/*
* Copyright 2009 Sun Microsystems, Inc. All rights reserved
* Use is subject to license terms.
+ *
+ * Copyright (c) 2012, Intel Corporation.
*/
/*
* lustre/lod/lod_lov.c
#include "lod_internal.h"
/*
- * Keep a refcount of lod->lod_osts usage to prevent racing with
+ * Keep a refcount of lod->ltd_tgts usage to prevent racing with
* addition/deletion. Any function that expects lov_tgts to remain stationary
* must take a ref.
*
* \param lod - is the lod device from which we want to grab a reference
*/
-void lod_getref(struct lod_device *lod)
+void lod_getref(struct lod_tgt_descs *ltd)
{
- cfs_down_read(&lod->lod_rw_sem);
- cfs_mutex_lock(&lod->lod_mutex);
- lod->lod_refcount++;
- cfs_mutex_unlock(&lod->lod_mutex);
+ down_read(<d->ltd_rw_sem);
+ mutex_lock(<d->ltd_mutex);
+ ltd->ltd_refcount++;
+ mutex_unlock(<d->ltd_mutex);
}
/*
*
* \param lod - is the lod device from which we release a reference
*/
-void lod_putref(struct lod_device *lod)
+void lod_putref(struct lod_device *lod, struct lod_tgt_descs *ltd)
{
- cfs_mutex_lock(&lod->lod_mutex);
- lod->lod_refcount--;
- if (lod->lod_refcount == 0 && lod->lod_death_row) {
- struct lod_ost_desc *ost_desc, *tmp;
+ mutex_lock(<d->ltd_mutex);
+ ltd->ltd_refcount--;
+ if (ltd->ltd_refcount == 0 && ltd->ltd_death_row) {
+ struct lod_tgt_desc *tgt_desc, *tmp;
int idx;
CFS_LIST_HEAD(kill);
- CDEBUG(D_CONFIG, "destroying %d lod desc\n",
- lod->lod_death_row);
+ CDEBUG(D_CONFIG, "destroying %d ltd desc\n",
+ ltd->ltd_death_row);
- cfs_foreach_bit(lod->lod_ost_bitmap, idx) {
- ost_desc = OST_TGT(lod, idx);
- LASSERT(ost_desc);
+ cfs_foreach_bit(ltd->ltd_tgt_bitmap, idx) {
+ tgt_desc = LTD_TGT(ltd, idx);
+ LASSERT(tgt_desc);
- if (!ost_desc->ltd_reap)
+ if (!tgt_desc->ltd_reap)
continue;
- cfs_list_add(&ost_desc->ltd_kill, &kill);
-
- lod_ost_pool_remove(&lod->lod_pool_info, idx);
- OST_TGT(lod, idx) = NULL;
- lod->lod_ostnr--;
- cfs_bitmap_clear(lod->lod_ost_bitmap, idx);
- if (ost_desc->ltd_active)
- lod->lod_desc.ld_active_tgt_count--;
- lod->lod_death_row--;
+ cfs_list_add(&tgt_desc->ltd_kill, &kill);
+ LTD_TGT(ltd, idx) = NULL;
+ /*FIXME: only support ost pool for now */
+ if (ltd == &lod->lod_ost_descs) {
+ lod_ost_pool_remove(&lod->lod_pool_info, idx);
+ if (tgt_desc->ltd_active)
+ lod->lod_desc.ld_active_tgt_count--;
+ }
+ ltd->ltd_tgtnr--;
+ cfs_bitmap_clear(ltd->ltd_tgt_bitmap, idx);
+ ltd->ltd_death_row--;
}
- cfs_mutex_unlock(&lod->lod_mutex);
- cfs_up_read(&lod->lod_rw_sem);
+ mutex_unlock(<d->ltd_mutex);
+ up_read(<d->ltd_rw_sem);
- cfs_list_for_each_entry_safe(ost_desc, tmp, &kill, ltd_kill) {
+ cfs_list_for_each_entry_safe(tgt_desc, tmp, &kill, ltd_kill) {
int rc;
- cfs_list_del(&ost_desc->ltd_kill);
- /* remove from QoS structures */
- rc = qos_del_tgt(lod, ost_desc);
- if (rc)
- CERROR("%s: qos_del_tgt(%s) failed: rc = %d\n",
- lod2obd(lod)->obd_name,
- obd_uuid2str(&ost_desc->ltd_uuid), rc);
- rc = obd_disconnect(ost_desc->ltd_exp);
+ cfs_list_del(&tgt_desc->ltd_kill);
+ if (ltd == &lod->lod_ost_descs) {
+ /* remove from QoS structures */
+ rc = qos_del_tgt(lod, tgt_desc);
+ if (rc)
+ CERROR("%s: qos_del_tgt(%s) failed:"
+ "rc = %d\n",
+ lod2obd(lod)->obd_name,
+ obd_uuid2str(&tgt_desc->ltd_uuid),
+ rc);
+ }
+ rc = obd_disconnect(tgt_desc->ltd_exp);
if (rc)
CERROR("%s: failed to disconnect %s: rc = %d\n",
lod2obd(lod)->obd_name,
- obd_uuid2str(&ost_desc->ltd_uuid), rc);
- OBD_FREE_PTR(ost_desc);
+ obd_uuid2str(&tgt_desc->ltd_uuid), rc);
+ OBD_FREE_PTR(tgt_desc);
}
} else {
- cfs_mutex_unlock(&lod->lod_mutex);
- cfs_up_read(&lod->lod_rw_sem);
+ mutex_unlock(<d->ltd_mutex);
+ up_read(<d->ltd_rw_sem);
}
}
-static int lod_bitmap_resize(struct lod_device *lod, __u32 newsize)
+static int ltd_bitmap_resize(struct lod_tgt_descs *ltd, __u32 newsize)
{
cfs_bitmap_t *new_bitmap, *old_bitmap = NULL;
int rc = 0;
/* grab write reference on the lod. Relocating the array requires
* exclusive access */
- cfs_down_write(&lod->lod_rw_sem);
- if (newsize <= lod->lod_osts_size)
+ down_write(<d->ltd_rw_sem);
+ if (newsize <= ltd->ltd_tgts_size)
/* someone else has already resize the array */
GOTO(out, rc = 0);
if (!new_bitmap)
GOTO(out, rc = -ENOMEM);
- if (lod->lod_osts_size > 0) {
+ if (ltd->ltd_tgts_size > 0) {
/* the bitmap already exists, we need
* to copy data from old one */
- cfs_bitmap_copy(new_bitmap, lod->lod_ost_bitmap);
- old_bitmap = lod->lod_ost_bitmap;
+ cfs_bitmap_copy(new_bitmap, ltd->ltd_tgt_bitmap);
+ old_bitmap = ltd->ltd_tgt_bitmap;
}
- lod->lod_osts_size = newsize;
- lod->lod_ost_bitmap = new_bitmap;
+ ltd->ltd_tgts_size = newsize;
+ ltd->ltd_tgt_bitmap = new_bitmap;
if (old_bitmap)
CFS_FREE_BITMAP(old_bitmap);
- CDEBUG(D_CONFIG, "ost size: %d\n", lod->lod_osts_size);
+ CDEBUG(D_CONFIG, "tgt size: %d\n", ltd->ltd_tgts_size);
EXIT;
out:
- cfs_up_write(&lod->lod_rw_sem);
+ up_write(<d->ltd_rw_sem);
return rc;
}
* \param osp - is the name of OSP device name about to be added
* \param index - is the OSP index
* \param gen - is the generation number
+ * \param tgt_index - is the group of the OSP.
+ * \param type - is the type of device (mdc or osc)
*/
int lod_add_device(const struct lu_env *env, struct lod_device *lod,
- char *osp, unsigned index, unsigned gen, int active)
+ char *osp, unsigned index, unsigned gen, int tgt_index,
+ char *type, int active)
{
struct obd_connect_data *data = NULL;
struct obd_export *exp = NULL;
struct lu_device *ldev;
struct dt_device *d;
int rc;
- struct lod_ost_desc *ost_desc;
- struct obd_uuid obd_uuid;
-
+ struct lod_tgt_desc *tgt_desc;
+ struct lod_tgt_descs *ltd;
+ struct obd_uuid obd_uuid;
ENTRY;
CDEBUG(D_CONFIG, "osp:%s idx:%d gen:%d\n", osp, index, gen);
data->ocd_version = LUSTRE_VERSION_CODE;
data->ocd_index = index;
+ if (strcmp(LUSTRE_OSC_NAME, type) == 0) {
+ data->ocd_connect_flags |= OBD_CONNECT_AT |
+ OBD_CONNECT_FULL20 |
+ OBD_CONNECT_INDEX |
+#ifdef HAVE_LRU_RESIZE_SUPPORT
+ OBD_CONNECT_LRU_RESIZE |
+#endif
+ OBD_CONNECT_MDS |
+ OBD_CONNECT_OSS_CAPA |
+ OBD_CONNECT_REQPORTAL |
+ OBD_CONNECT_SKIP_ORPHAN |
+ OBD_CONNECT_FID |
+ OBD_CONNECT_LVB_TYPE |
+ OBD_CONNECT_VERSION |
+ OBD_CONNECT_PINGLESS;
+
+ data->ocd_group = tgt_index;
+ ltd = &lod->lod_ost_descs;
+ } else {
+ struct obd_import *imp = obd->u.cli.cl_import;
+
+ data->ocd_ibits_known = MDS_INODELOCK_UPDATE;
+ data->ocd_connect_flags |= OBD_CONNECT_ACL |
+ OBD_CONNECT_MDS_CAPA |
+ OBD_CONNECT_OSS_CAPA |
+ OBD_CONNECT_IBITS |
+ OBD_CONNECT_MDS_MDS |
+ OBD_CONNECT_FID |
+ OBD_CONNECT_AT |
+ OBD_CONNECT_FULL20 |
+ OBD_CONNECT_PINGLESS;
+ /* XXX set MDS-MDS flags, remove this when running this
+ * on client*/
+ data->ocd_connect_flags |= OBD_CONNECT_MDS_MDS;
+ spin_lock(&imp->imp_lock);
+ imp->imp_server_timeout = 1;
+ spin_unlock(&imp->imp_lock);
+ imp->imp_client->cli_request_portal = MDS_MDS_PORTAL;
+ CDEBUG(D_OTHER, "%s: Set 'mds' portal and timeout\n",
+ obd->obd_name);
+ ltd = &lod->lod_mdt_descs;
+ }
+
rc = obd_connect(env, &exp, obd, &obd->obd_uuid, data, NULL);
OBD_FREE_PTR(data);
if (rc) {
d = lu2dt_dev(ldev);
/* Allocate ost descriptor and fill it */
- OBD_ALLOC_PTR(ost_desc);
- if (!ost_desc)
+ OBD_ALLOC_PTR(tgt_desc);
+ if (!tgt_desc)
GOTO(out_conn, rc = -ENOMEM);
- ost_desc->ltd_ost = d;
- ost_desc->ltd_exp = exp;
- ost_desc->ltd_uuid = obd->u.cli.cl_target_uuid;
- ost_desc->ltd_gen = gen;
- ost_desc->ltd_index = index;
- ost_desc->ltd_active = active;
+ tgt_desc->ltd_tgt = d;
+ tgt_desc->ltd_exp = exp;
+ tgt_desc->ltd_uuid = obd->u.cli.cl_target_uuid;
+ tgt_desc->ltd_gen = gen;
+ tgt_desc->ltd_index = index;
+ tgt_desc->ltd_active = active;
- lod_getref(lod);
- if (index >= lod->lod_osts_size) {
+ lod_getref(ltd);
+ if (index >= ltd->ltd_tgts_size) {
/* we have to increase the size of the lod_osts array */
__u32 newsize;
- newsize = max(lod->lod_osts_size, (__u32)2);
+ newsize = max(ltd->ltd_tgts_size, (__u32)2);
while (newsize < index + 1)
newsize = newsize << 1;
/* lod_bitmap_resize() needs lod_rw_sem
* which we hold with th reference */
- lod_putref(lod);
+ lod_putref(lod, ltd);
- rc = lod_bitmap_resize(lod, newsize);
+ rc = ltd_bitmap_resize(ltd, newsize);
if (rc)
GOTO(out_desc, rc);
- lod_getref(lod);
+ lod_getref(ltd);
}
- cfs_mutex_lock(&lod->lod_mutex);
- if (cfs_bitmap_check(lod->lod_ost_bitmap, index)) {
+ mutex_lock(<d->ltd_mutex);
+ if (cfs_bitmap_check(ltd->ltd_tgt_bitmap, index)) {
CERROR("%s: device %d is registered already\n", obd->obd_name,
index);
GOTO(out_mutex, rc = -EEXIST);
}
- if (lod->lod_ost_idx[index / OST_PTRS_PER_BLOCK] == NULL) {
- OBD_ALLOC_PTR(lod->lod_ost_idx[index / OST_PTRS_PER_BLOCK]);
- if (lod->lod_ost_idx[index / OST_PTRS_PER_BLOCK] == NULL) {
+ if (ltd->ltd_tgt_idx[index / TGT_PTRS_PER_BLOCK] == NULL) {
+ OBD_ALLOC_PTR(ltd->ltd_tgt_idx[index / TGT_PTRS_PER_BLOCK]);
+ if (ltd->ltd_tgt_idx[index / TGT_PTRS_PER_BLOCK] == NULL) {
CERROR("can't allocate index to add %s\n",
obd->obd_name);
GOTO(out_mutex, rc = -ENOMEM);
}
}
- rc = lod_ost_pool_add(&lod->lod_pool_info, index, lod->lod_osts_size);
- if (rc) {
- CERROR("%s: can't set up pool, failed with %d\n",
- obd->obd_name, rc);
- GOTO(out_mutex, rc);
- }
+ if (!strcmp(LUSTRE_OSC_NAME, type)) {
+ /* pool and qos are not supported for MDS stack yet */
+ rc = lod_ost_pool_add(&lod->lod_pool_info, index,
+ lod->lod_osts_size);
+ if (rc) {
+ CERROR("%s: can't set up pool, failed with %d\n",
+ obd->obd_name, rc);
+ GOTO(out_mutex, rc);
+ }
- rc = qos_add_tgt(lod, ost_desc);
- if (rc) {
- CERROR("%s: qos_add_tgt(%s) failed: rc = %d\n", obd->obd_name,
- obd_uuid2str(&ost_desc->ltd_uuid), rc);
- GOTO(out_pool, rc);
- }
-
- /* The new OST is now a full citizen */
- if (index >= lod->lod_desc.ld_tgt_count)
- lod->lod_desc.ld_tgt_count = index + 1;
- if (active)
- lod->lod_desc.ld_active_tgt_count++;
- OST_TGT(lod, index) = ost_desc;
- cfs_bitmap_set(lod->lod_ost_bitmap, index);
- lod->lod_ostnr++;
- cfs_mutex_unlock(&lod->lod_mutex);
- lod_putref(lod);
+ rc = qos_add_tgt(lod, tgt_desc);
+ if (rc) {
+ CERROR("%s: qos_add_tgt failed with %d\n",
+ obd->obd_name, rc);
+ GOTO(out_pool, rc);
+ }
+
+ /* The new OST is now a full citizen */
+ if (index >= lod->lod_desc.ld_tgt_count)
+ lod->lod_desc.ld_tgt_count = index + 1;
+ if (active)
+ lod->lod_desc.ld_active_tgt_count++;
+ }
+ LTD_TGT(ltd, index) = tgt_desc;
+ cfs_bitmap_set(ltd->ltd_tgt_bitmap, index);
+ ltd->ltd_tgtnr++;
+ mutex_unlock(<d->ltd_mutex);
+ lod_putref(lod, ltd);
if (lod->lod_recovery_completed)
ldev->ld_ops->ldo_recovery_complete(env, ldev);
out_pool:
lod_ost_pool_remove(&lod->lod_pool_info, index);
out_mutex:
- cfs_mutex_unlock(&lod->lod_mutex);
- lod_putref(lod);
+ mutex_unlock(<d->ltd_mutex);
+ lod_putref(lod, ltd);
out_desc:
- OBD_FREE_PTR(ost_desc);
+ OBD_FREE_PTR(tgt_desc);
out_conn:
obd_disconnect(exp);
out_free:
/*
* helper function to schedule OST removal from the device table
*/
-static void __lod_del_device(struct lod_device *lod, unsigned idx)
+static void __lod_del_device(struct lod_tgt_descs *ltd,
+ unsigned idx)
{
- LASSERT(OST_TGT(lod,idx));
- if (OST_TGT(lod,idx)->ltd_reap == 0) {
- OST_TGT(lod,idx)->ltd_reap = 1;
- lod->lod_death_row++;
+ LASSERT(LTD_TGT(ltd, idx));
+ if (LTD_TGT(ltd, idx)->ltd_reap == 0) {
+ LTD_TGT(ltd, idx)->ltd_reap = 1;
+ ltd->ltd_death_row++;
}
}
+int lod_fini_tgt(struct lod_device *lod, struct lod_tgt_descs *ltd)
+{
+ int idx;
+
+ if (ltd->ltd_tgts_size <= 0)
+ return 0;
+ lod_getref(ltd);
+ mutex_lock(<d->ltd_mutex);
+ cfs_foreach_bit(ltd->ltd_tgt_bitmap, idx)
+ __lod_del_device(ltd, idx);
+ mutex_unlock(<d->ltd_mutex);
+ lod_putref(lod, ltd);
+ CFS_FREE_BITMAP(ltd->ltd_tgt_bitmap);
+ for (idx = 0; idx < TGT_PTRS; idx++) {
+ if (ltd->ltd_tgt_idx[idx])
+ OBD_FREE_PTR(ltd->ltd_tgt_idx[idx]);
+ }
+ ltd->ltd_tgts_size = 0;
+ return 0;
+}
+
/*
* Add support for administratively disabled OST (through the MGS).
* Schedule a target for deletion. Disconnection and real removal from the
* \param gen - is the generation number, not used currently
*/
int lod_del_device(const struct lu_env *env, struct lod_device *lod,
- char *osp, unsigned idx, unsigned gen)
+ struct lod_tgt_descs *ltd, char *osp, unsigned idx,
+ unsigned gen)
{
struct obd_device *obd;
int rc = 0;
CDEBUG(D_CONFIG, "osp:%s idx:%d gen:%d\n", osp, idx, gen);
- obd = class_name2obd(osp);
+ obd_str2uuid(&uuid, osp);
+
+ obd = class_find_client_obd(&uuid, LUSTRE_OSP_NAME,
+ &lod->lod_dt_dev.dd_lu_dev.ld_obd->obd_uuid);
if (obd == NULL) {
CERROR("can't find %s device\n", osp);
RETURN(-EINVAL);
obd_str2uuid(&uuid, osp);
- lod_getref(lod);
- cfs_mutex_lock(&lod->lod_mutex);
+ lod_getref(ltd);
+ mutex_lock(<d->ltd_mutex);
/* check that the index is allocated in the bitmap */
- if (!cfs_bitmap_check(lod->lod_ost_bitmap, idx) || !OST_TGT(lod,idx)) {
+ if (!cfs_bitmap_check(ltd->ltd_tgt_bitmap, idx) ||
+ !LTD_TGT(ltd, idx)) {
CERROR("%s: device %d is not set up\n", obd->obd_name, idx);
GOTO(out, rc = -EINVAL);
}
/* check that the UUID matches */
- if (!obd_uuid_equals(&uuid, &OST_TGT(lod,idx)->ltd_uuid)) {
+ if (!obd_uuid_equals(&uuid, <D_TGT(ltd, idx)->ltd_uuid)) {
CERROR("%s: LOD target UUID %s at index %d does not match %s\n",
- obd->obd_name, obd_uuid2str(&OST_TGT(lod,idx)->ltd_uuid),
+ obd->obd_name, obd_uuid2str(<D_TGT(ltd,idx)->ltd_uuid),
idx, osp);
GOTO(out, rc = -EINVAL);
}
- __lod_del_device(lod, idx);
+ __lod_del_device(ltd, idx);
EXIT;
out:
- cfs_mutex_unlock(&lod->lod_mutex);
- lod_putref(lod);
+ mutex_unlock(<d->ltd_mutex);
+ lod_putref(lod, ltd);
return(rc);
}
RETURN(0);
}
+/*
+ * generate and write LOV EA for given striped object
+ */
+int lod_generate_and_set_lovea(const struct lu_env *env,
+ struct lod_object *lo, struct thandle *th)
+{
+ struct lod_thread_info *info = lod_env_info(env);
+ struct dt_object *next = dt_object_child(&lo->ldo_obj);
+ const struct lu_fid *fid = lu_object_fid(&lo->ldo_obj.do_lu);
+ struct lov_mds_md_v1 *lmm;
+ struct lov_ost_data_v1 *objs;
+ __u32 magic;
+ int i, rc, lmm_size;
+ ENTRY;
+
+ LASSERT(lo);
+ LASSERT(lo->ldo_stripenr > 0);
+
+ magic = lo->ldo_pool ? LOV_MAGIC_V3 : LOV_MAGIC_V1;
+ lmm_size = lov_mds_md_size(lo->ldo_stripenr, magic);
+ if (info->lti_ea_store_size < lmm_size) {
+ rc = lod_ea_store_resize(info, lmm_size);
+ if (rc)
+ RETURN(rc);
+ }
+
+ lmm = info->lti_ea_store;
+
+ lmm->lmm_magic = cpu_to_le32(magic);
+ lmm->lmm_pattern = cpu_to_le32(LOV_PATTERN_RAID0);
+ lmm->lmm_object_id = cpu_to_le64(fid_ver_oid(fid));
+ lmm->lmm_object_seq = cpu_to_le64(fid_seq(fid));
+ lmm->lmm_stripe_size = cpu_to_le32(lo->ldo_stripe_size);
+ lmm->lmm_stripe_count = cpu_to_le16(lo->ldo_stripenr);
+ lmm->lmm_layout_gen = 0;
+ if (magic == LOV_MAGIC_V1) {
+ objs = &lmm->lmm_objects[0];
+ } else {
+ struct lov_mds_md_v3 *v3 = (struct lov_mds_md_v3 *) lmm;
+ strncpy(v3->lmm_pool_name, lo->ldo_pool, LOV_MAXPOOLNAME);
+ objs = &v3->lmm_objects[0];
+ }
+
+ for (i = 0; i < lo->ldo_stripenr; i++) {
+ const struct lu_fid *fid;
+ struct lod_device *lod;
+ __u32 index;
+
+ lod = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
+ LASSERT(lo->ldo_stripe[i]);
+ fid = lu_object_fid(&lo->ldo_stripe[i]->do_lu);
+
+ rc = fid_ostid_pack(fid, &info->lti_ostid);
+ LASSERT(rc == 0);
+
+ objs[i].l_object_id = cpu_to_le64(info->lti_ostid.oi_id);
+ objs[i].l_object_seq = cpu_to_le64(info->lti_ostid.oi_seq);
+ objs[i].l_ost_gen = cpu_to_le32(0);
+ rc = lod_fld_lookup(env, lod, fid, &index, LU_SEQ_RANGE_OST);
+ if (rc < 0) {
+ CERROR("%s: Can not locate "DFID": rc = %d\n",
+ lod2obd(lod)->obd_name, PFID(fid), rc);
+ RETURN(rc);
+ }
+ objs[i].l_ost_idx = cpu_to_le32(index);
+ }
+
+ info->lti_buf.lb_buf = lmm;
+ info->lti_buf.lb_len = lmm_size;
+ rc = dt_xattr_set(env, next, &info->lti_buf, XATTR_NAME_LOV, 0,
+ th, BYPASS_CAPA);
+
+ RETURN(rc);
+}
+
int lod_get_lov_ea(const struct lu_env *env, struct lod_object *lo)
{
struct lod_thread_info *info = lod_env_info(env);
RETURN(rc);
}
+int lod_store_def_striping(const struct lu_env *env, struct dt_object *dt,
+ struct thandle *th)
+{
+ struct lod_thread_info *info = lod_env_info(env);
+ struct lod_object *lo = lod_dt_obj(dt);
+ struct dt_object *next = dt_object_child(dt);
+ struct lov_user_md_v3 *v3;
+ int rc;
+ ENTRY;
+
+ LASSERT(S_ISDIR(dt->do_lu.lo_header->loh_attr));
+
+ /*
+ * store striping defaults into new directory
+ * used to implement defaults inheritance
+ */
+
+ /* probably nothing to inherite */
+ if (lo->ldo_striping_cached == 0)
+ RETURN(0);
+
+ if (LOVEA_DELETE_VALUES(lo->ldo_def_stripe_size, lo->ldo_def_stripenr,
+ lo->ldo_def_stripe_offset))
+ RETURN(0);
+
+ /* XXX: use thread info */
+ OBD_ALLOC_PTR(v3);
+ if (v3 == NULL)
+ RETURN(-ENOMEM);
+
+ v3->lmm_magic = cpu_to_le32(LOV_MAGIC_V3);
+ v3->lmm_pattern = cpu_to_le32(LOV_PATTERN_RAID0);
+ v3->lmm_object_id = 0;
+ v3->lmm_object_seq = 0;
+ v3->lmm_stripe_size = cpu_to_le32(lo->ldo_def_stripe_size);
+ v3->lmm_stripe_count = cpu_to_le16(lo->ldo_def_stripenr);
+ v3->lmm_stripe_offset = cpu_to_le16(lo->ldo_def_stripe_offset);
+ if (lo->ldo_pool)
+ strncpy(v3->lmm_pool_name, lo->ldo_pool, LOV_MAXPOOLNAME);
+
+ info->lti_buf.lb_buf = v3;
+ info->lti_buf.lb_len = sizeof(*v3);
+ rc = dt_xattr_set(env, next, &info->lti_buf, XATTR_NAME_LOV, 0, th,
+ BYPASS_CAPA);
+
+ OBD_FREE_PTR(v3);
+
+ RETURN(rc);
+}
+
/*
* allocate array of objects pointers, find/create objects
* stripenr and other fields should be initialized by this moment
lo->ldo_stripes_allocated = lo->ldo_stripenr;
for (i = 0; i < lo->ldo_stripenr; i++) {
-
info->lti_ostid.oi_id = le64_to_cpu(objs[i].l_object_id);
- /* XXX: support for DNE? */
info->lti_ostid.oi_seq = le64_to_cpu(objs[i].l_object_seq);
idx = le64_to_cpu(objs[i].l_ost_idx);
fid_ostid_unpack(&info->lti_fid, &info->lti_ostid, idx);
-
+ LASSERTF(fid_is_sane(&info->lti_fid), ""DFID" insane!\n",
+ PFID(&info->lti_fid));
/*
* XXX: assertion is left for testing, to make
* sure we never process requests till configuration
* is completed. to be changed to -EINVAL
*/
- lod_getref(md);
+ lod_getref(&md->lod_ost_descs);
LASSERT(cfs_bitmap_check(md->lod_ost_bitmap, idx));
LASSERT(OST_TGT(md,idx));
LASSERTF(OST_TGT(md,idx)->ltd_ost, "idx %d\n", idx);
nd = &OST_TGT(md,idx)->ltd_ost->dd_lu_dev;
- lod_putref(md);
+ lod_putref(md, &md->lod_ost_descs);
+ /* In the function below, .hs_keycmp resolves to
+ * u_obj_hop_keycmp() */
+ /* coverity[overrun-buffer-val] */
o = lu_object_find_at(env, nd, &info->lti_fid, NULL);
if (IS_ERR(o))
GOTO(out, rc = PTR_ERR(o));
RETURN(rc);
}
+int lod_verify_striping(struct lod_device *d, const struct lu_buf *buf,
+ int specific)
+{
+ struct lov_user_md_v1 *lum;
+ struct lov_user_md_v3 *v3 = NULL;
+ struct pool_desc *pool = NULL;
+ int rc;
+ ENTRY;
+
+ lum = buf->lb_buf;
+
+ if (lum->lmm_magic != LOV_USER_MAGIC_V1 &&
+ lum->lmm_magic != LOV_USER_MAGIC_V3 &&
+ lum->lmm_magic != LOV_MAGIC_V1_DEF &&
+ lum->lmm_magic != LOV_MAGIC_V3_DEF) {
+ CDEBUG(D_IOCTL, "bad userland LOV MAGIC: %#x\n",
+ lum->lmm_magic);
+ RETURN(-EINVAL);
+ }
+
+ if ((specific && lum->lmm_pattern != LOV_PATTERN_RAID0) ||
+ (specific == 0 && lum->lmm_pattern != 0)) {
+ CDEBUG(D_IOCTL, "bad userland stripe pattern: %#x\n",
+ lum->lmm_pattern);
+ RETURN(-EINVAL);
+ }
+
+ /* 64kB is the largest common page size we see (ia64), and matches the
+ * check in lfs */
+ if (lum->lmm_stripe_size & (LOV_MIN_STRIPE_SIZE - 1)) {
+ CDEBUG(D_IOCTL, "stripe size %u not multiple of %u, fixing\n",
+ lum->lmm_stripe_size, LOV_MIN_STRIPE_SIZE);
+ RETURN(-EINVAL);
+ }
+
+ /* an offset of -1 is treated as a "special" valid offset */
+ if (lum->lmm_stripe_offset != (typeof(lum->lmm_stripe_offset))(-1)) {
+ /* if offset is not within valid range [0, osts_size) */
+ if (lum->lmm_stripe_offset >= d->lod_osts_size) {
+ CDEBUG(D_IOCTL, "stripe offset %u >= bitmap size %u\n",
+ lum->lmm_stripe_offset, d->lod_osts_size);
+ RETURN(-EINVAL);
+ }
+
+ /* if lmm_stripe_offset is *not* in bitmap */
+ if (!cfs_bitmap_check(d->lod_ost_bitmap,
+ lum->lmm_stripe_offset)) {
+ CDEBUG(D_IOCTL, "stripe offset %u not in bitmap\n",
+ lum->lmm_stripe_offset);
+ RETURN(-EINVAL);
+ }
+ }
+
+ if (lum->lmm_magic == LOV_USER_MAGIC_V3)
+ v3 = buf->lb_buf;
+
+ if (v3)
+ /* In the function below, .hs_keycmp resolves to
+ * pool_hashkey_keycmp() */
+ /* coverity[overrun-buffer-val] */
+ pool = lod_find_pool(d, v3->lmm_pool_name);
+
+ if (pool != NULL) {
+ __u16 offs = v3->lmm_stripe_offset;
+
+ if (offs != (typeof(v3->lmm_stripe_offset))(-1)) {
+ rc = lod_check_index_in_pool(offs, pool);
+ if (rc < 0) {
+ lod_pool_putref(pool);
+ RETURN(-EINVAL);
+ }
+ }
+
+ if (specific && lum->lmm_stripe_count > pool_tgt_count(pool)) {
+ CDEBUG(D_IOCTL,
+ "stripe count %u > # OSTs %u in the pool\n",
+ lum->lmm_stripe_count, pool_tgt_count(pool));
+ lod_pool_putref(pool);
+ RETURN(-EINVAL);
+ }
+
+ lod_pool_putref(pool);
+ }
+
+ RETURN(0);
+}
+
void lod_fix_desc_stripe_size(__u64 *val)
{
- if (*val < PTLRPC_MAX_BRW_SIZE) {
+ if (*val < LOV_DEFAULT_STRIPE_SIZE) {
LCONSOLE_WARN("Increasing default stripe size to min %u\n",
- PTLRPC_MAX_BRW_SIZE);
- *val = PTLRPC_MAX_BRW_SIZE;
+ LOV_DEFAULT_STRIPE_SIZE);
+ *val = LOV_DEFAULT_STRIPE_SIZE;
} else if (*val & (LOV_MIN_STRIPE_SIZE - 1)) {
*val &= ~(LOV_MIN_STRIPE_SIZE - 1);
LCONSOLE_WARN("Changing default stripe size to "LPU64" (a "
int lod_pools_init(struct lod_device *lod, struct lustre_cfg *lcfg)
{
- struct lprocfs_static_vars lvars = { 0 };
struct obd_device *obd;
struct lov_desc *desc;
int rc;
/* Set up allocation policy (QoS and RR) */
CFS_INIT_LIST_HEAD(&lod->lod_qos.lq_oss_list);
- cfs_init_rwsem(&lod->lod_qos.lq_rw_sem);
+ init_rwsem(&lod->lod_qos.lq_rw_sem);
lod->lod_qos.lq_dirty = 1;
lod->lod_qos.lq_rr.lqr_dirty = 1;
lod->lod_qos.lq_reset = 1;
if (rc)
GOTO(out_pool_info, rc);
- /* the OST array and bitmap are allocated/grown dynamically as OSTs are
- * added to the LOD, see lod_add_device() */
- lod->lod_ost_bitmap = NULL;
- lod->lod_osts_size = 0;
- lod->lod_ostnr = 0;
-
- lod->lod_death_row = 0;
- lod->lod_refcount = 0;
-
- lprocfs_lod_init_vars(&lvars);
- lprocfs_obd_setup(obd, lvars.obd_vars);
-
-#ifdef LPROCFS
- rc = lprocfs_seq_create(obd->obd_proc_entry, "target_obd",
- 0444, &lod_proc_target_fops, obd);
- if (rc) {
- CWARN("%s: Error adding the target_obd file %d\n",
- obd->obd_name, rc);
- GOTO(out_lproc, rc);
- }
- lod->lod_pool_proc_entry = lprocfs_register("pools",
- obd->obd_proc_entry,
- NULL, NULL);
- if (IS_ERR(lod->lod_pool_proc_entry)) {
- int ret = PTR_ERR(lod->lod_pool_proc_entry);
- lod->lod_pool_proc_entry = NULL;
- CWARN("%s: Failed to create pool proc file %d\n",
- obd->obd_name, ret);
- rc = lod_pools_fini(lod);
- RETURN(ret);
- }
-#endif
-
RETURN(0);
-out_lproc:
- lprocfs_obd_cleanup(obd);
- lod_ost_pool_free(&lod->lod_qos.lq_rr.lqr_pool);
out_pool_info:
lod_ost_pool_free(&lod->lod_pool_info);
out_hash:
pool = cfs_list_entry(pos, struct pool_desc, pool_list);
/* free pool structs */
CDEBUG(D_INFO, "delete pool %p\n", pool);
+ /* In the function below, .hs_keycmp resolves to
+ * pool_hashkey_keycmp() */
+ /* coverity[overrun-buffer-val] */
lod_pool_del(obd, pool->pool_name);
}
- if (lod->lod_osts_size > 0) {
- int idx;
- lod_getref(lod);
- cfs_mutex_lock(&lod->lod_mutex);
- cfs_foreach_bit(lod->lod_ost_bitmap, idx)
- __lod_del_device(lod, idx);
- cfs_mutex_unlock(&lod->lod_mutex);
- lod_putref(lod);
- CFS_FREE_BITMAP(lod->lod_ost_bitmap);
- for (idx = 0; idx < OST_PTRS; idx++) {
- if (lod->lod_ost_idx[idx])
- OBD_FREE_PTR(lod->lod_ost_idx[idx]);
- }
- lod->lod_osts_size = 0;
- }
-
cfs_hash_putref(lod->lod_pools_hash_body);
lod_ost_pool_free(&(lod->lod_qos.lq_rr.lqr_pool));
lod_ost_pool_free(&lod->lod_pool_info);
-
- /* clear pools parent proc entry only after all pools are killed */
- if (lod->lod_pool_proc_entry) {
- lprocfs_remove(&lod->lod_pool_proc_entry);
- lod->lod_pool_proc_entry = NULL;
- }
-
- lprocfs_obd_cleanup(obd);
-
OBD_FREE_PTR(lod->lod_qos.lq_statfs_data);
RETURN(0);
}
git://git.whamcloud.com / fs / lustre-release.git / blobdiff