*
* You should have received a copy of the GNU General Public License
* version 2 along with this program; If not, see
- * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
- *
- * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
- * CA 95054 USA or visit www.sun.com if you need additional information or
- * have any questions.
+ * http://www.gnu.org/licenses/gpl-2.0.html
*
* GPL HEADER END
*/
* Copyright (c) 2009, 2010, Oracle and/or its affiliates. All rights reserved.
* Use is subject to license terms.
*
- * Copyright (c) 2012, 2015, Intel Corporation.
+ * Copyright (c) 2012, 2016, Intel Corporation.
*/
/*
* This file is part of Lustre, http://www.lustre.org/
#define DEBUG_SUBSYSTEM S_OSD
-#include <lustre_ver.h>
#include <libcfs/libcfs.h>
#include <obd_support.h>
#include <lustre_net.h>
#include <obd_class.h>
#include <lustre_disk.h>
#include <lustre_fid.h>
-#include <lustre_param.h>
+#include <uapi/linux/lustre/lustre_param.h>
#include <md_object.h>
#include "osd_internal.h"
struct lu_env env;
int rc;
+ LASSERT(site->ls_obj_hash);
+
rc = lu_env_init(&env, LCT_SHRINKER);
if (rc) {
CERROR("%s: can't initialize shrinker env: rc = %d\n",
dt_txn_hook_commit(th);
/* call per-transaction callbacks if any */
- list_for_each_entry_safe(dcb, tmp, &oh->ot_dcb_list, dcb_linkage)
+ list_for_each_entry_safe(dcb, tmp, &oh->ot_dcb_list, dcb_linkage) {
+ LASSERTF(dcb->dcb_magic == TRANS_COMMIT_CB_MAGIC,
+ "commit callback entry: magic=%x name='%s'\n",
+ dcb->dcb_magic, dcb->dcb_name);
+ list_del_init(&dcb->dcb_linkage);
dcb->dcb_func(NULL, th, dcb, error);
+ }
/* Unlike ldiskfs, zfs updates space accounting at commit time.
* As a consequence, op_end is called only now to inform the quota slave
RETURN(rc);
}
-static int osd_unlinked_object_free(struct osd_device *osd, uint64_t oid);
-
-static void osd_unlinked_list_emptify(struct osd_device *osd,
+static void osd_unlinked_list_emptify(const struct lu_env *env,
+ struct osd_device *osd,
struct list_head *list, bool free)
{
struct osd_object *obj;
while (!list_empty(list)) {
obj = list_entry(list->next,
struct osd_object, oo_unlinked_linkage);
- LASSERT(obj->oo_db != NULL);
- oid = obj->oo_db->db_object;
+ LASSERT(obj->oo_dn != NULL);
+ oid = obj->oo_dn->dn_object;
list_del_init(&obj->oo_unlinked_linkage);
if (free)
- (void)osd_unlinked_object_free(osd, oid);
+ (void)osd_unlinked_object_free(env, osd, oid);
}
}
oh = container_of0(th, struct osd_thandle, ot_super);
INIT_LIST_HEAD(&unlinked);
list_splice_init(&oh->ot_unlinked_list, &unlinked);
+ /* reset OI cache for safety */
+ osd_oti_get(env)->oti_ins_cache_used = 0;
if (oh->ot_assigned == 0) {
LASSERT(oh->ot_tx);
dmu_tx_abort(oh->ot_tx);
osd_object_sa_dirty_rele(env, oh);
- osd_unlinked_list_emptify(osd, &unlinked, false);
+ osd_unlinked_list_emptify(env, osd, &unlinked, false);
/* there won't be any commit, release reserved quota space now,
* if any */
qsd_op_end(env, osd->od_quota_slave, &oh->ot_quota_trans);
RETURN(0);
}
- /* When doing our own inode accounting, the ZAPs storing per-uid/gid
- * usage are updated at operation execution time, so we should call
- * qsd_op_end() straight away. Otherwise (for blk accounting maintained
- * by ZFS and when #inode is estimated from #blks) accounting is updated
- * at commit time and the call to qsd_op_end() must be delayed */
- if (oh->ot_quota_trans.lqt_id_cnt > 0 &&
- !oh->ot_quota_trans.lqt_ids[0].lqi_is_blk &&
- !osd->od_quota_iused_est)
- qsd_op_end(env, osd->od_quota_slave, &oh->ot_quota_trans);
-
rc = dt_txn_hook_stop(env, th);
if (rc != 0)
CDEBUG(D_OTHER, "%s: transaction hook failed: rc = %d\n",
* by osd_trans_commit_cb already. */
dmu_tx_commit(oh->ot_tx);
- osd_unlinked_list_emptify(osd, &unlinked, true);
+ osd_unlinked_list_emptify(env, osd, &unlinked, true);
if (sync)
txg_wait_synced(dmu_objset_pool(osd->od_os), txg);
dmu_tx_t *tx;
ENTRY;
+ if (dt->dd_rdonly) {
+ CERROR("%s: someone try to start transaction under "
+ "readonly mode, should be disabled.\n",
+ osd_name(osd_dt_dev(dt)));
+ dump_stack();
+ RETURN(ERR_PTR(-EROFS));
+ }
+
tx = dmu_tx_create(osd->od_os);
if (tx == NULL)
RETURN(ERR_PTR(-ENOMEM));
RETURN(th);
}
-/* Estimate the number of objects from a number of blocks */
-uint64_t osd_objs_count_estimate(uint64_t refdbytes, uint64_t usedobjs,
+/* Estimate the total number of objects from a number of blocks */
+uint64_t osd_objs_count_estimate(uint64_t usedbytes, uint64_t usedobjs,
uint64_t nrblocks, uint64_t est_maxblockshift)
{
- uint64_t est_objs, est_refdblocks, est_usedobjs;
-
- /* Compute an nrblocks estimate based on the actual number of
- * dnodes that could fit in the space. Since we don't know the
- * overhead associated with each dnode (xattrs, SAs, VDEV overhead,
- * etc) just using DNODE_SHIFT isn't going to give a good estimate.
- * Instead, compute an estimate based on the average space usage per
- * dnode, with an upper and lower cap.
+ uint64_t est_totobjs, est_usedblocks, est_usedobjs;
+
+ /*
+ * If blocksize is below 64KB (e.g. MDT with recordsize=4096) then
+ * bump the free dnode estimate to assume blocks at least 64KB in
+ * case of a directory-heavy MDT (at 32KB/directory).
+ */
+ if (est_maxblockshift < 16) {
+ nrblocks >>= (16 - est_maxblockshift);
+ est_maxblockshift = 16;
+ }
+
+ /*
+ * Estimate the total number of dnodes from the total blocks count
+ * and the space used per dnode. Since we don't know the overhead
+ * associated with each dnode (xattrs, SAs, VDEV overhead, etc.)
+ * just using DNODE_SHIFT isn't going to give a good estimate.
+ * Instead, compute the current average space usage per dnode, with
+ * an upper and lower cap to avoid unrealistic estimates..
*
* In case there aren't many dnodes or blocks used yet, add a small
- * correction factor using OSD_DNODE_EST_SHIFT. This correction
- * factor gradually disappears as the number of real dnodes grows.
- * This also avoids the need to check for divide-by-zero later.
+ * correction factor (OSD_DNODE_EST_{COUNT,BLKSHIFT}). This factor
+ * gradually disappears as the number of real dnodes grows. It also
+ * avoids the need to check for divide-by-zero computing dn_per_block.
*/
CLASSERT(OSD_DNODE_MIN_BLKSHIFT > 0);
CLASSERT(OSD_DNODE_EST_BLKSHIFT > 0);
- est_refdblocks = (refdbytes >> est_maxblockshift) +
- (OSD_DNODE_EST_COUNT >> OSD_DNODE_EST_BLKSHIFT);
- est_usedobjs = usedobjs + OSD_DNODE_EST_COUNT;
-
- /* Average space/dnode more than maximum dnode size, use max dnode
- * size to estimate free dnodes from adjusted free blocks count.
- * OSTs typically use more than one block dnode so this case applies. */
- if (est_usedobjs <= est_refdblocks * 2) {
- est_objs = nrblocks;
-
- /* Average space/dnode smaller than min dnode size (probably due to
- * metadnode compression), use min dnode size to estimate the number of
- * objects.
- * An MDT typically uses below 512 bytes/dnode so this case applies. */
- } else if (est_usedobjs >= (est_refdblocks << OSD_DNODE_MIN_BLKSHIFT)) {
- est_objs = nrblocks << OSD_DNODE_MIN_BLKSHIFT;
-
- /* Between the extremes, we try to use the average size of
- * existing dnodes to compute the number of dnodes that fit
- * into nrblocks:
+ est_usedblocks = ((OSD_DNODE_EST_COUNT << OSD_DNODE_EST_BLKSHIFT) +
+ usedbytes) >> est_maxblockshift;
+ est_usedobjs = OSD_DNODE_EST_COUNT + usedobjs;
+
+ if (est_usedobjs <= est_usedblocks) {
+ /*
+ * Average space/dnode more than maximum block size, use max
+ * block size to estimate free dnodes from adjusted free blocks
+ * count. OSTs typically use multiple blocks per dnode so this
+ * case applies.
+ */
+ est_totobjs = nrblocks;
+
+ } else if (est_usedobjs >= (est_usedblocks << OSD_DNODE_MIN_BLKSHIFT)) {
+ /*
+ * Average space/dnode smaller than min dnode size (probably
+ * due to metadnode compression), use min dnode size to
+ * estimate object count. MDTs may use only one block per node
+ * so this case applies.
+ */
+ est_totobjs = nrblocks << OSD_DNODE_MIN_BLKSHIFT;
+
+ } else {
+ /*
+ * Between the extremes, use average space per existing dnode
+ * to compute the number of dnodes that will fit into nrblocks:
*
- * est_objs = nrblocks * (est_usedobjs / est_refblocks);
+ * est_totobjs = nrblocks * (est_usedobjs / est_usedblocks)
*
- * but this may overflow 64 bits or become 0 if not handled well
+ * this may overflow 64 bits or become 0 if not handled well.
*
- * We know nrblocks is below (64 - 17 = 47) bits from
- * SPA_MAXBLKSHIFT, and est_usedobjs is under 48 bits due to
- * DN_MAX_OBJECT_SHIFT, which means that multiplying them may
- * get as large as 2 ^ 95.
+ * We know nrblocks is below 2^(64 - blkbits) bits, and
+ * est_usedobjs is under 48 bits due to DN_MAX_OBJECT_SHIFT,
+ * which means that multiplying them may get as large as
+ * 2 ^ 96 for the minimum blocksize of 64KB allowed above.
*
- * We also know (est_usedobjs / est_refdblocks) is between 2 and
- * 256, due to above checks, we can safely compute this first.
+ * The ratio of dnodes per block (est_usedobjs / est_usedblocks)
+ * is under 2^(blkbits - DNODE_SHIFT) = blocksize / 512 due to
+ * the limit checks above, so we can safely compute this first.
* We care more about accuracy on the MDT (many dnodes/block)
* which is good because this is where truncation errors are
- * smallest. This adds 8 bits to nrblocks so we can use 7 bits
- * to compute a fixed-point fraction and nrblocks can still fit
- * in 64 bits. */
- } else {
- unsigned dnodes_per_block = (est_usedobjs << 7)/est_refdblocks;
+ * smallest. Since both nrblocks and dn_per_block are a
+ * function of blkbits, their product is at most:
+ *
+ * 2^(64 - blkbits) * 2^(blkbits - DNODE_SHIFT) = 2^(64 - 9)
+ *
+ * so we can safely use 7 bits to compute a fixed-point
+ * fraction and est_totobjs can still fit in 64 bits.
+ */
+ unsigned dn_per_block = (est_usedobjs << 7) / est_usedblocks;
- est_objs = (nrblocks * dnodes_per_block) >> 7;
+ est_totobjs = (nrblocks * dn_per_block) >> 7;
}
- return est_objs;
+ return est_totobjs;
}
static int osd_objset_statfs(struct osd_device *osd, struct obd_statfs *osfs)
{
struct objset *os = osd->od_os;
- uint64_t refdbytes, availbytes, usedobjs, availobjs;
+ uint64_t usedbytes, availbytes, usedobjs, availobjs;
uint64_t est_availobjs;
uint64_t reserved;
uint64_t bshift;
- dmu_objset_space(os, &refdbytes, &availbytes, &usedobjs, &availobjs);
+ dmu_objset_space(os, &usedbytes, &availbytes, &usedobjs, &availobjs);
memset(osfs, 0, sizeof(*osfs));
osfs->os_bsize = osd->od_max_blksz;
bshift = fls64(osfs->os_bsize) - 1;
- osfs->os_blocks = (refdbytes + availbytes) >> bshift;
+ osfs->os_blocks = (usedbytes + availbytes) >> bshift;
osfs->os_bfree = availbytes >> bshift;
osfs->os_bavail = osfs->os_bfree; /* no extra root reservation */
/* Take replication (i.e. number of copies) into account */
- osfs->os_bavail /= os->os_copies;
+ if (os->os_copies != 0)
+ osfs->os_bavail /= os->os_copies;
/*
* Reserve some space so we don't run into ENOSPC due to grants not
* issues like how much space is actually available in the pool.
* Compute a better estimate in udmu_objs_count_estimate().
*/
- est_availobjs = osd_objs_count_estimate(refdbytes, usedobjs,
+ est_availobjs = osd_objs_count_estimate(usedbytes, usedobjs,
osfs->os_bfree, bshift);
osfs->os_ffree = min(availobjs, est_availobjs);
*/
static int osd_sync(const struct lu_env *env, struct dt_device *d)
{
- struct osd_device *osd = osd_dt_dev(d);
- CDEBUG(D_CACHE, "syncing OSD %s\n", LUSTRE_OSD_ZFS_NAME);
- txg_wait_synced(dmu_objset_pool(osd->od_os), 0ULL);
- CDEBUG(D_CACHE, "synced OSD %s\n", LUSTRE_OSD_ZFS_NAME);
+ if (!d->dd_rdonly) {
+ struct osd_device *osd = osd_dt_dev(d);
+
+ CDEBUG(D_CACHE, "syncing OSD %s\n", LUSTRE_OSD_ZFS_NAME);
+ txg_wait_synced(dmu_objset_pool(osd->od_os), 0ULL);
+ CDEBUG(D_CACHE, "synced OSD %s\n", LUSTRE_OSD_ZFS_NAME);
+ }
+
return 0;
}
struct lu_context_key *key, void *data)
{
struct osd_thread_info *info = data;
+ struct osd_idmap_cache *idc = info->oti_ins_cache;
+ if (idc != NULL) {
+ LASSERT(info->oti_ins_cache_size > 0);
+ OBD_FREE(idc, sizeof(*idc) * info->oti_ins_cache_size);
+ info->oti_ins_cache = NULL;
+ info->oti_ins_cache_size = 0;
+ }
lu_buf_free(&info->oti_xattr_lbuf);
OBD_FREE_PTR(info);
}
/* shutdown quota slave instance associated with the device */
if (o->od_quota_slave != NULL) {
+ /* complete all in-flight callbacks */
+ osd_sync(env, &o->od_dt_dev);
+ txg_wait_callbacks(spa_get_dsl(dmu_objset_spa(o->od_os)));
qsd_fini(env, o->od_quota_slave);
o->od_quota_slave = NULL;
}
static int osd_objset_open(struct osd_device *o)
{
uint64_t version = ZPL_VERSION;
- uint64_t sa_obj;
+ uint64_t sa_obj, unlink_obj;
int rc;
ENTRY;
- rc = -dmu_objset_own(o->od_mntdev, DMU_OST_ZFS, B_FALSE, o, &o->od_os);
+ rc = -osd_dmu_objset_own(o->od_mntdev, DMU_OST_ZFS,
+ o->od_dt_dev.dd_rdonly ? B_TRUE : B_FALSE,
+ B_FALSE, o, &o->od_os);
+
if (rc) {
CERROR("%s: can't open %s\n", o->od_svname, o->od_mntdev);
o->od_os = NULL;
- goto out;
+
+ GOTO(out, rc);
}
/* Check ZFS version */
}
rc = -zap_lookup(o->od_os, MASTER_NODE_OBJ, ZFS_UNLINKED_SET,
- 8, 1, &o->od_unlinkedid);
+ 8, 1, &unlink_obj);
if (rc) {
CERROR("%s: lookup for %s failed: rc = %d\n",
o->od_svname, ZFS_UNLINKED_SET, rc);
DMU_GROUPUSED_DNODE(o->od_os)->dn_type != DMU_OT_USERGROUP_USED) {
CERROR("%s: Space accounting not supported by this target, "
"aborting\n", o->od_svname);
- GOTO(out, -ENOTSUPP);
+ GOTO(out, rc = -ENOTSUPP);
+ }
+
+ rc = __osd_obj2dnode(o->od_os, unlink_obj, &o->od_unlinked);
+ if (rc) {
+ CERROR("%s: can't get dnode for unlinked: rc = %d\n",
+ o->od_svname, rc);
+ GOTO(out, rc);
}
out:
if (rc != 0 && o->od_os != NULL) {
- dmu_objset_disown(o->od_os, o);
+ osd_dmu_objset_disown(o->od_os, B_FALSE, o);
o->od_os = NULL;
}
RETURN(rc);
}
-static int
-osd_unlinked_object_free(struct osd_device *osd, uint64_t oid)
+int osd_unlinked_object_free(const struct lu_env *env, struct osd_device *osd,
+ uint64_t oid)
{
+ char *key = osd_oti_get(env)->oti_str;
int rc;
dmu_tx_t *tx;
+ if (osd->od_dt_dev.dd_rdonly) {
+ CERROR("%s: someone try to free objects under "
+ "readonly mode, should be disabled.\n", osd_name(osd));
+ dump_stack();
+
+ return -EROFS;
+ }
+
rc = -dmu_free_long_range(osd->od_os, oid, 0, DMU_OBJECT_END);
if (rc != 0) {
- CWARN("%s: Cannot truncate "LPU64": rc = %d\n",
+ CWARN("%s: Cannot truncate %llu: rc = %d\n",
osd->od_svname, oid, rc);
return rc;
}
tx = dmu_tx_create(osd->od_os);
dmu_tx_hold_free(tx, oid, 0, DMU_OBJECT_END);
- dmu_tx_hold_zap(tx, osd->od_unlinkedid, FALSE, NULL);
+ osd_tx_hold_zap(tx, osd->od_unlinked->dn_object, osd->od_unlinked,
+ FALSE, NULL);
rc = -dmu_tx_assign(tx, TXG_WAIT);
if (rc != 0) {
- CWARN("%s: Cannot assign tx for "LPU64": rc = %d\n",
+ CWARN("%s: Cannot assign tx for %llu: rc = %d\n",
osd->od_svname, oid, rc);
goto failed;
}
- rc = -zap_remove_int(osd->od_os, osd->od_unlinkedid, oid, tx);
+ snprintf(key, sizeof(osd_oti_get(env)->oti_str), "%llx", oid);
+ rc = osd_zap_remove(osd, osd->od_unlinked->dn_object,
+ osd->od_unlinked, key, tx);
if (rc != 0) {
- CWARN("%s: Cannot remove "LPU64" from unlinked set: rc = %d\n",
+ CWARN("%s: Cannot remove %llu from unlinked set: rc = %d\n",
osd->od_svname, oid, rc);
goto failed;
}
rc = -dmu_object_free(osd->od_os, oid, tx);
if (rc != 0) {
- CWARN("%s: Cannot free "LPU64": rc = %d\n",
+ CWARN("%s: Cannot free %llu: rc = %d\n",
osd->od_svname, oid, rc);
goto failed;
}
zap_cursor_t zc;
zap_attribute_t *za = &osd_oti_get(env)->oti_za;
- zap_cursor_init(&zc, osd->od_os, osd->od_unlinkedid);
+ zap_cursor_init(&zc, osd->od_os, osd->od_unlinked->dn_object);
while (zap_cursor_retrieve(&zc, za) == 0) {
/* If cannot free the object, leave it in the unlinked set,
* until the OSD is mounted again when obd_unlinked_drain()
* will be called. */
- if (osd_unlinked_object_free(osd, za->za_first_integer) != 0)
+ if (osd_unlinked_object_free(env, osd, za->za_first_integer))
break;
zap_cursor_advance(&zc);
}
struct osd_device *o, struct lustre_cfg *cfg)
{
char *mntdev = lustre_cfg_string(cfg, 1);
+ char *str = lustre_cfg_string(cfg, 2);
char *svname = lustre_cfg_string(cfg, 4);
- dmu_buf_t *rootdb;
+ dnode_t *rootdn;
const char *opts;
int rc;
ENTRY;
if (rc >= sizeof(o->od_svname))
RETURN(-E2BIG);
+ str = strstr(str, ":");
+ if (str) {
+ unsigned long flags;
+
+ rc = kstrtoul(str + 1, 10, &flags);
+ if (rc)
+ RETURN(-EINVAL);
+
+ if (flags & LMD_FLG_DEV_RDONLY) {
+ o->od_dt_dev.dd_rdonly = 1;
+ LCONSOLE_WARN("%s: set dev_rdonly on this device\n",
+ svname);
+ }
+ }
+
if (server_name_is_ost(o->od_svname))
o->od_is_ost = 1;
RETURN(rc);
o->od_xattr_in_sa = B_TRUE;
- o->od_max_blksz = SPA_OLD_MAXBLOCKSIZE;
+ o->od_max_blksz = osd_spa_maxblocksize(o->od_os->os_spa);
- rc = osd_objset_register_callbacks(o);
+ rc = __osd_obj2dnode(o->od_os, o->od_rootid, &rootdn);
if (rc)
GOTO(err, rc);
+ o->od_root = rootdn->dn_object;
+ osd_dnode_rele(rootdn);
- rc = __osd_obj2dbuf(env, o->od_os, o->od_rootid, &rootdb);
+ rc = __osd_obj2dnode(o->od_os, DMU_USERUSED_OBJECT,
+ &o->od_userused_dn);
if (rc)
GOTO(err, rc);
- o->od_root = rootdb->db_object;
- sa_buf_rele(rootdb, osd_obj_tag);
+ rc = __osd_obj2dnode(o->od_os, DMU_GROUPUSED_OBJECT,
+ &o->od_groupused_dn);
+ if (rc)
+ GOTO(err, rc);
/* 1. initialize oi before any file create or file open */
rc = osd_oi_init(env, o);
if (rc)
GOTO(err, rc);
- /* Use our own ZAP for inode accounting by default, this can be changed
- * via procfs to estimate the inode usage from the block usage */
- o->od_quota_iused_est = 0;
+ rc = osd_objset_register_callbacks(o);
+ if (rc)
+ GOTO(err, rc);
rc = osd_procfs_init(o, o->od_svname);
if (rc)
GOTO(err, rc);
}
+#ifdef HAVE_DMU_USEROBJ_ACCOUNTING
+ if (!osd_dmu_userobj_accounting_available(o))
+ CWARN("%s: dnode accounting not enabled: "
+ "enable feature@userobj_accounting in pool\n",
+ o->od_mntdev);
+#endif
+
/* parse mount option "noacl", and enable ACL by default */
opts = lustre_cfg_string(cfg, 3);
if (opts == NULL || strstr(opts, "noacl") == NULL)
osd_unlinked_drain(env, o);
err:
- if (rc) {
- if (o->od_os)
- dmu_objset_disown(o->od_os, o);
+ if (rc && o->od_os) {
+ osd_dmu_objset_disown(o->od_os, B_FALSE, o);
o->od_os = NULL;
}
CERROR("%s: lost %d pinned dbuf(s)\n", o->od_svname,
atomic_read(&o->od_zerocopy_pin));
+ if (o->od_unlinked) {
+ osd_dnode_rele(o->od_unlinked);
+ o->od_unlinked = NULL;
+ }
+ if (o->od_userused_dn) {
+ osd_dnode_rele(o->od_userused_dn);
+ o->od_userused_dn = NULL;
+ }
+ if (o->od_groupused_dn) {
+ osd_dnode_rele(o->od_groupused_dn);
+ o->od_groupused_dn = NULL;
+ }
+
if (o->od_os != NULL) {
- /* force a txg sync to get all commit callbacks */
- txg_wait_synced(dmu_objset_pool(o->od_os), 0ULL);
+ if (!o->od_dt_dev.dd_rdonly)
+ /* force a txg sync to get all commit callbacks */
+ txg_wait_synced(dmu_objset_pool(o->od_os), 0ULL);
/* close the object set */
- dmu_objset_disown(o->od_os, o);
-
+ osd_dmu_objset_disown(o->od_os, B_FALSE, o);
o->od_os = NULL;
}
ENTRY;
- osd_shutdown(env, o);
- osd_oi_fini(env, o);
-
if (o->od_os) {
osd_objset_unregister_callbacks(o);
- osd_sync(env, lu2dt_dev(d));
- txg_wait_callbacks(spa_get_dsl(dmu_objset_spa(o->od_os)));
+ if (!o->od_dt_dev.dd_rdonly) {
+ osd_sync(env, lu2dt_dev(d));
+ txg_wait_callbacks(
+ spa_get_dsl(dmu_objset_spa(o->od_os)));
+ }
}
+ /* now with all the callbacks completed we can cleanup the remainings */
+ osd_shutdown(env, o);
+ osd_oi_fini(env, o);
+
rc = osd_procfs_fini(o);
if (rc) {
CERROR("proc fini error %d\n", rc);
LASSERT(&o->od_dt_dev);
rc = class_process_proc_param(PARAM_OSD, lprocfs_osd_obd_vars,
cfg, &o->od_dt_dev);
- if (rc > 0 || rc == -ENOSYS)
+ if (rc > 0 || rc == -ENOSYS) {
rc = class_process_proc_param(PARAM_OST,
lprocfs_osd_obd_vars,
cfg, &o->od_dt_dev);
+ if (rc > 0)
+ rc = 0;
+ }
break;
}
default:
git://git.whamcloud.com / fs / lustre-release.git / blobdiff