int out_tx_create_exec(const struct lu_env *env, struct thandle *th,
struct tx_arg *arg)
{
struct dt_object *dt_obj = arg->object;
int rc;
CDEBUG(D_OTHER, "%s: create "DFID": dof %u, mode %o\n",
dt_obd_name(th->th_dev),
PFID(lu_object_fid(&arg->object->do_lu)),
arg->u.create.dof.dof_type,
arg->u.create.attr.la_mode & S_IFMT);
dt_write_lock(env, dt_obj, MOR_TGT_CHILD);
rc = dt_create(env, dt_obj, &arg->u.create.attr,
&arg->u.create.hint, &arg->u.create.dof, th);
dt_write_unlock(env, dt_obj);
CDEBUG(D_INFO, "%s: insert create reply %p index %d: rc = %d\n",
dt_obd_name(th->th_dev), arg->reply, arg->index, rc);
update_insert_reply(arg->reply, NULL, 0, arg->index, rc);
return rc;
}
int out_tx_create_exec(struct mdt_thread_info *info, struct thandle *th,
struct tx_arg *arg)
{
struct dt_object *dt_obj = arg->object;
int rc;
LASSERT(dt_obj != NULL && !IS_ERR(dt_obj));
CDEBUG(D_OTHER, "create "DFID": dof %u, mode %o\n",
PFID(lu_object_fid(&arg->object->do_lu)),
arg->u.create.dof.dof_type,
arg->u.create.attr.la_mode & S_IFMT);
dt_write_lock(info->mti_env, dt_obj, MOR_TGT_CHILD);
rc = dt_create(info->mti_env, dt_obj, &arg->u.create.attr,
&arg->u.create.hint, &arg->u.create.dof, th);
dt_write_unlock(info->mti_env, dt_obj);
CDEBUG(D_INFO, "insert create reply mode %o index %d\n",
arg->u.create.attr.la_mode, arg->index);
update_insert_reply(arg->reply, NULL, 0, arg->index, rc);
return rc;
}
/**
* Allocate empty worker structures. With backptr and thread-number,
* from 0..numthread initialised. Used as user arguments to new threads.
* Creates the daemon random generator if it does not exist yet.
* The random generator stays existing between reloads with a unique state.
* @param daemon: the daemon with (new) config settings.
*/
static void
daemon_create_workers(struct daemon* daemon)
{
int i, numport;
int* shufport;
log_assert(daemon && daemon->cfg);
if(!daemon->rand) {
unsigned int seed = (unsigned int)time(NULL) ^
(unsigned int)getpid() ^ 0x438;
daemon->rand = ub_initstate(seed, NULL);
if(!daemon->rand)
fatal_exit("could not init random generator");
hash_set_raninit((uint32_t)ub_random(daemon->rand));
}
shufport = (int*)calloc(65536, sizeof(int));
if(!shufport)
fatal_exit("out of memory during daemon init");
numport = daemon_get_shufport(daemon, shufport);
verbose(VERB_ALGO, "total of %d outgoing ports available", numport);
daemon->num = (daemon->cfg->num_threads?daemon->cfg->num_threads:1);
if(daemon->reuseport && (int)daemon->num < (int)daemon->num_ports) {
log_warn("cannot reduce num-threads to %d because so-reuseport "
"so continuing with %d threads.", (int)daemon->num,
(int)daemon->num_ports);
daemon->num = (int)daemon->num_ports;
}
daemon->workers = (struct worker**)calloc((size_t)daemon->num,
sizeof(struct worker*));
if(!daemon->workers)
fatal_exit("out of memory during daemon init");
if(daemon->cfg->dnstap) {
#ifdef USE_DNSTAP
daemon->dtenv = dt_create(daemon->cfg->dnstap_socket_path,
(unsigned int)daemon->num);
if (!daemon->dtenv)
fatal_exit("dt_create failed");
dt_apply_cfg(daemon->dtenv, daemon->cfg);
#else
fatal_exit("dnstap enabled in config but not built with dnstap support");
#endif
}
for(i=0; i<daemon->num; i++) {
if(!(daemon->workers[i] = worker_create(daemon, i,
shufport+numport*i/daemon->num,
numport*(i+1)/daemon->num - numport*i/daemon->num)))
/* the above is not ports/numthr, due to rounding */
fatal_exit("could not create worker");
}
free(shufport);
}
/**
* Read the special file which contains the list of llog catalogs IDs
*
* This function reads the CATALOGS file which contains the array of llog
* catalogs IDs. The main purpose of this file is to store OSP llogs indexed
* by OST/MDT number.
*
* \param[in] env execution environment
* \param[in] d corresponding storage device
* \param[in] idx position to start from, usually OST/MDT index
* \param[in] count how many catalog IDs to read
* \param[out] idarray the buffer for the data. If it is NULL then
* function returns just number of catalog IDs
* in the file.
* \param[in] fid LLOG_CATALOGS_OID for CATALOG object
*
* \retval 0 on successful read of catalog IDs
* \retval negative value on error
* \retval positive value which is number of records in
* the file if \a idarray is NULL
*/
int llog_osd_get_cat_list(const struct lu_env *env, struct dt_device *d,
int idx, int count, struct llog_catid *idarray,
const struct lu_fid *fid)
{
struct llog_thread_info *lgi = llog_info(env);
struct dt_object *o = NULL;
struct thandle *th;
int rc, size;
ENTRY;
LASSERT(d);
size = sizeof(*idarray) * count;
lgi->lgi_off = idx * sizeof(*idarray);
lgi->lgi_fid = *fid;
o = dt_locate(env, d, &lgi->lgi_fid);
if (IS_ERR(o))
RETURN(PTR_ERR(o));
if (!dt_object_exists(o)) {
th = dt_trans_create(env, d);
if (IS_ERR(th))
GOTO(out, rc = PTR_ERR(th));
lgi->lgi_attr.la_valid = LA_MODE;
lgi->lgi_attr.la_mode = S_IFREG | S_IRUGO | S_IWUSR;
lgi->lgi_dof.dof_type = dt_mode_to_dft(S_IFREG);
rc = dt_declare_create(env, o, &lgi->lgi_attr, NULL,
&lgi->lgi_dof, th);
if (rc)
GOTO(out_trans, rc);
rc = dt_trans_start_local(env, d, th);
if (rc)
GOTO(out_trans, rc);
dt_write_lock(env, o, 0);
if (!dt_object_exists(o))
rc = dt_create(env, o, &lgi->lgi_attr, NULL,
&lgi->lgi_dof, th);
dt_write_unlock(env, o);
out_trans:
dt_trans_stop(env, d, th);
if (rc)
GOTO(out, rc);
}
rc = dt_attr_get(env, o, &lgi->lgi_attr, BYPASS_CAPA);
if (rc)
GOTO(out, rc);
if (!S_ISREG(lgi->lgi_attr.la_mode)) {
CERROR("%s: CATALOGS is not a regular file!: mode = %o\n",
o->do_lu.lo_dev->ld_obd->obd_name,
lgi->lgi_attr.la_mode);
GOTO(out, rc = -ENOENT);
}
CDEBUG(D_CONFIG, "cat list: disk size=%d, read=%d\n",
(int)lgi->lgi_attr.la_size, size);
/* return just number of llogs */
if (idarray == NULL) {
rc = lgi->lgi_attr.la_size / sizeof(*idarray);
GOTO(out, rc);
}
/* read for new ost index or for empty file */
memset(idarray, 0, size);
if (lgi->lgi_attr.la_size <= lgi->lgi_off)
GOTO(out, rc = 0);
if (lgi->lgi_attr.la_size < lgi->lgi_off + size)
size = lgi->lgi_attr.la_size - lgi->lgi_off;
lgi->lgi_buf.lb_buf = idarray;
lgi->lgi_buf.lb_len = size;
rc = dt_record_read(env, o, &lgi->lgi_buf, &lgi->lgi_off);
if (rc) {
CERROR("%s: error reading CATALOGS: rc = %d\n",
o->do_lu.lo_dev->ld_obd->obd_name, rc);
GOTO(out, rc);
}
EXIT;
out:
lu_object_put(env, &o->do_lu);
RETURN(rc);
}
/*! \brief Create unit. */
static inline dt_unit_t *dt_test_create(int size)
{
return dt_create(size);
}
/*! API: run tests. */
static int dt_tests_run(int argc, char *argv[])
{
// Register service and signal handler
struct sigaction sa;
sa.sa_handler = interrupt_handle;
sigemptyset(&sa.sa_mask);
sa.sa_flags = 0;
sigaction(SIGALRM, &sa, NULL); // Interrupt
/* Initialize */
srand(time(NULL));
struct timeval tv;
pthread_mutex_init(&_runnable_mx, NULL);
/* Test 1: Create unit */
dt_unit_t *unit = dt_test_create(2);
ok(unit != 0, "dthreads: create unit (optimal size %d)", unit->size);
skip(unit == 0, DT_TEST_COUNT - 1);
/* Test 2: Assign a single task. */
ok(dt_test_single(unit), "dthreads: assign single task");
/* Test 3: Start tasks. */
_runnable_i = 0;
ok(dt_test_start(unit), "dthreads: start single task");
/* Test 4: Wait for tasks. */
ok(dt_test_join(unit), "dthreads: join threads");
/* Test 5: Compare counter. */
int expected = _runnable_cycles * 1;
cmp_ok(_runnable_i, "==", expected, "dthreads: result ok");
/* Test 6: Repurpose threads. */
_runnable_i = 0;
ok(dt_test_coherent(unit), "dthreads: repurpose to coherent");
/* Test 7: Restart threads. */
ok(dt_test_start(unit), "dthreads: start coherent unit");
/* Test 8: Repurpose single thread. */
tv.tv_sec = 0;
tv.tv_usec = 4000 + rand() % 1000; // 4-5ms
note("waiting for %dus to let thread do some work ...",
tv.tv_usec);
select(0, 0, 0, 0, &tv);
ok(dt_test_repurpose(unit, 0), "dthreads: repurpose on-the-fly");
/* Test 9: Cancel blocking thread. */
tv.tv_sec = 0;
tv.tv_usec = (250 + rand() % 500) * 1000; // 250-750ms
note("waiting for %dms to let thread pretend blocking I/O ...",
tv.tv_usec / 1000);
select(0, 0, 0, 0, &tv);
ok(dt_test_cancel(unit, 0), "dthreads: cancel blocking thread");
/* Test 10: Wait for tasks. */
ok(dt_test_join(unit), "dthreads: join threads");
/* Test 11: Compare counter. */
int expected_lo = _runnable_cycles * (unit->size - 1);
cmp_ok(_runnable_i, ">=", expected_lo,
"dthreads: result %d is => %d", _runnable_i, expected_lo);
/* Test 12: Compare counter #2. */
/*! \note repurpose could trigger next run of the unit if both finished */
int expected_hi = _runnable_cycles * (unit->size + unit->size - 1);
cmp_ok(_runnable_i, "<=", expected_hi,
"dthreads: result %d is <= %d", _runnable_i, expected_hi);
/* Test 13: Reanimate dead threads. */
ok(dt_test_reanimate(unit), "dthreads: reanimate dead threads");
/* Test 14: Deinitialize */
dt_delete(&unit);
ok(unit == 0, "dthreads: delete unit");
endskip;
/* Test 15: Wrong values. */
unit = dt_create(-1);
ok(unit == 0, "dthreads: create with negative count");
unit = dt_create_coherent(dt_optimal_size(), 0, 0);
/* Test 16: NULL runnable. */
cmp_ok(dt_start(unit), "==", 0, "dthreads: start with NULL runnable");
/* Test 17: NULL operations crashing. */
int op_count = 14;
int expected_min = op_count * -1;
// All functions must return -1 at least
int ret = 0;
lives_ok( {
ret += dt_activate(0); // -1
ret += dt_cancel(0); // -1
ret += dt_compact(0); // -1
dt_delete(0); //
ret += dt_is_cancelled(0); // 0
ret += dt_join(0); // -1
ret += dt_repurpose(0, 0, 0); // -1
ret += dt_signalize(0, SIGALRM); // -1
ret += dt_start(0); // -1
ret += dt_start_id(0); // -1
ret += dt_stop(0); // -1
ret += dt_stop_id(0); // -1
ret += dt_unit_lock(0); // -1
ret += dt_unit_unlock(0); // -1
}, "dthreads: not crashed while executing functions on NULL context");
/*! API: run tests. */
int main(int argc, char *argv[])
{
plan(8);
// Register service and signal handler
struct sigaction sa;
sa.sa_handler = interrupt_handle;
sigemptyset(&sa.sa_mask);
sa.sa_flags = 0;
sigaction(SIGALRM, &sa, NULL); // Interrupt
/* Initialize */
srand(time(NULL));
pthread_mutex_init(&_runnable_mx, NULL);
pthread_mutex_init(&_destructor_mx, NULL);
/* Test 1: Create unit */
dt_unit_t *unit = dt_create(2, &runnable, NULL, NULL);
ok(unit != NULL, "dthreads: create unit (size %d)", unit->size);
if (unit == NULL) {
skip_block(7, "No dthreads unit");
goto skip_all;
}
/* Test 2: Start tasks. */
_runnable_i = 0;
ok(dt_start(unit) == 0, "dthreads: start single task");
/* Test 3: Wait for tasks. */
ok(dt_join(unit) == 0, "dthreads: join threads");
/* Test 4: Compare counter. */
int expected = _runnable_cycles * 2;
is_int(expected, _runnable_i, "dthreads: result ok");
/* Test 5: Deinitialize */
dt_delete(&unit);
ok(unit == NULL, "dthreads: delete unit");
/* Test 6: Wrong values. */
unit = dt_create(-1, NULL, NULL, NULL);
ok(unit == NULL, "dthreads: create with negative count");
/* Test 7: NULL operations crashing. */
int ret = 0;
ret += dt_activate(0);
ret += dt_cancel(0);
ret += dt_compact(0);
dt_delete(0);
ret += dt_is_cancelled(0);
ret += dt_join(0);
ret += dt_signalize(0, SIGALRM);
ret += dt_start(0);
ret += dt_stop(0);
ret += dt_unit_lock(0);
ret += dt_unit_unlock(0);
is_int(-1098, ret, "dthreads: correct values when passed NULL context");
/* Test 8: Thread destructor. */
_destructor_data = 0;
unit = dt_create(2, 0, destruct, 0);
dt_start(unit);
dt_stop(unit);
dt_join(unit);
is_int(2, _destructor_data, "dthreads: destructor with dt_create_coherent()");
dt_delete(&unit);
skip_all:
pthread_mutex_destroy(&_runnable_mx);
pthread_mutex_destroy(&_destructor_mx);
return 0;
}
int ofd_precreate_objects(const struct lu_env *env, struct ofd_device *ofd,
obd_id id, struct ofd_seq *oseq, int nr, int sync)
{
struct ofd_thread_info *info = ofd_info(env);
struct ofd_object *fo = NULL;
struct dt_object *next;
struct thandle *th;
struct ofd_object **batch;
struct lu_fid *fid = &info->fti_fid;
obd_id tmp;
int rc;
int i;
int objects = 0;
int nr_saved = nr;
ENTRY;
/* Don't create objects beyond the valid range for this SEQ */
if (unlikely(fid_seq_is_mdt0(ostid_seq(&oseq->os_oi)) &&
(id + nr) >= IDIF_MAX_OID)) {
CERROR("%s:"DOSTID" hit the IDIF_MAX_OID (1<<48)!\n",
ofd_name(ofd), id, ostid_seq(&oseq->os_oi));
RETURN(rc = -ENOSPC);
} else if (unlikely(!fid_seq_is_mdt0(ostid_seq(&oseq->os_oi)) &&
(id + nr) >= OBIF_MAX_OID)) {
CERROR("%s:"DOSTID" hit the OBIF_MAX_OID (1<<32)!\n",
ofd_name(ofd), id, ostid_seq(&oseq->os_oi));
RETURN(rc = -ENOSPC);
}
OBD_ALLOC(batch, nr_saved * sizeof(struct ofd_object *));
if (batch == NULL)
RETURN(-ENOMEM);
info->fti_attr.la_valid = LA_TYPE | LA_MODE;
/*
* We mark object SUID+SGID to flag it for accepting UID+GID from
* client on first write. Currently the permission bits on the OST are
* never used, so this is OK.
*/
info->fti_attr.la_mode = S_IFREG | S_ISUID | S_ISGID | 0666;
info->fti_dof.dof_type = dt_mode_to_dft(S_IFREG);
/* Initialize a/c/m time so any client timestamp will always
* be newer and update the inode. ctime = 0 is also handled
* specially in osd_inode_setattr(). See LU-221, LU-1042 */
info->fti_attr.la_valid |= LA_ATIME | LA_MTIME | LA_CTIME;
info->fti_attr.la_atime = 0;
info->fti_attr.la_mtime = 0;
info->fti_attr.la_ctime = 0;
LASSERT(id != 0);
/* prepare objects */
*fid = *lu_object_fid(&oseq->os_lastid_obj->do_lu);
for (i = 0; i < nr; i++) {
rc = fid_set_id(fid, id + i);
if (rc != 0) {
if (i == 0)
GOTO(out, rc);
nr = i;
break;
}
fo = ofd_object_find(env, ofd, fid);
if (IS_ERR(fo)) {
if (i == 0)
GOTO(out, rc = PTR_ERR(fo));
nr = i;
break;
}
ofd_write_lock(env, fo);
batch[i] = fo;
}
info->fti_buf.lb_buf = &tmp;
info->fti_buf.lb_len = sizeof(tmp);
info->fti_off = 0;
th = ofd_trans_create(env, ofd);
if (IS_ERR(th))
GOTO(out, rc = PTR_ERR(th));
th->th_sync |= sync;
rc = dt_declare_record_write(env, oseq->os_lastid_obj, &info->fti_buf,
info->fti_off, th);
if (rc)
GOTO(trans_stop, rc);
for (i = 0; i < nr; i++) {
fo = batch[i];
LASSERT(fo);
if (unlikely(ofd_object_exists(fo))) {
/* object may exist being re-created by write replay */
CDEBUG(D_INODE, "object "LPX64"/"LPX64" exists: "
DFID"\n", ostid_seq(&oseq->os_oi), id,
PFID(lu_object_fid(&fo->ofo_obj.do_lu)));
continue;
}
next = ofd_object_child(fo);
LASSERT(next != NULL);
rc = dt_declare_create(env, next, &info->fti_attr, NULL,
&info->fti_dof, th);
if (rc) {
nr = i;
break;
}
}
rc = dt_trans_start_local(env, ofd->ofd_osd, th);
if (rc)
GOTO(trans_stop, rc);
CDEBUG(D_OTHER, "%s: create new object "DFID" nr %d\n",
ofd_name(ofd), PFID(fid), nr);
LASSERT(nr > 0);
/* When the LFSCK scanning the whole device to verify the LAST_ID file
* consistency, it will load the last_id into RAM firstly, and compare
* the last_id with each OST-object's ID. If the later one is larger,
* then it will regard the LAST_ID file crashed. But during the LFSCK
* scanning, the OFD may continue to create new OST-objects. Those new
* created OST-objects will have larger IDs than the LFSCK known ones.
* So from the LFSCK view, it needs to re-load the last_id from disk
* file, and if the latest last_id is still smaller than the object's
* ID, then the LAST_ID file is real crashed.
*
* To make above mechanism to work, before OFD pre-create OST-objects,
* it needs to update the LAST_ID file firstly, otherwise, the LFSCK
* may cannot get latest last_id although new OST-object created. */
if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_SKIP_LASTID)) {
tmp = cpu_to_le64(id + nr - 1);
dt_write_lock(env, oseq->os_lastid_obj, 0);
rc = dt_record_write(env, oseq->os_lastid_obj,
&info->fti_buf, &info->fti_off, th);
dt_write_unlock(env, oseq->os_lastid_obj);
if (rc != 0)
GOTO(trans_stop, rc);
}
for (i = 0; i < nr; i++) {
fo = batch[i];
LASSERT(fo);
/* Only the new created objects need to be recorded. */
if (ofd->ofd_osd->dd_record_fid_accessed) {
lfsck_pack_rfa(&ofd_info(env)->fti_lr,
lu_object_fid(&fo->ofo_obj.do_lu));
lfsck_in_notify(env, ofd->ofd_osd,
&ofd_info(env)->fti_lr);
}
if (likely(!ofd_object_exists(fo) &&
!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_DANGLING))) {
next = ofd_object_child(fo);
LASSERT(next != NULL);
rc = dt_create(env, next, &info->fti_attr, NULL,
&info->fti_dof, th);
if (rc)
break;
LASSERT(ofd_object_exists(fo));
}
ofd_seq_last_oid_set(oseq, id + i);
}
objects = i;
/* NOT all the wanted objects have been created,
* set the LAST_ID as the real created. */
if (unlikely(objects < nr)) {
int rc1;
info->fti_off = 0;
tmp = cpu_to_le64(ofd_seq_last_oid(oseq));
dt_write_lock(env, oseq->os_lastid_obj, 0);
rc1 = dt_record_write(env, oseq->os_lastid_obj,
&info->fti_buf, &info->fti_off, th);
dt_write_unlock(env, oseq->os_lastid_obj);
if (rc1 != 0)
CERROR("%s: fail to reset the LAST_ID for seq ("LPX64
") from "LPU64" to "LPU64"\n", ofd_name(ofd),
ostid_seq(&oseq->os_oi), id + nr - 1,
ofd_seq_last_oid(oseq));
}
trans_stop:
ofd_trans_stop(env, ofd, th, rc);
out:
for (i = 0; i < nr_saved; i++) {
fo = batch[i];
if (fo) {
ofd_write_unlock(env, fo);
ofd_object_put(env, fo);
}
}
OBD_FREE(batch, nr_saved * sizeof(struct ofd_object *));
CDEBUG((objects == 0 && rc == 0) ? D_ERROR : D_OTHER,
"created %d/%d objects: %d\n", objects, nr_saved, rc);
LASSERT(ergo(objects == 0, rc < 0));
RETURN(objects > 0 ? objects : rc);
}
int ofd_precreate_objects(const struct lu_env *env, struct ofd_device *ofd,
obd_id id, struct ofd_seq *oseq, int nr, int sync)
{
struct ofd_thread_info *info = ofd_info(env);
struct ofd_object *fo = NULL;
struct dt_object *next;
struct thandle *th;
struct ofd_object **batch;
obd_id tmp;
int rc;
int i;
int objects = 0;
int nr_saved = nr;
ENTRY;
/* Don't create objects beyond the valid range for this SEQ */
if (unlikely(fid_seq_is_mdt0(ostid_seq(&oseq->os_oi)) &&
(id + nr) >= IDIF_MAX_OID)) {
CERROR("%s:"DOSTID" hit the IDIF_MAX_OID (1<<48)!\n",
ofd_name(ofd), id, ostid_seq(&oseq->os_oi));
RETURN(rc = -ENOSPC);
} else if (unlikely(!fid_seq_is_mdt0(ostid_seq(&oseq->os_oi)) &&
(id + nr) >= OBIF_MAX_OID)) {
CERROR("%s:"DOSTID" hit the OBIF_MAX_OID (1<<32)!\n",
ofd_name(ofd), id, ostid_seq(&oseq->os_oi));
RETURN(rc = -ENOSPC);
}
OBD_ALLOC(batch, nr_saved * sizeof(struct ofd_object *));
if (batch == NULL)
RETURN(-ENOMEM);
info->fti_attr.la_valid = LA_TYPE | LA_MODE;
/*
* We mark object SUID+SGID to flag it for accepting UID+GID from
* client on first write. Currently the permission bits on the OST are
* never used, so this is OK.
*/
info->fti_attr.la_mode = S_IFREG | S_ISUID | S_ISGID | 0666;
info->fti_dof.dof_type = dt_mode_to_dft(S_IFREG);
/* Initialize a/c/m time so any client timestamp will always
* be newer and update the inode. ctime = 0 is also handled
* specially in osd_inode_setattr(). See LU-221, LU-1042 */
info->fti_attr.la_valid |= LA_ATIME | LA_MTIME | LA_CTIME;
info->fti_attr.la_atime = 0;
info->fti_attr.la_mtime = 0;
info->fti_attr.la_ctime = 0;
/* prepare objects */
ostid_set_seq(&info->fti_ostid, ostid_seq(&oseq->os_oi));
for (i = 0; i < nr; i++) {
ostid_set_id(&info->fti_ostid, id + i);
rc = ostid_to_fid(&info->fti_fid, &info->fti_ostid, 0);
if (rc) {
if (i == 0)
GOTO(out, rc);
nr = i;
break;
}
fo = ofd_object_find(env, ofd, &info->fti_fid);
if (IS_ERR(fo)) {
if (i == 0)
GOTO(out, rc = PTR_ERR(fo));
nr = i;
break;
}
ofd_write_lock(env, fo);
batch[i] = fo;
}
info->fti_buf.lb_buf = &tmp;
info->fti_buf.lb_len = sizeof(tmp);
info->fti_off = 0;
th = ofd_trans_create(env, ofd);
if (IS_ERR(th))
GOTO(out, rc = PTR_ERR(th));
th->th_sync |= sync;
rc = dt_declare_record_write(env, oseq->os_lastid_obj, sizeof(tmp),
info->fti_off, th);
if (rc)
GOTO(trans_stop, rc);
for (i = 0; i < nr; i++) {
fo = batch[i];
LASSERT(fo);
if (unlikely(ofd_object_exists(fo))) {
/* object may exist being re-created by write replay */
CDEBUG(D_INODE, "object "LPX64"/"LPX64" exists: "
DFID"\n", ostid_seq(&oseq->os_oi), id,
PFID(&info->fti_fid));
continue;
}
next = ofd_object_child(fo);
LASSERT(next != NULL);
rc = dt_declare_create(env, next, &info->fti_attr, NULL,
&info->fti_dof, th);
if (rc) {
nr = i;
break;
}
}
rc = dt_trans_start_local(env, ofd->ofd_osd, th);
if (rc)
GOTO(trans_stop, rc);
CDEBUG(D_OTHER, "%s: create new object "DFID" nr %d\n",
ofd_name(ofd), PFID(&info->fti_fid), nr);
for (i = 0; i < nr; i++) {
fo = batch[i];
LASSERT(fo);
if (likely(!ofd_object_exists(fo))) {
next = ofd_object_child(fo);
LASSERT(next != NULL);
rc = dt_create(env, next, &info->fti_attr, NULL,
&info->fti_dof, th);
if (rc)
break;
LASSERT(ofd_object_exists(fo));
}
ofd_seq_last_oid_set(oseq, id + i);
}
objects = i;
if (objects > 0) {
tmp = cpu_to_le64(ofd_seq_last_oid(oseq));
rc = dt_record_write(env, oseq->os_lastid_obj,
&info->fti_buf, &info->fti_off, th);
}
trans_stop:
ofd_trans_stop(env, ofd, th, rc);
out:
for (i = 0; i < nr_saved; i++) {
fo = batch[i];
if (fo) {
ofd_write_unlock(env, fo);
ofd_object_put(env, fo);
}
}
OBD_FREE(batch, nr_saved * sizeof(struct ofd_object *));
CDEBUG((objects == 0 && rc == 0) ? D_ERROR : D_OTHER,
"created %d/%d objects: %d\n", objects, nr_saved, rc);
LASSERT(ergo(objects == 0, rc < 0));
RETURN(objects > 0 ? objects : rc);
}
