INFO, "src child "DFID" is on another MDT\n",
PFID(new_fid));
GOTO(out_unlock_old, rc = -EXDEV);
}
rc = mdt_object_lock(info, mnew, lh_newp,
MDS_INODELOCK_FULL, MDT_CROSS_LOCK);
if (rc != 0) {
mdt_object_put(info->mti_env, mnew);
GOTO(out_unlock_old, rc);
}
/* get and save version after locking */
mdt_version_get_save(info, mnew, 3);
mdt_set_capainfo(info, 3, new_fid, BYPASS_CAPA);
} else if (rc != -EREMOTE && rc != -ENOENT) {
GOTO(out_unlock_old, rc);
} else {
mdt_enoent_version_save(info, 3);
}
/* step 5: rename it */
mdt_reint_init_ma(info, ma);
mdt_fail_write(info->mti_env, info->mti_mdt->mdt_bottom,
OBD_FAIL_MDS_REINT_RENAME_WRITE);
/* Check if @dst is subdir of @src. */
rc = mdt_rename_sanity(info, old_fid);
if (rc)
GOTO(out_unlock_new, rc);
rc = mdo_rename(info->mti_env, mdt_object_child(msrcdir),
mdt_object_child(mtgtdir), old_fid, &slname,
(mnew ? mdt_object_child(mnew) : NULL),
lname, ma);
/* handle last link of tgt object */
if (rc == 0) {
mdt_counter_incr(req, LPROC_MDT_RENAME);
if (mnew)
mdt_handle_last_unlink(info, mnew, ma);
mdt_rename_counter_tally(info, info->mti_mdt, req,
msrcdir, mtgtdir);
}
EXIT;
out_unlock_new:
if (mnew)
mdt_object_unlock_put(info, mnew, lh_newp, rc);
out_unlock_old:
mdt_object_unlock_put(info, mold, lh_oldp, rc);
out_unlock_target:
mdt_object_unlock(info, mtgtdir, lh_tgtdirp, rc);
out_put_target:
mdt_object_put(info->mti_env, mtgtdir);
out_unlock_source:
mdt_object_unlock_put(info, msrcdir, lh_srcdirp, rc);
out_rename_lock:
if (lustre_handle_is_used(&rename_lh))
mdt_rename_unlock(&rename_lh);
return rc;
}
/*
* VBR: save parent version in reply and child version getting by its name.
* Version of child is getting and checking during its lookup. If
*/
static int mdt_reint_unlink(struct mdt_thread_info *info,
struct mdt_lock_handle *lhc)
{
struct mdt_reint_record *rr = &info->mti_rr;
struct ptlrpc_request *req = mdt_info_req(info);
struct md_attr *ma = &info->mti_attr;
struct lu_fid *child_fid = &info->mti_tmp_fid1;
struct mdt_object *mp;
struct mdt_object *mc;
struct mdt_lock_handle *parent_lh;
struct mdt_lock_handle *child_lh;
struct lu_name *lname;
int rc;
int no_name = 0;
ENTRY;
DEBUG_REQ(D_INODE, req, "unlink "DFID"/%s", PFID(rr->rr_fid1),
rr->rr_name);
if (info->mti_dlm_req)
ldlm_request_cancel(req, info->mti_dlm_req, 0);
if (OBD_FAIL_CHECK(OBD_FAIL_MDS_REINT_UNLINK))
RETURN(err_serious(-ENOENT));
if (fid_is_obf(rr->rr_fid1) || fid_is_dot_lustre(rr->rr_fid1))
RETURN(-EPERM);
/*
* step 1: Found the parent.
*/
mp = mdt_object_find(info->mti_env, info->mti_mdt, rr->rr_fid1);
if (IS_ERR(mp)) {
rc = PTR_ERR(mp);
GOTO(out, rc);
}
parent_lh = &info->mti_lh[MDT_LH_PARENT];
lname = mdt_name(info->mti_env, (char *)rr->rr_name, rr->rr_namelen);
if (mdt_object_remote(mp)) {
mdt_lock_reg_init(parent_lh, LCK_EX);
rc = mdt_remote_object_lock(info, mp, &parent_lh->mlh_rreg_lh,
parent_lh->mlh_rreg_mode,
MDS_INODELOCK_UPDATE);
if (rc != ELDLM_OK)
GOTO(put_parent, rc);
} else {
mdt_lock_pdo_init(parent_lh, LCK_PW, rr->rr_name,
rr->rr_namelen);
rc = mdt_object_lock(info, mp, parent_lh, MDS_INODELOCK_UPDATE,
MDT_LOCAL_LOCK);
if (rc)
GOTO(put_parent, rc);
rc = mdt_version_get_check_save(info, mp, 0);
if (rc)
GOTO(unlock_parent, rc);
}
/* step 2: find & lock the child */
/* lookup child object along with version checking */
fid_zero(child_fid);
rc = mdt_lookup_version_check(info, mp, lname, child_fid, 1);
if (rc != 0) {
/* Name might not be able to find during resend of
* remote unlink, considering following case.
* dir_A is a remote directory, the name entry of
* dir_A is on MDT0, the directory is on MDT1,
*
* 1. client sends unlink req to MDT1.
* 2. MDT1 sends name delete update to MDT0.
* 3. name entry is being deleted in MDT0 synchronously.
* 4. MDT1 is restarted.
* 5. client resends unlink req to MDT1. So it can not
* find the name entry on MDT0 anymore.
* In this case, MDT1 only needs to destory the local
* directory.
* */
if (mdt_object_remote(mp) && rc == -ENOENT &&
!fid_is_zero(rr->rr_fid2) &&
lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT) {
no_name = 1;
*child_fid = *rr->rr_fid2;
} else {
GOTO(unlock_parent, rc);
}
}
if (fid_is_obf(child_fid) || fid_is_dot_lustre(child_fid))
GOTO(unlock_parent, rc = -EPERM);
mdt_reint_init_ma(info, ma);
/* We will lock the child regardless it is local or remote. No harm. */
mc = mdt_object_find(info->mti_env, info->mti_mdt, child_fid);
if (IS_ERR(mc))
GOTO(unlock_parent, rc = PTR_ERR(mc));
child_lh = &info->mti_lh[MDT_LH_CHILD];
mdt_lock_reg_init(child_lh, LCK_EX);
if (mdt_object_remote(mc)) {
struct mdt_body *repbody;
if (!fid_is_zero(rr->rr_fid2)) {
CDEBUG(D_INFO, "%s: name %s can not find "DFID"\n",
mdt_obd_name(info->mti_mdt),
(char *)rr->rr_name, PFID(mdt_object_fid(mc)));
GOTO(put_child, rc = -ENOENT);
}
CDEBUG(D_INFO, "%s: name %s: "DFID" is another MDT\n",
mdt_obd_name(info->mti_mdt),
(char *)rr->rr_name, PFID(mdt_object_fid(mc)));
if (!mdt_is_dne_client(req->rq_export))
/* Return -EIO for old client */
GOTO(put_child, rc = -EIO);
if (info->mti_spec.sp_rm_entry) {
struct lu_ucred *uc = mdt_ucred(info);
if (!md_capable(uc, CFS_CAP_SYS_ADMIN)) {
CERROR("%s: unlink remote entry is only "
"permitted for administrator: rc = %d\n",
mdt_obd_name(info->mti_mdt),
-EPERM);
GOTO(put_child, rc = -EPERM);
}
ma->ma_need = MA_INODE;
ma->ma_valid = 0;
mdt_set_capainfo(info, 1, child_fid, BYPASS_CAPA);
rc = mdo_unlink(info->mti_env, mdt_object_child(mp),
NULL, lname, ma, no_name);
GOTO(put_child, rc);
}
/* Revoke the LOOKUP lock of the remote object granted by
* this MDT. Since the unlink will happen on another MDT,
* it will release the LOOKUP lock right away. Then What
* would happen if another client try to grab the LOOKUP
* lock at the same time with unlink XXX */
mdt_object_lock(info, mc, child_lh, MDS_INODELOCK_LOOKUP,
MDT_CROSS_LOCK);
repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
LASSERT(repbody != NULL);
repbody->fid1 = *mdt_object_fid(mc);
repbody->valid |= (OBD_MD_FLID | OBD_MD_MDS);
GOTO(unlock_child, rc = -EREMOTE);
} else if (info->mti_spec.sp_rm_entry) {
rc = -EPERM;
CDEBUG(D_INFO, "%s: no rm_entry on local dir '%s': rc = %d\n",
mdt_obd_name(info->mti_mdt), (char *)rr->rr_name, rc);
GOTO(put_child, rc);
}
/* We used to acquire MDS_INODELOCK_FULL here but we can't do
* this now because a running HSM restore on the child (unlink
* victim) will hold the layout lock. See LU-4002. */
rc = mdt_object_lock(info, mc, child_lh,
MDS_INODELOCK_LOOKUP | MDS_INODELOCK_UPDATE,
MDT_CROSS_LOCK);
if (rc != 0)
GOTO(put_child, rc);
mdt_fail_write(info->mti_env, info->mti_mdt->mdt_bottom,
OBD_FAIL_MDS_REINT_UNLINK_WRITE);
/* save version when object is locked */
mdt_version_get_save(info, mc, 1);
/*
* Now we can only make sure we need MA_INODE, in mdd layer, will check
* whether need MA_LOV and MA_COOKIE.
*/
ma->ma_need = MA_INODE;
ma->ma_valid = 0;
mdt_set_capainfo(info, 1, child_fid, BYPASS_CAPA);
mutex_lock(&mc->mot_lov_mutex);
rc = mdo_unlink(info->mti_env, mdt_object_child(mp),
mdt_object_child(mc), lname, ma, no_name);
mutex_unlock(&mc->mot_lov_mutex);
if (rc == 0 && !lu_object_is_dying(&mc->mot_header))
rc = mdt_attr_get_complex(info, mc, ma);
if (rc == 0)
mdt_handle_last_unlink(info, mc, ma);
if (ma->ma_valid & MA_INODE) {
switch (ma->ma_attr.la_mode & S_IFMT) {
case S_IFDIR:
mdt_counter_incr(req, LPROC_MDT_RMDIR);
break;
case S_IFREG:
case S_IFLNK:
case S_IFCHR:
case S_IFBLK:
case S_IFIFO:
case S_IFSOCK:
mdt_counter_incr(req, LPROC_MDT_UNLINK);
break;
default:
LASSERTF(0, "bad file type %o unlinking\n",
ma->ma_attr.la_mode);
}
}
EXIT;
unlock_child:
mdt_object_unlock(info, mc, child_lh, rc);
put_child:
mdt_object_put(info->mti_env, mc);
unlock_parent:
mdt_object_unlock(info, mp, parent_lh, rc);
put_parent:
mdt_object_put(info->mti_env, mp);
out:
return rc;
}
/*
* Handler for: getattr, lookup and revalidate cases.
*/
int lmv_intent_lookup(struct obd_export *exp, struct md_op_data *op_data,
void *lmm, int lmmsize, struct lookup_intent *it,
int flags, struct ptlrpc_request **reqp,
ldlm_blocking_callback cb_blocking,
__u64 extra_lock_flags)
{
struct obd_device *obd = exp->exp_obd;
struct lmv_obd *lmv = &obd->u.lmv;
struct lmv_tgt_desc *tgt = NULL;
struct mdt_body *body;
struct lmv_stripe_md *lsm = op_data->op_mea1;
int rc = 0;
ENTRY;
tgt = lmv_locate_mds(lmv, op_data, &op_data->op_fid1);
if (IS_ERR(tgt))
RETURN(PTR_ERR(tgt));
if (!fid_is_sane(&op_data->op_fid2))
fid_zero(&op_data->op_fid2);
CDEBUG(D_INODE, "LOOKUP_INTENT with fid1="DFID", fid2="DFID
", name='%s' -> mds #%d lsm=%p lsm_magic=%x\n",
PFID(&op_data->op_fid1), PFID(&op_data->op_fid2),
op_data->op_name ? op_data->op_name : "<NULL>",
tgt->ltd_idx, lsm, lsm == NULL ? -1 : lsm->lsm_md_magic);
op_data->op_bias &= ~MDS_CROSS_REF;
rc = md_intent_lock(tgt->ltd_exp, op_data, lmm, lmmsize, it,
flags, reqp, cb_blocking, extra_lock_flags);
if (rc < 0)
RETURN(rc);
if (*reqp == NULL) {
/* If RPC happens, lsm information will be revalidated
* during update_inode process (see ll_update_lsm_md) */
if (op_data->op_mea2 != NULL) {
rc = lmv_revalidate_slaves(exp, NULL, op_data->op_mea2,
cb_blocking,
extra_lock_flags);
if (rc != 0)
RETURN(rc);
}
RETURN(rc);
} else if (it_disposition(it, DISP_LOOKUP_NEG) &&
lsm != NULL && lsm->lsm_md_magic == LMV_MAGIC_MIGRATE) {
/* For migrating directory, if it can not find the child in
* the source directory(master stripe), try the targeting
* directory(stripe 1) */
tgt = lmv_find_target(lmv, &lsm->lsm_md_oinfo[1].lmo_fid);
if (IS_ERR(tgt))
RETURN(PTR_ERR(tgt));
ptlrpc_req_finished(*reqp);
it->d.lustre.it_data = NULL;
*reqp = NULL;
CDEBUG(D_INODE, "For migrating dir, try target dir "DFID"\n",
PFID(&lsm->lsm_md_oinfo[1].lmo_fid));
op_data->op_fid1 = lsm->lsm_md_oinfo[1].lmo_fid;
it->d.lustre.it_disposition &= ~DISP_ENQ_COMPLETE;
rc = md_intent_lock(tgt->ltd_exp, op_data, lmm, lmmsize, it,
flags, reqp, cb_blocking, extra_lock_flags);
}
/*
* MDS has returned success. Probably name has been resolved in
* remote inode. Let's check this.
*/
body = req_capsule_server_get(&(*reqp)->rq_pill, &RMF_MDT_BODY);
if (body == NULL)
RETURN(-EPROTO);
/* Not cross-ref case, just get out of here. */
if (unlikely((body->valid & OBD_MD_MDS))) {
rc = lmv_intent_remote(exp, lmm, lmmsize, it, NULL, flags,
reqp, cb_blocking, extra_lock_flags);
if (rc != 0)
RETURN(rc);
body = req_capsule_server_get(&(*reqp)->rq_pill, &RMF_MDT_BODY);
if (body == NULL)
RETURN(-EPROTO);
}
RETURN(rc);
}
