#include "ofd_internal.h"
#include <obd_cksum.h>
#include <lustre_quota.h>
+#include <lustre_lfsck.h>
static int ofd_export_stats_init(struct ofd_device *ofd,
struct obd_export *exp, void *client_nid)
ENTRY;
+ LASSERT(obd->obd_uses_nid_stats);
+
if (obd_uuid_equals(&exp->exp_client_uuid, &obd->obd_uuid))
/* Self-export gets no proc entry */
RETURN(0);
stats = exp->exp_nid_stats;
LASSERT(stats != NULL);
- num_stats = (sizeof(*obd->obd_type->typ_dt_ops) / sizeof(void *)) +
- LPROC_OFD_LAST - 1;
-
+ num_stats = NUM_OBD_STATS + LPROC_OFD_LAST;
stats->nid_stats = lprocfs_alloc_stats(num_stats,
LPROCFS_STATS_FLAG_NOPERCPU);
if (stats->nid_stats == NULL)
fed->fed_group = data->ocd_group;
data->ocd_connect_flags &= OST_CONNECT_SUPPORTED;
- exp->exp_connect_data = *data;
data->ocd_version = LUSTRE_VERSION_CODE;
/* Kindly make sure the SKIP_ORPHAN flag is from MDS. */
data->ocd_grant_extent = ofd->ofd_dt_conf.ddp_grant_frag >> 10;
}
- if (exp_connect_flags(exp) & OBD_CONNECT_GRANT)
+ if (data->ocd_connect_flags & OBD_CONNECT_GRANT)
data->ocd_grant = ofd_grant_connect(env, exp, data->ocd_grant,
new_connection);
}
CDEBUG(D_HA, "%s: get connection from MDS %d\n", obd->obd_name,
- data->ocd_group);
+ data ? data->ocd_group : -1);
out:
if (rc != 0) {
struct lov_stripe_md *lsm)
{
struct ofd_device *ofd;
- int rc = 0;
+ int rc = 0;
ENTRY;
RETURN(-EINVAL);
}
+ /* Because ofd_get_info might be called from
+ * handle_request_in as well(see LU-3239), where env might
+ * not be initilaized correctly, and le_ses might be in
+ * an un-initialized state, so only refill le_ctx here */
+ rc = lu_env_refill((struct lu_env *)env);
+ if (rc != 0)
+ RETURN(rc);
+
ofd = ofd_exp(exp);
if (KEY_IS(KEY_BLOCKSIZE)) {
__u32 *blocksize = val;
if (blocksize) {
if (*vallen < sizeof(*blocksize))
- RETURN(-EOVERFLOW);
+ GOTO(out, rc = -EOVERFLOW);
*blocksize = 1 << ofd->ofd_dt_conf.ddp_block_shift;
}
*vallen = sizeof(*blocksize);
__u32 *blocksize_bits = val;
if (blocksize_bits) {
if (*vallen < sizeof(*blocksize_bits))
- RETURN(-EOVERFLOW);
+ GOTO(out, rc = -EOVERFLOW);
*blocksize_bits = ofd->ofd_dt_conf.ddp_block_shift;
}
*vallen = sizeof(*blocksize_bits);
if (val == NULL) {
*vallen = sizeof(obd_id);
- RETURN(0);
+ GOTO(out, rc = 0);
}
ofd_info_init(env, exp);
oseq = ofd_seq_load(env, ofd,
if (last_id) {
if (*vallen < sizeof(*last_id)) {
ofd_seq_put(env, oseq);
- RETURN(-EOVERFLOW);
+ GOTO(out, rc = -EOVERFLOW);
}
*last_id = ofd_seq_last_oid(oseq);
}
ofd_seq_put(env, oseq);
*vallen = sizeof(*last_id);
} else if (KEY_IS(KEY_FIEMAP)) {
- struct ofd_thread_info *info;
struct ofd_device *ofd = ofd_exp(exp);
struct ofd_object *fo;
struct ll_fiemap_info_key *fm_key = key;
+ struct lu_fid fid;
if (val == NULL) {
*vallen = fiemap_count_to_size(
fm_key->fiemap.fm_extent_count);
- RETURN(0);
+ GOTO(out, rc = 0);
}
- info = ofd_info_init(env, exp);
- rc = ostid_to_fid(&info->fti_fid, &fm_key->oa.o_oi, 0);
+ rc = ostid_to_fid(&fid, &fm_key->oa.o_oi, 0);
if (rc != 0)
- RETURN(rc);
+ GOTO(out, rc);
CDEBUG(D_INODE, "get FIEMAP of object "DFID"\n",
- PFID(&info->fti_fid));
+ PFID(&fid));
- fo = ofd_object_find(env, ofd, &info->fti_fid);
+ fo = ofd_object_find(env, ofd, &fid);
if (IS_ERR(fo)) {
CERROR("%s: error finding object "DFID"\n",
- exp->exp_obd->obd_name, PFID(&info->fti_fid));
+ exp->exp_obd->obd_name, PFID(&fid));
rc = PTR_ERR(fo);
} else {
struct ll_user_fiemap *fiemap = val;
*((__u32 *) val) = ofd->ofd_sync_lock_cancel;
*vallen = sizeof(__u32);
} else if (KEY_IS(KEY_LAST_FID)) {
- struct lu_env env;
struct ofd_device *ofd = ofd_exp(exp);
struct ofd_seq *oseq;
struct lu_fid *fid = val;
if (fid == NULL) {
*vallen = sizeof(struct lu_fid);
- RETURN(0);
+ GOTO(out, rc = 0);
}
if (*vallen < sizeof(*fid))
- RETURN(-EOVERFLOW);
+ GOTO(out, rc = -EOVERFLOW);
- rc = lu_env_init(&env, LCT_DT_THREAD);
- if (rc != 0)
- RETURN(rc);
- ofd_info_init(&env, exp);
+ ofd_info_init(env, exp);
fid_le_to_cpu(fid, fid);
- oseq = ofd_seq_load(&env, ofd, ostid_seq((struct ost_id *)fid));
+ oseq = ofd_seq_load(env, ofd,
+ ostid_seq((struct ost_id *)fid));
if (IS_ERR(oseq))
- GOTO(out_fini, rc = PTR_ERR(oseq));
+ GOTO(out, rc = PTR_ERR(oseq));
rc = ostid_to_fid(fid, &oseq->os_oi,
ofd->ofd_lut.lut_lsd.lsd_osd_index);
PFID(fid));
*vallen = sizeof(*fid);
out_put:
- ofd_seq_put(&env, oseq);
-out_fini:
- lu_env_fini(&env);
+ ofd_seq_put(env, oseq);
} else {
CERROR("Not supported key %s\n", (char*)key);
rc = -EOPNOTSUPP;
}
-
+out:
RETURN(rc);
}
if (rc)
GOTO(out_unlock, rc);
- res = ldlm_resource_get(ns, NULL, &info->fti_resid, LDLM_EXTENT, 0);
- if (res != NULL) {
- ldlm_res_lvbo_update(res, NULL, 0);
- ldlm_resource_putref(res);
- }
-
- oinfo->oi_oa->o_valid = OBD_MD_FLID;
-
- /* Quota release needs uid/gid info */
- rc = ofd_attr_get(env, fo, &info->fti_attr);
- obdo_from_la(oinfo->oi_oa, &info->fti_attr,
- OFD_VALID_FLAGS | LA_UID | LA_GID);
ofd_info2oti(info, oti);
ofd_counter_incr(exp, LPROC_OFD_STATS_SETATTR, oti->oti_jobid, 1);
out_unlock:
ofd_object_put(env, fo);
out:
+ if (rc == 0) {
+ /* we do not call this before to avoid lu_object_find() in
+ * ->lvbo_update() holding another reference on the object.
+ * otherwise concurrent destroy can make the object unavailable
+ * for 2nd lu_object_find() waiting for the first reference
+ * to go... deadlock! */
+ res = ldlm_resource_get(ns, NULL, &info->fti_resid, LDLM_EXTENT, 0);
+ if (res != NULL) {
+ ldlm_res_lvbo_update(res, NULL, 0);
+ ldlm_resource_putref(res);
+ }
+ }
+
return rc;
}
{
struct ofd_thread_info *info = ofd_info(env);
struct lustre_handle lockh;
- __u64 flags = LDLM_AST_DISCARD_DATA, rc = 0;
+ __u64 flags = LDLM_FL_AST_DISCARD_DATA;
+ __u64 rc = 0;
ldlm_policy_data_t policy = {
.l_extent = { 0, OBD_OBJECT_EOF }
};
fo = ofd_object_find(env, ofd, fid);
if (IS_ERR(fo))
RETURN(PTR_ERR(fo));
+ if (!ofd_object_exists(fo))
+ GOTO(out, rc = -ENOENT);
/* Tell the clients that the object is gone now and that they should
* throw away any cached pages. */
LASSERT(fo != NULL);
rc = ofd_object_destroy(env, fo, orphan);
-
+ EXIT;
+out:
ofd_object_put(env, fo);
RETURN(rc);
}
skip_orphan = !!(exp_connect_flags(exp) & OBD_CONNECT_SKIP_ORPHAN);
last = ofd_seq_last_oid(oseq);
- LCONSOLE_INFO("%s: deleting orphan objects from "LPX64":"LPU64
- " to "LPU64"\n", ofd_name(ofd), ostid_seq(&oa->o_oi),
- end_id + 1, last);
+ LCONSOLE(D_INFO, "%s: deleting orphan objects from "DOSTID
+ " to "DOSTID"\n", ofd_name(ofd), ostid_seq(&oa->o_oi),
+ end_id + 1, ostid_seq(&oa->o_oi), last);
for (ostid_set_id(&oi, last); ostid_id(&oi) > end_id;
ostid_dec_id(&oi)) {
GOTO(out_put, rc);
rc = ofd_destroy_by_fid(env, ofd, &info->fti_fid, 1);
if (rc && rc != -ENOENT) /* this is pretty fatal... */
- CEMERG("error destroying precreated id "DOSTID": %d\n",
- POSTID(&oi), rc);
+ CEMERG("%s: error destroying precreated id "DOSTID
+ ": rc = %d\n", ofd_name(ofd), POSTID(&oi), rc);
if (!skip_orphan) {
ofd_seq_last_oid_set(oseq, ostid_id(&oi) - 1);
/* update last_id on disk periodically so that if we
ofd_seq_last_oid_write(env, ofd, oseq);
}
}
- CDEBUG(D_HA, "%s: after destroy: set last_objids"DOSTID"\n",
+ CDEBUG(D_HA, "%s: after destroy: set last_id to "DOSTID"\n",
ofd_obd(ofd)->obd_name, POSTID(&oa->o_oi));
if (!skip_orphan) {
rc = ofd_seq_last_oid_write(env, ofd, oseq);
}
}
+ /* This can happen if a new OST is formatted and installed
+ * in place of an old one at the same index. Instead of
+ * precreating potentially millions of deleted old objects
+ * (possibly filling the OST), only precreate the last batch.
+ * LFSCK will eventually clean up any orphans. LU-14 */
+ if (diff > 5 * OST_MAX_PRECREATE) {
+ diff = OST_MAX_PRECREATE / 2;
+ LCONSOLE_WARN("%s: precreate FID "DOSTID" is over %u "
+ "larger than the LAST_ID "DOSTID", only "
+ "precreating the last %u objects.\n",
+ ofd_name(ofd), POSTID(&oa->o_oi),
+ 5 * OST_MAX_PRECREATE,
+ POSTID(&oseq->os_oi), diff);
+ ofd_seq_last_oid_set(oseq, ostid_id(&oa->o_oi) - diff);
+ }
+
while (diff > 0) {
next_id = ofd_seq_last_oid(oseq) + 1;
count = ofd_precreate_batch(ofd, diff);
struct ofd_device *ofd = ofd_exp(exp);
struct ofd_thread_info *info;
struct ofd_object *fo;
- __u64 curr_version;
int rc = 0;
ENTRY;
fo = ofd_object_find(env, ofd, &info->fti_fid);
if (IS_ERR(fo))
GOTO(out, rc = PTR_ERR(fo));
+ if (!ofd_object_exists(fo))
+ GOTO(out_put, rc = -ENOENT);
+
LASSERT(fo != NULL);
rc = ofd_attr_get(env, fo, &info->fti_attr);
oinfo->oi_oa->o_valid = OBD_MD_FLID;
- if (rc == 0)
+ if (rc == 0) {
+ __u64 curr_version;
+
obdo_from_la(oinfo->oi_oa, &info->fti_attr,
OFD_VALID_FLAGS | LA_UID | LA_GID);
- /* Store object version in reply */
- curr_version = dt_version_get(env, ofd_object_child(fo));
- if ((__s64)curr_version != -EOPNOTSUPP) {
- oinfo->oi_oa->o_valid |= OBD_MD_FLDATAVERSION;
- oinfo->oi_oa->o_data_version = curr_version;
+ /* Store object version in reply */
+ curr_version = dt_version_get(env, ofd_object_child(fo));
+ if ((__s64)curr_version != -EOPNOTSUPP) {
+ oinfo->oi_oa->o_valid |= OBD_MD_FLDATAVERSION;
+ oinfo->oi_oa->o_data_version = curr_version;
+ }
}
+
+out_put:
ofd_object_put(env, fo);
out:
RETURN(rc);
if (rc == 0)
rc = dt_ro(&env, ofd->ofd_osd);
break;
+ case OBD_IOC_START_LFSCK: {
+ struct obd_ioctl_data *data = karg;
+ struct lfsck_start_param lsp;
+
+ if (unlikely(data == NULL)) {
+ rc = -EINVAL;
+ break;
+ }
+
+ lsp.lsp_start = (struct lfsck_start *)(data->ioc_inlbuf1);
+ lsp.lsp_namespace = ofd->ofd_namespace;
+ rc = lfsck_start(&env, ofd->ofd_osd, &lsp);
+ break;
+ }
+ case OBD_IOC_STOP_LFSCK: {
+ rc = lfsck_stop(&env, ofd->ofd_osd, false);
+ break;
+ }
case OBD_IOC_GET_OBJ_VERSION:
rc = ofd_ioc_get_obj_version(&env, ofd, karg);
break;
GOTO(out, rc = -EROFS);
#ifdef USE_HEALTH_CHECK_WRITE
- OBD_ALLOC(info->fti_buf.lb_buf, CFS_PAGE_SIZE);
+ OBD_ALLOC(info->fti_buf.lb_buf, PAGE_CACHE_SIZE);
if (info->fti_buf.lb_buf == NULL)
GOTO(out, rc = -ENOMEM);
- info->fti_buf.lb_len = CFS_PAGE_SIZE;
+ info->fti_buf.lb_len = PAGE_CACHE_SIZE;
info->fti_off = 0;
th = dt_trans_create(&env, ofd->ofd_osd);
}
dt_trans_stop(&env, ofd->ofd_osd, th);
- OBD_FREE(info->fti_buf.lb_buf, CFS_PAGE_SIZE);
+ OBD_FREE(info->fti_buf.lb_buf, PAGE_CACHE_SIZE);
CDEBUG(D_INFO, "write 1 page synchronously for checking io rc %d\n",rc);
#endif
git://git.whamcloud.com / fs / lustre-release.git / blobdiff